UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Understanding human balance through applied robotics : exploring the roles of ankle motion and the vestibular… Pospisil, Eric Robert 2014

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2014_november_pospisil_eric.pdf [ 2.73MB ]
Metadata
JSON: 24-1.0167625.json
JSON-LD: 24-1.0167625-ld.json
RDF/XML (Pretty): 24-1.0167625-rdf.xml
RDF/JSON: 24-1.0167625-rdf.json
Turtle: 24-1.0167625-turtle.txt
N-Triples: 24-1.0167625-rdf-ntriples.txt
Original Record: 24-1.0167625-source.json
Full Text
24-1.0167625-fulltext.txt
Citation
24-1.0167625.ris

Full Text

    UNDERSTANDING HUMAN BALANCE THROUGH APPLIED ROBOTICS: EXPLORING THE ROLES OF ANKLE MOTION AND THE VESTIBULAR SYSTEM IN MAINTAINING STANDING BALANCE   by  Eric Robert Pospisil  B.A.Sc., The University of British Columbia, 2010    A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF APPLIED SCIENCE  in  The Faculty of Graduate and Postdoctoral Studies  (Mechanical Engineering)    THE UNIVERSITY OF BRITISH COLUMBIA  (Vancouver)  October 2014   © Eric Robert Pospisil, 2014 ii  Abstract Abstract This thesis details the implementation and application of a robotic system for investigating the role of somatosensory feedback and the human vestibular apparatus in maintaining standing balance. A 6 degree-of-freedom Stewart platform is employed to explore the human balance system in ways not possible during normal standing conditions. This robotic system, RISER (Robot for Interactive Sensory Engagement and Rehabilitation), uses a physics-based model to simulate of a variety of balance conditions for participants while they are secured to the system, making it possible to modify or isolate aspects of the balance control system for study. The first study explores the role of somatosensory feedback using a robotic “ankle-tilt” platform, which was designed and implemented on the RISER system. The new platform enables independent manipulation of the ankles during balance simulations. Results demonstrate that providing accurate somatosensory feedback plays a significant role in improving balance control during standing simulations through reduction of sway amplitude and smoother motion during deliberate sway. The addition and validation of this platform opens new avenues for research involving incorrect, delayed, or partial somatosensory feedback, to study the effects of varying these parameters on balance performance. In the second set of studies, a new technique is developed for investigating the gains and delays of the vestibular organs, employing the ankle-tilt platform. These studies utilized sinusoidal Galvanic Vestibular Stimulation (GVS) to generate an isolated vestibular error signal, producing sensations of motion. The RISER system is used to relate the response to GVS with the response to physical motions. The author investigates the perception and reflex responses to GVS using the RISER system, and demonstrates that sinusoidal GVS and rotation can be combined to produce superimposed perceptions or reflex responses. The author also compares the relationships frequency-dependant phase relationship between GVS and rotation, and finds that they do not conform to prior model expectations. Possible reasons for these discrepancies are examined, and repercussions on the existing understanding of the human balance model are considered. iii  Preface  Preface Throughout my research, my supervisors Dr. Elizabeth Croft, Dr. Machiel Van der Loos, and Dr. Jean-Sébastien Blouin provided guidance, support, and review of my experimental procedures and written works. Dr. Blouin also provided direction as to relevant literature and suggested research avenues that would be interesting to the academic community. Much of this work builds off the development and research performed by Thomas Huryn on the RISER platform. His balance simulation, human body models, and LabVIEW control code served as a development base for the research and work presented in this thesis. This thesis attributes background work and baseline research to Tom through references to (Huryn, Luu, Van der Loos, Blouin, & Croft, 2010) and (Luu, Huryn, Van der Loos, Croft, & Blouin, 2011). Some figures from Huryn’s thesis: (Huryn, 2012) are reprinted with permission and attribution. Chapter 2 details the design, implementation, and validation of a novel somatosensory feedback device, which was an addition to the lab’s robotic balance simulator: RISER (robot for interactive sensory engagement and rehabilitation). The design, implementation and validation of the novel somatosensory feedback device added to the RISER device detailed in chapter 2 is largely reprinted from "Independent Ankle Motion Control Improves Robotic Balance Simulator,"1(Pospisil, Luu, Blouin, Van der Loos, & Croft, 2012) which was presented at the IEEE Engineering in Medicine and Biological Systems Conference in September 2012. Mechanical design of the platform was completed by an undergraduate team, and excerpts from their final report are included directly in Appendix A.4 (Bell, Groves, Lanfranchi, Nyuli, & Tomsett, 2010). I was responsible for the mechanical assembly, electrical and software design, controller design/tuning, integration with the existing robot, design and conduction of the validation studies, data analysis (with guidance from Dr. Billy Luu), full authorship of the manuscript’s first draft, and presentation at the conference. The remaining chapters describe presently unpublished work. I took the primary role in the conception, design, and implementation of the experiments. All experiments were conducted with approval from the UBC Clinical Research Ethics Board (H09-00987: Study of Human Balance Physiology using a Robotic Motion Simulator).                                                            1 © 2012 IEEE. Reprinted, with permission, from Eric R. Pospisil, Billy L. Luu, Jean-Sébastien Blouin, H. F. Machiel Van der Loos, and Elizabeth A. Croft, "Independent Ankle Motion Control Improves Robotic Balance Simulator." Proceedings of the 34th Annual International Conference of the IEEE EMBS, September 2012, pages 6487-6491. iv  Table of Conte nts  Table of Contents  Abstract ......................................................................................................................................................... ii Preface .......................................................................................................................................................... iii Table of Contents ......................................................................................................................................... iv List of Tables ............................................................................................................................................... viii List of Figures ................................................................................................................................................ ix Glossary ........................................................................................................................................................ xi Acknowledgements .................................................................................................................................... xiv Dedication ................................................................................................................................................... xv 1 Introduction and Literature Review ......................................................................................................1 1.1 Motivation .....................................................................................................................................1 1.2 Standing Balance ...........................................................................................................................1 1.2.1 Biomechanics & Physiology of Human Standing Balance .....................................................2 1.2.2 Primary Sensory Systems in Balance .....................................................................................4 1.2.3 Passive, Reflexive, and Conscious Control of Balance ...........................................................6 1.2.4 Measurement Techniques .....................................................................................................8 1.2.5 Vestibular System Research ............................................................................................... 10 1.2.6 Ankle-proprioception Research .......................................................................................... 12 1.3 Balance Simulation ..................................................................................................................... 13 1.4 Research Goals ........................................................................................................................... 16 1.5 Summary ..................................................................................................................................... 19 2 “Ankle-Tilt” Platform Design .............................................................................................................. 20 2.1 Introduction ................................................................................................................................ 20 v  2.2 Design ......................................................................................................................................... 20 2.2.1 Functional Requirements for Balance Studies ................................................................... 21 2.2.2 Physical Functional Requirements ..................................................................................... 22 2.2.3 Instrumentation, Software, and Controls Requirements ................................................... 22 2.2.4 Physical Implementation .................................................................................................... 23 2.2.5 Software Implementation .................................................................................................. 26 2.2.6 Control Implementation ..................................................................................................... 26 2.2.7 Forceplate Compensation and Passive Stiffness Simulation .............................................. 27 2.2.8 Technical Validation ........................................................................................................... 29 2.3 Validation with Human Subjects ................................................................................................ 29 2.3.1 Experimental Validation Goals ........................................................................................... 30 2.3.2 Procedures .......................................................................................................................... 30 2.4 Analysis and Results ................................................................................................................... 31 2.4.1 Analysis and Results of Technical Validation ...................................................................... 31 2.4.2 Analysis of Validation with Human Subjects ...................................................................... 32 2.4.3 Results of Validation with Human Subjects ........................................................................ 33 2.5 Discussion ................................................................................................................................... 35 2.5.1 Technical Performance ....................................................................................................... 35 2.5.2 Validation with Human Subjects ........................................................................................ 36 2.6 Summary ..................................................................................................................................... 36 3 Exploring the Perception and Reflex of Galvanic Vestibular Stimulation .......................................... 38 3.1 Introduction ................................................................................................................................ 38 3.1.1 Background ......................................................................................................................... 38 3.2 Procedures .................................................................................................................................. 42 3.2.1 Perception Study ................................................................................................................ 43 3.2.2 Reflex Response Study........................................................................................................ 47 vi  3.3 Results ........................................................................................................................................ 51 3.3.1 Perception Study ................................................................................................................ 51 3.3.2 Reflex Study ........................................................................................................................ 53 3.4 Discussion ................................................................................................................................... 57 3.5 Limitations .................................................................................................................................. 59 3.6 Summary ..................................................................................................................................... 60 4 Conclusions and Future Work ............................................................................................................ 61 4.1 Overall Contributions ................................................................................................................. 61 4.1.1 Contribution 1: Ankle-Tilt Platform .................................................................................... 62 4.1.2 Contribution 2: Sinusoidal GVS-Rotation Cancellation....................................................... 63 4.1.3 Contribution 3: Comparison of GVS Response in Perception and Reflex........................... 64 4.2 Future Work ............................................................................................................................... 64 4.2.1 Ankle-Tilt Platform Research Opportunities ...................................................................... 64 4.2.2 GVS Reflex and Perception ................................................................................................. 65 4.3 Summary ..................................................................................................................................... 66 References .................................................................................................................................................. 67 A Appendix A: Ankle-Tilt System Design ................................................................................................ 76 A.1 Hardware .......................................................................................................................................... 76 A.2 Software Implementation ................................................................................................................ 80 A.3 Safety ................................................................................................................................................ 82 A.4 Capstone Team Report ..................................................................................................................... 84 A.4.1 Introduction .............................................................................................................................. 84 A.4.2 Requirements ............................................................................................................................ 84 A.4.3 System Overview ....................................................................................................................... 85 A.4.4 Subsystems ................................................................................................................................ 86 A.4.5 Recommendations .................................................................................................................... 93 vii  A.4.6 Conclusion ................................................................................................................................. 96 B Appendix B: Reflex Study Implementation......................................................................................... 97 B.1 Pilot Study Implementation and Issues ............................................................................................ 97 B.2 Head Tracking ................................................................................................................................. 101 B.3 Forceplate Compensation .............................................................................................................. 101 C Appendix C: Supplementary Results ................................................................................................ 103 C.1 Height and Amplitude Perception Tuning ...................................................................................... 103 C.2 Perceptual Flip ................................................................................................................................ 104 C.3 Auditory Perception Phase Results ................................................................................................ 105     viii  List of Ta bles  List of Tables Table 3-1 – Perception study phase leads and reported confidence ......................................................... 53 Table 3-2 – Reflex study cancellation, addition, and quality metrics. ........................................................ 54 Table A-1 – Bill of materials for ankle-tilt platform .................................................................................... 76 Table A-2 – Tilt platform actuator specifications ....................................................................................... 89 Table A-3 – Tilt platform actuator rod end specifications .......................................................................... 90 Table A-4 – Tilt platform amplifier specifications ...................................................................................... 90 Table A-5 – Tilt platform encoder specifications ........................................................................................ 92 Table A-6 – Tilt platform timing belt specifications ................................................................................... 92 Table A-7 – Tilt platform encoder pulley specifications ............................................................................. 93 Table C-1 – Mean selected rotation leads, standard deviations, and mean reported confidence. ......... 107    ix  List of Figure s  List of Figures Figure 1-1 – Inverted pendulum model of human balance. ..........................................................................3 Figure 1-2 – Organs located in the labyrinth of the inner ear. ......................................................................6 Figure 1-3 – Examples of ankle manipulation devices. .............................................................................. 13 Figure 1-4 – Test participant standing on balance simulator. .................................................................... 14 Figure 1-5 – Inverted pendulum model of 1DOF human balance, as implemented on the RISER system. 15 Figure 1-6 – Ankle-tilt platform enabling independent ankle manipulation during balance simulations. 18 Figure 2-1 – Exploded rendering of the ankle-tilt platform assembly. ...................................................... 24 Figure 2-2 – Ankle-tilt platform motion illustration. .................................................................................. 24 Figure 2-3 – Balance simulator with ankle-tilt platform improvement. .................................................... 25 Figure 2-4 – Ankle-tilt platform trajectory generation protocol. ............................................................... 27 Figure 2-5 – Stewart platform and ankle-tilt platform closed loop transfer function after tuning. .......... 32 Figure 2-6 – Load stiffness plots for a representative participant, with the ankle-tilt platform locked (left) and engaged (right). ................................................................................................................................... 34 Figure 2-7 – RMS of load stiffness results. ................................................................................................. 35 Figure 3-1 – Model of the perceptual response to galvanic and physical vestibular stimulations. ........... 40 Figure 3-2 – Findings by Goldberg relating the frequency dependant vestibular afferent response to physical and electrical stimulation ............................................................................................................. 42 Figure 3-3 - Experimental setup for vestibular studies. ............................................................................. 44 Figure 3-4 – Perception of motion induced through superposition of orthogonal GVS and rotation. ...... 45 Figure 3-5 – Superposition of balance simulation and head rotations. ..................................................... 48 Figure 3-6 – (Left) Selected phase leads for galvanic vestibular stimulation (GVS) relative to angular velocity in perception. (Right) inferred phase lead of afferents re: GVS stimulation. ............................... 52 Figure 3-7 – Fast Fourier transform results of soleus muscle EMG for a participant exhibiting complete cancellation at 0.8125 Hz. .......................................................................................................................... 54 Figure 3-8 – Computed gains and phase leads for galvanic vestibular stimulation (GVS) relative to angular velocity in reflexes. ..................................................................................................................................... 55 Figure 3-9 – EMG reflex response amplitudes grouped by trial number, with corresponding correlation coefficients. ................................................................................................................................................ 56 Figure A-1 – General layout of the ankle-tilt platform control system. ..................................................... 78 x  Figure A-2 – Exploded rendering of the ankle-tilt platform assembly. ...................................................... 79 Figure A-3 – Simplified model of RISER software architecture with ankle-tilt platform addition. ............ 81 Figure A-4 – Ankle-tilt platform emergency stop block diagram. .............................................................. 83 Figure A-5 – Tilt platform system schematic. ............................................................................................. 86 Figure A-6 – Tilt platform mechanical design 3D model. ........................................................................... 86 Figure A-7 – Tilt platform main tower mechanical component ................................................................. 87 Figure A-8 – Tilt platform bearing and shaft assembly .............................................................................. 88 Figure A-9 – Tilt platform and force plate assembly. ................................................................................. 88 Figure A-10 – Motor interference with back support ................................................................................ 95 Figure A-11 – Motor interference with foot platform................................................................................ 95 Figure B-1 – Head rotation-tracking method. ............................................................................................ 98 Figure B-2 – Raw forceplate measurements from reflex study trials....................................................... 100 Figure B-3 – EMG responses for superimposed passive torques instead of head rotations. .................. 100 Figure B-4 – 2.5 Hz torque response to passive rotations, before and after compensation. .................. 102 Figure C-1 – Example of data collected for indirect comparison of perception, and fitting of sinusoids. 107 Figure C-2 – Phase leads of rotation and GVS relative to metronome signal. ......................................... 108 Figure C-3 – Comparison of perception study results. ............................................................................. 109     xi  Glossary  Glossary Afferents: Nerves that carry signals from the sensory organs to the central nervous system. Ankle-Tilt Platform: A new addition to the Robot for Interactive Sensory Engagement and Rehabilitation system that permits independent manipulation of the feet and ankles relative to the rest of the body. Anode/Cathode: The cathode is the positively charged electrode. In conventional current, electricity flows from cathode to anode. A-P Sway: Anterior-Posterior sway, when referring to balance control, is front-to-back body motion. CNS: Central Nervous System, including the brain and spinal cord. COM: Centre of Mass, the object (or individual) has an equal distribution of mass about this location. COP: Centre of Pressure, the location of the instantaneous centre of the forces being applied to a surface, which is different from the COM projection in dynamic situations. DAQ: Data Acquisition System. A DAQ is an electronic device capable of sampling analog signals from sensors or other devices and converting the sampled data into digitized measurements for use by a computer. DOF: Degrees of Freedom, the number of independent joints that can be manipulated. Dorsiflexion: Contracting (primarily) the tibialis anterior muscle to apply a torque that would tend to lift the toes upwards or cause the body to tilt forwards when balancing.  Efferents: Nerves that carry signals from the central nervous system to the peripheral organs (to execute motor commands or modify sensitivity of afferent signals). EMG: Electromyography, a technique for measuring muscle activation by monitoring muscular electrical activity. Gastrocnemius: A bipennate muscle that runs on top of the soleus in the back of the lower leg, and is heavily involved in balance control through plantarflexion. xii  GVS: Galvanic Vestibular Stimulation, the procedure of activating the vestibular system through externally imposed electrical currents. In humans, this is accomplished by passing small (1-5mA) currents across the mastoids using surface electrodes. Typical types of stimulation include sinusoidal, stepped, pulsed, or stochastic (known as SVS) currents. Interaural Line: A physiological reference line that connects the ears across the head. Relevant in brain studies and vestibular research.  Inverted Pendulum: A simple mathematical model in which a point or distributed mass is suspended above a pivot joint. A fundamentally unstable system that can be balanced by applying torques through the pivot joint. Frequently used as a model of humans maintaining standing balance. Jitter: Oscillations of a control system caused by system delays or variances in control timing. Can lead high-frequency position tracking error, or even system instability in extreme cases. Load Stiffness: The relationship between required ankle torque (to maintain position) and standing angle. Linear for quasi-static sway. LTI: Linear time invariant. A type of system in which the response to a given input scales linearly with the magnitude of the input and the response to multiple inputs is the superposition of each individual input. Mastoids: The prominent skull projections behind the ears. Motoneuron: An abbreviation for ‘motor neuron,’ which is a type of efferent nerve designed to transmit signals to muscles. M-L Sway: Medial-Lateral sway, again in balance control terminology, is left to right body motion. Otoliths: Located in the labyrinth of the inner ear, otoliths are small calcium carbonate crystals that flow in a gelatinous fluid inside the utricle and saccule organs (which are hair-filled ducts inside the inner ears). The relative motion of the otoliths against the hairs allows the nervous system to detect accelerations of the head, including gravitational accelerations. Passive Stiffness: Elastic torque induced through the rotation of a joint primarily due to the stretching of tendons. Plantarflexion: Contracting (primarily) the gastrocnemius and soleus muscles to apply a torque that would push the toes downwards or cause the body to tilt backwards when balancing. xiii  Proprioception: The ability to sense the relative position and orientation of one’s body, as well as the amount of effort being expended during muscle activity. RISER: Robot for Interactive Sensory Engagement and Rehabilitation, a control system and 6 DOF Stewart platform customized for use in exploring human balance. RMS: Root Mean Square SD: Standard Deviation Semicircular Canals: Located in the labyrinth of the inner ear, these organs are small semi-round tubes filled with fluid. Rotational motion causes the fluid to move in the canals, stimulating the hair cells that pervade the organ and allowing the nervous system to detect angular motions of the head. Soleus: A powerful muscle on the back of the lower leg that runs from the heel to just below the knee that plays a major role in balance control through plantarflexion. Somatosensory Feedback: A more complete term for the sense of ‘touch’, which encompasses a variety of sensory modalities including pressure, temperature, and proprioception. Sway: The motion of the body during standing balance. Even in quiet stance, there is measurable sway in the anterior-posterior direction, with the majority of rotation taking place about the ankle joint. Tibialis Anterior (TA): A muscle located on the front of the lower leg that runs from the upper part of the tibia down to the foot, and is the primary muscle responsible for dorsiflexion. Vestibular System: A set of internal organs located in the labyrinth of the inner ear that can detect linear (otoliths) and rotational (semicircular canals) motion of the head. These organs are critical for maintaining balance. xiv  Acknowledge me nts  Acknowledgements I have many people to thank. First, I would like to thank my supervisors, Dr. Croft, Dr. Van der Loos, and Dr. Blouin, who provided spectacular support, guidance, and direction throughout this process. I would like to acknowledge Tom Huryn and Dr. Luu (Billy), the ones who came before me, and put in a huge amount of work bringing the project to the point where I picked it up. I would particularly like to thank Tom for putting a lot of effort into transition, and Billy for asking hard questions that helped me to learn. Philip, the one to whom I pass the RISER torch, I wish you the best of luck and success in your experiments. May your Master’s degree be more expedient than mine. Thank you to Hina, Max, Ian, Kiran, and Andrew, the co-ops and interns who participated in the project, and contributed both hard work and brainpower. I enjoyed working with all of you and I know you will all be successful. A big thanks to my lab-mates and friends who supported and encouraged me. Thank you for telling me not to give up on the days when nothing was working. Jenny, a special thanks for helping me finish this and for just generally being an awesome person. Christina, thank you. I am not sure I could have done this without your help and support. I have been incredibly lucky to have the unwavering support and love of my family. Mom, Dad, Adam, and Kelly, you have helped in too many ways to count. I hope that I make you proud.  xv  Dedication  Dedication To my Api; I know you would have read this cover to cover.1  1 Introduction and Literature Review Introduction and Literature Review 1.1 Motivation Standing balance plays a pivotal role in daily life. While most take this innate ability for granted, the simple act of standing can be a major challenge for older persons and those living with pathologies including stroke (Niam, Cheung, Sullivan, Kent, & Gu, 1999), vestibular impairment (Pothula, Chew, Lesser, & Sharma, 2004), and Parkinson’s disease (Adkin, Bloem, & Allum, 2005). Such pathologies contribute to the risk of falls, a leading cause of death among older adults (van Asseldonk et al., 2006). Improved understanding of balance biomechanics and physiology is important for developing appropriate therapies for these clinical populations. While the main inputs (vision, vestibular and somatosensory information) and outputs (motoneuron activity) of the human balance system are generally well understood (Allum & Pfaltz, 1985; Diener, Dichgans, Guschlbauer, & Mau, 1984), the on-going control that takes place to integrate the input information and produce an appropriate response remains an area of active investigation (Mahboobin et al., 2008; Milton et al., 2009; Qu, Nussbaum, & Madigan, 2007). This work contributes to the understanding of the human balance model, and helps to explain how the individual senses contribute to this control system. 1.2 Standing Balance Human balance is a well-researched field, due to its significant implications in day-to-day life. The following subsections provide a review of the current understanding of human standing balance biomechanics & physiology and examine the literature relevant to the studies detailed in Chapters 2 and 3. The first subsection provides an overview of the current understanding of human balance. The second subsection examines the current understanding of the primary senses involved in balance, with a focus on proprioception and vestibular sensory information. The third subsection delves into the current understanding of how this sensory information is processed and used by the nervous system to control standing balance. The fourth subsection details a variety of measurement techniques used to study and 2  explore the human balance system. Finally, the fifth and sixth subsections explore current work in the areas of proprioceptive and vestibular research, respectively, and identify gaps in current understanding that the present work endeavours to address. 1.2.1  Biomechanics & Physiology of Human Standing Balance The act of maintaining standing balance is a complicated interaction between numerous sensory inputs, sensorimotor processing, and motor output responses. With a centre of mass located well above the ground and only two legs for support, humans maintain balance in a fundamentally unstable state. A slight lean in any direction will lead to an accelerating fall that can only be counteracted through active muscular action. Balance responses are produced through a complex integration of muscular myostatic (stretch) responses to motion (Carpenter, Allum, & Honegger, 1999), low-level (unconscious) reflexive processing in the central nervous system (CNS) (Horak, 2006), planned (feed forward) movements based on an internal balance model (Green & Angelaki, 2010; Merfeld, Zupan, & Peterka, 1999), and high-level conscious control (Krishnamoorthy & Latash, 2005). To determine an appropriate balance response, the body relies on a vast suite of sensory information: vision, the somatosensory system, and the vestibular system have all been shown to provide critical sensory information for balance control (Fitzpatrick & McCloskey, 1994). All balance responses are limited by the capacity of the individual’s physiology to produce the desired motion. Tendons, muscles, and the skeletal system are all involved in producing the torques and forces needed to maintain balance. The skeleton provides structure and an attachment point for the tendons and muscles that modulate sway and control balance. In ‘quiet stance’ (standing on a stable surface with no external perturbations), the majority of sway takes place across the ankle joint and in the AP plane. The soleus and medial gastrocnemius (Héroux, Dakin, Luu, Inglis, & Blouin, 2014) muscles, located in the back of the leg, are the primary muscles responsible for plantarflexion, and are connected and supported by the gastrocnemius and Achilles tendons. The tibialis anterior (TA) muscle, located in the front of the leg, is the workhorse for dorsiflexion. These three muscles together produce the majority of muscle activity during quiet stance, and all three act about the ankle joint (Di Giulio, Maganaris, Baltzopoulos, & Loram, 2009). The motion of these muscles, at first glance, appears paradoxical in nature: as the body sways forward, the soleus and gastrocnemius muscles contract, while they extend and the TA retracts during posterior sway (Loram, Maganaris, & Lakie, 2004). This contradiction, however, is a consequence of and a testament to the human balance system: as the body begins to sway forward, the muscles 3  contract to provide opposing torques and begin pulling the body in the posterior direction. Without this anticipatory action, the passive stiffness of the muscles and tendons alone would be insufficient to maintain standing (Loram & Lakie, 2002a; Loram et al., 2004). More complex balance tasks involve activation of the knees, hips, and arms for additional power and inertial compensation, but this work is focused on the maintenance of quiet stance. Developing an effective and accurate model of the human balance system remains a difficult challenge due to the complexity of the physiology and the intricacy of the control. Nonetheless, simplifications have been developed and tested to permit an incomplete modelling of the system without trivialization. As a key example, a standing human closely resembles an inverted pendulum in the anterior-posterior (AP) direction (Figure 1-1). The use of an inverted pendulum as a model for AP human balance was validated for quiet stance by Gage et al. (Gage, Winter, Frank, & Adkin, 2004), and remains a central component to human balance studies. In much of the current literature discussed in this chapter, the focus of the research is on attempting to explain how the human body controls and stabilizes this inverted-pendulum system.  Figure 1-1 – Inverted pendulum model of human balance. In this model, the body is treated as rigid, so the system can be modelled as a point (or, for more advanced models, distributed) mass (weight = mg) located at the subject’s centre of gravity. All torques (T) are generated and act about the ankles. The feet are treated as pinned to the ground, so the reaction forces (FRZ and FRY) do not affect the system. 4  1.2.2 Primary Sensory Systems in Balance The human body relies primarily on vision, somatosensory, and vestibular inputs to provide sensory information while maintaining quiet stance. There is typically a noticeable impact on balance performance and postural stability when even one system is removed or impaired (Diener et al. 1984; Krishnan and Aruin 2011; McCall and Yates 2011), depending on the extent of that system’s involvement in the current balance task. The body’s ability to sense pressures under the feet, as well as the capacity to determine the current rotation of the ankle joint, collectively form the primary somatosensory components involved in standing balance. The somatosensory balance system includes a variety of sensory subsystems: muscle spindle fibres innervate the leg muscles, and respond primarily to static changes in length of the muscle; Golgi tendon organs are receptors located at the connection between the muscles and the tendon, and respond to changes in load and length; cutaneous and ligament mechanoreceptors are sensitive to pressure, touch, and stretching. Fitzpatrick et al. showed that sensory information from the feet and ankles (that is, cutaneous and ligament mechanoreceptors) has a “small but significant effect on stability,” and further, that combination of passive mechanics with lower-limb muscle-spindle and Golgi afferents were together sufficient to maintain quiet stance (Fitzpatrick, Rogers, & Mccloskey, 1994). Feuerbach et al. reinforced Fitzpatrick’s findings by concluding that ligament mechanoreceptors are generally unimportant in maintaining standing balance, however they placed a higher importance on cutaneous sensors (Feuerbach, Grabiner, Koh, & Weiker, 1994). In a study of walking subjects, Klint et al. found that the load sensitive receptors played the biggest role in afferent feedback; however, this finding may not be generalizable to quiet stance (af Klint, Mazzaro, Nielsen, Sinkjaer, & Grey, 2010). Proprioception is capable of detecting very small ankle movements: Fitzpatrick et al. demonstrated that human subjects could perceive ankle rotations as small as 0.003 rad while standing (Fitzpatrick & McCloskey, 1994). Studies into the role of the somatosensory system in balance typically involve the removal of pressure under the feet (Hansson, Beckman, & Håkansson, 2010; Patel, Fransson, Johansson, & Magnusson, 2011), locking the ankles relative to the rest of the body (Feuerbach et al., 1994), or applying an anaesthetic to eliminate sensation (Feuerbach et al., 1994; Patel et al., 2011; Perrin, 1996). Lee and Aronson demonstrated that vision is one of the first senses infants rely on to maintain balance (Lee & Aronson, 1974), and vision remains important in adult life by significantly reducing overall postural sway during balance (Krishnan & Aruin, 2011). Vision appears to be primarily involved in low frequency adjustments to standing position (Buchanan & Horak, 1999), but does not perform well at detecting 5  extremely slow changes. Under normal conditions, it can detect approximately the same amount of angular displacement about the ankles as can be sensed by proprioception (about 0.003 rad)(Fitzpatrick & McCloskey, 1994). Vision is capable of providing rich information for an internal balance model, so it is unsurprising that research has found it is instrumental in assisting in compensation for incoming or anticipated balance disturbances, such as standing on a moving surface or catching a ball (Mohapatra, Krishnan, & Aruin, 2012). The vestibular system is a set of organs located in the labyrinth of the inner ear, behind the mastoid process. Each side of the head holds three semicircular canals and two otolith organs, that are sensitive to angular and linear acceleration, respectively (Tascioglu, 2005), as shown in Figure 1-2, below. The semicircular canals are filled with fluid, and motion of this fluid in the canal causes bundles of hairs to bend in the direction of flow. This causes a chemical process which increases or inhibits the firing rates of the attached afferent nerves, depending on the direction that the hairs are bent (Colclasure & Holt, 2003). For the otolith organs, several thin layers of cells slide relative to each other, inhibiting or increasing hair cell firing rates and allowing each organ to sense accelerations in two planes (Fitzpatrick and Day 2004). Fundamentally, both the otolith and semicircular canal organs encode physical motions as neural impulse rates on their respective vestibular afferents. The signals from all of the afferents are then combined in the central nervous system to permit sensation of movement of the head in all six degrees-of-freedom (DOF) (Tascioglu, 2005). Through the vestibulo-ocular reflex, the vestibular system also provides important information to help stabilize the visual field during motion (Straka & Dieringer, 2004). Vestibular information has nearly an order of magnitude higher detection threshold for ankle rotation during standing when compared to vision or proprioception (Fitzpatrick & McCloskey, 1994). As a consequence of this lower sensitivity, vestibular signals can be expected to contribute less to the balance response during quiet stance; studies have shown that vestibular information is increasingly important when standing on an unstable surface (Fitzpatrick, Burke, & Gandevia, 1994) or when the consequences of a fall are increased (Osler, Tersteeg, Reynolds, & Loram, 2013). In general, vestibular information also appears to have a comparatively high latency of approximately 60-100ms in developing a reflex response (Cathers, Day, & Fitzpatrick, 2005). These delays can be attributed to the neural processing that takes place to transform the vestibular signals from a set of head-oriented accelerations and rotations into a relevant muscular response in the legs. This neural processing has been shown to consider both the current balance task (Fitzpatrick, Burke, et al., 1994), and the orientation of the head relative to the rest of the body (Lund & Broberg, 1983), in determining an appropriate balance response. 6    Figure 1-2 – Organs located in the labyrinth of the inner ear. The three semicircular canals (anterior, posterior and horizontal) and the two otolith organs (utricle and saccule) are sensitive to physical disturbances, and excite or inhibit the firing rates of the attached vestibular afferents2. 1.2.3 Passive, Reflexive, and Conscious Control of Balance Maintenance of balance is facilitated by a wide range of physiological mechanisms. All of the sources of sensory information discussed in the previous section have been shown to play an important role in balance, however the extent to which each sense is used in a particular task remains an active field of research, and may change substantially depending on the conditions (Mahboobin et al., 2008). Further, balance control does not rely solely on sensory feedback. Other relevant factors for the maintenance of balance include the generation of restoring torques from natural muscle and tendon stiffness (Fitzpatrick, Taylor, & Mccloskey, 1992), the internal model of the current balance task (Merfeld, Park, Gianna-Poulin, Black, & Wood, 2005; Merfeld et al., 1999), and the influence of voluntary conscious control (Reynolds, 2010). This section examines the variety of physiological mechanisms that have been shown to influence the maintenance of standing balance and discusses their significance, based on current understanding. The first of these mechanisms is passive stiffness. Passive stiffness is the natural resistance to plantarflexion and dorsiflexion provided by the mechanical properties of the muscles, tendons, and skeletal structure that make up the ankle joint. This resistance to motion provides a stabilizing effect on the human pendulum model, reduces the effort required by active muscular action to maintain standing.                                                           2 Reprinted under the terms of the GNU free documentation licence. Obtained from Wikipedia, published 19 November 2008 by user Thomas.haslwanter from Sensesweb. http://en.wikipedia.org/wiki/Vestibular_system#mediaviewer/File:VestibularSystem.gif. 7  While arguments have been previously made that passive stiffness is sufficient to maintain standing (Horak & Macpherson, 1996; Winter, Patla, Prince, Ishac, & Gielo-Perczak, 1998), in general researchers have found that while passive stiffness provides a significant contribution to static stability, this stiffness alone is insufficient to maintain standing balance (Loram & Lakie, 2002a; Morasso & Schieppati, 1999). Loram and Lakie demonstrated that passive stiffness can provide approximately 90% of the static load needed to maintain balance (Loram & Lakie, 2002a). The remainder of the static load and any dynamic load is borne by active muscular contractions, resulting from sensory activity and higher level processes (Loram & Lakie, 2002b). Active muscular responses are controlled by both local sensory information (reflexes) and motor information from the CNS (commands). As discussed in the previous section, Golgi tendon organs, muscle spindle receptors, and cutaneous mechanoreceptors all contribute proprioceptive information about the ankle’s motion and loading. This information is transferred to the central nervous system for higher-level balance control, but also contributes to short circuit responses that can provide rapid reflex control to resist muscle motion (Sorensen, Hollands, and Patla 2002). These reflexes are believed to be less important in maintaining quiet standing balance than in more dynamic situations, but nonetheless play a role in overall stability (Richard Fitzpatrick, Rogers, and Mccloskey 1994).  Higher-level responses are organized by the central nervous system (CNS). The CNS handles unconscious control of balance, including reflexes and natural standing sway. Present research in human balance tends to be interested in CNS responses, given that the lower levels of balance control are much better understood (as they are easy to correlate to specific stimuli). These studies are beginning to provide a broader view of the multisensory integration that takes place during unconscious balance tasks, such as maintaining stable posture, responding to a balance disturbance, and reflexively anticipating an upcoming disturbance (granting the ability to adapt to standing or walking on a boat, for example). In general, the research has supported the idea that the human body unconsciously maintains an internal model of the current balance state based on available sensory information (Luu et al., 2012), and uses this model to plan and issue muscle commands in a feed-forward manner to minimize the effort required to maintain balance (Green & Angelaki, 2010; Loram et al., 2004; Merfeld et al., 1999). The research community has also shown increased interest in the highest level of balance control: conscious or deliberate movement. The CNS is capable of consciously perceiving inputs from all three balance-related senses (vision, proprioception and vestibular) and can produce balance responses that incorporate or almost completely disregard them. Consequently, conscious balance is difficult to 8  incorporate into any balance model, and often serves as a confounding but important factor in balance research. One aspect of conscious control that has been explored more significantly is voluntary sway during standing (Danna-Dos-Santos, Degani, Zatsiorsky, & Latash, 2008). In one aspect of this research, the intent is to understand the extent to which humans can consciously minimize sway, by focusing their attention on holding a specific position, and also whether this extent varies with age or specific pathologies (Green & Angelaki, 2010; Stelmach, Phillips, DiFabio, & Teasdale, 1989; Tucker, Kavanagh, Morrison, & Barrett, 2010). Other researchers are more interested in whether attention during sway affects reflex responses and other balance outcomes (Horak, 2006; Reynolds, 2010). While it is generally possible to group balance control into the categories of passive, unconscious, and conscious control, there are crossovers and interactions that make it more difficult to completely separate the layers. Much of this is due to the internal model component of the balance system, which makes use of both high and low level sensory information in determining balance responses, particularly for the vestibular system (Green & Angelaki, 2010; Luu et al., 2012; Merfeld et al., 2005). This can be observed in balance reflexes; Fitzpatrick showed that vestibular reflexes are dynamic, and are modified by the current balance task (Fitzpatrick, Burke, et al., 1994). As example of a mechanism through which modified reflexes could occur is via vestibular efferents, which Sadeghi has shown to be capable of inhibiting or exciting the afferent responses (Sadeghi, Goldberg, Minor, & Cullen, 2009), and which could consequently affect the reflexive vestibular response. The result is a large and complex control system consisting of many (inaccessible) low and high-level control loops, as well as the possibility for interaction between them; this complexity is the primary challenge of studying human balance control. 1.2.4 Measurement Techniques In their efforts to study this closed loop, multifaceted control system, researchers have developed a vast variety of measurement techniques. Some approaches directly measure stability and muscle action to provide a ‘big-picture’ view of balance control, while other techniques examine peripheral responses, such as eye reflexes, with the goal of gleaning more information about neural processing and control strategies. One straightforward and popular measurement for AP balance is direct measurement of body sway angle relative to the ankles. This metric is commonly used, particularly in clinical research, due to its ease of measurement and strong correlations with a variety of balance disorders (Bergin et al. 1995; Corporaal et al. 2013). Larger sway velocities tend to be correlated with aging and increased risk of falls, as the responsiveness of the balance reflexes begin to decrease (Fernie et al. 1982). A wide variety of 9  measurement methods are designed to examine the angular position, velocity, and acceleration during standing. Sway measurement systems include accelerometers (Seimetz et al. 2012), inclinometers (Smith, Coppieters, and Hodges 2005), and laser distance sensors (Luu et al. 2011). On its own, however, sway often does not provide significant insight into the human balance control system, as an individual’s sway pattern is the cumulative effect of all reflexes and conscious movements, making it difficult to correlate outcomes to a specific stimulus. The high inertial effects of the body and multiple degrees of freedom can also serve to attenuate the effects of any given muscular response, which necessitates simplifications in the measurement of sway. To help mitigate the effects of body dynamics, instrumented forceplates can be used to provide a more direct measurement of balance responses. The most common metric where forces are involved is the tracking of the participant’s Centre of Pressure (COP), the dynamic location that represents the net reaction force between the participant and the ground. As the participant sways, all of the forces and torques applied through their feet can be summarized as an X, Y, and Z reaction force at some position on the force plate. In static (unmoving) situations, the X and Y position of the COP is the same as the projection of the participant’s centre of mass onto the plate, however in dynamic situations this is not the case; the COP is affected by the stiffness and damping present in the participant’s joints, as well as inertial effects of their moving mass. Larger excursions of the centre of pressure from the central balance point have been shown to correspond to decreased balance control and an increased likelihood of falls (Prince, Corriveau, Hébert, & Winter, 1997). As the forces and torques upon the plate represent the net response of the body, whole body control of balance can be examined, and anterior-posterior (AP) responses can be examined simultaneously with medial-lateral (ML) responses. These force measurements are a combination of active, static, and inertial loads, making measurement more complicated. COP also still represents a cumulative balance response, limiting its utility in extracting the response to a particular input. Despite the challenges, forceplate measurements remain an important form of data collection in human balance studies (Karlsson & Frykberg, 2000; Önell, 1999). If additional information is desired, the activity of the major AP balance muscles (soleus, tibialis anterior, and gastrocnemius) can be measured through surface electromyography (EMG). This procedure involves placing differential electrodes on the surface of the skin above the target muscle. Muscle activity will produce a potential difference between the electrodes, which can be amplified and recorded. EMG can be performed using intramuscular electrodes as well, which allows more direct measurement of a specific muscle and reduces crosstalk from other muscle groups. By measuring the muscles directly, it becomes 10  possible to separate active responses from passive responses induced by muscle stretching, allowing greater insight into the active processes taking place during balance. Eye movements are influenced by the vestibular system through a set of direct nervous system connections. The well-known vestibulo-ocular reflex is characterized by eye movements in response to movements of the head in space. This is widely considered a stabilization mechanism to allow the eyes to compensate for external movements and still function effectively (Straka & Dieringer, 2004). Due to this link, high speed recordings of eye movements can be used to examine the reflex response to vestibular inputs (Goldberg, 2000; Straka & Dieringer, 2004). Another valuable measurement technique is the reported perception of participants during balance trials. Research in this field typically has participants report binary outcomes (moving vs. not moving, left vs. right, etc.) (Wardman, Taylor, & Fitzpatrick, 2003), rank the amplitude of their perception on a subjective points scale (St George & Fitzpatrick, 2011), or relate the observer’s success at detection to some physical quality of the stimulus (such as stimulus amplitude or duration)(Grabherr, Nicoucar, Mast, & Merfeld, 2008; Wichmann & Hill, 2001). This technique allows researchers to directly access the participants’ perceptions of their motion through space, but may not account for reflexive actions that bypass their awareness or for the individual biases inherent in self-reported results. Further, simply asking participants to pay attention to their balance senses may modify their reflex responses. Most importantly, measuring perception does not tell the entire story, as numerous works have shown that perception and balance response are not always congruent, particularly in vestibular studies (Luu et al., 2012; Merfeld et al., 2005). Each of these measurement techniques provide insight and raise new questions about the understanding of the balance control system, and the role of the various senses in maintaining a stable upright posture. The following two subsections detail active research in the vestibular system and proprioception, respectively, and examine some of the open questions researchers are still trying to address. 1.2.5 Vestibular System Research It has been shown that (e.g. (Fitzpatrick and Day 2004)), by applying a small electrical current across the mastoids it is possible to produce a physiological response that affects standing balance. The mechanism through which this acts is not fully understood, however it has been demonstrated in monkeys (Goldberg, Smith, & Fernández, 1984) that electrical current can affect the firing rates of the vestibular afferents, which is theorized to result in the stimulation being interpreted as a vestibular input. This stimulation method is known as galvanic vestibular stimulation, or GVS. 11  Participants receiving GVS report an illusion of motion, and exhibit independent ocular and postural reflex responses if they are engaged in some form of balance task (Fitzpatrick et al. 1994; Kim 2013). The direction of this response depends on the polarity of the electrodes and head orientation. When freely standing, participants consistently sway towards the anode (positively charged electrode) along the interaural line (Lund & Broberg, 1983), and are able to accurately report the direction they are swaying (Wardman et al. 2003). When constrained (unable to sway), participants perceive a much smaller illusory motion towards the cathode along the interaural line (Wardman et al. 2003). There has been significant effort by researchers to characterize the electrical vestibular stimulation, both in perception and reflex, however it remains an active field of debate (Cohen, Yakushin, & Holstein, 2011; Curthoys & Macdougall, 2012). In the study of GVS, it is unclear which of the vestibular organs’ afferents are affected by the stimulation, or if they are all affected to the same extent. Studies based on human perception tend to find that the sensation is perceived as predominantly a rotation about the head (Wardman et al. 2003), based on reports from the participants. Studies of human physiology have proposed that all of the afferents are stimulated equally, and the net result (after equal and opposite signals cancel) is a rotation about the head plus a linear acceleration vector (Day, Marsden, Ramsay, Mian, & Fitzpatrick, 2010; Kaptein & Van Gisbergen, 2006; Wardman et al., 2003). Still others, primarily those researching the role of vision in human balance, suggest that the reflexive eye movement patterns characteristic of GVS stimulation (discussed below) are indicative of only a linear acceleration, tying these responses more closely to the otolith afferents (Kim, 2013), however other eye researchers have shown evidence of angular responses (Kaptein & Van Gisbergen, 2006; Schneider, Glasauer, & Dieterich, 2000). As the input methods and measurement techniques vary greatly across studies, it is difficult to develop a cohesive explanation for the differences between the results. Being able to consistently predict how an individual will experience and respond to GVS is important to the field of vestibular physiology. Specifically, it would greatly expand the usefulness of GVS in serving as a pure vestibular disturbance for studying human balance, and it would help clarify whether GVS affects the semicircular canal or otolith afferents, or both (Cohen et al., 2011; Curthoys & Macdougall, 2012). Demonstrating how the vestibular signals are processed and combined with other signals would improve the research community’s understanding of how the CNS is affected by vestibular inputs. 12  1.2.6 Ankle-proprioception Research A key element of the somatosensory balance system is ankle motion, that is, the capability of the body to detect the angle of the feet relative to the rest of the body. Research into ankle-proprioception has typically focused on its role in maintaining stable balance posture, with a goal of reducing falls and helping to understand how aging, disability, and other impairments or diseases can affect the ability to balance (Bergin, Bronstein, Murray, Sancovic, & Zeppenfeld, 1995; Niam et al., 1999). Results for these clinical populations are already being realized, with studies demonstrating improved posture control and gait after targeted proprioceptive rehabilitation (Park, Kim, & Lee, 2013). One research avenue is the comparison of clinical populations with healthy participants, to look for differences in balance performance (Nardone et al., 2000; Niam et al., 1999). Research effort has been put into developing devices that study the role of ankle motion and proprioception in balance without producing other sensory inputs. For example, Fitzpatrick et al. isolated ankle motion by having participants use their feet to balance an inverted pendulum mass that is theoretically equivalent to their body (Figure 1-3), which eliminates any vestibular and visual inputs (Fitzpatrick, Rogers, et al., 1994). Examples of a variety of devices developed for specifically studying ankle motion (Fitzpatrick, Rogers, et al., 1994; Refshauge & Fitzpatrick, 1995) are shown in Figure 1-3. The challenge in employing many of these systems to study standing balance, however, is that by isolating proprioception using these methods, the balance task has been inherently changed; the participant no longer needs to maintain standing balance, so their balance responses may be altered. This limits the ability to relate the results to balance performance during free standing. Consequently, a device capable of closely mimicking normal standing while providing the ability to modify proprioceptive feedback would be a valuable research tool. To this end, this thesis details the implementation of a new device (Chapter 2) to help explore and quantify the role of ankle proprioception in human balance.  13   Figure 1-3 – Examples of ankle manipulation devices. Reprinted with permission from (Fitzpatrick, Rogers, et al., 1994) and (Refshauge & Fitzpatrick, 1995)3. Left: In (A), participants balance in normal stance with pressure cuffs anaesthetizing the feet and ankles. In (B), participants are fixed vertically and instead balance an inverted pendulum bob through the ankle platform to eliminate vestibular feedback. In (C), participants are free to move but have the same backboard as (B), to mimic the somatosensory conditions but include vestibular feedback. Right: participants’ perceptual thresholds of ankle motion are measured in a variety of standing and seated positions using an ankle manipulator. 1.3 Balance Simulation As discussed earlier in this chapter, a common technique in human balance research is distortion or elimination of sensory information, as the differences observed in the balance response can be compared to normal conditions. There are, however, limitations to modifying the sensory information available to participants standing freely, as participants have inherent balance parameters such as mass and height. Moreover, when balance system inputs are sufficiently distorted, participants may fall, putting them at risk for injury. These considerations significantly limit the potential to explore the human balance system in freely standing participants. This presents a significant problem for balance research, as studies have shown that balance responses are task-dependant (Fitzpatrick, Burke, et al., 1994), so any improvements for safety purposes must modify the balance environment as little as possible.                                                           3 Left: Reprinted from The Journal of physiology, volume 480(2), pages 395–403, Copyright (1994), Fitzpatrick, R., Rogers, D.K., Mccloskey, D.I., “Stable human standing with lower-limb muscle afferents providing the only sensory input.” with permission from Wiley. Right: Reprinted from The Journal of physiology, volume 488(1), pages 243–248, Copyright (1995), Refshauge, K.M., Fitzpatrick, R.C., “Perception of movement at the human ankle: effects of leg position.” with permission from Wiley. 14  The Collaborative Advanced Robotics and Intelligent Systems (CARIS) Lab has previously demonstrated that the task of maintaining standing balance in the sagittal plane can be simulated by having a participant stand on a force plate mounted on a 6 degree-of-freedom Stewart platform (MOOG Inc., East Aurora, NY) (Huryn et al., 2010; Luu et al., 2011). The system is collectively referred to as RISER (Robot for Interactive Sensory Engagement and Rehabilitation). The RISER system is configured such that, with the participant safely secured to the platform via a backboard, ankle moments applied to the force-plate cause the robotic platform (and participant) to rotate about an axis passing through the ankle joints (Figure 1-4).   Figure 1-4 – Test participant standing on balance simulator. Axis of rotation for 1DOF balance simulator (pitch direction) passes through both ankles. Reprinted and modified with permission from (Huryn, 2012).  Axis of rotation during balance Forceplate Stewart Platform Adjustable backboard (secures subject) 15  The RISER system implements a physics-based inverted pendulum model of the participant, with parameters based on the participant’s mass and body type (Figure 1-5). Participants are able to provide input to the simulation by applying ankle torques to the forceplate, while their current standing angle is fed back to them through the motion of the Stewart platform. The system is designed such that balancing the platform should feel like normal standing. The fact that participants are balancing a computer model of themselves (as implemented on the simulator) rather than their own body will not be apparent, because the inputs (ankle torques) and outputs (sway) of normal standing are still present, and carry the same mathematical relationship (Huryn et al., 2010; Luu et al., 2011).   Figure 1-5 – Inverted pendulum model of 1DOF human balance, as implemented on the RISER system. Reprinted with permission from (Huryn, 2012). M_ankle denotes the participant’s applied ankle moment, as measured by the forceplate. M_gravity denotes the moment that gravity would apply on an inverted pendulum for a given standing angle. L, m, Im, and g denote the length to the participant’s centre of mass, the participant’s mass, the participant’s computed intertia, and gravity, respectively. An angular acceleration (?̈?) is calculated by dividing the sum of the two moments by the participant’s inertia. This is then integrated twice to produce an angle (𝜽), which is passed to the balance simulator. With the inertial (“plant”) portion of the system converted into a simulation, parameters such as mass and gravitational acceleration are no longer fixed, allowing exploration of the effect of varying these parameters on the balance response. Further, having the participant tightly secured to a platform eliminates the risk of falling that is normally present in freestanding trials. This balance simulator opens many new opportunities for human balance research. It permits great flexibility in the exploration of the human balance system, in particular through the modification of the sensory feedback provided to the participant. While the system provided a very similar task to normal standing (Luu et al., 2011), the existing simulator did not accurately reproduce a key element in balance control: ankle proprioception (Fitzpatrick, Rogers, et al., 1994). As both the force plate and the participant were fixed to the Stewart platform at a constant angle (at the participant’s natural standing angle), the 16  participant performed near isometric contractions during the balance simulations. This situation eliminated the contribution of ankle somatosensory feedback to the balance control loop (Diener et al., 1984) as well as the contributions from the passive mechanics of the ankle, potentially resulting in diminished balance control. In (Luu et al., 2011), the constant ankle angle was raised as the primary reason that the RMS of sway angle (a measure of balance stability) in the simulation was approximately twice that observed in free standing. This motivated the development of a mechanism to provide this feedback through an independent ankle manipulator, as described in Chapter 2. The introduction of separate manipulation of the ankles presents new research opportunities. A device that allows independent ankle manipulation during a balance task permits the decoupling and exploration of the role of passive mechanics and ankle-somatosensory feedback in balance control. Further, improving the device to the point that it is nearly indistinguishable from standing balance allows it to be used as a versatile system for studies into other elements of human balance. As an example, the RISER platform with the ankle-tilt platform is used in Chapter 3 to study the effects of GVS on the human vestibular system by comparing and contrasting participants’ reflex responses to GVS with their responses to physical perturbations of the platform during balance simulations. 1.4 Research Goals In pursuing a richer understanding of the human balance system, this thesis seeks to answer three questions raised in the literature: 1) What role does ankle angle play in regulating quiet stance? 2) Does galvanic vestibular stimulation affect the semicircular canals, otolith afferents, or both? 3) Can the perception of galvanic vestibular stimulation be related to the reflex response? Chapter 2 explores Question 1 through the design and implementation of a robotic ‘ankle-tilt’ system (Pospisil et al., 2012), which the author has integrated into the existing balance simulation. Chapter 2 presents the design of the platform, as well as a human-participant validation experiment that, through two different measures, demonstrates the importance of ankle motion during quiet stance. Chapter 3 describes two studies that explore the perception and reflex response of sinusoidal GVS, and attempt to relate it to a physical rotation at several frequencies.  17  - In the first study, addressing Question 2 (above), participants directly relate their perception of sinusoidal GVS to their perception of sinusoidal rotation.  - In the second study and addressing both Questions 2 and 3, participants engage in the balance simulation while sinusoidal GVS or rotation is superimposed, and the reflex response in their soleus muscle is measured. Then, both GVS and rotation are applied at the same time in an attempt to cancel the reflex.  This work makes use of the ankle-tilt platform from Chapter 2 in order to help explore the reflex response of sinusoidal GVS. The findings in Chapter 3 demonstrate that the GVS reflex response can be at least partially cancelled out when combined with a physical rotation, and that there is a strong but imperfect relationship between the perception of and reflex response to GVS. This chapter explores the differences between these results and their relevance to existing literature. 18   Figure 1-6 – Ankle-tilt platform enabling independent ankle manipulation during balance simulations. Chapter 4 presents a discussion of the contributions of this thesis. This chapter elaborates on the results of both chapters, proposes additional research stemming from this work, and summarizes the main conclusions of my thesis. Appendix A: Ankle-Tilt Control System details the hardware and software implementation of the robotic ankle-tilt platform and explores considerations for forceplate compensation and passive stiffness simulation. The report by the UBC Mechanical Engineering capstone design team is included here as well. Appendix B: Reflex Study Implementation presents technical information on the software implementation of the reflex study, such as the superposition of head rotations and the implementation of the forceplate compensation algorithm. 19  Appendix C: Supplementary Results details the outcomes of pilot studies and other work that supports or expands upon the results and conclusions of the work presented in this thesis. 1.5 Summary This thesis explores aspects of the human balance system using a robotic motion simulator, while presenting technical details of improvements to the simulation. The impact of the vestibular and proprioceptive feedback systems are studied through the addition or removal of stimuli during the balance simulation, providing insights into their roles in maintaining balance.  20  2 “Ankle-Tilt” Platform Design “Ankle-Tilt” Platform Design 2.1 Introduction This chapter discusses the addition of a robotic platform to provide passive ankle torque and ankle-somatosensory feedback to the RISER system. In Chapter 1, the fixed ankle angle was identified as a significant limiting factor in using the platform to study human balance, as the motion of the platform about the ankles did not accurately replicate human body motion during free standing. In that chapter, Figure 1-4 depicts the balance simulator setup without independent ankle manipulation. To facilitate ankle motion while still maintaining the experimental and safety benefits of the balance simulator, an additional degree of freedom was required to manipulate the ankles independently from the rest of the body. To meet this need, the research team working with the robotic platform proposed to develop an ‘ankle-tilt’ platform permitting independent ankle position control. This chapter covers the design, implementation, and validation of the platform. The author led this work, collaborated with an undergraduate capstone-project student team, and supervised five student interns over the course of the design and implementation phases. Mechanical assembly, instrumentation, and system integration (including software) were completed by the author. Additional technical information on the hardware and software implementation, including the design report by the undergraduate project team, assembly instructions, and a bill of materials, can be found in Appendix A. 2.2 Design The following sections detail the design of the ankle-tilt platform. First, existing literature on human balance is examined to determine the minimum performance criteria that will allow the ankle-tilt platform accurately simulate human standing. Second, the physical implementation of the platform is described along with functional considerations regarding its installation. The third subsection provides the requirements of the software and its implementation, which ensures seamless integration with the existing balance simulator. The software was developed to be versatile and permit a variety of modes of operation beyond the scope of the research presented in this thesis. With this consideration in mind, the 21  forceplate and passive stiffness compensation algorithms that accommodate some of these modes of operation are presented below. Finally, a technical validation study is detailed, the results of which are discussed later in this chapter. 2.2.1 Functional Requirements for Balance Studies To simulate the task of human standing accurately, the ankle-tilt platform must be capable of producing ankle-centred accelerations, velocities, and positions that are typical of anterior-posterior sway, while synchronizing with the base Stewart platform and remaining stiff enough to represent a solid surface accurately. Existing literature, discussed below, provided the basis for a set of minimum performance criteria for the physical operation of the platform. The goal of these criteria is to ensure that the ankle-tilt platform will be compatible with all anterior-posterior balance studies. While studying Parkinson’s disease, Adkin et al. (Adkin et al., 2005) reported angular velocities of up to 10 °/s for body sway in healthy subjects. Consequently, as a minimum requirement, the device must be capable of achieving a velocity of at least 10 °/s. Increasing this value further is ideal, as higher peak velocities will likely be achieved in clinical populations, or studies where the participant is unable to maintain quiet stance. As discussed by Luu (Luu et al., 2011), the maximum angular acceleration of the Stewart platform is approximately 500 °/s2. This acceleration was set as a minimum requirement for the new ankle-tilt platform to permit easier controller tuning, and to ensure that the new platform was not a limiting factor in overall system performance. Refshauge and Fitzpatrick reported that 0.2° was the smallest detectable angular displacement during passive rotations of the ankle joint (Refshauge & Fitzpatrick, 1995). This necessitates a minimum angular accuracy of 0.2° to prevent detectable deviations from the correct ankle position. In addition to being undetectable, the forces induced by such a deviation must be small enough to be negligible. Luu et al. proposed an equation for passive stiffness as a function of ankle angle (Luu et al., 2011). A simplified version is presented as Equation (2.1), where 𝜃 denotes ankle angle (rad) and 𝑚 is subject mass (kg). A 100 kg subject would experience less than 5 mN of force generated by passive stiffness for a 0.2° deviation. This force is negligible when compared to the typical torque output of 20-60 Nm during standing balance. 𝑇𝑝 = 0.19𝑚(𝑒−0.05609(𝜃) − 1)  (2.1) 22  Further to this, the original RISER configuration was configured to permit rotational pitch about the participant’s ankles. To maintain the single-degree of freedom pendulum model, the axis of rotation of the ankle-tilt platform must pass through the ankles as well. 2.2.2 Physical Functional Requirements In addition to the balance requirements discussed in the previous section, the ankle-tilt platform must conform to physical requirements that would permit integration with the existing system. Fundamental to these functional requirements is the capability for the ankle-tilt platform to both be secured to and move independently of the Stewart platform. Further, it must still be possible to secure the participant comfortably to the backboard (as shown in Figure 1-4), and the forceplates must still be capable of measuring the ankle torques applied by the participant. Regardless of the method of actuation, the primary purpose of the platform is to produce accurate angular pitch motion about the ankles. Consequently, measurement of angular pitch is necessary as a feedback source for the control system. To permit the possibility of removal or modification, the design of the ankle-tilt platform should permit disassembly. Where possible (without compromising structural integrity), removable or temporary fasteners should be used in place of more permanent methods such as welding or press fits. 2.2.3 Instrumentation, Software, and Controls Requirements For the ankle-tilt platform to interact with the balance simulation and meet the functional requirements, a suitable control system and means of interfacing with LabVIEW™ are required.  Control of the RISER system was accomplished in LabVIEW© (National Instruments, Austin, TX), necessitating ankle-tilt platform instrumentation capable of interfacing with a National Instruments PXI-8108 real-time controller mounted in a PXI-1042 chassis. Power availability in the CARIS lab is limited to single phase 120VAC (15A). For the safety of the participants, power to the actuators must be limited to 24V, driving a need for a 24VDC actuator solution. Smooth and repeatable movement of the ankle-tilt platform as well as synchronization with the Stewart platform are the primary goals of the software/controls design. To accomplish this, high PID rates and a controller that is sufficiently damped to eliminate jitter are required in the implementation. 23  As the design would likely necessitate new forceplate positioning, a modified forceplate compensation algorithm (adapted from (Huryn, 2012)) is required to attenuate the ‘false’ forces generated on the plates by moving them through space. Further to this, algorithms for simulated dynamics that assume a fixed ankle position (such as passive stiffness compensation, discussed below) must be modified to account for the new degree of freedom. In addition to the performance specifications discussed above, the device must be capable of safe operation before, during, and after balance studies. Important safety considerations are discussed in Appendix A. 2.2.4 Physical Implementation The ankle-tilt platform was designed to conform to the requirements in the above three sections, and the completed mechanical model, as presented in the capstone team’s final report, is shown in Figure 2-1. This section outlines the basic functionality and physical implementation of the ankle-tilt platform. Further information on the design methodology employed by the capstone team can be found in Appendix A. Extension and retraction of a linear actuator rotates the platform about the support bearings, which are aligned with the rotation centre of the ankles. By changing the angular position of the platform, it is possible to manipulate the position of the feet and ankles relative to the rest of the body in AP motions (Figure 2-2). Encoder feedback is collected from the support bearing shafts via a timing belt. This design decision sacrifices a small amount of controller linearity, as platform angle is sinusoidally related to actuator length, but offers a more direct measurement of the design variables (angular position, velocity, and acceleration). The ankle-tilt platform is designed such that it can be run in “coupled” mode, in which a single linear actuator drives both ankles, or “dual-actuator” mode, in which two linear actuators control each ankle independently. Some of the redundancies apparent in the design, such as the four bearings, are present to accommodate the dual-actuator mode of operation. At present, only the coupled mode is implemented, so all references to the ankle-tilt platform in this thesis refer to coupled-mode operation. The dual-actuator implementation is discussed further in the design report in Appendix A. 24   Figure 2-1 – Exploded rendering of the ankle-tilt platform assembly. Linear actuator (1) is mounted via ball bearing rod ends below the force plates (2); encoder (3) is connected via timing belt (belt not shown). The entire balance system rotates about the support shafts (4), which are located at the ankles.   Figure 2-2 – Ankle-tilt platform motion illustration. Left: Model of participant standing on RISER with ankle-tilt platform improvement. Centre: Platform can now rotate while maintaining a planar foot position, more accurately resembling normal standing. Right: Tilt-platform gain can be varied, permitting research into the effects of amplified or attenuated proprioceptive feedback. 25   The device incorporates separate force plates for each foot (AMTI, Watertown, MA), supported by a pair of custom machined aluminum ‘foot plates’. The plates are supported by four NTN bearings (NTN Inc., Mississauga, Ontario), and driven by a custom linear actuator (Ultra Motion, Cutchogue, NY). The tower supports are also custom machined and welded. The platform was bolted directly onto the existing Stewart platform, and electronics were integrated into the control rack (Figure 2-3).  Figure 2-3 – Balance simulator with ankle-tilt platform improvement. Subject is secured to backboard and standing on ankle-tilt platform. The Stewart platform’s axis of rotation has been adjusted to pass through the new (more elevated) ankle position, which is aligned with ankle-tilt platform rotation axis. The Stewart platform controls the position of the participant’s body via the backboard while the ankle-tilt platform controls the ankles and feet. Axis of rotation during balance Forceplates Stewart Platform Adjustable backboard (secures subject directly to the Stewart platform) 26  2.2.5 Software Implementation Control integration was accomplished with LabVIEW 2010 and a NI7354 Motion Card (National Instruments, Austin, TX). The motion card performs PID control at 2 kHz. To produce a smooth trajectory profile, the LabVIEW controller writes motion commands to a three-sample buffer in the Motion Card at 60 Hz.  The author configured the axis of rotation for the inverse kinematics of the Stewart platform to coincide with the rotational axis of the ankle-tilt platform, as shown in Figure 2-3. The forceplates were positioned 0.071 m below this axis, corresponding to the average ankle height measured for ten adult human participants in a pilot trial to the study described later in this chapter (SD 0.007 m). This axis placement ensured that the entire system (Stewart platform and ankle-tilt platform) rotated about the participant’s ankles, a location that has been shown by Fitzpatrick et al. to be a good approximation of the pivot point for an inverted pendulum model of anterior-posterior postural sway (Fitzpatrick et al., 1992). The desired angle of the ankle-tilt platform was calculated from the commanded angle of the Stewart platform relative to the global (fixed) frame of reference, so a commanded angle of 0° will cause the ankle-tilt platform to move in such a manner that it remains level regardless of the Stewart platform’s pitch angle. 2.2.6 Control Implementation Successful performance of the ankle-tilt platform depended on not only a high quality frequency response, but also synchronization with the Stewart platform. Phase leads or lags can result in differential motion at the ankles, creating inaccurate proprioceptive feedback. The control systems of the Stewart platform and ankle-tilt platform are fundamentally different; the Stewart platform is controlled internally (‘black boxed’) and the controller requests new motion positions through a callback function at a rate of approximately 60.2 Hz. From the time the new command is sent, there is a 100 ms delay before the motion is carried out. Huryn (Huryn, 2012), in earlier work using RISER, introduced a predictive controller to compensate for part of this delay, reducing the motion delay of the Stewart platform to approximately 41 ms. This known delay was incorporated into the synchronization of the ankle-tilt platform, as discussed below. The ankle-tilt platform is controlled by the NI 7354 Motion Card, providing flexibility in control elements such as PID parameters, update rate, and control scheme. The controller is configured to perform 3-point buffered moves (updated at 62.5 Hz), allowing the trajectory generator to spline the commands together and generate continuous acceleration and jerk trajectories. Controller tuning was performed manually, 27  with a goal of matching the frequency-gain response of the Stewart platform while minimizing delays, which could be manually added if needed. As there is no callback function available, the LabVIEW interface was configured to poll the buffer and write a new trajectory point when space became available in the buffer (Figure 2-4). Consequently, it was not possible to exactly synchronize the timing of the Stewart platform’s trajectory generation with that of the ankle-tilt platform. Instead, an extrapolation scheme was implemented to calculate the required ankle-tilt platform position.  The buffering introduces 50 ms of pure delay on top of the controller and plant delays, so a spline-fitting extrapolation algorithm on the commanded angle was developed to compensate for both delays (Boor, 1978). The extrapolation algorithm accepts the three most recent position commands sent to the Stewart platform and their corresponding time stamps, and splines them together to predict the new command. Cross-correlation of the encoder feedback signals from step responses of the ankle-tilt platform and Stewart platform revealed that a feed forward prediction of 22 ms is required to minimize overall delay errors in the coordinated motion.  Figure 2-4 – Ankle-tilt platform trajectory generation protocol. A polling loop is utilized for this buffered operation. The LabVIEW program waits until there is space in the buffer, and then generates a new position command and writes the command to the open space in the motion controller’s three-point buffer. The buffer is used by the motion controller to generate a trajectory for the ankle-tilt platform. As there are still two commands remaining to be read, there will be a 100 ms delay before the new command is executed, so extrapolation is performed when generating the new position. 2.2.7 Forceplate Compensation and Passive Stiffness Simulation  When the ankle-tilt platform maintains a flat position (gain = 1) the subject’s feet and ankles will interact with the forceplates in a manner that is equivalent to standing on the ground. However, any ankle-tilt gain 28  other than 1 will produce movement of the forceplates relative to the world frame. The movement of the forceplates will have two effects on the physics of the simulation: the motion of the forceplates through space will induce torques due to the inertia of the plates themselves, and the ankles will not produce the appropriate amount of passive stiffness since the ankle angle will not change in the same manner as normal standing. To produce a realistic simulation of balance, these phenomena must be predicted and compensated for in software. In Huryn’s work (Huryn, 2012), compensation algorithms were presented to remove the torques induced from the inertia of the forceplates, and to simulate passive stiffness in lieu of the ankle-tilt platform, which had not been implemented at the time. While these parameters should remain unchanged when the ankle-tilt platform is locked (gain = 0), the amount of compensation should drop to zero when the ankle-tilt platform is fixed relative to the ground (gain = 1), as the forceplates are not moving, and the passive stiffness is fully provided by the ankle flexion. However, for exploratory cases where the platform is set to give only partial feedback (e.g. gain = 0.5), or enhanced feedback (e.g., gain = 2, as in Figure 2-2 - Right), the compensation algorithms must be modified to account for the new combinations of forceplate and ankle motion. For the forceplate compensation, the movement of the plates depends solely on the gain of the ankle-tilt platform. If the gain is 0.25, then when the Stewart platform pitches forward by 4°, the ankle-tilt platform will pitch backward by 1°, resulting in a net 3° forward pitch of the forceplates. Equations (2) and (3) from Luu (Luu et al., 2011) were designed to compensate for the torques induced on the forceplate by calculating a set of compensation constants, and then using the current angular position and acceleration of the forceplate to calculate the compensation value. The compensation constants are different for the revised forceplate positioning, but can be calculated in the same manner. The only necessary change to the compensation algorithm is to replace the 𝜃 and ?̈? of the Stewart platform with the relative terms, as shown in Equation (2.2). The only new term, ‘gain’, is the gain of the ankle-tilt platform. 𝑀𝑎𝑛𝑘𝑙𝑒𝑃 = −𝑀𝑚𝑒𝑎𝑠𝑃 + 𝐶𝑀0(1 − 𝑔𝑎𝑖𝑛)𝜃 + 𝐶𝑀2(1 − 𝑔𝑎𝑖𝑛)?̈? + (𝑧0 + 𝑧𝑎𝑛𝑘𝑙𝑒) ∗ (𝐶𝐹0(1 − 𝑔𝑎𝑖𝑛)𝜃 + 𝐶𝐹2(1 − 𝑔𝑎𝑖𝑛)?̈? − 𝐹𝑚𝑒𝑎𝑠𝑆 ) (2.2) Passive ankle stiffness, however, does not change linearly with sway angle (Moseley, Crosbie, & Adams, 2001). The original passive stiffness torque calculation is shown in Equation (2.3), where 𝜃𝑛  is the individual’s natural standing angle, 𝑚 is mass, and 𝜃 is the current angle of the Stewart platform. 𝑇𝑝 = 0.19𝑚(𝑒−0.05609(𝜃+𝜃𝑛) − 𝑒−0.05609(𝜃𝑛)) (2.3) 29   Consequently, the passive stiffness that is ‘missing’ (and must be simulated) in the 0.25 gain case described above is the difference between the passive stiffness that would be expected at the Stewart platform’s pitch and the amount of passive stiffness that would be expected at the ankle-tilt platform’s pitch. Continuing with the previous example, the Stewart platform angle is 4°, however the first 3° of rotation do not need passive stiffness compensation, because the ankle-tilt platform’s rotation has produced this passive stiffness naturally by rotating the ankle joint. Thus, the only passive stiffness that must be simulated is the difference between a 3° ankle flexion and a 4° ankle flexion. The modified equation (2.4) is shown below.  𝑇𝑝 = 0.19𝑚(𝑒−0.05609(𝜃+𝜃𝑛) − 𝑒−0.05609(𝜃(𝑔𝑎𝑖𝑛−1)+𝜃𝑛)) (2.4) Other literature, such as the work by Loram et al. in 2004 (Loram et al., 2004), has suggested that the passive stiffness involved in quiet stance may be much greater than that presented in Equation (2.3) (Moseley et al., 2001). Despite these newer findings, the equation by Moseley et al. was instead retained and modified, to permit a more direct comparison between the findings in Luu et al.’s work and the human subject validation described below (Luu et al., 2011). 2.2.8 Technical Validation After completing the installation and controller design, the author performed a technical validation to verify that the completed system was performing according to the original design specifications. The ankle-tilt platform’s closed loop system performance was examined by generating sinusoidal position commands and recording the actuator’s responses. The frequencies tested were 0.1, 0.2, 0.4, 0.8, 1.6, 3.2, and 6.4 Hz.  After recording 1 minute of data at each frequency, the sinusoid amplitude was increased until one of angular position, velocity, or acceleration limits of the ankle tilt platform was saturated. The maximum values obtained at each frequency were used to determine the performance limits of the actuator. Analysis and results of the technical validation procedure are discussed later in this chapter. 2.3 Validation with Human Subjects  To examine the platform’s improvements in simulating balance, ten healthy adults (M/F = 6/4) participated in a validation study. The University of British Columbia’s Clinical Research Ethics Board approved all experimental procedures, and all participants provided written informed consent before 30  taking part. All group data is reported as means ± standard deviations. Participants were 27.9 ± 9.2 years old, had a mass of 64.3 ± 8.6 kg, and a centre of mass located 0.89 ± 0.06 m above the ankles. 2.3.1 Experimental Validation Goals The relationship between ankle torque and sway angle (i.e., load stiffness) has been presented as an important criterion in validating the balance simulation (Luu et al., 2011). Load stiffness is an indicator of subject stability; when a participant sways slowly, output torque should change smoothly with standing angle. Large abrupt changes in a load stiffness curve indicate that the participant is having difficulty maintaining a static position. The author hypothesized that: (2.1) the smoothness of the load stiffness curve, as measured by the amount of error about the anticipated curve, would improve with the addition of the ankle-tilt platform. In Luu’s work, the amount of sway during quiet stance on RISER without the ankle-tilt platform was found to be approximately twice that observed during normal standing (Luu et al., 2011). The author further hypothesized that: (2.2) the addition of the ankle-tilt platform would reduce sway during quiet stance, when compared to the previous simulation setup. 2.3.2 Procedures Participants were asked to lie flat on a balance board and adjust their body position until the board was balanced (that is, their centre of mass was directly above the pivot point). The centre of mass height was recorded as the distance from the pivot point to the ankles.  The experimenter positioned the participants’ feet on the ankle-tilt platform force plates so that their ankles were in line with the rotational axis of the platforms, and adjusted the backboard until it gently pressed against the shoulders and lower back without modifying the participants’ preferred standing angle. The experimenter recorded the weight of the participant using the force plates, and then secured the participant to the backboard using chest and waist seatbelts, as shown in Figure 2-3. The participants balanced the simulation in two randomly ordered conditions. Condition A presents a locked ankle-tilt platform actuator, replicating the condition presented in Luu’s work in which the ankle joint was fixed and constrained to the motion of the Stewart platform (Luu et al., 2011). In this case, the dynamic properties of the ankle joint, such as passive ankle torque (Moseley et al., 2001) and ankle damping (Loram & Lakie, 2002a), were simulated in the inverted pendulum model. In Condition B, the 31  ankle-tilt platform was programmed to maintain a constant angle of 0° relative to horizontal, to replicate standing on a flat surface (Figure 2-2 – Centre). As the platform no longer held the ankles in a fixed orientation, the simulated dynamics of the ankle joint provided in Condition A were disabled for Condition B. In each condition, before collecting data, the experimenter gave participants 2 minutes to become accustomed to balancing the system. For each condition, the experimenter instructed participants to sway the simulator within a comfortable balance range at approximately 0.1 Hz for 1 minute. Their motion was guided by an audible metronome. Sway angle and forceplate moments were recorded to permit calculation of load stiffness. This sway frequency is low enough to minimize dynamic effects on the load stiffness that come from the participant’s inertia and muscle activation (Luu et al., 2011). This block of recordings is referred to as the ‘Load Stiffness’ data. The participants then balanced the simulator for 2 minutes while maintaining quiet stance. Sway angle was recorded; this block of recordings is referred to as the ‘Sway’ data. During the balance trials, the author collected additional data for the technical validation. As the ankle-tilt platform was commanded to maintain a 0° angle in Condition B, any deviations measured through the encoder feedback were recorded. These results provided an indication of the tracking errors present during typical operation. Qualitative feedback was collected by asking participants, “What did you think about that condition?” Subjects had consistently reported in pilot studies that balancing the simulator (prior to the ankle-tilt platform installation) was more difficult than maintaining free standing balance, suggesting that participant perceptions are another valuable measure of the realism of the balance-task simulation. 2.4 Analysis and Results 2.4.1 Analysis and Results of Technical Validation The closed-loop response to the sinusoidal position commands was analyzed in MATLAB (Mathworks, Natick, MA). Positional errors were calculated as the difference between the Stewart platform angle and the ankle-tilt platform angle, which accounts for both magnitude and temporal distortions. The raw sinusoids were also converted to the frequency domain for an overall examination of system performance. The frequency analysis showed that the ankle-tilt platform met all performance requirements for the frequency ranges observed in standing balance (0 – 3 Hz) (Moseley et al., 2001). During testing, the 32  maximum positional error was 0.127°, which meets the 0.2° requirement. The platform met the acceleration and velocity performance requirements (500°/s2 and 10°/s, respectively) with peak accelerations of 720°/s2 (measured with a 0.5° amplitude sine wave at 6.4 Hz) and a peak velocity of 18.5°/s (measured at 6° amplitude, 0.8 Hz).  Figure 2-5 – Stewart platform and ankle-tilt platform closed loop transfer function after tuning. Ankle-tilt platform is shown as black squares, Stewart platform as grey circles. Error and phase deviations between the platforms remain minimal within 0-3Hz, the typical frequency range of the platforms. 2.4.2  Analysis of Validation with Human Subjects The balance simulation produced by Huryn et al. models participants as an inverted pendulum with mass m at a distance l from their ankles to their centre of mass (Luu et al., 2011). The static relationship between the required ankle moment M and the angular position 𝜃 from the vertical can be expressed as: 𝑀 = 𝑚𝑔𝑙 sinθ (2.5) Load stiffness was examined by plotting the measured ankle moment against the angular position θ of the Stewart platform, which corresponds to the angle of the inverted pendulum simulation. From Equation 10-1100101-60-40-20020Stewart Platform and Ankle-Tilt Platform Closed Loop ResponsesGain (dB)10-1100101-400-300-200-1000Phase (deg)Frequency (Hz)  Stewart PlatformAnkle-Tilt Platform33  (2.5), the load stiffness curve should remain approximately linear within the 6° limits of the balance simulation (sin(0.1) ≈ 0.1). Consequently, the slope of the load stiffness curve should be 𝑚 ∗ 𝑔 ∗ 𝑙, as per Equation (2.6), regardless of the test condition.  𝑀𝜃[−0.1 < 𝜃 < 0.1𝑟𝑎𝑑] =𝑚𝑔𝑙 sin (𝜃)𝜃≈ 𝑚𝑔𝑙 (2.6) Two performance metrics were extracted from the ‘Load Stiffness’ data: the slope of the best-fit line to the data and the RMS error of this line fit. For the purposes of hypothesis 2.1, the RMS error of the load stiffness line was used as a metric for comparison. The RMS of the ‘Sway’ data for hypothesis 2.2 was calculated after removing the mean standing angle from the trial. The signal was also converted into the frequency domain for analysis and comparison to other reported sway frequency data (Luu et al., 2011)(Loram, Gawthrop, & Lakie, 2006). After extracting the performance metrics from each trial, the results for the two conditions were compared. Statistical analysis was performed using pair-wise t-tests with a significance level of p = 0.05. 2.4.3  Results of Validation with Human Subjects Figure 2-6 shows the load stiffness curves for a single participant. By inspection, the addition of the ankle-tilt platform produced a reduction in the number and amplitude of the “torque loops” (hysteresis) occurring as the participant swayed back and forth on the simulator. A similar result was observed for all participants. This corresponded to a significant decrease in the load stiffness RMS (Figure 2-7) about the best fit line (t(9) = 6.17, p < 0.001), and a significant decrease in the RMS sway trials from 0.625° ± 0.159° to 0.423° ± 0.153° when the ankle-tilt platform was enabled (t(9) = 5.35, p < 0.02), providing support for hypothesis 2.1 and hypothesis 2.2, respectively. As expected, no significant difference in the slope of the load stiffness curves was apparent when the ankle-tilt platform was enabled (t(9) = 0.57, p = 0.29). Similar to quiet standing (Loram et al., 2006), the maximum frequency content (signal attenuated to 1% of maximum) identified in the sway profiles was less than 2.5 Hz for all participants in both conditions. This finding further supports the 0 – 3 Hz performance requirements for the ankle-tilt platform. The maximum angular errors measured for the ankle-tilt platform during the balance trials were 0.048° ± 0.016° and did not exceed 0.07° for any participant. When asked, “What did you think about that condition?” a majority (6/10) of participants reported finding the ankle-tilt platform-locked condition “challenging.” After trying both conditions, all ten participants stated that they found it easier to balance with the ankle-tilt platform engaged. The reasons given were 34  similar: 8 participants reported that they did not have to concentrate as much on the task to retain balance, 5 participants reported finding the system easier to balance, and 3 participants reported that the task required less effort (all participants gave at least one of these responses).  Figure 2-6 – Load stiffness plots for a representative participant, with the ankle-tilt platform locked (left) and engaged (right). Subjects swayed within a comfortable range at 0.1 Hz to approximate static conditions. The slope of the best fit curve was similar across conditions, and only slightly steeper than the expected slope of mgl predicted by an inverted pendulum model, as described in Equation (2.6) (102.2% and 102.3% of mgl, respectively). The reduced deviations from the line (RMS error) with the ankle-tilt platform enabled (right) indicate improved balance control. © 2012 IEEE (Pospisil et al., 2012)  35   Figure 2-7 – RMS of load stiffness results. Individual participant data are shown in grey, and the mean and standard deviation for the population are shown in black. Group mean data for the load stiffness trials (left) and free sway trials (right) both show a significant (p < 0.05) decrease in RMS amplitude with the ankle-tilt platform engaged. © 2012 IEEE (Pospisil et al., 2012) 2.5 Discussion 2.5.1 Technical Performance The technical performance requirements for the ankle-tilt platform have been met in this implementation. Angular position errors of the ankle-tilt platform were four times below reported thresholds of human detection (Refshauge & Fitzpatrick, 1995). While the actuator performed according to specifications, several recommended improvements may offer even better performance for future work. The NI 7354 motion card, while providing a highly usable interface and good integration with LabVIEW, is not well suited for real time motion control. To produce smooth motion, the buffering system was needed, which introduces unnecessary delays in the actuator control. While these delays have been reduced by spline fitting, any predictive controller will introduce error in the motion profile. Instead, the ankle-tilt platform would be better controlled directly through LabVIEW, employing the PID toolkit and producing torque commands through a DAQ. This will allow greater flexibility in the control and tuning of the platform by offering the ability to modify all of the controller parameters. As the eventual goal is to move to a dual-actuator setup, the author recommends that the transition to a LabVIEW-based controller be made at that time. Adding the second actuator should further improve the 36  performance capabilities, as the load will be cut in half and the platform will no longer be over constrained, since each footplate will be supported by two bearings. While there are other platform designs for investigating the effects and perceptions of ankle motion (Mahboobin et al., 2008; Refshauge & Fitzpatrick, 1995), this novel design provides both independent control of the ankle motion and permits whole-body movement. This flexibility offers new research avenues for experimentally manipulating proprioception during balance tasks, which may also be valuable for investigating rehabilitation techniques (Abdollahi et al., 2011).  2.5.2 Validation with Human Subjects  Luu showed that when the ankle joint was constrained to the motion of the simulator, the mean-removed RMS of sway for the balance simulator was 60% greater than the RMS observed in freestanding sway (Luu et al., 2011). With the addition of the ankle-tilt platform in the present study, the RMS of sway for the balance simulator has been reduced by 32%, indicative of improvement in the balance simulation. This improvement is also reflected in the reduction of hysteresis in the load stiffness curves when the ankle-tilt platform is engaged, and a consequent decrease in RMS amplitude. Furthermore, the participants’ perception of the system indicates that the addition of ankle motion reduced the effort and concentration required to maintain quiet stance. All three measures indicate that the ankle-tilt platform’s addition has improved control and fidelity for simulating quiet stance.  The ankle-tilt platform has brought the simulator closer to replicating normal standing; however, there remains areas in which improvements can be made, as evidenced by the remaining discrepancies in sway and load stiffness RMS. These discrepancies are primarily attributed to the system delays. As with any control system, response delays can result in overcorrection, which in this case would result in increased sway movement and decreased stability. Investigation into methods of reducing this control delay is strongly recommended. 2.6 Summary The results of the study strongly support the hypothesis from Luu (Luu et al., 2011) that enabling ankle motion via the ankle-tilt platform improves the robotic balance simulation. While it was possible to simulate passive ankle stiffness and ankle damping without the ankle-tilt platform (the slopes of the load-stiffness curves were the same), the addition of ankle motion improved the balance performance, with a significant reduction in sway, a significant decrease in load stiffness RMS, and all participants reporting this simulation as being easier to control. 37  The RISER platform with the ankle-tilt platform modification is used in the studies described in Chapter 3, where participants are required to maintain balance while blindfolded and receiving galvanic vestibular stimulation, resulting in a heavy reliance on somatosensory feedback and ankle motion; the ankle-tilt platform plays a critical role in helping the participants maintain balance.   38  3 Exploring the Perception and Reflex of Galvanic Vestibular Stimulation Exploring the Perception and Reflex of Galvanic Vestibular Stimulation This chapter details the author’s research into the human vestibular system. Two studies were performed, with the goal of relating the perception and reflex response to GVS during quiet standing to that of a physical motion of the body. To accomplish this, the work in this chapter makes use of the RISER system described in Chapter 1, and the ankle-tilt platform, described and validated in Chapter 2, served a pivotal role in the reflex portion of this study.  3.1 Introduction The previous chapter presented the design and implementation of an independent ankle manipulator for the RISER platform. This chapter describes a series of studies that examine current models of the vestibular system’s role in standing balance. The research employs the newly integrated ankle-tilt platform, described in the previous chapter, to assist participants in maintaining balance on the RISER balance simulator despite external perturbations to their vestibular system. In the two studies described in this chapter, the perceptual and reflexive responses to physical and electrical vestibular stimuli are examined, and the results are compared and contrasted with current models of human balance. 3.1.1 Background As discussed in Chapter 1, the vestibular system plays an important role in maintaining balance, employing the semicircular canals and otolith organs to detect rotations and linear accelerations of the head. Clinical populations with vestibular deficits are subject to vertigo, disorientation, dizziness, and falls (Herdman, Blatt, Schubert, & Tusa, 2000; McCall & Yates, 2011). Current work shows significant progress towards the development of a vestibular prosthesis to assist individuals with a variety of vestibular impairments (Dakin, Elmore, & Rosenberg, 2013). A major goal in this research is developing an accurate model of the vestibular system’s response to a given input. A robust vestibular model will permit a better understanding of the role of the vestibular system in balance and could facilitate the development of prosthesis 39  technologies. Pioneer work in this field was performed when Goldberg studied the vestibular-nerve afferent responses of squirrel monkeys to rotation (Fernández & Goldberg, 1971), linear acceleration (Fernández & Goldberg, 1976), and galvanic stimulation (Goldberg, Fernández, & Smith, 1982). These transfer functions served as the basis for more integrated models of the vestibular system, which serve to predict the visual, perceptual, and/or reflex response to a given vestibular stimulus. Galvanic Vestibular Stimulation (GVS) serves as an important tool in the development and evaluation of these vestibular models on humans. GVS is used to provide an isolated vestibular error signal for study by inducing a simulated vestibular disturbance without requiring physical motion. GVS has provided valuable insight into the role of the vestibular system in a variety of balance systems, as reflex responses to GVS are only observed when the vestibular system is actively involved in maintaining whole-body balance (Fitzpatrick, Burke, et al., 1994). While it was demonstrated by Goldberg that GVS affects vestibular afferents (Goldberg et al., 1982), several important questions remain as to the physiological effects of GVS when it is applied to an intact human vestibular system. In Goldberg’s work, the canal afferents were stimulated directly, while with human participants electrodes are normally placed on the mastoids, resulting in a much broader stimulation area. Consequently, it remains an active subject of discussion whether GVS primarily acts upon the otolith afferents (Cohen et al., 2011), the semicircular canal afferents (Reynolds & Osler, 2012)(Mian, Dakin, Blouin, Fitzpatrick, & Day, 2010), or both (Fitzpatrick & Day, 2004)(Schneider et al., 2000). Further, it is not clear that the afferents are the only system affected by GVS. Aw et al. (Aw, Todd, Aw, Weber, & Halmagyi, 2008) demonstrated that patients with vestibular hair cell death (but otherwise intact vestibular systems) exhibited different ocular reflex responses to GVS from those exhibited by healthy participants, implying that GVS may also act through activation of the basal membrane or hair cells. Understanding which elements of the vestibular system are affected by GVS is an important step in employing the tool for developing improved models of the vestibular system. It is generally accepted that GVS is indistinguishable from physical motion once the signals have arrived on the vestibular afferents, as both forms of stimulation induce excitation (or inhibition) patterns that encode the signal in the firing rates of the afferents (Goldberg et al., 1982). However, it is unclear whether there exists an equivalent physical stimulus that activates the afferents in the same pattern as GVS. Comparing the response to GVS with the response to physical motions can help reveal the mechanisms through which GVS acts on the vestibular system. For example, demonstrating that the effects of GVS can 40  be equated to a rotation about the head would show that GVS is primarily encoded as a disturbance to the semicircular canals. St. George et al. proposed and tested a model of the perceptual response to GVS and to a physical rotation about the head (St George, Day, & Fitzpatrick, 2011). This model (Figure 3.1) was based on current models of semicircular canal dynamics (which GVS should bypass) (Oman, Marcus, & Curthoys, 1987), Goldberg’s work on vestibular afferents (Fernández & Goldberg, 1971; Goldberg et al., 1982), and observations in previous work that GVS induced a perception that was primarily interpreted as a head-centred angular velocity (Fitzpatrick, Marsden, Lord, & Day, 2002). This model proposes that the semicircular canal vestibular afferents are the primary location where GVS activations occur.  Figure 3-1 – Model of the perceptual response to galvanic and physical vestibular stimulations. Produced in Simulink (The MathWorks, Inc. Natick, Massachusetts, USA) based on the model developed by St. George and presented in (St George et al., 2011). I_stim and Omega are the amplitude-time series of galvanic vestibular stimulation (GVS) and physical rotation of the participant, respectively. The model proposes that GVS has a fixed gain (kg) in the transduction step (with no phase dependency); while angular velocity is low pass filtered with time constant tc due to canal-cupula fluid dynamics. According to this model, it should not be possible to differentiate GVS from Angular Velocity past the transduction stage. Consequently, it should be possible to produce an electrical stimulation that exactly mimics a physical rotation in space. St. George’s model does not provide for any frequency-dependent transfer function in the transduction of GVS, instead modelling the afferent response to GVS as a fixed gain with no phase dependency. In Goldberg’s work (Goldberg et al., 1982), the semicircular canal vestibular afferents of deafferented squirrel monkeys were found to respond to direct GVS stimulation with an increasing (though small) gain with frequency, as well as an approximately constant phase lead of 15°-20° across the frequency range (Figure 3-2). Although standard GVS stimulation in humans employs a different method of application (electrodes on the mastoids), the theory that GVS acts primarily through afferent activation means that 41  afferent responses can be expected to exhibit similar leads and gains to those found in Goldberg’s work. Assuming that GVS primarily activates the semicircular canals, it is possible to check if the predictions of the model match the outcome of a direct comparison between GVS and physical motion. Based on model predictions, the response to GVS should have little to no frequency dependence, and the stimulation should be perceived as a pure rotation about the head (as the model assumes that only semicircular canals are activated). To this end, this work explores three research questions:  (3.1) Is there a frequency-dependent transfer function from GVS to the vestibular afferents? (3.2) Does galvanic vestibular stimulation affect the semicircular canals, otolith afferents, or both? (3.3) Can the perception of galvanic vestibular stimulation be related to the reflex response? Two studies involving sinusoidal stimulation are detailed in this chapter. In the first study, participants relate the perceptual response to GVS to that of a physical rotation, providing a direct comparison of the propagation delays of the two stimuli to help answer research question (3.1). In the second study, sinusoidal GVS and rotation are superimposed with the goal of eliminating the vestibular reflex response, which would provide compelling support that St. George’s model is correct in proposing that GVS can be equated to a pure physical rotation (question 3.2). Several pilot studies were performed in advance of these experiments (see Appendix B: Reflex Study Implementation) that assisted in the formation and development of the study procedures.  42   Figure 3-2 – Findings by Goldberg relating the frequency dependant vestibular afferent response to physical and electrical stimulation in a deafferented Squirrel Monkey. 4  From Goldberg: “Comparison of the responses to sinusoidal head rotations (A,C) and to sinusoidally modulated galvanic currents (B,D) for nine semicircular-canal units innervating the superior or horizontal canals. Gain and phase are expressed re peak excitatory head velocity (A,C) or re peak cathodal current (B,D). Included are 3 irregular (●), 2 intermediate (X) and 4 regular (○) units…” (Goldberg et al., 1982). In the model described in this chapter, the phase of the transfer function from sinusoidal angular velocity to firing rates, shown in (C), is subtracted from the perceived phase reported by participants between sinusoidal angular velocity and GVS. The result of this subtraction should represent the phase of the transfer function from sinusoidal GVS to vestibular afferent firing rates, which, the author hypothesized, should be comparable to the phase response shown in (D). 3.2 Procedures In the following sections, the procedures and data analysis techniques for each study are detailed. Results and discussion for these studies are presented in Section 3.3 in the same ordering. Ten participants (M/F = 6/4) took part in the first experiment, and nine participants (M/F = 6/3) took part in the second experiment, with four participants taking part in both studies. Participants were between                                                           4 Reprinted from Brain Research, volume 252(1), pages 156–60, Copyright (1984), Goldberg, J. M., Fernández, C., & Smith, C. E., “Responses of vestibular-nerve afferents in the squirrel monkey to externally applied galvanic currents”, with permission from Elsevier. 43  19 and 56 years old and were able-bodied with no history of balance disorders. The University of British Columbia’s Clinical Research Ethics Board approved all experimental procedures, and all participants provided written informed consent before participating.  Participants in the second experiment had a mass of 68.6 ± 9.1 kg (mean ± standard deviation), and a centre of mass located 0.91 ± 0.04 m above the ankles, relevant for the balance simulation. 3.2.1 Perception Study To study the relationship between sinusoidal motion and sinusoidal GVS in perception, the first experiment was designed to permit participants to adjust the relative phase of the two sinusoids until the sensations matched. The sinusoids were presented orthogonally, with rotation in the AP direction and GVS in the ML direction. In pilot studies, gain was found to be too challenging for participants to match, so only phase was adjusted by the participants; GVS amplitude was set to 2mA, based on the findings from pilot studies. 3.2.1.1 Setup Participants were fitted with 14 cm2 electrodes over the mastoids, and GVS was generated by a Biopac STMISOL (Biopac Systems Inc, Goleta, CA) stimulus isolator. Participants were secured by seatbelts on the RISER (Huryn et al., 2010) platform with the head facing forward and pitched 18 degrees above the horizontal, putting the head in the position which previous literature has found to elicit the maximum ML response while minimizing the response in other planes (Day & Fitzpatrick, 2005). To minimize external sensory input, participants wore a blindfold and earplugs during all trials (Figure 3-3).The experimenter measured the vertical and horizontal displacement from a landmark on the RISER platform to the subject’s mastoids. These values were programmed into the LabVIEW controller, allowing the platform to produce pure sinusoidal rotation in the AP direction with a centre of rotation at the mastoids. Trials consisted of simultaneous input of sinusoidal GVS and a sinusoidal pitch rotation about the inter mastoid-axis. Three trials were performed at each frequency for each participant, and the frequencies tested were 0.05 Hz, 0.1 Hz, 0.2 Hz, 0.4 Hz, 0.8 Hz, 1.6 Hz, and 2.5 Hz, capturing a similar range of frequencies as those covered in Goldberg’s studies (Goldberg et al., 1984). Pilot studies showed that frequencies outside of this range were too difficult to perceive reliably. GVS amplitude was held constant at 2 mA, unless the participant reported pain or an absence of perception (most commonly at 0.05 Hz and 2.5 Hz, respectively), in which case the trial was restarted, and the amplitude was increased or decreased 44  by up to 1 mA. Peak angular velocity was fixed at 1.88°/s for all frequencies, providing the maximum amplitude without exceeding the motion extents of the RISER platform during the 0.05 Hz stimulation.   Figure 3-3 – Experimental setup for vestibular studies. The participant is secured to the RISER balance simulator for both studies. Blindfold and earplugs are used to minimize external disturbances. Inset: For reflex study, subject balances with head turned 90 degrees, to align GVS stimulation with robot motion.  3.2.1.2 Perception of Combined Stimulation Prior to the primary experiment, all subjects completed an initial orientation trial, performed at 0.8 Hz, in which the experimenter adjusted the phase of the rotation in increments of 45°, and asked the participants to describe the motions of their feet. The motions described by all participants were similar, and fall into four major categories (Figure 3-4-left), each 90° of phase apart. As the phase was increased to 360°, participants reported a perception of the same foot motion they described at 0°. These perceived motions are believed to occur through the linear combination of the two sinusoidal signals. With the head pointed forward and the body secured to the platform, the sinusoidal GVS induces a medial-lateral sinusoidal motion perception of the feet, frequently described by study participants as a “swinging motion” that appears to be primarily centred at the head. The physical sinusoidal rotation of 45  the platform is purely in the AP direction, and so does not interact directly with the GVS (Figure 3-4-right). A linear combination model predicts that the perception of foot motion should be the vector sum of the two sinusoids, producing different illusions of foot motion paths depending on the relative phase between the two sinusoids.   Figure 3-4 – Perception of motion induced through superposition of orthogonal GVS and rotation. Left: Perceived motion through combination of sinusoidal galvanic vestibular stimulation (GVS) and sinusoidal vestibular-axis rotation. COM indicates centre of head position (fixed throughout study). Orange arrow shows perceived motion path, as reported by participants. These perceptions are believed to be elicited from the vector sum of the anterior-posterior (rotation) sinusoid perception with the medial-lateral (GVS) sinusoid perception. Right: Sign conventions for stimulation. Positive conventional current induces a perception of foot travel to the right, while negative current corresponds to foot travel to the left. Positive angular velocity corresponds to the feet travelling backwards (nose pitching downwards). 3.2.1.3 Phase Matching Trials After describing and experiencing all of the motions, participants were instructed that, during the trials, they were to find the phase at which their feet moved in a straight line from their back right to their front left, with no circular motion. Participants were provided with a wireless mouse, which permitted them to adjust the phase of the Stewart platform’s sinusoidal motion in increments of 10° by scrolling the mouse wheel. Participants were informed that they could scroll as far as they wished in either direction, but were reminded that 36 clicks of the scroll wheel in either direction would bring them back to where they had 46  started. The GVS sinusoid was assigned a random phase delay of 0-360 degrees, to randomize the relative starting point of the two sinusoids. To encourage freedom of exploration of their perception, participants were permitted to place ‘bookmarks’ at phases they felt were close to the desired perception, so that they could return to them later or compare them directly. Neither the experimenter nor the participant was aware of the ‘absolute’ phase difference (due to the randomized GVS start phase), however the experimenter was able to view the relative phase change from the start position, facilitating bookmark placement and navigation without introducing experimental bias. Once participants selected a phase for a trial, the experimenter asked them for their confidence in the selection on a scale of 1-10, and then concluded the trial. Trials were at most five minutes long; if participants had not selected a phase by the end of the five minutes, they were given a break and asked to re-attempt the trial. Mandatory breaks were taken after each block of seven trials, and the participant could request additional breaks at any time. The testing order was pseudo-randomized for each participant such that the order the frequencies were presented was random, but every frequency was tested once before any were tested a second time. 3.2.1.4 Analysis Vestibular stimuli and angular velocity were collected on an NI PXI-6229 DAQ board (National Instruments, Austin, TX), sampled at 2048 Hz. Angular velocity was sensed using a Systron Donner SDG500-00100-100 (Systron Donner Inertial, Concord, CA) sensor mounted on the foot platform. The current delivered to the vestibular system was sensed and transmitted to the DAQ board using a Tektronix TCPA300 inductive probe (Tektronix inc., Beaverton, OR), with a 3% DC accuracy and 7 ns rise time. Phases were extracted from the signals measured on the DAQ using a least squares sinusoid fit. The absolute phase difference of the participants’ perception was encoded as the phase lead of the GVS sinusoid relative to the lead of the angular velocity sinusoid. Positive GVS corresponded to a positive (anodal) voltage on the right mastoid and cathodal voltage on the left mastoid. A positive GVS produced an illusion of the feet travelling to the right, which is consistent with previous literature (Wardman et al., 2003). A positive angular velocity signal corresponded to the feet travelling in the posterior direction. As participants were asked to select a phase such that they would perceive their feet travelling backwards as they simultaneously perceived them travelling right, a selected absolute phase difference of 0° corresponds to perceiving GVS with no phase lead (or lag) relative to the angular velocity of the physical rotation. 47  Using the assumption that the frequency dependent phases from angular velocity to canal afferents in squirrel monkeys (Figure 3-2) (Goldberg et al., 1982) hold in humans, the phase leads from GVS to the vestibular afferents were inferred by subtracting the phase leads found in Goldberg’s study from the relative phase leads between angular velocity and GVS found in this study. From Goldberg’s work, this is expected to result in a small and generally flat phase lead of the GVS stimulation relative to the vestibular afferent responses across the studied frequency range, as seen in Figure 3-2. 3.2.2 Reflex Response Study The reflex study offers a second comparison between sinusoidal GVS and sinusoidal rotation, examining whether there is a frequency-dependent transfer function from GVS to the vestibular afferents (research question 3.1). In this study, measurements of both phase and gain of the soleus reflex were measured across frequencies in response to both GVS and physical rotation. The relative phases between the two sinusoids provided a reflex-based estimate of the relative delays between GVS and rotation, for comparison with the perception-based estimates from the previous study. The two sinusoids were also presented in combination to cancel the reflex response. Successful cancellation at all frequencies would provide evidence that GVS can be modelled as a head-centred whole-body rotation (research question 3.2). In order to elicit a reflex response to the stimuli, participants needed to be actively engaged in a balance task [Fitzpatrick-94]. Consequently, the experimental setup needed to be capable of delivering sinusoidal rotation about the head, while still permitting the participant to engage in standing balance. The human balance simulation was modified such that sinusoidal head rotations could be superimposed on top of the balance simulation (Figure 3-5). See Appendix B: Reflex Study Implementation, for further implementation details.  As the stimuli were intended to be vestibular in nature, the motions of the head rotations were not applied to the ankle-tilt platform, which meant that the ankle-tilt platform (Pospisil et al., 2012) was sway referenced to only the balance simulation’s orientation. As an additional benefit, this configuration ensured that the ankle-tilt platform was always providing accurate ankle-somatosensory feedback for the current state of the balance simulation, assisting the participant with maintaining balance. 3.2.2.1 Setup For this experiment, due to a longer procedure for each stimulation frequency, only three frequencies were tested: 0.125 Hz, 0.8125 Hz, and 2.5 Hz. Frequencies were selected with the goal of examining the 48  same range as the perception experiments, while placing the sinusoidal EMG measurements at the centre of their respective bins in the Fourier transform. Surface electromyography (EMG) was collected from the right soleus using AG-AgCl surface electrodes (Blue Sensor M; Ambu A/S, Ballerup, Denmark). The skin was cleaned prior to adhesion. Using a 2-cm inter-electrode distance, one electrode pair was positioned along the longitudinal axis over the soleus. The signal was referenced to a ground electrode placed on the right patella. EMG readings were collected using a Grass Telefactor P511 1000X pre-amplifier (Grass Technologies, West Warwick, RI), band pass filtered between 30 Hz and 1 kHz, and digitized at 2048 Hz on the same DAQ board as the angular velocity and galvanic stimulation. The EMG signal was rectified and mean removed in post processing. Participants were secured to the RISER platform in a manner similar to the perception experiment, except that the helmet was rotated 90° to the participant’s left (head pitched back at 18°), causing the GVS to produce an AP reflex response, along the line of action of the soleus muscle (Figure 3-3, inset). Participants rotated their upper bodies approximately 30° to reduce neck strain, and the experimenter placed a pillow behind the right side of their back before securing the participant to the backboard. Feet were in the forward facing position on the ankle tilt platform (the same as the first experiment). Participants wore a blindfold and earplugs during the experiments.  Figure 3-5 – Superposition of balance simulation and head rotations. Left: Simplified model of participant secured on the RISER system. Centre: Balance simulation engaged; participant can control body angle. Ankle-tilt platform keeps feet parallel to floor. Right: Sinusoidal rotation is applied about the head. Ankle-tilt platform does not 49  compensate for head rotation motion. Head rotation and balance simulation motions are superimposed during the reflex study. 3.2.2.2 Calibration After securing the participant, the experimenter started the balance simulation and gave the participant time to adjust to the balance simulation. Once participants stated they were comfortable and in good control of their balance, the experimenter recorded their natural standing angle θ0 (0.029 ± 0.012 rad) for calibration purposes (described below), as well as their baseline EMG (for comparison with the stimulus trials) over a period of 3 minutes. When a participant is rotated about the head, the sinusoidal motion of the Stewart platform induces sinusoidal torques on the forceplates that are not due to muscle activation, but take place at the frequency being studied. To prevent these torques from affecting the balance simulation (and by extension, the reflex response), it is necessary to compensate for these additional torques in real time. After the participant completed the baseline trial, the experimenter ended the balance simulation, pitched the robot (and participant) to θ0, instructed the participant to relax (not try to balance), and then sinusoidally rotated the participant about the head without the balance simulation engaged, at each of the frequencies to be studied. The inertial torques induced on the forceplates were measured for each frequency so that they could be subtracted in real time from future trials. Soleus EMG was monitored to verify relaxation during the calibration. Due to the complicated dynamics of the environment (with the subject standing on the forceplates but secured to the backboard), this method proved more reliable than using a kinematic model to predict and cancel the forces in a feed forward manner (See Appendix B, Reflex Study Implementation). Data collection was broken into two distinct phases of trials: the discovery trials, where only one type of sinusoidal stimulation was applied, and the combination trials, where both stimulations were applied simultaneously to attempt to cancel (or combine) the signals. 3.2.2.3 Discovery Trials In the discovery phase, participants maintained balance on the simulator while the experimenter applied either sinusoidal GVS or a sinusoidal head rotation (but not both) at one of the three frequencies (0.125 Hz, 0.8125 Hz, or 2.5 Hz). Estimates of phase and gain were extracted from the Fourier transform of the EMG data, at the frequency of the stimulation. 50  Trials consisted of 64 s of data collection, with a 5-minute break every 7 trials. Participants were tasked with maintaining balance despite the external perturbations. The head rotation peak velocity was fixed at 1.88°/s for consistency with the perception experiments. The GVS amplitude was initially set to 2mA and then increased or decreased iteratively in subsequent trials in an attempt to match the amplitude of the EMG response to that from the current average of the rotation trial responses at that frequency (order alternated between GVS and rotation). GVS increments and decrements were performed in 0.25 mA or 0.5 mA steps (depending on the relative difference between the reflex responses), not exceeding a maximum stimulation amplitude of 3.5 mA or a minimum of 0.25 mA. When an acceptable match was found (<10% difference between GVS and rotation EMG), two more trials were collected at this amplitude to ensure repeatability. For the determined stimulus amplitude, a minimum of three 64 s trials were collected for each stimulus type at each frequency. If the phase or amplitude of a stimulus response had a large standard deviation across trials, (>20° or >25% for phase or amplitude, respectively), additional trials were performed until the standard deviation dropped into this range, or the maximum of 8 trials for the frequency was reached. 3.2.2.4 Combination Trials After finding a match between the amplitude of the muscle responses evoked by GVS and rotation, additional 64 s trials were performed at each frequency with the sinusoidal head rotation and the matched GVS amplitude applied simultaneously. For each frequency, three ‘cancellation’ trials were performed, where the initial phase of the GVS was set so that the expected GVS reflex response would be 180° of the expected head rotation reflex response. In addition, one ‘additive’ trial was performed, where the GVS phase was set so that the expected reflex responses would align, and ideally combine. In both cases, the combined signal should result in a change in the EMG amplitude of the soleus muscle response at the stimulus frequency. In the cancellation trials, this should result in a soleus response that is indistinguishable from that of balancing without any stimulus, while in the additive trial the soleus response amplitude should double (by a linear assumption). 3.2.2.5 Analysis Two metrics were used to provide indications of the quality of the match found during the discovery trials, specifically, the percentile difference between the two amplitudes, as per Equation (3.1), and the average variance between the estimates of phase for all trials, as per Equation (3.2). Significant amplitude differences or variations in phase affect the expected success of the cancellation and addition. In the 51  following equations, C, B, G, and R represent the average EMG amplitude at the target frequency of the three trials for Cancellation, Baseline, (independent) GVS, and (independent) Rotation. % 𝐴𝑚𝑝𝑙𝑖𝑡𝑢𝑑𝑒 𝐷𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒 = 100 ∗|𝐺 − 𝑅|(𝐺 + 𝑅)/2 (3.1) 𝑃ℎ𝑎𝑠𝑒 𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 = √𝑆𝑡𝑑𝐷𝑒𝑣(𝐺𝑉𝑆 𝑃ℎ𝑎𝑠𝑒2) + 𝑆𝑡𝑑𝐷𝑒𝑣(𝑅𝑜𝑡 𝑃ℎ𝑎𝑠𝑒2) (3.2) To quantify the degree of cancellation, the average amplitude of the EMG reflex during the cancellation trials was compared to the amplitude of the EMG reflex during the independent trials, as per Equation (3.3).  % 𝐶𝑎𝑛𝑐𝑒𝑙𝑙𝑎𝑡𝑖𝑜𝑛 = 100 ∗ (1 −𝐶 − 𝐵max(𝐺, 𝑅) − 𝐵) (3.3) Similarly, the addition trials were accessed using a similar metric, as per Equation (3.4), except that A represents the EMG amplitude at the target frequency of the single addition trial (rather than the average of three trials). % 𝐴𝑑𝑑𝑖𝑡𝑖𝑜𝑛 = 100 ∗ (𝐴 − 𝐵max(𝐺, 𝑅) − 𝐵− 1) (3.4) Linear regression analysis was carried out using SPSS (IBM, Armonk, New York) to examine the linearity of the GVS gain. Qualitative evidence of habituation was observed during pilot studies, so temporal relationships were also analyzed. Linear regression was performed to look for relationships between the number of trials carried out and the amplitude of the response. All regressions were performed with significance of p = 0.05, with outlier rejection at 3 standard deviations. 3.3 Results 3.3.1 Perception Study Participants in the perception study demonstrated a consistent trend of selecting an increasing lead for the GVS sinusoid relative to the angular velocity sinusoid across the frequency range, as shown in Figure 3-6, left. At 0.05 Hz, participants selected a mean of -39.6° phase lead for GVS (i.e., a lead for angular velocity), increasing to a mean of 62.6° at 2.5 Hz. Mean and standard deviations of subject confidence are summarized in Table 3-1. Frequencies with a higher variance of phase lead tended to exhibit a lower subject confidence. This relationship was also supported by qualitative participant feedback, which indicated that the highest and lowest frequencies were very difficult to perceive reliably. 52  After including Goldberg’s model (Fernández & Goldberg, 1971) of velocity-to-afferents phases, the inferred GVS to afferent phases are shown in Figure 3-6, right. Results show a low frequency lead of the vestibular afferents over the GVS stimulation, decreasing to a slight lag by 2.5 Hz (corresponding to approximately 100 ms of delay relative to the 0.05 Hz measurement). For comparison, Goldberg’s work on galvanic stimulation of vestibular afferents (Goldberg et al., 1982) phase leads lie within one standard deviation of the estimates from this this study at all frequencies, but did not exhibit a significant frequency dependency, remaining constant at 15°-20° across frequencies.  Figure 3-6 – (Left) Selected phase leads for galvanic vestibular stimulation (GVS) relative to angular velocity in perception. (Right) inferred phase lead of afferents re: GVS stimulation. The black line illustrates the means and standard deviations for all participants, while the grey lines denote individual results. The dashed line in the right figure is the GVS to afferent phases predicted by Goldberg et al. (Fernández & Goldberg, 1971). Participants reliably chose an increasing phase lead for GVS with increasing frequency, demonstrating either a decreasing delay in the perception of angular velocity with frequency or an increasing delay in the perception of GVS (or some combination thereof). Using Goldberg’s model for canal afferents (Fernández & Goldberg, 1971) the transfer function from GVS stimulation to afferent firing rates was inferred, revealing a frequency dependent phase lead for the afferent response to GVS.      10-210-1100101-150-100-50050100Sinusoid Frequency (Hz)Per Subject Mean Selected Phase Lead of GVS (deg)Selected Phase Leads of GVS Sinusoid Relative to Angular Velocity: Perception StudyInferred GVS to Vestibular Afferent Leads  Sinusoid Frequency (Hz) 10-210-1100101-150-100-50050100150Frequency (Hz)Phase Lead (deg)Inferred GVS to afferent phases  GVS to Afferent Phase LeadsPhase Lead, GVS vs Afferent Response (deg)  53  Table 3-1 – Perception study phase leads and reported confidence. All data are reported as mean ± standard deviation. The phase leads, as shown in Figure 3-1, left, are the relative leads of GVS over angular velocity that participants selected to line up their perception of the two signals. Participant confidences were highest at 0.4 Hz and 0.8 Hz, which corresponded to the lowest standard deviations in phase selection; suggesting that higher standard deviations at the peripheral frequencies may be attributable to participants having difficulty in reliably perceiving the stimulations. Frequency Phase Lead (0) Subject Confidence 0.05 Hz -39.6 ± 56.4 6.0 ± 1.5 0.1 Hz -24.9 ± 46.4 6.9 ± 1.0 0.2 Hz -11.1 ± 40.9 7.2 ± 1.2 0.4 Hz 10.8 ± 28.5 7.8 ± 1.0 0.8 Hz 26.1 ± 24.9 7.9 ± 1.1 1.6 Hz 37.4 ± 34.8 7.4 ± 1.4 2.5 Hz 62.6 ± 37.7 6.0 ± 1.3  3.3.2 Reflex Study All participants exhibited responses to both stimuli for all frequencies, with the most consistency both across and within subjects observed at 0.8125 Hz. For all participants, soleus reflex responses could be reliably amplified or attenuated by presenting both sinusoidal GVS and sinusoidal rotation simultaneously during balance simulation. Figure 3-7 demonstrates EMG results for a participant exhibiting complete cancellation (smaller than background EMG) of the reflex response at 0.8125 Hz. In general, only partial cancellation was achieved, with the greatest success at 0.8125 Hz, where 84% attenuation of the reflex response was observed. Table 3-2 summarizes the results and quality metrics for all subjects. Quality was very high at 0.8125 Hz, with an average of 11% variation between amplitude estimates and 19° between phase estimates. 0.125 Hz exhibited very high phase variance (33°), while 2.5 Hz exhibited a high variance in amplitudes between trials (30.6%), which likely contributed to the respectively poorer cancellations at these frequencies, with both frequencies exhibiting approximately 56% cancellation on average. Required GVS gain to match a fixed angular velocity was shown to increase with frequency (Figure 3-8), consistent with expectations from Goldberg et al. (Goldberg et al., 1982), as shown in Figure 3-2 (A) and (B). Phases at 0.05 and 0.8125 Hz were consistent with the results from the perception study; however, the phase at 2.5 Hz dropped instead of continuing to increase as expected from perception results.  54   Figure 3-7 – Fast Fourier transform results of soleus muscle EMG for a participant exhibiting complete cancellation at 0.8125 Hz. The participant exhibited an EMG reflex response at the stimulus frequency (0.8125 Hz) that was smaller for the cancellation trials (where participants balanced while receiving both GVS and a head rotation) than for the baseline trials (where participants balanced without external stimulation). Such results demonstrate that it is possible to combine sinusoidal GVS with sinusoidal rotation to produce a net reflex response that is larger (addition trial) or smaller (cancellation trials) than the response evoked from the presentation of either stimuli independently, and that the increase or decrease is governed by the relative phase between the two stimuli. All plots are the average result of three trials except for the addition, of which only one trial was performed.  Table 3-2 – Reflex study cancellation, addition, and quality metrics. When attempting to match the reflex responses for GVS to angular velocity, the quality of the matches was typically highest at 0.8125 Hz. Correspondingly, 0.8125 Hz demonstrated the most cancellation, with an average of 84 % of the signal eliminated in the cancellation trials. 0.125 Hz exhibited an unexpectedly high average increase in the addition EMG response (157 %), a result that is not consistent with a linear combination model. Frequency 0.125 Hz 0.8125 Hz 2.5 Hz GVS Amplitude (mA) 0.7 ± 0.3 1.3 ± 0.2 3.3 ± 0.4 % Amplitude Difference 13.6 ± 8.6 11.3 ± 11.1 30.6 ± 30.6 Phase Variance (deg) 32.6 ± 16.5 19.3 ± 2.3 16.2 ± 4.8 % Cancellation 56.4 ± 17.8 84.1 ± 22.9 56.6 ± 13.1 % Addition 156.8 ± 77.1 38.7 ± 62 10.4 ± 25.5 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 212345x 10-3Frequency (Hz)EMG FFT Amplitude (V)0.8125Hz EMG Results, Subject 1 (full cancellation)  Baseline EMGCancellationRotation OnlyGVS OnlyAddition55     Figure 3-8 – Computed gains and phase leads for galvanic vestibular stimulation (GVS) relative to angular velocity in reflexes. The black line illustrates the means and standard deviations for all participants, while the grey lines denote individual results. The required GVS amplitude to match the soleus reflex response to the response from angular velocity was found to increase with frequency, consistent with expectations. GVS and angular velocity leads were measured with respect to the observed EMG responses, and the relative phase leads were computed from these measures. 10-210-110010100.511.52Computed GVS gain relative to angular velocity in reflex studyFrequency (Hz)GVS gain re: Angular Velocity (mA/(deg/s))10-210-1100101-150-100-5050100150Selected Phase Leads of GVS Sinusoid Relative to Angular Velocity: Reflex StudyFrequency (Hz)Phase Lead (deg)56  Reflex responses also exhibited strong evidence of habituation at higher frequencies. Figure 3-9 shows the mean and standard deviation of the EMG response amplitudes with respect to trial number. The downward trends indicate a reduction in response amplitude over repeated exposures. A statistically significant relationship between trial number and EMG amplitude was observed for both rotation and GVS at 2.5 Hz and 0.8125 Hz. GVS amplitudes did change between trials, in an attempt to match the GVS response with the rotation response, however, controlling for this change actually increased the strength of the negative correlation. This result is contrary to most models and predictions (Goldberg et al., 1984; St George et al., 2011), but provides support for the observations of sinusoidal habituation in a study by Balter (Balter et al., 2004).   Figure 3-9 – EMG reflex response amplitudes grouped by trial number, with corresponding correlation coefficients. EMG responses show a statistically significant decrease in amplitude for GVS and Rotation stimulation in the 2.5 Hz and 0.8125 Hz trials. This decrease suggests that habituation may be taking place over the course of the study. Balter et al. have previously identified habituation taking place at similar frequencies (0.5 Hz) and stimulation amplitudes (2 mA) (Balter et al., 2004) 0 2 4 6 800.511.5x 10-32.5Hz GVS0 2 4 6 8024x 10-30.8125Hz GVSEMG Gain (normalized mV)0 2 4 6 802468x 10-30.125Hz GVSTrial Number0 2 4 6 8-505x 10-30.125Hz RotationTrial Number0 2 4 60246x 10-30.8125Hz RotationEMG Gain (mV)0 2 4 60246x 10-32.5Hz RotationEMG Reflex Amplitudes Grouped by Trial Number   p<0.001           p<0.01    p<0.001           p=0.025    p=0.573           p=0.255 57  3.4 Discussion The purpose of this study was to examine current models of GVS activation, and compare the predicted delays to those found in perception and reflex. While perceptual feedback and reflex cancellation results support the hypothesis that GVS activates the semicircular canal afferents, phase and gain results from both studies suggest that the GVS transfer function is more complicated that the afferent activation pattern presented by Goldberg (Goldberg et al., 1982).  Participants in the perception study reliably selected an increasing lead for the GVS sinusoid relative to the angular velocity sinusoid. As mentioned, all resulting GVS-to-afferent leads are within one standard deviation of those identified by Goldberg (Goldberg et al., 1982), however the consistency of the trend suggests that the phase lead of GVS transduction is actually decreasing with frequency (research question 3.1), and not remaining constant as Goldberg’s results would suggest. Unlike Goldberg’s work, however, this work examines intact vestibular systems, which, as shown by Highstein (Highstein, Rabbitt, & Boyle, 1996), contain both angular velocity and angular acceleration sensitive afferents, as well as other electrically sensitive physiology such as the efferent nerves and vestibular hair cells. It is possible that, in addition to the vestibular afferents, the GVS stimulation is also activating additional locations in the vestibular system; this is supported by Aw’s findings that vestibular hair cell death altered the ocular response to GVS (Aw et al., 2008). While it remains likely that the vestibular afferents directly respond to GVS, it is plausible that hair cells, the basal membrane, or even vestibular efferents may be affected as well, resulting in a cumulative response on the afferents that is different from that predicted by Goldberg’s work. To explore this possibility, further experimentation is recommended. Direct measurement of afferent responses in an intact vestibular system would be ideal; however this remains impractical due to the inaccessibility of the vestibular nerve. Slightly modified experiments similar to the perception study described in this chapter may help to further elucidate the dynamics of the GVS response. One example would be having participants lie on a large turntable with their head positioned at the centre of rotation. Participants could then be asked to modify the phase and gain of a GVS sinusoid until they perceived cancellation of a sinusoidal motion of the turntable. This would eliminate gravity as a confounding factor, as well as a significant amount of somatosensory feedback. In the reflex study, the 0.8125 Hz trials demonstrate that it is possible to cancel the majority of the soleus reflex to sinusoidal GVS using a sinusoidal head rotation. The remaining 16% of the signal can be primarily 58  attributed to issues with matching the amplitude and phase perfectly; the mean amplitude difference between the stimuli was 11% on average, which on its own can account for the majority of the remaining sinusoid. The mean phase variance between trials was 19° at this frequency, where even a 10° error between the two input sinusoids would generate a 17% error when adding the two sinusoids linearly. This result, along with the perceptual responses indicating that the sinusoidal GVS produced an illusion of a “swinging” motion at the feet (not the head), appears to eliminate the possibility that the otolith organs’ afferents are producing an illusion of linear acceleration at 0.8125 Hz (research question 3.2). As a much larger proportion of the sinusoidal reflex remained at the other two frequencies, it is not possible to draw conclusions about the presence or absence of an otolith response; however, it is also quite possible that the remaining reflex is a result of the larger amplitude and phase variances at these frequencies. Nonetheless, the cancellations and additions remained clear indicators that it was possible to at least partially cancel the reflex responses. This serves to help validate the original phase and gain measurements, as they reliably produced addition and cancellation across all participants. There is a noteworthy difference between the relative phase leads at 2.5 Hz for the perception and reflex studies (research question 3.3). In theory, the phase leads should be the same for both studies, as the afferent signals for GVS and angular velocity should be indistinguishable. This result may suggest that different afferents are involved in generating a perception than those involved in the reflex response (resulting in different phases and gains for GVS). However, it is also possible that at high frequencies, the larger accelerations of the Stewart platform provide proprioceptive feedback and/or interfere with the balance simulation, resulting in a phase shifted input signal for the velocity sinusoids. Further investigation into comparisons between perception and reflex would be required to determine if any significant differences exist. Finally, the observations of habituation to sinusoidal stimulation are potentially problematic for related research. While long-term habituation to sinusoidal body rotations can be easily observed (such as after an extended period at sea), the majority of research into GVS has shown no evidence of habituation (St George & Fitzpatrick, 2011). It must be noted, however, that sinusoidal GVS is less common, and sway habituation to sinusoidal stimulation was observed by Balter for GVS sinusoids (Balter et al., 2004). It is possible that human subjects are able to habituate to purely sinusoidal stimulations, attenuating their response over extended periods. Additional testing is recommended to verify these observations, which can be accomplished by having participants stand and receive sinusoidal GVS for extended periods. 59  Modelling the exact function of GVS remains a challenging task. Results demonstrate that not only is the GVS response frequency dependent, but that it can be time dependent as well. These studies show that GVS activation does not occur solely in the semicircular canal afferents, but that it is still possible to relate it to a physical motion. Further work in the field should focus on developing a better understanding of the activation patterns of GVS, without assuming that the transfer functions provided by Goldberg tell the entire story. Taking another critical look at the underlying assumptions of modern vestibular models may provide new insights in the quest to improve the model of human balance. 3.5 Limitations While consistency was typically high across participants, it is possible that systematic concerns could modify the outcomes and be responsible for some of the observed phenomenon.  While the balance simulator has been shown to replicate human standing, the motion of the Stewart platform is fundamentally less smooth than natural sway. There are vibrations present that are easily discernable by the subject. The vibrations are not correlated with a specific part of the movement, and consequently should not introduce a bias into the results. These vibrations, however, may be responsible in part for the relatively high standard deviations observed in the perception studies, as they have the potential to disrupt the sinusoid and make it more difficult to perceive. Another related source of feedback for the participants is the sound of the Stewart platform’s motion. During platform rotation, there is an audible sound produced by the platform that changes in pitch over the sinusoidal motion. This information could affect the participants’ perception of motion. Earplugs were used to help mitigate this issue, however one participant still reported being able to detect the noise. While steps (such as the helmet and padded backboard) have been taken to mitigate the effects of somatosensory feedback, the simple act of moving the participants through space will necessarily generate forces on their body (including a changing gravity vector), which can be detected through somatosensory mechanisms. This information could be used in conjunction with the vestibular information to produce a different perception and response to the physical rotations. Finally, Fitzpatrick and McCloskey identified that there is a certain amplitude threshold of as much as 0.01 rad (0.6°) below which vestibular signals cannot be reliably perceived (Fitzpatrick & McCloskey, 1994). This phenomenon could translate to a period of no perception that takes place at every zero crossing of vestibular input or GVS. While below the perceptual threshold, it would be difficult for the subject to detect exactly where the zero crossing occurs. This phenomenon may have been responsible for the larger 60  standard deviations on phase estimates at the lower frequencies, as the signal spends a longer period below this threshold. Further, it is possible that this threshold could affect the perception of phase. It is recommended in future studies employing sinusoidal stimulation that a step be applied in the stimulation when crossing the zero threshold, which should result in the perception of a more pure sinusoid. 3.6 Summary This chapter described a series of experiments designed to help further the understanding of the human vestibular system. Perceptual and reflexive responses evoked by sinusoidal galvanic vestibular stimulation and sinusoidal body rotations were compared. Results demonstrated that it was possible to relate GVS to a physical rotation about the head in reflex, however the relationship remains complicated, and additional studies are proposed to further examine the activation patterns of GVS.  61  4 Conclusions and Future Work Conclusions and Future Work Chapter 4 summarizes the major findings and contributions of this thesis. This chapter also identifies a number of new research questions raised by this work, which are presented along with proposed experiments for further investigation. 4.1 Overall Contributions This thesis offers several contributions to the research community. The most notable of these contributions are: 1) The design and validation of a robotic ankle-tilt platform to provide passive ankle torque and ankle-somatosensory feedback in the RISER balance simulation, and characterization of the importance of this ankle information in maintaining standing balance. 2) Demonstration that it is possible to cancel a galvanic vestibular stimulation (GVS) reflex with a physical rotation (and vice versa). 3) The development of new quantitative methods to directly compare the frequency responses of physical rotation to GVS in both perception and reflex. Chapter 2 explored the role of ankle motion in maintaining standing balance. The chapter described the development, implementation, and validation of a new ankle manipulator to permit the modification of ankle motion feedback during balance simulations. Using this new ‘ankle-tilt’ platform, the author performed a study that demonstrated a statistically significant improvement in several balance control metrics for participants when the platform was enabled. The development of the ankle-tilt platform was primarily intended to help examine research question 1: Q1: What role does ankle angle play in regulating quiet stance? In Chapter 3, the author examined the means by which motion is sensed by the human vestibular system, and studied the effects of GVS on reflex and perception. This was accomplished by superimposing physical 62  rotations with GVS, and attempting to quantify the frequency-dependant relationship between the two through novel reflex and perception experiments. In this chapter, the goal was to explore research questions 2 and 3: Q2: Does galvanic vestibular stimulation affect the semicircular canals, otolith afferents, or both? Q3: Can the perception of galvanic vestibular stimulation be related to the reflex response? This work has provided insight into these three research questions through the three corresponding contributions listed above. The following subsections elaborate on each of the contributions and their significance to the research community. 4.1.1 Contribution 1: Ankle-Tilt Platform The ankle-tilt platform was employed to provide a direct comparison between balance with and without corresponding ankle motion. Using the RISER balance simulation, study participants maintained balance in two test conditions: C1, where the ankles remained at a fixed angle to the rest of the body, and C2, where the ankle-tilt platform kept the feet parallel to the floor regardless of body angle.  The results of the research showed a statistically significant reduction in sway RMS when the ankle-tilt platform was enabled (C2), as well as a significant reduction in RMS about the load stiffness line. In Luu’s work on the RISER system, prior to ankle-tilt platform implementation, it was identified that sway RMS during normal standing balance was 40-50% less than that observed when balancing on RISER (Luu et al., 2011). The implementation of the ankle-tilt platform reduced RMS sway on RISER by a mean of 32% suggesting a level of balance control more closely resembling that of normal stance. The load stiffness RMS reduction also demonstrates increased platform control; participants demonstrated a reduced variability in ankle-torque output during deliberate sway, suggesting they were making fewer errors in predicting the appropriate amount of torque for a given standing angle. Taken together, the results demonstrate that ankle motion during standing balance on the platform significantly improves stability. Other literature has supported the importance of ankle feedback in maintaining standing balance (Diener et al., 1984; Fitzpatrick, Rogers, et al., 1994). This work provides further evidence through a direct comparison of balance with and without ankle motion. Additionally, when queried, participants indicated that they found balancing to be easier with the modified system, noting that maintaining balance on the simulator required less active mental concentration. This work strongly supports literature suggesting that the balance stability offered by ankle motion is passive and/or highly reflexive in nature, and assists 63  humans in maintaining standing with minimal attention. Lack of accurate ankle feedback during quiet stance correlates to a significant increase in mental effort required to maintain balance. 4.1.2 Contribution 2: Sinusoidal GVS-Rotation Cancellation Chapter 3 provided a new method of comparing the reflex response to sinusoidal GVS to that of a sinusoidal head-centred whole-body rotation. These sinusoidal stimuli were presented to the subject separately, and the phase and gain from stimulation to soleus EMG reflex were measured. Then, both stimuli were presented simultaneously, with the gains and phases of the two signals designed to either attenuate or amplify the reflex response, based on the measurements from the independent trials. This study showed that across a range of frequencies it was possible to attenuate and amplify the reflex response in this manner, which in turn provided evidence to support the phases and gains that were measured in the independent trials. These findings, while promising, were not conclusive. They do, however, provide a strong motivation and direction for further study. While it appears that a statistically significant reflex cancellation occurred at all frequencies, the differences in the experimental results from what would be expected from the literature cannot be ignored. The change in phase leads across frequencies did not align with those expected from Goldberg’s work with squirrel monkeys (Goldberg et al., 1984), and the sinusoidal signals did not appear to combine in a fully linear manner for addition and cancellation, despite literature suggesting they should (Fernandez and Goldberg 1971). Many of the experiments in this thesis were designed with the hypothesis that the linear time invariant assumption would hold for sinusoidal stimulation, and were not specifically designed to test the validity of this assumption.  Also of concern were observations of habituation at higher frequencies and over longer trials. This runs contradictory to the findings by St. George, which specifically refuted perceptual habituation to vestibular stimulation (St George et al., 2011), however findings by Balter support these findings of GVS habituation (Balter et al., 2004). Further work is needed to conclusively detect or exclude this phenomenon and to characterize its temporal effects on perception and reflex, as it may have a significant effect on any studies involving sustained GVS stimulation. Despite these limitations, the success in using GVS to both attenuate and amplify the reflex response to a physical movement provides a new research vector into studying the relationship between GVS and the vestibular system. By superimposing different types of motions, such as accelerations or rotation centred at a different location on the body, it may be possible to compare the success in cancellation with GVS. 64  Improved cancellation through more complex signals would suggest that GVS affects the otolith afferents as well, while a stronger cancellation for a pure rotation about the head would provide strong evidence that GVS is interpreted by the nervous system as a signal originating in the semicircular canals. 4.1.3 Contribution 3: Comparison of GVS Response in Perception and Reflex Chapter 3 also demonstrated a novel method of comparing the perceptual response to sinusoidal GVS to that of a sinusoidal head-centred whole-body rotation. By presenting the two stimuli simultaneously, but in orthogonal planes, the participants were able to directly compare the sensations, and select a relative phase that caused the two movements to ‘align’ in their perception. This provided a quantitative measure of the perceptual phase difference between GVS and rotation, and produced highly repeatable results across participants. The results for phase shifts from the perception and reflex studies reported in Chapter 3 support the conclusion that perception of GVS and the corresponding reflex response are closely related. While gain was not directly modified in the perception study, the gains used (as determined from the pilot study, see Appendix C) were similar to those determined in the reflex study. The phase leads from GVS to perception or reflex were similar at the lower frequencies, and diverged at high frequencies. Understanding the relationship between perception and reflex response is an important step in understanding vestibular stimulation. As discussed in the introduction, researchers have examined reflex responses, perception, and eye movements, and each body of work seems to yield different conclusions as to the function and effects of GVS. This work has demonstrated, however, that it is possible to produce a reliable measurement of both the perception of and reflex to sinusoidal GVS, opening the door to further studies into establishing the relations between responses. The differences in the relationships may be due to differences in the tasks that were presented during the two different trials, or due to underlying physiological differences in transduction that have not yet been discovered. Either way, the author anticipates that the findings from this work will serve as a starting point for further research in the field. 4.2 Future Work 4.2.1 Ankle-Tilt Platform Research Opportunities One of the many research opportunities inspired by the capabilities of the ankle-tilt platform is research into stroke rehabilitation (Abdollahi et al., 2011), in which modifying the amount of feedback given to a patient has been found to assist in recovery. As the ankle-tilt platform is capable of providing varying 65  levels of feedback, it may be possible to use the platform to help retrain sensitivity to somatosensory feedback for some clinical populations with balance impairment. Another avenue for exploration is the role of proprioceptive information in perception of the vertical. By rotating the ankle-tilt platform or RISER separately, it may be possible to discern how reliant individuals are on their somatosensory system for determining their standing angle. An example experiment would be rotating participants forward on RISER, but rotating the ankle-tilt platform as if they had been rotated backwards. Without vision, this experiment will help demonstrate which senses play a dominant role in the participant’s report of their standing angle. In a similar vein, the ankle-tilt platform can be used to examine how much of the somatosensory balance response is automatic. It is possible to reverse the balance simulation in such a way that plantarflexion causes the platform to pitch forward, while dorsiflexion causes the platform to pitch backwards. Without proprioceptive information, this simulation is very difficult to balance, as has now been shown in recent research involving the ankle-tilt platform (Dalton et al., 2014). Interestingly, this research has shown that in this condition, programming the ankle-tilt platform to remain parallel to the floor actually increases the difficulty. In this case, the natural reflexive balance response is working against the participant, as the reflex is opposite of the necessary muscle activation. This and related research into this phenomenon may help explain how automatic the ankle-somatosensory balance reflex is, and whether it is possible to suppress this reflex when it is not beneficial to the current task. 4.2.2 GVS Reflex and Perception Further work is needed to investigate the perception and reflex response to GVS. Several important questions have been raised. First, in order to better characterize the differences and similarities between perception and reflex response, additional data must be collected at other frequencies. At present, it is difficult to draw strong conclusions from the reflex data, while the relative perception data appear reliable and repeatable. Collecting additional data at 0.25Hz, 1.25Hz, and 1.6Hz will help to fill in some of the gaps in the data to better pinpoint where the results being to diverge. This thesis has demonstrated that further research into the human transfer function for GVS is necessary to build a better model of the vestibular system. Finding a method of extracting the phase of both GVS and rotation relative to an external signal could provide further insight into which transfer function does not align with Goldberg’s findings (see Appendix C.3). A study involving longer-term exposure to sinusoidal GVS of a variety of frequencies will help to examine any short-term habituation. 66  Many of the studies involving GVS (including this one), assume an open-loop model of input. This discounts any possible effects of vestibular efferents, which have remained a peripheral focus of attention over the years (Goldberg, 1991; Highstein & Baker, 1985; Holt, Lysakowski, & Goldberg, 2011). It is possible that, especially for perception, the participants are ‘tuning’ their perception of phase to bring all of their senses in line. As an example, after several rotations at the same sinusoid, it is possible that participants will begin predicting their movement better, thus reducing or eliminating any phase lag in their perception of rotation. Several studies examining visual responses to GVS (Larsby, Hyden, & Odkvist, 1984)(Rude & Baker, 1988) show close to zero phase lag for sinusoidal rotations in the frequency range studied. It is unclear whether such feedback systems could have any impact on GVS responses and perceptions, but it is clear that they have the potential to modify the relative phase leads and lags. 4.3  Summary This thesis has presented original research into the significance and function of the ankle-somatosensory and vestibular system in maintaining standing balance. These questions were approached from a controls perspective in order to provide insight into the functional relationship between vestibular stimulation, physical motion, and reflex response.  A robotic platform was employed and improved upon to assist in the study of balance, and the improvements were validated from both a technical and functional perspective. Through these improvements, the author demonstrated the importance of ankle feedback in reducing attention requirements and maintaining stable balance. Through the use of the robotic platform, the author has demonstrated that it is possible to cancel out reflexive responses to physical rotations using GVS, but identified that additional work is necessary to fully characterize the effects of GVS. In this thesis, the author employed a variety of techniques to examine the human balance system and performed a set of studies providing additional insight into the role of proprioception and the vestibular system in maintaining standing balance in healthy humans. It is the author’s hope that these contributions will benefit the research community in helping to further our understanding of the capabilities and limitations of the human balance system.  67  Refere nce s  References Abdollahi, F., Rozario, S. V, Kenyon, R. V, Patton, J. L., Case, E., Kovic, M., & Listenberger, M. (2011). Arm control recovery enhanced by error augmentation. In IEEE International Conference on Rehabilitation Robotics: [proceedings] (Vol. 2011). doi:10.1109/ICORR.2011.5975504 Adkin, A. L., Bloem, B. R., & Allum, J. H. J. (2005). Trunk sway measurements during stance and gait tasks in Parkinson’s disease. Gait & Posture, 22(3), 240–249. doi:10.1016/j.gaitpost.2004.09.009 af Klint, R., Mazzaro, N., Nielsen, J. B., Sinkjaer, T., & Grey, M. J. (2010). Load rather than length sensitive feedback contributes to soleus muscle activity during human treadmill walking. Journal of Neurophysiology, 103(5), 2747–56. doi:10.1152/jn.00547.2009 Allum, J. H. J., & Pfaltz, C. (1985). Visual and vestibular contributions to pitch sway stabilization in the ankle muscles of normals and patients with bilateral peripheral vestibular deficits. Experimental Brain Research, 58, 82–94. Retrieved from http://link.springer.com/article/10.1007/BF00238956 Aw, S. T., Todd, M. J., Aw, G. E., Weber, K. P., & Halmagyi, G. M. (2008). Gentamicin vestibulotoxicity impairs human electrically evoked vestibulo-ocular reflex. Neurology, 71(22), 1776–82. doi:10.1212/01.wnl.0000335971.43443.d9 Balter, S. G. T., Stokroos, R. J., Eterman, R. M. A., Paredis, S. A. B., Orbons, J., & Kingma, H. (2004). Habituation to galvanic vestibular stimulation. Acta Otolaryngol., 124(8), 941–5. doi:10.1080/00016480410017350 Bell, I., Groves, J., Lanfranchi, P., Nyuli, C., & Tomsett, C. (2010). MECH 458 Tilt Platform - Final Report (p. 130). Bergin, P. S., Bronstein, A. M., Murray, N. M., Sancovic, S., & Zeppenfeld, D. K. (1995). Body sway and vibration perception thresholds in normal aging and in patients with polyneuropathy. Journal of Neurology, Neurosurgery & Psychiatry, 58(3), 335–340. doi:10.1136/jnnp.58.3.335 Boone, D. C., & Azen, S. P. (1979). Normal range of motion of joints in male subjects. The Journal of Bone and Joint Surgery. American Volume, 61(5), 756–9. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/457719 Boor, C. D. (1978). A Practical Guide to Splines. NY: Springer. Retrieved from http://www.researchgate.net/publication/200744645_A_Practical_Guide_to_Splines/file/72e7e51700ff609454.pdf Buchanan, J. J., & Horak, F. B. (1999). Emergence of postural patterns as a function of vision and translation frequency. J Neurophysiol, 81(5), 2325–2339. Retrieved from http://jn.physiology.org/content/81/5/2325.short 68  Carpenter, M. G., Allum, J. H. J., & Honegger, F. (1999). Directional sensitivity of stretch reflexes and balance corrections for normal subjects in the roll and pitch planes. Experimental Brain Research, 129(1), 93–113. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10550507 Cathers, I., Day, B. L., & Fitzpatrick, R. C. (2005). Otolith and canal reflexes in human standing. The Journal of Physiology, 563(Pt 1), 229–34. doi:10.1113/jphysiol.2004.079525 Cohen, B., Yakushin, S. B., & Holstein, G. R. (2011). What does galvanic vestibular stimulation actually activate? Frontiers in Neurology, 2(January), 90. doi:10.3389/fneur.2011.00090 Colclasure, J. C., & Holt, J. R. (2003). Transduction and adaptation in sensory hair cells of the mammalian vestibular system. Gravitational and Space Biology Bulletin : Publication of the American Society for Gravitational and Space Biology, 16(2), 61–70. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/12959133 Curthoys, I. S., & Macdougall, H. G. (2012). What galvanic vestibular stimulation actually activates. Frontiers in Neurology, 3(July), 117. doi:10.3389/fneur.2012.00117 Dakin, C. J., Elmore, L. C., & Rosenberg, A. (2013). One step closer to a functional vestibular prosthesis. The Journal of Neuroscience : The Official Journal of the Society for Neuroscience, 33(38), 14978–80. doi:10.1523/JNEUROSCI.3221-13.2013 Dakin, C. J., Luu, B. L., van den Doel, K., Inglis, J. T., & Blouin, J.-S. (2010). Frequency-specific modulation of vestibular-evoked sway responses in humans. Journal of Neurophysiology, 103(2), 1048–56. doi:10.1152/jn.00881.2009 Dalton, B. H., Inglis, J. T., Koehle, M. S., Van der Loos, H. F. M., Croft, E. A., & Blouin, J.-S. (2014). Does ankle torque/ankle angle relationship facilitate the on-going modulation of upright standing balance? In 43rd annual meeting of the Society for Neuroscience (p. 1). Danna-Dos-Santos, A., Degani, A. M., Zatsiorsky, V. M., & Latash, M. L. (2008). Is voluntary control of natural postural sway possible? Journal of Motor Behavior, 40(3), 179–85. doi:10.3200/JMBR.40.3.179-185 Day, B. L., & Fitzpatrick, R. C. (2005). Virtual head rotation reveals a process of route reconstruction from human vestibular signals. The Journal of Physiology, 567(Pt 2), 591–7. doi:10.1113/jphysiol.2005.092544 Day, B. L., Marsden, J. F., Ramsay, E., Mian, O. S., & Fitzpatrick, R. C. (2010). Non-linear vector summation of left and right vestibular signals for human balance. The Journal of Physiology, 588(Pt 4), 671–82. doi:10.1113/jphysiol.2009.181768 Di Giulio, I., Maganaris, C. N., Baltzopoulos, V., & Loram, I. D. (2009). The proprioceptive and agonist roles of gastrocnemius, soleus and tibialis anterior muscles in maintaining human upright posture. The Journal of Physiology, 587(Pt 10), 2399–416. doi:10.1113/jphysiol.2009.168690 69  Diener, H. C., Dichgans, J., Guschlbauer, B., & Mau, H. (1984). The significance of proprioception on postural stabilization as assessed by ischemia. Brain Research, 296(1), 103–9. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/6713202 Fernández, C., & Goldberg, J. M. (1971). Physiology of peripheral neurons innervating semicircular canals of the squirrel monkey. II. Response to sinusoidal stimulation and dynamics of peripheral vestibular system. Journal of Neurophysiology, 34(4), 661–75. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/5000363 Fernández, C., & Goldberg, J. M. (1976). Physiology of peripheral neurons innervating otolith organs of the squirrel monkey. III. Response dynamics. Journal of Neurophysiology, 39(5), 996–1008. Retrieved from http://jn.physiology.org/content/39/5/996.short Feuerbach, J. W., Grabiner, M. D., Koh, T. J., & Weiker, G. G. (1994). Effect of an Ankle Orthosis and Ankle Ligament Anesthesia on Ankle Joint Proprioception. The American Journal of Sports Medicine, 22(2), 223–229. doi:10.1177/036354659402200212 Fitzpatrick, R. C., Burke, D., & Gandevia, S. C. (1994). Task-dependent reflex responses and movement illusions evoked by galvanic vestibular stimulation in standing humans. The Journal of Physiology, 478 ( Pt 2, 363–72. Retrieved from http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1155693&tool=pmcentrez&rendertype=abstract Fitzpatrick, R. C., & Day, B. L. (2004). Probing the human vestibular system with galvanic stimulation. Journal of Applied Physiology (Bethesda, Md. : 1985), 96(6), 2301–16. doi:10.1152/japplphysiol.00008.2004 Fitzpatrick, R. C., Marsden, J., Lord, S. R., & Day, B. L. (2002). Galvanic vestibular stimulation evokes sensations of body rotation. Neuroreport, 13(18), 2379–2383. doi:10.1097/01.wnr.0000048002.96487 Fitzpatrick, R. C., & McCloskey, D. I. (1994). Proprioceptive, visual and vestibular thresholds for the perception of sway during standing in humans. The Journal of Physiology, 478 ( Pt 1, 173–86. Retrieved from http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1155655&tool=pmcentrez&rendertype=abstract Fitzpatrick, R. C., Rogers, D. K., & Mccloskey, D. I. (1994). Stable human standing with lower-limb muscle afferents providing the only sensory input. The Journal of Physiology, 480(2), 395–403. Retrieved from http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1155855&tool=pmcentrez&rendertype=abstract Fitzpatrick, R. C., Taylor, J. L., & Mccloskey, D. I. (1992). Ankle stiffness of standing humans in response to imperceptible perturbation: reflex and task-dependent components. The Journal of Physiology, (1992), 533–547. Retrieved from http://jp.physoc.org/content/454/1/533.short 70  Gage, W. H., Winter, D. A., Frank, J. S., & Adkin, A. L. (2004). Kinematic and kinetic validity of the inverted pendulum model in quiet standing. Gait & Posture, 19(2), 124–32. doi:10.1016/S0966-6362(03)00037-7 Goldberg, J. M. (1991). The vestibular end organs: morphological and physiological diversity of afferents. Current Opinion in Neurobiology, 1(2), 229–35. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/1821186 Goldberg, J. M. (2000). Afferent diversity and the organization of central vestibular pathways. Experimental Brain Research, 130(3), 277–97. Retrieved from http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3731078&tool=pmcentrez&rendertype=abstract Goldberg, J. M., Fernández, C., & Smith, C. E. (1982). Responses of vestibular-nerve afferents in the squirrel monkey to externally applied galvanic currents. Brain Research, 252(1), 156–60. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/6737029 Goldberg, J. M., Smith, C. E., & Fernández, C. (1984). Relation between discharge regularity and responses to externally applied galvanic currents in vestibular nerve afferents of the squirrel monkey. Journal of Neurophysiology, 51(6), 1236–56. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/6737029 Grabherr, L., Nicoucar, K., Mast, F. W., & Merfeld, D. M. (2008). Vestibular thresholds for yaw rotation about an earth-vertical axis as a function of frequency. Experimental Brain Research, 186(4), 677–81. doi:10.1007/s00221-008-1350-8 Green, A. M., & Angelaki, D. E. (2010). Internal models and neural computation in the vestibular system. Experimental Brain Research, 200(3-4), 197–222. doi:10.1007/s00221-009-2054-4 Hansson, E. E., Beckman, A., & Håkansson, A. (2010). Effect of vision, proprioception, and the position of the vestibular organ on postural sway. Acta Otolaryngol., 130(12), 1358–63. doi:10.3109/00016489.2010.498024 Herdman, S. J., Blatt, P., Schubert, M. C., & Tusa, R. J. (2000). Falls in patients with vestibular deficits. The American Journal of Otology, 21(6), 847–51. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/11078074 Héroux, M. E., Dakin, C. J., Luu, B. L., Inglis, J. T., & Blouin, J.-S. (2014). Absence of lateral gastrocnemius activity and differential motor unit behavior in soleus and medial gastrocnemius during standing balance. Journal of Applied Physiology (Bethesda, Md. : 1985), 116(2), 140–8. doi:10.1152/japplphysiol.00906.2013 Highstein, S. M., & Baker, R. (1985). Action of the efferent vestibular system on primary afferents in the toadfish, Opsanus tau. Journal of Neurophysiology, 54(2), 370–84. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/4031993 71  Highstein, S. M., Rabbitt, R. D., & Boyle, R. (1996). Determinants of semicircular canal afferent response dynamics in the toadfish, Opsanus tau. J Neurophysiol., 75(2), 575–596. Holt, J. C., Lysakowski, A., & Goldberg, J. M. (2011). Auditory and Vestibular Efferents. (D. K. Ryugo & R. R. Fay, Eds.) (Vol. 38, pp. 135–186). New York, NY: Springer New York. doi:10.1007/978-1-4419-7070-1 Horak, F. B. (2006). Postural orientation and equilibrium: what do we need to know about neural control of balance to prevent falls? Age and Ageing, 35 Suppl 2, ii7–ii11. doi:10.1093/ageing/afl077 Horak, F. B., & Macpherson, J. M. (1996). Postural Orientation and Equilibrium. In L. B. Rowell & J. T. Sheperd (Eds.), Handbook of Physiology, Exercise: Regulation and Integration of Multiple Systems (Vol. Supplement, pp. 255–292). Oxford University Press, New York. Huryn, T. P. (2012). A New Platform for Studying Human Balance Control: Design, Validation, and Experiments. University of British Columbia. Huryn, T. P., Luu, B. L., Van der Loos, H. F. M., Blouin, J.-S., & Croft, E. A. (2010). Investigating human balance using a robotic motion platform. 2010 IEEE International Conference on Robotics and Automation, 5090–5095. doi:10.1109/ROBOT.2010.5509378 Kaptein, R. G., & Van Gisbergen, J. A. M. (2006). Canal and otolith contributions to visual orientation constancy during sinusoidal roll rotation. Journal of Neurophysiology, 95(3), 1936–48. doi:10.1152/jn.00856.2005 Karlsson, A., & Frykberg, G. (2000). Correlations between force plate measures for assessment of balance. Clinical Biomechanics (Bristol, Avon), 15(5), 365–9. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10758298 Kim, J. (2013). Tonic eye movements induced by bilateral and unilateral galvanic vestibular stimulation (GVS) in guinea pigs. Brain Research Bulletin, 90, 72–8. doi:10.1016/j.brainresbull.2012.09.010 Krishnamoorthy, V., & Latash, M. L. (2005). Reversals of anticipatory postural adjustments during voluntary sway in humans. The Journal of Physiology, 565(Pt 2), 675–84. doi:10.1113/jphysiol.2005.084772 Krishnan, V., & Aruin, A. S. (2011). Postural control in response to a perturbation: role of vision and additional support. Exp Brain Res., 212(3), 385–397. Larsby, B., Hyden, D. A. G., & Odkvist, L. M. (1984). Gain and phase characteristics of compensatory eye movements in light and darkness. A study with a broad frequency-band rotatory test. Acta Otolaryngol., 97(3-4), 223–232. Lee, D. N., & Aronson, E. (1974). Visual proprioceptive control of standing in human infants. Perception & Psychophysics, 15(3). Retrieved from http://link.springer.com/article/10.3758/BF03199297 72  Loram, I. D., Gawthrop, P. J., & Lakie, M. (2006). The frequency of human, manual adjustments in balancing an inverted pendulum is constrained by intrinsic physiological factors. The Journal of Physiology, 577(Pt 1), 417–32. doi:10.1113/jphysiol.2006.118786 Loram, I. D., & Lakie, M. (2002a). Direct measurement of human ankle stiffness during quiet standing: the intrinsic mechanical stiffness is insufficient for stability. The Journal of Physiology, 545(3), 1041–1053. doi:10.1113/jphysiol.2002.025049 Loram, I. D., & Lakie, M. (2002b). Human balancing of an inverted pendulum: position control by small, ballistic-like, throw and catch movements. The Journal of Physiology, 540(3), 1111–1124. doi:10.1113/jphysiol.2001.013077 Loram, I. D., Maganaris, C. N., & Lakie, M. (2004). Paradoxical muscle movement in human standing. The Journal of Physiology, 556(Pt 3), 683–9. doi:10.1113/jphysiol.2004.062398 Lund, S., & Broberg, C. (1983). Effects of different head positions on postural sway in man induced by a reproducible vestibular error signal. Acta Physiologica Scandinavica, 117(2), 307–9. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/6603098 Luu, B. L., Huryn, T. P., Van der Loos, H. F. M., Croft, E. A., & Blouin, J.-S. (2011). Validation of a robotic balance system for investigations in the control of human standing balance. IEEE Transactions on Neural Systems and Rehabilitation Engineering : A Publication of the IEEE Engineering in Medicine and Biology Society, 19(4), 382–90. doi:10.1109/TNSRE.2011.2140332 Luu, B. L., Inglis, J. T., Huryn, T. P., Van der Loos, H. F. M., Croft, E. A., & Blouin, J.-S. (2012). Human standing is modified by an unconscious integration of congruent sensory and motor signals. The Journal of Physiology, 590(Pt 22), 5783–94. doi:10.1113/jphysiol.2012.230334 Mahboobin, A., Loughlin, P. J., Redfern, M. S., Anderson, S. O., Atkeson, C. G., & Hodgins, J. K. (2008). Sensory adaptation in human balance control: lessons for biomimetic robotic bipeds. Neural Networks : The Official Journal of the International Neural Network Society, 21(4), 621–7. doi:10.1016/j.neunet.2008.03.013 McCall, A. A., & Yates, B. J. (2011). Compensation following bilateral vestibular damage. Frontiers in Neurology, 2, 88. doi:10.3389/fneur.2011.00088 Merfeld, D. M., Park, S., Gianna-Poulin, C., Black, F. O., & Wood, S. (2005). Vestibular perception and action employ qualitatively different mechanisms. I. Frequency response of VOR and perceptual responses during Translation and Tilt. Journal of Neurophysiology, 94(1), 186–98. doi:10.1152/jn.00904.2004 Merfeld, D. M., Zupan, L., & Peterka, R. J. (1999). Humans use internal models to estimate gravity and linear acceleration. Nature, 398(6728), 615–8. doi:10.1038/19303 Mian, O. S., Dakin, C. J., Blouin, J.-S., Fitzpatrick, R. C., & Day, B. L. (2010). Lack of otolith involvement in balance responses evoked by mastoid electrical stimulation. The Journal of Physiology, 588(Pt 22), 4441–51. doi:10.1113/jphysiol.2010.195222 73  Milton, J., Cabrera, J. L., Ohira, T., Tajima, S., Tonosaki, Y., Eurich, C. W., & Campbell, S. A. (2009). The time-delayed inverted pendulum: implications for human balance control. Chaos (Woodbury, N.Y.), 19(2), 026110. doi:10.1063/1.3141429 Mohapatra, S., Krishnan, V., & Aruin, A. S. (2012). The effect of decreased visual acuity on control of posture. Clinical Neurophysiology : Official Journal of the International Federation of Clinical Neurophysiology, 123(1), 173–82. doi:10.1016/j.clinph.2011.06.008 Morasso, P. G., & Schieppati, M. (1999). Can muscle stiffness alone stabilize upright standing? J Neurophysiol, 82(3), 1622–1626. Retrieved from http://jn.physiology.org/content/82/3/1622.short Moseley, A. M., Crosbie, J., & Adams, R. (2001). Normative data for passive ankle plantarflexion–dorsiflexion flexibility. Clinical Biomechanics, 16, 514–521. Retrieved from http://www.sciencedirect.com/science/article/pii/S0268003301000304 Nardone, A., Tarantola, J., Miscio, G., Pisano, F., Schenone, A., & Schieppati, M. (2000). Loss of large-diameter spindle afferent fibres is not detrimental to the control of body sway during upright stance: evidence from neuropathy. Experimental Brain Research, 135(2), 155–62. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/11131499 Niam, S., Cheung, W., Sullivan, P., Kent, S., & Gu, X. (1999). Balance and physical impairments after stroke. Archives of Physical …, 80(October). Retrieved from http://www.sciencedirect.com/science/article/pii/S0003999399900205 Oman, C. M., Marcus, E. N., & Curthoys, I. S. (1987). The influence of semicircular canal morphology on endolymph flow dynamics. An anatomically descriptive mathematical model. Acta Otolaryngol., 103(1-2), 1–13. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/3494374 Önell, A. (1999). Quantifying human balance : Analysis of force plate data. Acta Universitatis Upsaliensis. Retrieved from http://www.diva-portal.org/smash/record.jsf?pid=diva2:160551 Osler, C. J., Tersteeg, M. C. A., Reynolds, R. F., & Loram, I. D. (2013). Postural threat differentially affects the feedforward and feedback components of the vestibular-evoked balance response. The European Journal of Neuroscience, 38(8), 3239–47. doi:10.1111/ejn.12336 Park, Y.-H., Kim, Y.-M., & Lee, B.-H. (2013). An ankle proprioceptive control program improves balance, gait ability of chronic stroke patients. Journal of Physical Therapy Science, 25(10), 1321–4. doi:10.1589/jpts.25.1321 Patel, M., Fransson, P. A., Johansson, R., & Magnusson, M. (2011). Foam posturography: standing on foam is not equivalent to standing with decreased rapidly adapting mechanoreceptive sensation. Experimental Brain Research, 208(4), 519–27. doi:10.1007/s00221-010-2498-6 Perrin, D. H. (1996). Effect of lateral ankle joint anesthesia, center of balance, postural sway, and joint position sense. Journal of Sports Rehabilitation, 2, 111–119. Retrieved from http://libres.uncg.edu/ir/uncg/listing.aspx?id=572 74  Pospisil, E. R., Luu, B. L., Blouin, J.-S., Van der Loos, H. F. M., & Croft, E. A. (2012). Independent ankle motion control improves robotic balance simulator. In Conf Proc IEEE Engineering in Medicine and Biology (pp. 6487–6491). Retrieved from http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=6347480 Pothula, V. B., Chew, F., Lesser, T. H. J., & Sharma, A. K. (2004). Falls and vestibular impairment. Clinical Otolaryngology and Allied Sciences, 29(2), 179–82. doi:10.1111/j.0307-7772.2004.00785.x Prince, F., Corriveau, H., Hébert, R., & Winter, D. A. (1997). Gait in the elderly. Gait & Posture, 5(2), 128–135. doi:10.1016/S0966-6362(97)01118-1 Qu, X., Nussbaum, M. A., & Madigan, M. L. (2007). Interpretation of human balance control based on an optimal control model. Journal of Biomechanics, 40(16), 3590–3597. Refshauge, K. M., & Fitzpatrick, R. C. (1995). Perception of movement at the human ankle: effects of leg position. The Journal of Physiology, 488(1995), 243–8. Retrieved from http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1156717&tool=pmcentrez&rendertype=abstract Reynolds, R. F. (2010). The effect of voluntary sway control on the early and late components of the vestibular-evoked postural response. Experimental Brain Research, 201(2), 133–9. doi:10.1007/s00221-009-2017-9 Reynolds, R. F., & Osler, C. J. (2012). Galvanic vestibular stimulation produces sensations of rotation consistent with activation of semicircular canal afferents. Frontiers in Neurology, 3(June), 104. doi:10.3389/fneur.2012.00104 Rude, S. A., & Baker, J. F. (1988). Dynamic otolith stimulation improves the low frequency horizontal vestibulo-ocular reflex. Experimental Brain Research, 116(73), 357–363. Retrieved from http://link.springer.com/article/10.1007/BF00248228 Sadeghi, S. G., Goldberg, J. M., Minor, L. B., & Cullen, K. E. (2009). Efferent-mediated responses in vestibular nerve afferents of the alert macaque. Journal of Neurophysiology, 101(2), 988–1001. doi:10.1152/jn.91112.2008 Schneider, E., Glasauer, S., & Dieterich, M. (2000). Central processing of human ocular torsion analyzed by galvanic vestibular stimulation. Neuroreport, 11(7), 1559–63. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10841377 St George, R. J., Day, B. L., & Fitzpatrick, R. C. (2011). Adaptation of vestibular signals for self-motion perception. The Journal of Physiology, 589(Pt 4), 843–53. doi:10.1113/jphysiol.2010.197053 St George, R. J., & Fitzpatrick, R. C. (2011). The sense of self-motion, orientation and balance explored by vestibular stimulation. The Journal of Physiology, 589(Pt 4), 807–13. doi:10.1113/jphysiol.2010.197665 75  Stelmach, G. E., Phillips, J., DiFabio, R. P., & Teasdale, N. (1989). Age, functional postural reflexes, and voluntary sway. Journal of Gerontology, 44(4), B100–6. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/2738304 Straka, H., & Dieringer, N. (2004). Basic organization principles of the VOR: lessons from frogs. Progress in Neurobiology, 73(4), 259–309. Retrieved from http://www.sciencedirect.com/science/article/pii/S030100820400084X Tascioglu, A. B. (2005). Brief review of vestibular system anatomy and its higher order projections. Neuroanatomy, 4, 24–27. Retrieved from http://www.researchgate.net/publication/26415423_Brief_review_of_vestibular_system_anatomy_and_its_higher_order_projections Tucker, M. G., Kavanagh, J. J., Morrison, S., & Barrett, R. S. (2010). What are the relations between voluntary postural sway measures and falls-history status in community-dwelling older adults? Archives of Physical Medicine and Rehabilitation, 91(5), 750–8. doi:10.1016/j.apmr.2010.01.004 Van Asseldonk, E. H. F., Buurke, J. H., Bloem, B. R., Renzenbrink, G. J., Nene, A. V, van der Helm, F. C. T., & van der Kooij, H. (2006). Disentangling the contribution of the paretic and non-paretic ankle to balance control in stroke patients. Experimental Neurology, 201(2), 441–51. doi:10.1016/j.expneurol.2006.04.036 Wardman, D. L., Taylor, J. L., & Fitzpatrick, R. C. (2003). Effects of galvanic vestibular stimulation on human posture and perception while standing. The Journal of Physiology, 551(Pt 3), 1033–42. doi:10.1113/jphysiol.2003.045971 Wichmann, F. A., & Hill, N. J. (2001). The psychometric function: I. Fitting, sampling, and goodness of fit. Perception & Psychophysics, 63(8), 1293–313. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/11800458 Winter, D. A., Patla, A. E., Prince, F., Ishac, M., & Gielo-Perczak, K. (1998). Stiffness control of balance in quiet standing. J Neurophysiol, 80(3), 1211–1221. Retrieved from http://jn.physiology.org/content/80/3/1211.short A 76  Appendi x A: Ankle-Tilt System D esign  Appendix A: Ankle-Tilt System Design This appendix presents more in-depth details of the ankle-tilt system design and control scheme. The unique challenge to the design of the ankle-tilt control system was the necessity of integration and synchronization with the Stewart platform. Design considerations, assembly details, and technical validation results are presented. Section A.1 provides an overview of the design, with the bill of materials, general assembly, and electrical block diagram of the system. Section A.2 details the software implementation in LabVIEW. Section A.3 covers several important safety considerations that are integrated into the architecture. Section A.4 contains excerpts from the Capstone Team’s final report, with occasional revisions and notes (changes that I implemented) shown in italics. A.1 Hardware The original hardware design was performed by the MECH 45X Capstone Team as mentioned in the acknowledgements. After some refinements, the final bill of materials for the single-actuator assembly is presented in Table A-1. Wiring, electrical connectors, and enclosures are not included in this list, as they were customized to the exact assembly. The wiring block diagram of the platform is presented in Figure A-1, and the mechanical assembly in Figure A-2. The purpose of the 2 capacitors (added after testing) is to buffer the input of the power supply, which would otherwise periodically shut down due to overvoltage from motor regeneration during rapid retraction (gravity assisted) moves. Table A-1 – Bill of materials for ankle-tilt platform Part Part # Qty Manufacturer Encoder System       Rotary Incremental Encoder DRS20-1E507200 1 Stegmann Encoder Pulley (Encoder shaft) A 6D16-020DF2508 1 SDP Encoder Pulley (Platform shaft) A 6A16-120NF2512 1 SDP Timing Belt A 6B16-200025 1 SDP Actuator System       Linear Actuator "The Bug" 2-B.M82C7-DC427_24-8-2NO-ST4/4 1 Ultramotion 25A8 motor driver 25A8 1 A.M.C Power Supply SWS600 1 TDK Lambda Buffer Capacitors ECE-T1EP104FA 2 Panasonic Internal thread rod ends RAP3M42FS464 1 RBC Bearings 77  Part Part # Qty Manufacturer External thread rod ends REP3F4FS428 1 RBC Bearings Support Structure       Platform Supporting Bearings UCP211-200D1 4 NTN Bearings 6" x 10" x 0.25" steel-hot rolled structural tube rectangle HTRT106250 1   steel-hot rolled flat A36 (0.5" x 6" steel-hot rolled A36 flat bar) HF126 1   12" x 20.5" x 1" 6061-T6 Aluminium plate AP6061/1000 1   12" x 20.5" x 0.75" 6061-T6 Aluminium plate AP6061/750 1   aluminum square 6061T6 (4" x 4" aluminum 6061 T6 square bar) ASQ6061/4 1   aluminum flat 6061T6 (4" x 5" aluminum 6061 T6 square bar) AF6061/45 1   steel-hot rolled plate A36 (0.5" steel-hot rolled A36 plate) HP44W500 1   (2.125" steel-cold rolled 1018 round) CR1018/218 1   aluminum flat 6061T6 (0.5" x 1" aluminum 6061 T6 flat bar)  AF6061/121 1   steel-hot rolled flat A36 (0.5" x 4" x 16") HF124 1  Fasteners/Hardware       Standard Hex Head Cap Screw 5/8"-18 x 1-1/2" 92620A824 1   Standard Hex Head Cap Screw 5/8"-18 x 2" 92620A825 2   Standard Hex Head Cap Screw 3/8"-24 x 1-3/4" 91257A658 1   Standard Hex Head Cap Screw 3/8"-24 x 1" 92620A655 1   Standard Hex Head Cap Screw 3/8"-24 x 2-1/2" 91257A661 1   Toothed External Push On Ring DTX-4.0 8   Socket Cap Screw M4 - 60mm 91290A188 1   Nylon-Insert Hex Flange Locknuts 3/8"-24 93298A135 1   Hex Nut 1-5/8"-12 90496A061 2   Standard Hex Head Cap Screw 1/2"-13 x 3-3/4" 91309A727 2   Washers - Round Hole 5/8" 91083A035 1   Washers - Round Hole 3/8" 91201A031 2   Washers - Round Hole 1/2" 91083A033 1   Metric Hex Nuts M4 92497A250 1   Steel Standoffs 0.166" I.D. x 2" 92414A403 8   Control System       8-slot 3U PXI Chassis with Universal AC NI PXI-1042 1  National Instruments 4-Axis High-Performance Stepper/Servo Motion Controller NI PXI-7354 1  National Instruments 4-Axis Universal Motion Interface for Industrial Applications NI UMI-7774 1  National Instruments  78   Figure A-1 – General layout of the ankle-tilt platform control system. The PXI-1042 generates motion commands for both the MOOG Stewart platform (sent via the network switch), and the ankle-tilt platform (generated by the onboard 7354 motion controller and sent via the UMI 7774 Motion Interface). 79     Figure A-2 – Exploded rendering of the ankle-tilt platform assembly. Linear actuator (1) is mounted via ball bearing rod ends below the force plates (2), encoder (3) is connected via timing belt (belt not shown). The entire balance system rotates about the support shafts (4), which are located at the ankles.80  A.2 Software Implementation Software integration for the ankle-tilt platform was accomplished through LabVIEW©. As discussed in Chapter 2, the system was buffered to permit better synchronization with the Stewart platform, while permitting smooth profiling. The software architecture diagram on the following page (Figure A-3) highlights the general software integration method of the ankle tilt platform. For a detailed description of the functionality of the RISER Stewart platform and balance control algorithm, refer to Thomas Huryn’s thesis (Huryn, 2012). 81   Figure A-3 – Simplified model of RISER software architecture with ankle-tilt platform addition. The ankle-tilt platform is driven as a ‘slave’ to the Stewart platform. In the main loop of the Stewart Platform, forceplate torque data is passed into the inverted pendulum state-space model, and then sent to the riser platform and stored for the ankle-tilt platform. The ankle-tilt platform then uses the current and past commands from the Stewart platform to extrapolate a new target position when space is available in the buffer (see Figure 2-4). Current position information from the ankle-tilt platform is also used for forceplate compensation, as discussed in Chapter 2.82  A.3 Safety To ensure safe operation of the ankle-tilt platform, integration with the emergency stop system of the Stewart platform was required. When either the participant or the experimenter presses one of the emergency stop buttons, the following events must take place: - Activation of the emergency stop routine of the Stewart platform - Activation of the ankle-tilt platform actuator brake (to prevent back driving of the actuator when power is cut) - Removal of power from the ankle-tilt platform actuator - Emergency stop detected in software and “Kill” command issued to NI motion planner The Stewart platform emergency stop system is a normally closed 24V relay circuit with a 1A breaker. To integrate the ankle-tilt platform into the system, this circuit was used to power two small normally open single-pole single-throw relays (Figure A-4). The first relay controls the ankle-tilt platform brake circuit (brake is active when circuit is broken), while the second relay controls the fault input on the power amplifier, cutting the power from the actuator. The software Emergency Stop logic was accomplished through detection of the “Emergency Stop” state of the Stewart platform. Detection of this state produces an interrupt routine that issues the “Kill” command to the Motion Card. An additional safety consideration was ensuring that the range of motion of the system did not exceed the range of motion of the participant’s ankles in plantar flexion or dorsiflexion. While under normal operating conditions, the ankle should flex approximately the same amount as in standing balance. However, it is in theory possible for the platform to produce a motion that exceeds a comfortable range of motion for a participant with low ankle flexibility. Boone showed that typical angles for maximum voluntary plantar flexion and dorsiflexion were 56.2° ± 6.1° and 12.6° ± 4.4°, respectively (Boone & Azen, 1979). In plantar flexion, the maximum range of the platform is 20 degrees, more than three standard deviations below the mean range of motion. In dorsiflexion, the maximum range is 18°. As this amount of dorsiflexion could be uncomfortable for participants, the motion is software limited to 6°.  83  Ankle Tilt Platform Emergency Stop Block Diagram  Figure A-4 – Ankle-tilt platform emergency stop block diagram. The ankle-tilt platform will enter an emergency stop condition and apply the platform brake when the Stewart platform (MOOG) enters an emergency stop state, or when either E-Stop button is pressed. Pressing either E-Stop button will also disable the ‘E-Stop Out’ on the MOOG until the stop is cleared in software; this ensures the ankle-tilt platform will not resume motion unexpectedly.84  A.4 Capstone Team Report The following section contains excerpts from the final report created by the UBC Mech 458 Capstone team, who produced the mechanical design of the ankle-tilt platform. Significant text not from the original report is highlighted using italics, information deemed extraneous has been removed, and formatting (including captions) has been modified to fit with this thesis. Otherwise, the majority of this section has been directly drawn from that report (Bell et al., 2010). Note that the students refer to the ankle-tilt platform as the ‘Tilt Platform.’ A.4.1 Introduction Within the CARIS Lab, there exists the need for a device that can supplement existing research into human proprioception by providing insight to the relationships between balance, gait, and posture. The device must be able to accurately measure the positions of both feet of a human user with respect to the rotating ankle. The device must also be able to measure the reaction forces of a human user’s feet with respect to torques about the respective ankle. Finally, the device must be able to interact with a human user by providing unstable ground input and measuring the user’s reaction in an attempt to regain balance. We have designed a Tilt Platform that meets the needs previously stated. This report details the proposed design, the path taken in arriving at that design, and analysis supporting the decisions made in arriving at the design. A.4.2 Requirements The functional requirements of the Tilt Platform project include precise positioning, enhanced stiffness, robust design, and reliable data acquisition. With respect to precision, the requirements of the client were that the tilt platform be able to provide angular position accuracy to within 0.01°, and an angular deflection limit of 0.05°. This angular deflection limit was described as being a maximum angle between foot platform vertical section and foot platform horizontal section measured from the static position. Robustness requirements include the ability to withstand the force of any user with a mass of less than 100 kg. This requirement was further supplemented by peak torque loading values of 400 Nm in plantar-flexion, and 200 Nm in dorsiflexion applied by the user about their ankle. Positional requirements include the ability to accelerate up to 500°/s², and provide an angular velocity of up to 10°/s over a range of +/- 15° about horizontal. Finally, it was required that the axis of rotation be adjustable from between 7 cm and 11 cm above the platform horizontal section. These conclude the 85  quantified performance requirements for the Tilt Platform project. Several functional requirements were also set forth, including the need for the Tilt Platform to provide smooth motion to the user, with no discernable stepping in motion. In addition, it was necessary for the Tilt Platform to fit within the available real estate on the Moog Platform without interfering with any existing components. The final functional requirement was that the platform be able to acquire user force and torque data through sensory measurement. The functional and performance requirements are tabulated below: Performance Requirements  Angular position accuracy: 0.01°  Maximum angular deflection: 0.05°  Angular range about horizontal: +/- 15°  Angular velocity: 10 °/s  Angular acceleration: 500 °/s²  400 Nm plantarflexion max torque  200 Nm dorsiflexion torque  Adjustable rotation axis position: 7 cm to 11 cm above foot platform Functional Requirements  Smooth motion is necessary, stepped motion is unacceptable  Tilt platform must conform to the available real estate atop the Moog Hex Plate  Tilt platform must provide torque feedback for analysis A.4.3 System Overview The diagram on the following page shows an overall schematic of the system, including the control components, actuator, mechanical structure, and sensors. It demonstrates the links that exist between various components, and how they will interact during operation of the system. 86   Figure A-5 – Tilt platform system schematic.  The figure below shows an overview of the mechanical system.   Figure A-6 – Tilt platform mechanical design 3D model. A.4.4 Subsystems The Tilt Platform is comprised of several subsystems, as described in the previous section. These subsystems in turn consist of several individual components, which are presented below. 87  A.4.4.1 Structure The mechanical structure, including the main tower, bearings, shafts, foot platforms, and ankle height adjustment system is designed to support the weight of the test subject while allowing the actuator to move the structure in a smooth motion as specified in the requirements. The structure also ensures that the medial, dorsal, and plantar flexion deflection limits are not exceeded. Main Tower The main tower, which can be seen in the figure below, is designed to support the bearings, shaft and foot platform. The A36 Structural Steel main channel is 6” by 10”, and ¼” thick. It is sized to adequately mount the bearings, and minimize deflection. The slots in the side of the tower provide for location and positioning of the encoder mounting.  Figure A-7 – Tilt platform main tower mechanical component Bearings and Shaft In order to meet the specified deflection and load requirements, a 2” diameter cold rolled 1045 steel shaft was designed to support the foot platform. This shaft is stepped and threaded on one end to allow for precise axial alignment of the foot platform on the shaft. The other end of the shaft is stepped to allow mounting of the pulley that connects to the encoder. A bending and stress analysis was also performed on the shaft. 88  The bearings chosen are NTN ball bearings with a 2” bore diameter in pillow block housings. These bearings were sized to cope with the loading requirements of the system.   Figure A-8 – Tilt platform bearing and shaft assembly Foot Platform The image below shows the foot platform with the force plate mounted in the lowest possible position. The vertical section of the foot platform is fabricated from 1” aluminum plate, while the horizontal section is fabricated from ¾” aluminum plate. The diagonal struts are in designed to keep the deflection of the platform within the allowable limits.  Figure A-9 – Tilt platform and force plate assembly.  89  Ankle Height Adjustment The ankle height adjustment system consists of a variety of shims and supports, designed to adjust the height of the force plate over a range of 4cm with a resolution finer than 2mm.  This system was not used in the final implementation of the platform, due to the difficulty in removing or adding the shims. Instead, I designed a single shim that set the ankle height at the average for human ankles, as discussed in chapter 2. A.4.4.2 Linear Actuator The linear actuator chosen for this application is “The Bug,” manufactured by Ultra Motion LLC. In order to achieve the high precision requirements of the design, a custom low-backlash ball screw is required. Ultra Motion will supply the finished linear actuator unit, complete with the ball screw, and Pittman Express DC427 brushed DC motor. The following table represents the key actuator specifications for this design. Table A-2 – Tilt platform actuator specifications Manufacturer Ultra Motion Model The Bug Motor DC427_24 Lead screw Custom ball screw and nut with 2mm/rev and 20µm backlash  Belt drive ratio 2:1 The final implementation used a 1:1 drive ratio, as the 2:1 drive ratio was insufficient to reach the required speeds. Stroke length 8” Nose fitting ¼-28 UNF threaded hole Base fitting ¼-28 UNF threaded stud Tube fitting None Motor add-ons Power off solenoid brake Position switches 2 normally open position switches    Rod Ends Each end of the linear actuator must be mounted using a ball bearing rod end to allow for rotation due to the change in angle of the actuator throughout its range of motion. Because the rod ends are directly in line with the actuators, any play in the rod ends would translate directly to backlash in the system. For this reason, precision rod ends were chosen that satisfy the backlash allowances. Two different rod ends 90  are required – an external thread and an internal thread – due to the different mounting points on the actuator. On the base end of the actuator, a threaded stud will allow an internally threaded rod end to be mounted. On the nose end of the actuator, a threaded hole will allow an externally threaded rod end to be mounted. The thread sizes of the rod ends were chosen to match the threaded hole and stud provided by Ultra Motion on the actuators. The key rod end specifications are as follows. Table A-3 – Tilt platform actuator rod end specifications Location Base end Nose end Manufacturer RBC Bearings RBC Bearings Model number REP3F4 RAP3M4-2 Bore 0.1900“ 0.1900” Thread ¼-28RH ¼-28RH Radial load limit 1000 lbf 1000 lbf  A.4.4.3 Control Control Card The CARIS lab is supplying a National Instruments PXI-7344 card to use for motion control of the Tilt Platform system. The National Instruments board is a mid-range stepper/servo motion controller, with PID capability, encoder input, and standard input/output pins. It will be used in the Tilt Platform project to interpret reference and encoder signals, implement a discrete controller, and output an analogue signal in the range of ±10V to the motor driver. The final implementation made use of a PXI-7354 motion card instead, which offers a higher control frequency and buffered position capture. Motor Driver Based on the maximum operating current found in the analysis of the requirements, a driver has been sourced from Advanced Motion Controls. The driver (25A8) has the following key specifications: Table A-4 – Tilt platform amplifier specifications Supply Voltage  20 – 80 [VDC] Maximum Continuous Output Current 12.5 [A] PWM Switching Frequency 22 [kHz] Command Input +/- 10 [V] Analog 91   The driver comes with directionally inhibiting inputs for the limit switches of the linear actuator. Additionally, the existing emergency stop in the CARIS lab will be connected to a fully inhibiting switch on the driver. This inhibit switch must be maintained at 5 [V] in order for the driver to output. In the case of an emergency the inhibit switch should be pulled down to 0 [V].  See the implementation of this safety system in Appendix A.3. In order to ensure that the current output of the driver is maintained below the motor’s maximum current the driver must be run in current mode. The controller must then apply saturation limits to its output in order to maintain the current at acceptable levels. As an additional failsafe, a limiting resistor will be connected between the continuous current inputs in order to reduce the maximum continuous current.  A pair of buffering capacitors with a total capacitance of 500 mF was added in parallel with the actuator. This was done to prevent the power supply from shutting down due to over or under voltage errors, which were occurring due to the large current capacity of the actuator. As the range of motion is limited, the 500mF capacitance provides a sufficient buffer to the power supply. Power Supply The motor driver requires a DC input voltage of 24 [V] and a continuous current capability of 12.5 [A]. The SWS300 AC-DC power supply from TDK-Lambda meets these specifications and provides a reliable low cost solution. It has built in over-current, over-voltage, and over-temperature protection. The power supply was upgraded to a SWS600 to provide improve actuator torque performance. A.4.4.4 Encoder System The encoder system, which is necessary to guarantee the system position to within 0.01° as required, consists of an encoder that is connected to the foot platform shaft through a belt and pulley system. The components of this subsystem are described below. Encoder The encoder specified is a SICK|STEGMANN DRS20-1EM07200. It has a resolution of 7200PPR (pulses per revolution). Post quadrature, the signal will have 28800 counts per revolution. Some of the key specifications can be seen in the figure below. 92  Table A-5 – Tilt platform encoder specifications Part Number DRS20-1EM07200 Number of lines per revolution 7200 Shaft Size 1/4" Cable/Connector Radial Cable 5.0 m Resistance to Shocks 5g Resistance to Vibration 20g  Since the encoder disk is made of glass and is only capable of shocks up to 5g, care must be taken to prevent shattering of the encoder disk. The MOOG’s maximum acceleration specification is 6 m/s2 (0.611g), which is well under this 5g limit.  In order to prevent damage to the encoder during assembly, flexible washers (e.g. rubber washers) should be placed on the mounting screw on both sides of the encoder flange. In addition, the encoder mounting is designed so that the encoder can be mounted after the tower has been mounted to the hex plate to prevent damage to the encoder disk during assembly. Belt A timing belt connects the two pulleys. Given that timing belts are inherently constant speed and non-slip with zero backlash, this belt is well suited to this application. This belt is also Kevlar-reinforced to prevent stretching issues, and allows for a 1:6 gear ratio. Some of the key belt specifications can be seen in the table below. Table A-6 – Tilt platform timing belt specifications Part Number A 6B16-200025 Pitch .080" (MXL) Belt Width 1/4" Material Urethane Tension Member Kevlar Pitch Length 16.00"   The belt system connects the platform ankle shaft to the encoder. The centre-to-centre distance between the two pulleys is set to be 5“, but this is variable based on the exact mounting location of the encoder in the main tower slots. The gear ratio is 1:6, with the encoder shaft rotating faster than the main shaft. 93  Pulleys The timing belt connects two pulleys, one on the end of the shaft connected to the foot platform, and one connected to the shaft of the encoder. The ratio between the pitch diameters is 1:6. Some key specifications of these pulleys can be seen in the table below. Table A-7 – Tilt platform encoder pulley specifications Location Main Shaft Encoder Shaft Part Number A 6A16-120NF2512 A 6D16-020DF2508 Pitch 0.080” (MXL) 0.080” (MXL) No. Of Teeth 120 20 Belt Width 0.25" 0.25" Bore Size 0.375" 0.250" Flange & Hub No Flange / With Hub 2 Flanges / Fairloc Hub Pitch Dia. 3.056" 0.509"  Encoder Shock Proofing If the encoder were rigidly attached to the tower while the tower is installed onto the MOOG, there is a risk that the encoder disk could shatter during installation. In order to prevent this from occurring, the encoder mounting is designed so that the encoder can be mounted after the tower has been mounted to the hex plate. Additionally, flexible washers (e.g. rubber washers) should be placed on both sides of the encoder flange to reduce vibrations transmitted to the encoder during normal operation.  A.4.4.5 Fasteners There is a variety of fasteners used in the assembly of the full system. These are all standard types and sizes that can be easily obtained from any fastener supplier. Please refer to the bill of materials for a full list of fastener requirements. A.4.5 Recommendations While the system has been successfully designed to meet the client specifications, there are several issues that could arise during assembly of the device and future improvements that may be made. These are discussed below. 94  A.4.5.1 MOOG Motion System Preparation In order to install the Tilt Platform, several steps must be taken to prepare the MOOG Motion System. These include drilling and tapping holes in the steel hex plate. In addition, the existing paint finish on the hex plate should be removed to ensure the best surface finish possible for the contact between the plate and the Tilt Platform system. A.4.5.2 Fabrication of Tilt Platform While the system has been designed and the engineering drawings submitted to the machine shop, it will be necessary to follow up with the machine shop to determine when the fabrication is complete and the system is ready to be collected. The only significant issue that arose during fabrication was fastening the bottom of the footplates to the sides. As both plates were large pieces of aluminum, it proved extremely difficult to weld the pieces. Instead, the two plates were connected using four ¼-20x2” countersunk hex head bolts and two Φ3/8”x1.5” dowel pins. A.4.5.3 Actuator Clearance Due to the space constraints on the MOOG Motion System, the Tilt Platform needed to be designed with little clearance between various components both within the Tilt Platform and on the MOOG device, particularly the linear actuator. While this component was successfully placed based on the available dimensions during design, slight variations in these dimensions in the final product could result in interference issues, which could be rectified with minor design modifications. It is advised that the Tilt Platform be moved over the full range of motion by hand, before powering with the motor, to ensure that these clearances are not an issue. One possible interference location is between the DC motor and the back support structure as seen in the figure below. This could possible occur in the single actuator configuration, when the clevis mount on the foot platform is lower due to the coupling plate. If this clearance is a concern, it is recommended that a ½” shim be placed under the large clevis mount in order to raise the motor and avoid any interference. Alternatively, the back support structure could be modified.  This interference was resolved by mounting the motor so that all of the wires travelled out of the side. The motor itself was not found to interfere with the support structure. 95   Figure A-10 – Motor interference with back support Another possible source of interference is between the plunger of the linear actuator and the foot platforms. In the dual actuator configuration, this area can be seen in the figure below. If this interference becomes a problem, it would be advisable to either adjust the stroke limits of the actuator or reduce the height of the large clevis mount, while ensuring the motor clearance with respect to the back support is not an issue.  This interference is outside of the 6° software limit for dorsiflexion, and so does not occur.  Figure A-11 – Motor interference with foot platform  96  A.4.5.4 Controller Design A preliminary controller for the system has been designed. This controller was based off theoretical values, and was not used in the final implementation.  The recommended procedure to perform this optimization is as follows. Firstly, a plant response will need to be identified by performing a frequency sweep of the system. This will provide an experimental model of the system, which will be more accurate than the theoretical model used in the report and include additional effects such as unexpected damping and friction, the motor driver bandwidth, and more. Once this response has been obtained, it will be necessary to modify the controller to optimize the performance of the system, and implement this modified controller on the NI card. The procedure used for tuning the controlled and synchronizing the motor is described in Chapter 2.  A.4.5.5 Dual Actuator System In order to prepare the system for independent, dual actuator operation it will be necessary to purchase and/or fabricate several additional components. These are listed below.  Small clevis mount  Large clevis mount  Linear actuator  Encoder  Belt/pulley system  Power supply  Motor driver  Male and female rod ends A.4.6 Conclusion Although a functioning prototype of the tilt platform was not constructed by projects end, it was felt that a successful and functional design had been achieved. All functional and performance requirements have been designed towards and a significant amount of analysis has been completed with the aim of verifying the chosen design.97  B Appendi x B: Re flex Study Impl eme ntation  Appendix B: Reflex Study Implementation This appendix presents the technical details of the superposition of head rotations on top of the balance simulation for the RISER system. In a first iteration of the reflex study experiment detailed in Chapter 3, several issues with the study design were identified after testing a number of participants. Consequently, the study was repeated after the issues were resolved. The two primary concerns that prompted the study repeat were ensuring that the superimposed head rotations took place about the actual location of the participant’s head, and preventing forces induced by the head rotations from affecting the balance simulation. The two identified issues are discussed in subsection B.1, and the solutions are presented in subsections B.2 and B.3. B.1 Pilot Study Implementation and Issues In the first implementation of the reflex study the sinusoidal head rotation was about a fixed location in world coordinates, meaning that as participants swayed forward or backwards on the robot, the position of the rotation relative to their head would change (Figure B-1). The superposition for this design was simple, purely adding the rotation angles of the balance simulation and head rotation together, and calculating the changes to RISER’s heave and surge position commands to locate the rotations correctly. The equations are shown in simplified form (neglecting roll and yaw angles, as they are zero in this study) in Equation (B.1) and Equation (B.2), below. Here, θb and θh represent the angles of the balance simulation and head rotation, respectively, while Zh, Zb, and Xh are the vertical and horizontal offsets from the Stewart platform’s centre of rotation to the centre of rotation of the balance simulation (about the ankle-tilt platform axis) and the head (about the interaural line). Note that there is no Xb as the horizontal position of the Stewart platform’s centre of rotation is the same as that of the balance simulation. H0 and S0 are the heave and surge of the Stewart platform when it is engaged and at rest. 𝐻𝑒𝑎𝑣𝑒 =  Z𝑏 ∗ (1 − cos(θb)) + Zh ∗ (1 − cos(θh))  + Xh ∗ sin(θh) + H0  (B.1) 𝑆𝑢𝑟𝑔𝑒 =  Zb ∗ sin(θb) + Zh ∗ sin(θh) + Xℎ ∗ (1 − cos(θh)) + S0  (B.2) In this implementation, it was possible that linear accelerations at the head could occur because the head rotations were fixed relative to world coordinates, not relative to head position. This was deemed unacceptable for the purposes of this study (which required a “pure” head rotation). In the balance 98  simulation, similar to normal standing, participants would typically stand at an angle approximately 3° forward from the vertical, as determined by an average of the standing angle for the freestanding trials of all participants. This natural standing angle will be referred to as θ. The tangential linear accelerations at the head ‘at’ can be approximated by Equation (B.3).  Figure B-1 – Head rotation-tracking method. Left: participant is shown standing vertically with both rotation points in the correct positions. Centre: Subject has leaned forward 5 degrees in balance simulation. Fixed frame head rotation is no longer in correct location, head experiences linear accelerations. Right: Head rotation is performed in the current frame, rotation point remains at the centre of head. 𝑎𝑡 ≈ (𝑍ℎ − 𝑍𝑏) ∗ sin(𝜃) ∗ 𝐴 ∗ (2𝜋𝑓)2  (B.3) For a typical vestibular height of 1.5m, the sinusoid amplitude of 0.12° (0.002 rad), and a frequency of 2.5Hz, this would induce approximately 0.04 m/s2. Similarly, centripetal head accelerations ‘ac’ can be approximated by Equation (B.4). 𝑎𝑐 ≈ ℎ𝑐 sin(𝜃) ∗ (𝐴 ∗ 2𝜋𝑓)2  (B.4) Which, for the same conditions, gives 8E-5 m/s2. While the centripetal acceleration is small enough to be negligible, it was possible that the tangential acceleration could be on the verge of significance. Goldberg et al. determined that the sensitivity of the irregular units in the otolith organs to sinusoidal accelerations 99  was approximately 200 spikes/s-g (Fernández & Goldberg, 1976). Using the tangential acceleration from above, the 2.5Hz case yields approximately 0.8 spikes/s. This is less than 1 spike per sinusoid, but that is only for a single unit. Consequently, it is possible that the linear accelerations could have affected the pilot study results. It was also identified that the rotation of the platform about the head had the potential to induce forces on the force plates due to the presence of the participant. The dynamics of the situation are difficult to model due to the presence of the backboard and seatbelts, which can impart accelerations on the participant. Instead, the implications of this issue were explored from an experimental standpoint. First, pilot participants were secured in the same position on the robot as they would stand for the reflex study. Without the balance simulation engaged, they were passively oscillated at the 0.125 Hz, 0.8125 Hz, and 2.5 Hz frequencies. This permitted measurement of the passively induced forces that were not a result of participant activation. As an example, the forces measured at each frequency for the passive rotations are shown in Figure B-2. As is evident in the figure, the induced forces are sinusoidal, and match the frequency of the rotation. Showing that the forces were present was a cause of concern, but if the forces were not affecting the EMG response, then they could be neglected. To test this, the balance simulation was engaged, and instead of superimposing a head rotation, the system was set up to superimpose the induced forces measured on the previous trial on top of the forces measured by the forceplate. This way, any EMG response observed at the stimulus frequencies could be attributed to these forces. The results for one participant are shown in Figure B-3. These results showed that the induced forces were indeed affecting the EMG responses, and so they would need to be removed from the torque measurements to prevent them from affecting the balance simulation.  100   Figure B-2 – Raw forceplate measurements from reflex study trials.   Figure B-3 – EMG responses for superimposed passive torques instead of head rotations. The presence of peaks at the stimulus frequencies (0.125Hz, 0.8125Hz, and 2.5Hz) indicates that the passive torques must be removed in real time, as they are affecting the output of the balance simulation (and consequently modifying the participant’s EMG response). 101  B.2 Head Tracking To resolve the issue of head accelerations, a head-tracking algorithm was implemented to ensure that the point of rotation always remained at or near the head’s position. There is a trade-off here, as increasing the rate at which the head rotation tracks the head could introduce instability into the system by interacting with the balance simulation. Instead of ensuring that the head rotation position is always exactly at the head’s centre, the goal of the head tracking algorithm is to prevent the rotation position from having a large or unilaterally biased offset from the target position (the centre of the head). Changes to standing angle set point during balance typically occur every 1-2 seconds, so it is not necessary for the head rotation position to change at a substantially higher rate. The position of the head changes as a function of the balance simulation’s angle, θb. As participants are secured to the robot, the distance from their head to the balance simulation’s centre of rotation remains fixed. This makes the transformation from the world frame to the balance simulation’s frame straightforward, as shown in Equation (B.5) and Equation (B.6). Zh1 and Xh1 represent the current Z and X position of the head in the world frame (fixed at the Stewart platform’s centre of rotation). 𝑍ℎ1 = (𝑍ℎ − 𝑍𝑏) cos(−𝜃𝑏) − 𝑋ℎ sin(−𝜃𝑏) + 𝑍𝑏  (B.5) 𝑋ℎ1 = (𝑍ℎ − 𝑍𝑏) sin(−𝜃𝑏) + 𝑋ℎ cos(−𝜃𝑏)  (B.6) The code was modified to calculate these updated positions in real time. The positions were mean filtered at 0.2 Hz to ensure that the calculated head position would gradually follow the actual head position without introducing any new system interactions. B.3 Forceplate Compensation As discussed above, the experimental setup negated the possibility of an accurate mathematical prediction of the forceplate torques. However, the gain and phase of the sinusoidal torques were found to be very consistent across trials with the same participant. As an example, pilot participant 1, on 4 separate 2.5 Hz passive rotation trials, had a mean 2.5 Hz torque amplitude of 1.99 Nm with a standard deviation of only 0.026 Nm. The phase lead from peak rotation position command to peak torque had a standard deviation of only 0.75°. With such a high repeatability it is possible to compensate for the torques by first rotating the participant passively, measuring the phase and gain of the torques, and then subtracting this sinusoidal torque from the measured torque before it is passed into the balance 102  simulation. As a proof of concept, a pilot participant was rotated passively once at each frequency. A best-fit sinusoid was found for the measured forceplate torques, and this sinusoid was subtracted in real time from the torque readings of a second set of passive rotations. An example of the before-subtraction and after-subtraction torque plot is shown in Figure B-4. In this case, the best-fit sinusoid to the compensated data was still 2.5 Hz; however, the amplitude was just 2% of the original value, a typical result across test participants. When this new sinusoidal torque was fed directly into the balance simulation, the induced motion produced no detectable peak in the EMG response.  Figure B-4 – 2.5 Hz torque response to passive rotations, before and after compensation. Amplitude of best fit sinusoid at 2.5 Hz is 2% of original amplitude.   103  C Appendi x C: Supple me ntary Results  Appendix C: Supplementary Results This appendix presents additional work and pilot studies, which offer supplementary information related to the work in Chapter 3. Much of this work was exploratory in nature. Therefore, drawing conclusions from these findings would be overstepping their limitations. Nonetheless, some of the findings are noteworthy or supportive of conclusions and decisions in the body of this thesis, and are presented here to inspire further research in the area. C.1 Height and Amplitude Perception Tuning In the development and pilot testing of the perception study described in Chapter 3, the original experiment permitted the participant to modify two variables: phase and amplitude of GVS. Participants could switch between these variables by clicking the right mouse button, and scrolling the mouse wheel would adjust the currently selected variable. In pilot testing, however, participants found gain difficult to tune and were reluctant to increase the gain (as it resulted in increased GVS amplitude). As higher gains increased discomfort for the participants, the gain adjustment was discarded from the final version of the study due to the possibility of systematic bias in the estimates. It was observed, however, that the GVS amplitude required to produce a sensation corresponding to the same amount of angular velocity increased with frequency, supportive of existing research in the field (Dakin, Luu, van den Doel, Inglis, & Blouin, 2010) and reflective of the results from the reflex study. In another perception study, which was only piloted, participants were secured with their head turned to the left, in the same manner as the reflex study from Chapter 3. Sinusoidal GVS and rotation were superimposed, and participants were asked to use the mouse to select a GVS amplitude, centre of rotation (vertical position that the platform would rotate about), and rotation phase in an attempt to eliminate any perception of motion. Participants could switch between variables through a right click of the mouse. This experiment also proved too difficult for participants, particularly due to the vibrations of the robot, and coupling between the three variables, which made it difficult for participants to determine which variable needed to be adjusted. This considered, all participants selected centre of rotation points close to the centre of the head (as opposed to their centre of mass or ankles), which supports the hypothesis that GVS is interpreted as a pure rotation about the head, without any linear acceleration components from stimulation of the otoliths.  104  This pilot was performed at 0.8Hz, which was also the frequency (0.8125Hz) in Chapter 3 at which the best reflex cancellation was observed when superimposing GVS to a pure head rotation. This provides further support that, at least for this frequency, sinusoidal GVS is perceived as a sinusoidal head-centred rotation. It is possible, however, that the perception of otolith motions does not align in phase with the perception of canal motions, meaning that only permitting participants to adjust accelerations by adjusting centre of rotation height may have biased the outcomes of this study. If participants were permitted to adjust the phase and amplitude of both linear acceleration and head rotations independently (a daunting 4-variable task), it is possible that the participants may have chosen a motion that involved linear accelerations. C.2 Perceptual Flip In another attempt to discern the perceptual gain of GVS, participants were blindfolded and secured with their head turned to the left, in the same manner as the reflex study from Chapter 3, and were initially stimulated with a sinusoidal head rotation and no GVS. Participants were then asked to indicate their feet’s direction of motion using their index finger. All participants were able to correctly indicate their direction of travel at all frequencies. Sinusoidal GVS was then applied with the phase participants selected to align their perceptions in the first (perceptual) study described in Chapter 3. With the new head orientation, the GVS induced perceptions of motion in the same direction (AP) as the head rotation, but with a phase that would, in theory, cause the two forms of stimulation to cancel. The amplitude of the sinusoidal GVS was then increased in a ‘staircase’ style until participants no longer correctly indicated their direction of travel. For the most part, participants were still able to identify when they were moving (as opposed to when they had no velocity at the ends of the sinusoid), however as the GVS stimulation increased they became more likely to indicate a direction of motion opposite to their actual physical motion. This indicated that the perception of the GVS was now overriding the participants’ perception of the physical motion. After three successive trials that showed a reversed direction, the GVS was stepped down until the participants indicated the correct direction of motion again. In this pilot study, participants had a very large variance and range at which the perceptual reversal occurred, and there was a wide band of gains for which they were unable to determine their direction. Even as the GVS was increased, participants would occasionally go back to indicating the correct motion before reversing again. This unpredictability resulted in a wide range of gain estimates, and required participants to remain on the robot for an unreasonable length of time to narrow down the estimate. 105  One contributing factor to the range of gains was that the average phase across all of the participants of the previous experiment was used as the GVS phase for all pilot participants in this experiment. This was performed because selecting a phase for each participant was a 4-hour study in itself, and not all previous participants were available to return for this study. Consequently, the phase between GVS and rotation may not have been completely 180 degrees for each participant, which would have a substantial impact on the gain required for cancellation. Nonetheless, this study provided several significant insights. It provided strong observational evidence that it was possible to cancel or reverse perception of motion using GVS, just with a large amount of variability. It also supported the notion of GVS being perceived as a head-centred rotation, as it was difficult for participants to detect any differences between GVS and physical rotation. C.3 Auditory Perception Phase Results While the direct comparison perception study yielded consistent phases across subjects (see Chapter 3), the relative phases between GVS and rotation did not conform to the hypothetical model derived from Goldberg’s work (Goldberg et al., 1982). In Goldberg’s work, however, GVS was applied to squirrel monkeys, and directly across the vestibular afferent nerves, while in humans the stimulation spreads broadly across the vestibular system. To investigate the possibility that the GVS transfer function in humans is different from that obtained by Goldberg, a second perception study was performed in which the phase of GVS and rotation were each independently measured relative to an audible metronome click. Five participants (M = 5) participated in this experiment, all of whom participated in the other perception study described in Chapter 3. The University of British Columbia’s Clinical Research Ethics Board approved all experimental procedures, and all participants provided written informed consent before participating. The procedure for this study was similar to that described for the perception study presented in Chapter 3 (see that chapter for a more detailed review of procedures). Six trials (three GVS, three rotations) were performed at 0.05 Hz, 0.1 Hz, 0.2 Hz, 0.4 Hz, 0.8 Hz, and 1.6 Hz (2.5 Hz was found to be too fast for participants to perceive reliably). In this study, a computer speaker and a condenser microphone were installed on the backboard immediately behind the participant’s head. A LabVIEW© subroutine was programmed to emit an audible 40 ms, 1.5 kHz pulse from the PC speaker, repeated at the same frequency as the stimulus. 106  The blindfolded participants were instructed to scroll the mouse, as before, until they perceived the sound as occurring simultaneously with their feet reaching the extent of their perception of motion (as far forward as possible for rotation, or as far to the left as possible for GVS). Participants could again request phase ‘bookmarks’ that they could return to, and were asked to report their confidence on a scale of 1-10 once they made their final selection. The condenser microphone was wired to the NI PXI-6229 DAQ board where the angular velocity and GVS signals were collected, and sampled at 2048 Hz. Once the participant selected a phase, the DAQ was used to record the metronome as well as the stimulus signal (angular velocity or GVS). The phase difference between the signals was encoded as the difference between the peak of the sound pulse and the positive zero crossing of the sinusoidal stimulation. In the second perception study, participants selected a phase lead so that their perception of either rotation or GVS would ‘align’ with the metronome click. Once they had selected a value, the signals were recorded to determine the relative phase between the metronome and the input signal. A least squares sinusoid fit was performed on the input signal, and a sinusoid was fit to the metronome data. An example of the data collected, and the sinusoid fitting, is shown in Figure C-1. 107   Figure C-1 – Example of data collected for indirect comparison of perception, and fitting of sinusoids. In this example, the phase difference of the sinusoids was 81 degrees. Participants in this study were generally consistent in their reported perceptions of angular motion, however all participants struggled to align the metronome signal with the GVS signal, often reporting that the sound was interfering with their perception of motion. Means, standard deviations, and reported confidence for rotation and GVS are summarized in Table C-1. Table C-1 – Mean selected rotation leads, standard deviations, and mean reported confidence. Leads are reported positive when metronome signal led input signal. Frequency GVS Lead (0) Confidence Rotation Lead (0) Confidence 0.05 Hz 3.5 ± 14.9 7.0 -8.3 ± 102.9 5.2 0.1 Hz 4.5 ± 19.2 7.3 -15.1 ± 86.1 5.3 0.2 Hz -0.9 ± 20.6 8.0 -17.2 ± 89.4 6.2 0.4 Hz 19.1 ± 44.5 8.0 24.6 ± 78.3 7.1 0.8 Hz 14.5 ± 56.4 7.6 23.0 ± 87.5 6.6 1.6 Hz 41.5 ± 51.4 6.1 1.9 ± 81.6 5.0  4 4.5 5 5.5 6 6.5 7 7.5 8-0.1-0.0500.050.10.150.2Time (s)Amplitude (V)Rotation to Sound Comparison, Subject 5, Trial 1, 0.8 Hz  Metronome SpikesMetronome Best FitAngular Velocity SensorVelocity Best Fit108  Figure C-2 left and Figure C-2 right shows the per-subject leads for rotation and GVS, respectively.   Figure C-2 – Phase leads of rotation and GVS relative to metronome signal. Left: Phase leads selected by participants when comparing audible metronome click to perception of angular velocity. A positive lead indicates that the metronome click occurred prior to the positive zero crossing of the angular velocity sinusoid. Right: Phase leads selected by participants when comparing audible metronome click to perception of GVS. A positive lead indicates that the metronome click occurred prior to the positive zero crossing of the GVS sinusoid. Figure C-3 left shows the combined results, for comparison with the results from the direct perception study, repeated in Figure C-3 right. 10-210-1100101-200-150-100-50050100150Selected Phase Leads of Audible Metronome Click Over GVS SinusoidSinusoid Frequency (Hz)Per Subject Selected Phase Lead of Metronome Click (deg)10-210-1100101-200-150-100-50050100150Selected Phase Leads of Audible Metronom  Click Over GVS SinusoidSin soid Frequency (Hz)Per Subject Selected Phase Lead of Metronome Click (deg) Velocity Sinusoid 109   Figure C-3 – Comparison of perception study results. Left: Inferred GVS to angular velocity leads from the independent GVS and rotation measurements in the metronome study. Right: Results from perception study in Chapter 3. Note vertical scale differences. While some participants in the metronome study showed the same trends that were observed in Chapter 3, the variance at any given frequency was quite large, and the sample size very small. Some participants showed similar trends to those from the perception study in Chapter 3, but the low participant confidence and high standard deviations make drawing any conclusions quite difficult. The standard deviations were higher primarily in the GVS trials. While a high standard deviation for both measures would likely indicate issues with the study design, a high standard deviation for just GVS indicates that participants were unable to reliably perceive the motion. As GVS amplitudes were the same as the first study, the GVS should be no more difficult to perceive in this study. The author hypothesizes that the most likely cause is that in this study, the participant was not moving at all while receiving the GVS. The absence of any motion, combined with the security of being strapped to a large platform, may have impeded participants’ abilities to reliably perceive the illusion of motion, when faced with reliable proprioceptive information confirming that they were not, indeed, actually moving.  Alternatively, it is possible that the close relationship between the auditory nerves and the vestibular nerves could cause an interaction between the vestibular stimulation and the audible click. Four participants verbally reported the metronome signal to be distracting to their perception, particularly during GVS, which could have resulted in the larger standard deviation during GVS trials. Further experiments could try other forms of external signals, such as a tactile pulse, to minimize interference with the vestibular system. 10-210-1100101-150-100-50050100150200Inferred Selection Of GVS Lead over Angular VelocitySinusoid Frequency (Hz)Per Subject Selected Phase Lead of GVS (deg)10-210-1100101-150-100-50050100Sinusoid Frequency (Hz)Per Subject Mean Selected Phase Lead of GVS (deg)Selected Phase Leads of GVS Sinusoid Relative to Angular Velocity: Perception Study

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0167625/manifest

Comment

Related Items