UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

A study of haptic icons Enriquez, Mario 2002

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

Item Metadata

Download

Media
831-ubc_2002-0389.pdf [ 12.27MB ]
Metadata
JSON: 831-1.0051677.json
JSON-LD: 831-1.0051677-ld.json
RDF/XML (Pretty): 831-1.0051677-rdf.xml
RDF/JSON: 831-1.0051677-rdf.json
Turtle: 831-1.0051677-turtle.txt
N-Triples: 831-1.0051677-rdf-ntriples.txt
Original Record: 831-1.0051677-source.json
Full Text
831-1.0051677-fulltext.txt
Citation
831-1.0051677.ris

Full Text

A STUDY OF HAPTIC ICONS by MARIO ENRIQUEZ B.Sc, Instituto Tecnologico de Monterrey, 1997 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES DEPARTMENT OF COMPUTER SCIENCE We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA October 2002 © Mario J Enriquez, 2002 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purpose may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis in whole or in part for financial gain shall not be allowed without my written permission. Thesis Title: A Study of Haptic Icons Author: Mario Enriquez (date) Department of Computer Science The University of British Columbia 2366 Main Mall Vancouver, BC, Canada V6T 1Z4 A B S T R A C T The goal for this research is to study a new class of force feedback applications based on abstract messages we call haptic icons. With the introduction of active haptic displays, a single knob or joystick can be used to control several different, sometimes non-related, functions. The functions associated with these multi-function handles can no longer be identified from one another by position, shape or texture differences. Haptic icons are brief programmed forces applied to a user through a haptic interface conveying an object's or event's state, function or content in a manner similar to visual or auditory icons. This thesis begins with a presentation of several tools that were developed to aid this research. It then describes a series of psychophysical tests designed to obtain the basic perceptual limits for our haptic interface. Knowing these perceptual limits is a prerequisite for proper haptic icon design. We analyzed a set of synthetically constructed haptic icons using Multidimensional Scaling, in order to discover the underlying perceptual processes in identifying different haptic stimuli. Results show that a set of icons constructed by varying the frequency, magnitude and shape of 2-sec, time-invariant waveforms map to two perceptual axes, which differ depending on the signals' frequency range, and suggest that expressive capability is maximized in one frequency subspace. I finish by proposing future work to be done on this area. Table of Contents Abstract » Table of Contents iii List of Figures viii Acknowledgements xi Chapter 1 l Introduction and General Methodology 1.1 Introduction 1 1.2 Methodology and General approach 2 1.3 Structure of Document 3 Chapter 2 5 Previous Work 2.1 Iconography 5 2.2 Applied Multidimensional Scaling Techniques 6 2.1 Haptics In General 8 Chapter 3 10 Experiment Setup and Standard Procedures 3.1 Equipment 10 iii 3.2 Hardware 10 3.3 Software 12 3.4 Pre-Experiment Procedures 13 3.5 Post-Experiment Procedures 13 Chapter 4 14 Step 1: Determining the Basic Perceptual Limits 4.1 Tools for Basic Perceptual Limits Testing 14 4.1.1 Two Alternative Forced Choice 14 4.1.2 Parameter Estimation by Sequential Testing 15 4.2 Detection Threshold Experiment 20 4.2.1 Methodology 20 4.2.2 Stimuli Selection 20 4.2.3 Considerations 21 4.2.4 Experimental Procedure 22 4.2.4.1 Participants 22 4.2.4.2 Stimuli Presentation 23 4.2.4.3 Instruction of Participants 23 4.2.4.4 The Experiment Session 24 4.2.5 Data Collection and Experiment Output 24 4.2.6 Results for Detection Threshold Experiment 25 4.2.7 Conclusions 30 4.3 Frequency Differentiation Experiment 31 iv 4.3.1 Methodology 31 4.3.2 Stimuli Selection 32 4.3.3 Considerations 32 4.3.3.1 Completeness of Cycles 32 4.3.3.2 Counting of Cycles 34 4.3.4 Experimental Procedure 35 4.3.4.1 Participants 35 4.3.4.2 Stimuli Presentation 36 4.3.4.3 Instruction of Participants 36 4.3.4.4 Experiment Session 36 4.3.5 Data Collection and Experiment Output 37 4.3.6 Results for the Frequency Differentiation Experiment . . . . 37 4.3.7 Conclusions 40 Chapter 5 43 Step 2: Mapping the Perceptual Space 5.1 Multidimensional Scaling Analysis Techniques 44 5.1.1 Introduction to Multidimensional Scaling Analysis (MDS) .44 5.1.2 Types of MDS 48 5.1.3 Data Gathering Techniques for MDS Tests 49 5.1.3.1 Traditional Direct Comparison Method 49 5.1.3.2 Icon Grouping Test Method 51 5.2 MDS Pilot Test 52 v 5.3 Multidimensional Scaling Experiment 1 54 5.3.1 Methodology 54 5.3.2 Stimuli Selection 54 5.3.3 Experimental Procedure 55 5.2.3.1 Participants 55 5.2.3.2 Stimuli Presentation 56 5.2.3.3 Instruction of Participants 56 5.2.3.4 The Experiment 57 5.3.4 Data Collection and Experiment Output 58 5.3.5 Results for MDS Experiment 1 59 5.4 Multidimensional Scaling Experiment 2 65 5.4.1 Methodology 65 5.4.2 Stimuli Selection 65 5.4.3 Experimental Procedure 66 5.4.3.1 Participants 66 5.4.3.2 Stimuli Presentation 67 5.4.3.3 Instruction of Participants 67 5.4.3.4 The Experiment 67 5.4.4 Data Collection and Experiment Output 67 5.4.5 MDS Results for Experiment 2 67 5.5 Individual Results for Experiments 1 and 2 78 5.6 Finding the Frequency With Most Expressive Capabilities 87 5.7 Discussion of MDS Experiments 88 vi 5.7.1 Experiment 2 Dimensionality 88 5.7.2 Participant Extremes 89 5.7.3 Interpretation of Dimensions 90 5.7.4 Maximizing Expressive Capacity through Frequency 91 5.7.5 Effect of Shape: Smooth vs. Jerky 92 5.7.6 The Range Effect - A Possible Explanation for Our Results . 92 Chapter 6 94 Conclusions and Future Work 6.1 Conclusions 94 6.2 Future Work 95 Bibliography 96 Appendix A Individual Results for MDS Experiment 1 99 Appendix B Individual Results for MDS Experiment 2 107 Appendix C Dissimilarity Matrices for Experiments 1 and 2 115 Appendix D Participant Instructions for Experiments 1 and 2 117 Appendix E Consent Forms 118 List of Figures Figure 1. The Haptic Display Motor 11 Figure 2. Participant arm position while grasping the haptic knob 12 Figure 3. Amplitude Changes and Steps of PEST Procedure 18 Figure 4. PEST operation using Rapid Adaptive Tracking 20 Figure 3. Participant involvement in Detection Threshold Experiment 24 Figure 6. Dialog box presented to participants for identification of interval holding the specified stimulus 25 Figure 7. Detection Threshold test results for the sine waveform 26 Figure 8. Detection Threshold results for the square waveform 27 Figure 9. Detection Threshold results for the triangle waveform 28 Figure 10. Detection Threshold results for the sawtooth waveform 29 Figure 11. Comparative Detection Threshold for all tested waveforms 30 Figure 12. Haptic sensation with complete cycles 34 Figure 13. Haptic sensation with incomplete cycles 34 Figure 14. Participant involvement in Frequency Differentiation Experiment.... 37 Figure 15. Frequency Differentiation test results for the sine waveform 39 Figure 16. Frequency Differentiation test results for the square waveform 39 Figure 17. Frequency Differentiation test results for the triangle waveform 40 Figure 18. Frequency Differentiation test results for the sawtooth waveform . . . 40 Figure 19. Frequency Differentiation Experiment result summary 41 Figure 20. Scree plot of stress values for MDS solutions in several dimensions . . 47 Figure 27.Interface used to gather similarity data for a test set using direct comparison . . 51 Figure 22. Two-dimensional MDS representation for the simple haptic icons used in the pilot study 53 Figure 23. Interface used to obtain the dissimilarity data for the haptic icons . . . . 57 Figure 24. Stress values for the average results of MDS Experiment 1 59 Figure 25. MDS Experiment 1 results 62 Figure 26. Projection of design parameters onto the Experiment 1 MDS solution. .63 Figure 27. Portion of Dissimilarity Matrix for sine waveform obtained in MDS Experiment 1 64 Figure 28. Stress values for the average results for MDS Experiment 2 68 Figure 29. Two-dimensional representation of the results for MDS Experiment 2 . 69 Figure 30. Projection of design parameters onto the two-dimensional Experiment 2 MDS solution 70 Figure 31. View 1 of the three-dimensional representation of the results for MDS Experiment 2 71 Figure 32. Projection of design parameters for view 1 of the three-dimensional representation of Experiment 2 MDS solution 72 Figure 33. View 2 of the three-dimensional representation of the results for MDS Experiment 2 73 Figure 34. Projection of design parameters for view 2 of the three-dimensional representation of Experiment 2 MDS solution 74 Figure 35. View 3 of the three-dimensional representation of the results for MDS Experiment 2 75 Figure 36. Projection of design parameters for view 3 of the three-dimensional representation of Experiment 2 MDS solution 76 Figure 37. Portion of Dissimilarity Matrix for sine waveform obtained in MDS Experiment 2 77 Figure 38. MDS Experiment 1 results for Participant 1 79 Figure 39. Projection of design parameters onto the Experiment 1 MDS solution for Participant 1 80 Figure 40. Experiment 1 stress values for Participant 1 81 Figure 41. MDS Experiment 1 results for Participant 2 82 Figure 42. Projection of design parameters onto the Experiment 1 MDS solution for Participant 2 83 Figure 43. Experiment 1 stress values for Participant 2 84 Figure 44. MDS Experiment 2 results for Participant 1 85 Figure 45. Projection of design parameters onto the Experiment 2 MDS solution for Participant 1 86 Figure 46. Average results for the Discriminability Analysis 88 Acknowledgments Many people have made my studies at UBC an unforgettable experience. I owe all of them my gratitude. I extend sincere appreciation to Dr. Karon MacLean, my faculty supervisor, for her valuable guidance and encouragement, reviewing my work and suggesting needed improvements. Much appreciation goes to Dr. Vince DiLollo for his insight and invaluable knowledge. I am also very grateful to Dr. Jim Little for his help and everlasting enthusiasm, which sheds off to everyone in the lab. Special thanks go to Dr. Jason Harrison for his help and comments on my writing and experiment design. Many thanks for the LCI lab community and particularly for the SPIN group, particularly, Tim Beamish, Michael Shaver and Giusi Di Pietro. I shall make a special thank you note for Valerie McRae for her help, support, humor and general T L C that go way beyond her duty. Finally, I would like to thank my family and friends who have had to listen to the words "haptic icons" a googolplex more times than they wanted. Mario Enriquez The University of British Columbia September 2002 xi Chapter 1 Introduction and General Methodology 1.1 Introduction Human-machine interaction has long used icons of different sorts to aid in identifying the different functions of a machine. These icons can be used instead of words to help keep the interface compact, concise and language-free. Traditionally, icons are small graphic representations of real things. These icons should be easily identified by the user and are used to relate specific functions to abstract physical controls. For example, $ can be placed next to a switch to let the user know its function is to turn on a light. A visual icon is defined as an image that represents an application, a capability, or some other concept or specific entity with meaning for the user [1]. In the world of natural touch we use characteristics such as shape, texture and relative position (from some fixed location to the user or other reference point) to identify different functions and states of knobs, switches and other controls. However, with the introduction of active haptic displays, a single knob or joystick can be used to control several different and sometimes unrelated functions which might now become indistinguishable from one another by position, shape or texture differences. "Haptic icons" offer a new challenge and opportunity. With the addition of haptic output, a controller (knob, button or joystick) also becomes a display, since it can be made to provide a different haptic behavior when manipulating different functions. For example, the hapticized knob might allow us to convey information to the user as to what function is currently being adjusted by the handle 1 without the need for a separate display. Thus, a car-stereo knob might inform the user if it is currently adjusting volume, bass or balance levels by displaying a different haptic behavior for each function. Another example is a cell phone buzzer that feels different depending on who is calling. These are only two of many ways that haptic icons might come in handy. The main topic for this thesis is the study of haptic icons. We will define haptic icons as brief computer-generated signals displayed to a user through force feedback with the role of communicating a simple idea in a similar manner as visual or auditory icons. Using haptic icons, we may be able to reduce the visual attention and cognitive load otherwise required to identify what the device is controlling at that time. We believe that these haptic icons will help the user control functions by touch and manipulation, leaving his/her visual system free to interact with their environment or visual displays. In this work, we present an innovative approach to study the perception of haptic icons, relying on an exploratory mathematical procedure called Multidimensional Scaling Analysis (MDS). The goal of this research is to understand and characterize haptic perception of complex stimuli in terms of key input dimensions for artificially generated haptic icons. 1.2 Methodology and General Approach There are several possible approaches to studying haptic icons. We present the following two and explain our choice: • Abstract / Perceptual. 2 Explore the underlying psychophysics of synthetic, complex haptic stimuli, with the goal of determining and generalizing the perceptual effect of various icon design parameters: for example, the discriminability and salience that result from the parameter settings used to produce a particular set of icons. In this approach, we are concerned simply with the user's ability to discriminate among stimuli, rather than associate meaning to it. It does not feel "like" anything - it feels different from that thing. • Applied / Semantic Through a process of iterative design and evaluation, create a set of haptic icons that seem naturally assignable to a corresponding set of functions in a specific interface (e.g., a stereo). For the study presented in this thesis we began with the abstract, perceptually based approach, with the assumption that the ability to associate icons with specific meanings must be eventually addressed. For this purpose we employed a single degree-of-freedom (DOF) haptic device (a knob) as being most representative of the level of haptic feedback that would be available in typical applications. Haptic icons can be defined as either passive or explorative; this refers to whether the users merely hold the haptic display and obtain information from the forces being exerted on them, or actively explore the virtual haptic world by voluntarily moving or pushing on the haptic display. For the scope of this thesis, we concentrate on passive haptic icons. Participants in the experiments will receive information from the haptic display without having to voluntarily move the display. 3 1.3 Document Structure We begin by presenting previous work in several related areas (Chapter 2) and explain several methodologies we adopted. Our study was carried out in 2 basic steps: Step 1: Determine the Basic Perceptual Limits (Chapter 4) Design, develop and perform basic psychophysical experiments to obtain the perceptual limits for our setup utilizing an adaptive procedure, Parameter Estimation by Sequential Testing (PEST) [2]. These limits are needed to create a balanced stimulus set for our formal MDS analysis. Step 2: Mapping the Perceptual Space (Chapter 5) The Multidimensional Scaling techniques and methodology used for our tests are explained in detail (Chapter 5.1). Experiment 1 (Chapter 5.3) Using the results from Step 1, we defined the haptic icons to be analyzed using MDS. We present the design and implementation of the MDS experiment for the chosen haptic icons and present the results obtained. Experiment 2 (Chapter 5.4) The MDS analysis for Experiment 1 (Chapter 5.3) raised several questions that we addressed with a second experiment using a re-designed haptic icon set. We present this re-designed haptic icon set, the MDS analysis performed on it and the results. We finish by presenting our conclusions and suggested future work (Chapter 6). 4 Chapter 2 Previous Work 2.1 On Iconography Visual and auditory icons have long been integral to computer interfaces, as a means of indicating functionality, location and other low-dimensional information more efficiently than does displayed text [3,4]. Graphic icons, for example, are small and concise graphic representations of real or abstract objects. The auditory and haptic icon design spaces share key attributes: they are both temporally sequential and have limits to amplitude and period discrimination abilities. Thus in our haptic icon research program, we have found it most productive to consider auditory icon design as an example. In the case of using audio to iconify information, there have been two general approaches: • Gaver et al. [5,6] studied "Auditory Icons". These are essentially representations of objects or notions that embody a literal, direct meaning: for example, using the sound of a paper being crushed to indicate deleting a computer file. Most of us are familiar with both the sound of a wadded paper and the action of deleting a file, and easily make the association. Auditory Icons are auditory representations of real objects and actions imitated by computer interfaces. • Stephen Brewster et al. [7,8] took a different approach. "Earcons" are sounds and rhythms with no intrinsic or cultural meaning: their target or meaning has to be learned to be effective. Their study focused on understanding and quantifying the 5 different "Earcons" that could be differentiated by people at a perceptual level. Amongst other things, they found that Earcons were better-differentiated than unstructured bursts of sound and that musical timbres were more effective than simple tones. Another important aspect was determining which sounds are more perceptually salient or even whether certain sounds are appropriate for a given application. For example, a quiet sound might literally represent a very urgent or dangerous event because that event does not generate much sound in the real world. However, in a different application, this might be inappropriate. This work differs from Gaver's work, which does not address whether users can differentiate the icons or how many icons they can tell apart. Our own study shares more philosophically with Brewster's; it relies on perceptual science and studies to determine where haptic icons should lie perceptually. However, we have a long-term aim of adding the intuitive benefits of Gaver's approach when we better understand the perception of complex haptic stimuli. 2.2 On Applied Multidimensional Scaling (MDS) Techniques In this work, we will be using Multidimensional Scaling to analyze the perception of haptic icons. MDS is a powerful tool for analyzing complex scenarios. It simplifies the understanding of complex preference data by uncovering potential hidden structure. A fairly detailed explanation of this technique is given in Chapter 5.1. Many studies have been carried out in different disciplines utilizing this technique. We looked at several of them to observe their different approaches, ideas and pitfalls. 6 Mark Hollins et al. analyzed the tactile perception of real surface textures using MDS [9]. These textures were presented by moving them across the index finger of the subjects who sorted them into categories on the basis of perceived similarity. Their test set consisted of 17 textures such as wood, sandpaper, and velvet. They obtained results mapped into a 3-dimensional space. Two axes were roughly associated with hard/soft and rough/smooth; the third was difficult to interpret. Their work shows an interpretation of the results based on the groupings in the MDS solution space that we adopted for our interpretations. Aleksandra Mojsilovic et al. utilized MDS in a set of experiments on the visual perception of color patterns [10]. In this work twenty patterns from an interior design catalog were presented to the participants. The stimuli were rated on a scale from 0 for "very different" to 100 for "very similar". No instruction was given to the participants concerning the characteristics on which these similarity judgments were to be made. This allowed the experimenters to discover what characteristics were important for the participants in differentiating between the textures. The results obtained mapped onto a 2-dimensional space. The two axes were associated with "color" and "color purity" respectively. From this work, we adopted the general approach of giving the participants freedom as to the characteristics to consider forjudging the stimuli. Lawrence Ward used MDS to study the perception of a set of pictures containing images of different natural and artificial (human generated) environments [11,12]. For his work, Ward utilizes an innovative approach to obtain the dissimilarity data for the MDS analysis. Participants in these experiments were asked to rank the images five times using a different number of categories for each sort. The perceived dissimilarity for the picture set is calculated based on these sortings. The results obtained mapped to a space where one axis 7 represented "naturalness" and another "scale". The sorting methodology used in this work increases the efficiency of evaluating large sample sets, and improves repeatability and accuracy by avoiding the need to judge each item pair individually. We utilized this sorting methodology for our own experiments. Terri L. Bonebright utilized MDS for a different task: a study of everyday sounds [13]. In this work, the dissimilarity data required for the MDS tests was obtained using a similar approach to the category sorting method used by Lawrence Ward. The participants were asked to sort the sound stimuli presented into categories. There were 74 different stimuli. Each stimulus had a visual representation on the screen that was used to perform the sorting into categories. The visual representation was a small graphical icon in the shape of a randomly numbered box, which the participant could repeatedly test and move about. The use of graphic icons allows the analysis of large data sets without requiring exceedingly long tasks. The results for these tests were mapped in a 3-D solution space where dimension 1 is defined by 5 perceptual attributes and 3 acoustic measures; dimension 2 is explained by 1 perceptual attribute and 1 acoustic measure; and finally dimension 3 is characterized by 3 perceptual attributes and 1 acoustic measurement. We adopted the idea of representing the stimuli as icons on the screen to allow graphic sorting for our own experiments (Chapter 5). 2.3 On Haptics in General With few exceptions, haptic force feedback research has been devoted to rendering virtual environments. To our knowledge, synthetic haptic stimuli have not been studied in the context of supplying iconic information prior to the research presented here. There are, however, a great number of papers and research articles that focus on haptic feedback 8 applications, their design and use. We were particularly interested in those that addressed some simple questions that we felt were relevant to our work. The work done by Martha A. Rinker et al. [14] was particularly interesting to us. Their study focused on determining the sensory capacity of a human hand to perceive haptic stimuli through the fingers, in particular the ability to discriminate finger movements along the dimensions of amplitude and frequency for the purpose of processing speech stimuli. Their study involved determining the difference thresholds for a series of stimuli using an adaptive, two-interval temporal forced choice procedure. The results obtained from this experiments show thresholds in a range from 10% to 18% for amplitude discrimination and 6% to 16% for frequency discrimination. We adopted the general methodology followed in these tests to determine the basic perceptual limits for our specific setup (Chapter 4). Hong Z. Tan et al. performed a series of experiments to determine human perceptual capabilities for parameters such as pressure, stiffness, position sensing resolution, force sensing, human force control and force output range and resolution [15]. Their idea was to provide simple rules to be used as a guideline for the mechanical requirements in the design of haptic interfaces. Of particular interest to us, the Just Noticeable Differences (JND) for force sensing of slowly varying forces is presented as being 7%. The JND for vibro-tactile stimulation is roughly 28 dB below 30 Hz and decreases at a rate of roughly -12 dB/octave from 30 to 300 Hz. After that, the threshold rises again. We used the results presented here as a rough guideline to authenticate the experiments done for our own system. 9 Chapter 3 Experiment Setup and Standard Procedures All of the experiments carried out throughout this work and described in Chapters 4 and 5 utilize the same haptic display, setup and general procedures. We proceed to explain them in detail here. 3.1 Equipment All participant sessions took place in a quiet corner of the UBC Computer Science Department's "Laboratory for Computational Intelligence". 3.2 Hardware A generic 1.2 GHz Pentium 3 computer running Windows 2000 in real-time mode was used for all experiments. An Immersion Impulse Drive I/O Board v. 1.0 provided I/O and amplification for the haptic interface. The haptic interface was a direct drive actuated knob, shown in Figure 1. The knob was rubber-covered brass with an outer diameter of 10.5 mm and a length of 16.5 mm. The 1 mm thick rubber coating prevented slipping and allowed a better grip of the knob while minimizing compliance. The knob was mounted directly on the shaft of a 20-W Maxon DC motor model 118752. This 24-volt motor has a stall torque of 240 mNm and a mechanical time constant of 5 ms allowing a maximum frequency output of 200 Hz. A Hewlett Packard model HEDS-5500 optical encoder with 4000 post-quadrature counts per revolution provides positional feedback. 10 All force magnitudes presented through our haptic knob are expressed as peak-to-peak torque values in mNm (Newton/millimeters). These torque values are approximate, derived from the electrical current multiplied by the motor torque constant, since we could not easily measure displayed absolute force. Figure 1. The haptic display motor. Shown here is the DC motor used as a haptic display. The knob on the motor shaft was covered with 1-mm thick rubber to prevent slipping and facilitate grasp during the experiments. The adjustable vise allowed us to re-position the motor for each participant. The motor/knob assembly was held horizontally with an adjustable vise on a table. A padded arm support was built to hold the participants' forearm comfortably while performing the experiments. The purpose of this armrest was twofold; the participants remained comfortable even for the one-hour long tests and they were forced to grasp the 11 knob in a specific manner. This setup was designed to provide the participants with a single comfortable grasping position of the knob (Figure 2). Figure 2. Participant arm position while grasping the haptic knob. The participants were asked to hold the knob lightly between their thumb and index finger as shown here. The haptic knob was placed horizontally in a position easily reachable by the participant while resting the forearm on a cushioned armrest. 3.3 Software The experiment control GUI and the I/O routines were written by the author in Visual Basic and C++ respectively. The I/O board control routines were developed based on sample code provided by Dr. Karon MacLean and Immersion Co. The experiment results were stored in *.CSV (Comma separated value) files that can be read using any spreadsheet software or text editor. 12 3.4 Pre-experiment procedures Prior to beginning each experiment, participants signed a consent form as required by the university ethics review board. The form gave the participants some basic information about the type of experiment they were to perform, as well as an outline of the payment and withdrawal conditions. A copy of this form can be found in Appendix E. After receiving instructions about the task to be performed, participants were required to wear noise-canceling headphones to block any audible artifacts from the haptic display. 3.5 Post experiment procedures At the end of every session, the participants' questions regarding the purpose and application of the experiment were answered. Participants were paid $10 per hour spent in the experiments. 13 Chapter 4 Determining the Basic Perceptual Limits In order to effectively compare the perception of haptic icons composed by varying multiple haptic signal parameters, we must first determine the perceptual limits for those signal parameters in isolation when displayed using our haptic hardware. We performed a series of experiments to determine what we considered to be the two most important perceptual parameters for our setup: • the minimum amplitude required to detect the presence of a haptic stimulus for a set of different frequencies and waveforms (Chapter 4.2) • the minimum detectable change in frequency for haptic stimuli generated using different frequencies and waveforms (Chapter 4.3) 4.1 Tools for Basic Perceptual Limits Testing The experiments used to determine the perceptual parameters for our setup required the use of several different methodologies and techniques. These are presented in Chapters 4.1.1 and 4.1.2. 4.1.1 Two Alternative Forced Choice (2 AFC) When presenting a stimulus to participants with the purpose of inquiring whether it holds a specific characteristic, the responses obtained might be affected by differences in the participants' personalities. Because the task is hard, there is always uncertainty as to what was there or not. Either there was a signal (signal plus noise) or there was none (noise 14 alone). There are four possible outcomes for this type of test: hit (signal present and participant says "yes"), miss (signal present and participant says "no"), false alarm (signal absent and participant says "yes"), and correct rejection (signal absent and participant says "no") [16]. One problem with this approach is that the participants are asked to use their own criteria in making the decision. Different participants might have a different idea of what the researcher is asking for. Daring participants might claim to feel the particular characteristic when they are not sure, conversely, timid participants might be more inclined to conservative judgments and require a positive identification before declaring they perceive the characteristic. Another approach to measure performance independently from any participant criterion is to use a two-alternative forced choice method (2AFC). This technique avoids these response biases and prevents obtaining different results for conservative and liberal observers. 2AFC eliminates such problems by avoiding the need to answer a "yes-no" question, presenting instead 2 non-intimidating options: the characteristic was present in the first stimulus or in the second stimulus. For our experiments, this method is implemented by presenting two intervals, only one of which contains a signal. The participant is then asked which interval holds the signal. The order of the intervals is randomized so the participant has to identify the location of the stimulus in order to answer correctly. We utilized this methodology for all of our psychophysical experiments (Chapter 4.2 and 4.3). 15 4.1.2 Parameter Estimation by Sequential Testing (PEST) The Two Alternative Forced Choice (2AFC) methodology explained in Chapter 4.1.1 is essential for the implementation of Parameter Estimation by Sequential Testing or PEST [16]. PEST is an adaptive procedure for rapid and efficient psychophysical testing used extensively in our experiments. It allows the researcher to determine the level of an independent variable that leads to some predetermined probability that a related event will occur on a single discrete trial. An example is the signal loudness required for a participant to hear a beep in a quiet room 75% of the times the signal is presented. Here, for example, we used it to determine the lowest possible amplitude at which a specific waveform with certain frequency will be detected 80% of the time when displayed using our haptic knob. PEST was designed to obtain information using trial-by-trial sequential 2AFC decisions at each stimuli level in a sequence converging on a selected target level. Let us suppose we are trying to determine the amplitude at which a sine waveform with a specific frequency is detected 80% of the time when displayed through the haptic knob. We want the signal level to be as low as possible but still large enough to be noticed by the human observer 80% of the time it is displayed. PEST allows you to determine the signal intensity at which the participant will be able to determine the "location" of the stimulus with a predetermined probability level, typically 80% or greater. PEST is an adaptive procedure, which means its behaviour is continuously affected and determined by the ongoing test. The system follows a set of steps that eventually lead to the determination of a stimulus intensity that meets a specified criterion for detecting its presence. 16 PEST OPERATION PEST begins by presenting the participant with two intervals in random order: • One of them is a signal large enough so it is easily noticeable. • The other is a "blank" signal. Following this, the participant is asked for the "location" of said signal amongst 2 interval choices. This constitutes a single trial. After every trial, PEST takes the collected responses for the previous trials and decides whether they are statistically better or worse than a pre-specified percentage of correct responses (80% for our example). If the results are statistically inconclusive, PEST runs another trial at the same level. If the results merit a change, PEST specifies the amount of change to be made according to the following rules: For our scenario: 1. If the results are above the desired correct response percentage (test too easy), the amplitude of the waveform to be presented is decremented by a specified step size. This should make the test harder. If the results are above the desired correct response percentage for two trials in a row, the step size is doubled for the next change to be made. If the previous results had determined that the test was too hard, the step size is halved before changing the amplitude of the waveform. 2. If the results are below the desired correct response percentage (test too hard), the amplitude of the waveform is incremented by a specified step size. This should make the test easier. If the results are below the desired correct response percentage for two trials in a row, the step size is doubled for the next change. If the previous 17 results determined the test was too easy, the step size is halved before changing the amplitude of the waveform. The initial value for the step size is pre-defined by the researcher and is dynamically altered by PEST thereafter. Figure 3 shows a sample graph of the amplitude changes ordered by PEST to eventually converge at a level (0.34 for this example) that meets the percentage of correct responses specified for the task. PEST Operation 0.9 0.8 ~ 0.7 E i °-6 % 0.5 1 0.3 •SP K 0.2 0.1 0 ; j | If i \ —L-L It \ T L L "T 1 r j I [ • V i i 1 i—r T T i i — i ] 1 \ 1 ,,.} L ~ i r L 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 Trials Figure 3. Amplitude Changes and Steps of PEST Procedure. The vertical axis represents the amplitude of the stimulus presented. The horizontal axis represents the sequential trials. This graph presents the amplitudes presented by PEST in 48 trials converging to a final value of 0.34 mNm. PEST repeats the previous steps: presenting the stimulus, collecting data and possibly changing the stimulus amplitude, until a certain criterion is met. There are two methods for deciding when to stop PEST: 1. Minimum Overshoot and Undershoot Sequential Estimates (MOUSE) 18 This method works by monitoring the step size magnitude ordered by PEST. When the step size is smaller than a pre-specified minimum, the process stops. The result for the test is calculated by averaging the values at the last 5 reversals in amplitude change direction. A reversal occurs when PEST commands a change in stimulus level using a step size in a direction opposite to the previous one (Figure 4). 2. Rapid Adaptive Tracking (RAT) R A T monitors the number of reversals in amplitude change direction. This method stops PEST after a specified number of step size reversals plus a specified number of additional trials. After the 5th reversal, the participant is presented with the stimuli 16 times and the values obtained from these 16 trials are averaged to get a final result (Figure 4). For our experiments we used the RAT criterion to decide when to stop the test. PEST was used in both of our perceptual limits experiments (Chapters 4.2 and 4.3). 19 Pest Operation ' " [••]• ' —-j | • Reversal \ i , j T T 1 X « r 4 - It iTi rial s A ' i III 1 * ' — -1 L j L 1" \ k Reversal "H—i—i—i—i—i—i—i—4* J ! ! I i i Reversal • - M - 4 i i i i i: 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 Sequential Trials Figure 4. PEST operation using Rapid Adaptive Tracking. For this example, Rapid Adaptive Tracking (RAT) was used to determine the target level. The final value is the average (0.34 for this example) of the 16 trials following the fifth reversal in magnitude change direction. 4.2 Detection Threshold Experiment In order to properly quantify what humans can perceive through our haptically enabled knob, we must first understand humans' perceptual limits. As with all physical systems, humans have different gains and sensitivity levels for varying frequencies, amplitudes and waveforms [17]. The purpose of the Detection Threshold (DT) experiment is to determine these sensitivity levels for our specific setup, i.e., for a set of simple waveforms displayed through the knob. Obtaining these thresholds is a prerequisite to determining appropriate amplitude levels for the stimuli to be used as haptic icons. Given these requirements, we designed and completed a set of experiments to collect data from participants regarding their perceptual limits to a set of haptic sensations. 20 4.2.1 Methodology Participants were presented with the haptic stimuli using a 2 Alternative Forced Choice (2AFC) paradigm. The stimuli were presented in two intervals: one that held a haptic sensation with a given frequency and amplitude and a second "blank" interval. The order of the intervals was randomized. The participant was asked to identify which of the intervals contained the haptic sensation. 4.2.2 Stimuli Selection In our experiments we used a combination of frequencies and simple waveforms to generate the haptic sensations to be tested. The stimuli presented consisted of 4 waveforms: sine, square, triangle and sawtooth. These waveforms were presented at different frequencies and amplitudes throughout the experiment. The hardware used (Chapter 3) limits the test frequencies to those 200Hz and under. We selected frequencies for this experiment to be spread over the frequency range allowed by our existing setup. The four waveforms were presented at 10 different frequencies: 0.1, 0.25, 0.5, 1,7,10, 20, 40, 100 and 200 Hz. This gives us a total of 40 stimulus detection levels to be determined. The duration of the stimuli to be used was set at 10 seconds per interval for each trial. All force magnitudes presented through our haptic knob are expressed as peak-to-peak torque values in mNm (Newton/millimeters). These torque values are approximate, derived from the electrical current multiplied by the motor torque constant, since we could not easily measure displayed absolute force. 21 4.2.3 Considerations To be able to effectively measure the participants' sensitivity at a given frequency, we had to confirm that the intended signal was the only factor determining the participants' responses. Due to physical (electrical) limitations of our setup, a small magnitude "bump" movement is generated on our haptic knob anytime the output device is enabled. To prevent this noise from having a negative effect on our results, we present the participant with two stimuli: a blank presentation and the stimulus that holds the specified waveform. For both cases, the output device is enabled. This causes the participant to feel the "bump" at both intervals and thus makes the judgment whether there is a specific sensation present in the stimulus more precise. The duration for each stimulus was chosen to be 10 seconds as to allow a single complete cycle for the lowest frequency to be tested (0.1 Hz); the "blank" stimuli had a 10 second duration as well. The presentation order for the stimuli was randomized. 4.2.4 Experimental Procedure A complete session yields the Detection Threshold for a specific waveform at all 10 selected frequencies. Each session was divided into 10 blocks, each with a duration of approximately 8 minutes. The Detection Threshold for a single frequency is determined in each block. This frequency is presented in a number of sequential trials with different force amplitudes as determined by the experiment system. We used Parameter Estimation by Sequential Testing (PEST) to determine the Detection Threshold. PEST adaptively determines the number of trials required for each block. All trials were self-paced with breaks allowed between blocks and when the participant felt the need for relaxing the grip 22 on the knob. Participants were told that the experiment was not timed. A complete session had a duration of approximately 1.5 hours. 4.2.4.1 Participants Due to the nature of this experiment, which focuses on determining basic perceptual limits, we recruited participants without regard to background or abilities on the assumption that basic tactile perceptual abilities would be consistent across the general population. For this experiment, six participants were recruited through a posting in a bulletin board in the computer science building. The participants included one female and five males. One was left-handed and five right-handed, all of them between 20 and 29 years of age. None of the participants had any disabilities or limitations in the sight or touch senses. The participants were naive to haptic perceptual experiments and were paid cash for their involvement. Figure 5 shows a table explaining the details of the participants' involvement in the DT experiments. Three participants completed a session for each waveform tested; and some participants completed more than one session. 23 Part ic ipat ion in Detect ion Thresho ld Exper iments Sine Tr iangle Square Sawtooth Part ic ipant W a v e f o r m W a v e f o r m W a v e f o r m W a v e f o r m 1 X X X X 2 X 3 X X 4 X X 5 , X X 6 X Figure 5. Participant involvement in Detection Threshold experiment. The X denotes participation for the experiment in the respective column. Each experiment was completed by 3 participants. 4.2.4.2 Stimuli Presentation In each trial, two randomly-ordered 10-second intervals were presented: a haptic sensation and a "blank" presentation. During the stimuli presentation, the screen was cleared and the participants were required to wear noise canceling headphones to ensure the haptic stimuli was the only factor determining their responses. 4.2.4.3 Instruction of Participants Each session began by presenting a set of simple instructions on the screen. The instructions were then read to the participant by the experimenter. Copies of these instructions are included in appendix D. After making sure that the participant was seated comfortably and the haptic display was adjusted to be within easy reach, we observed the participant interact with the system and answered any questions. Since PEST is an adaptive procedure, the first couple of trials can be used for training without affecting the end results of the experiment. 24 Participants were instructed to perform the first two trials under our supervision. After these trials, the participants were again asked if they had any questions. Having answered any further questions, the participants were required to wear the noise canceling headphones and continue with the experiment. After each block the participants were reminded by the experiment software to take a break. 4.2.4.4 Experiment Session For the duration of the session, the participants were asked to determine the location of a stimulus amongst two intervals. During the presentation of the stimuli, the screen was blank. After each stimulus pair, the participants were presented with a dialog box with two options: the stimulus was located in interval one or in interval two (Figure 6). The participants answered by typing 1 or 2 in the keyboard and pressing the <ENTER> key or clicking on the <OK> button to continue to the next trial. The <Cancel> button had no functionality. Figure 6. Dialog box presented to participants for identification of interval holding the specified stimulus. This dialog box was presented to the participants after each trial. The participants had to type their response using the keyboard. 4.2.5 Data Collection and Experiment Output The data gathered from each session was stored into individual files. The name of these files was formed by combining the date and time of the session with the participant's name. 25 The file contained the waveform being tested, the participant's name, age and gender and the Detection Threshold results. These results were stored as 10 values that represent the amplitude required for detection of a signal at each frequency tested. The results were graphed using Microsoft Excel. 4.2.6 Results for the Detection Threshold Test The Detection Threshold Experiment results are displayed in five graphs (Figures 7 to 11). The results for each waveform tested (sine, square, triangle and sawtooth) are shown in four separate graphs, a fifth graph presents a comparative of the results. Figure 7 shows the average results of the Detection Threshold for the sine waveform. The error bars represent the standard deviation of the 3 samples taken. There is a noticeable increase in sensitivity for the higher frequencies. We can also notice that standard deviation decreases with the increment in frequency. DT for Sine Waveform o o 1 o 0 -> e D r o 9-0.1 1 10 100 1000 Frequency (Log10 Hz) Figure 7. The Detection Threshold experiment results for the sine waveform. Each point in the graph represents the average for 3 participants. The graph represents the average sensitivity to the haptic sensations presented at the different frequencies. The horizontal axis is the frequency (scaled in logio) being tested and the vertical axis is the force output required for detection. The error bars are the standard deviation of the samples taken. 26 Figure 8 shows the average results of the Detection Threshold for the square waveform. The error bars represent the standard deviation of the 3 samples taken. There is a subtle increase in sensitivity for the higher frequencies. However, the change is smaller than for the sine waveform (Figure 7). We can observe the standard deviation decreases with the increment in frequency. Note that the scale of the vertical axis in this graph is different from that used in Figure 7. DT for Square Waveform o o O. i n 1 0 1 6 ^ A •4 rt • 1— 0.6 •—0.4-e-X 0.1 1 10 100 1000 Frequency (loglO Hz) Figure 8. Detection Threshold results for the square waveform. Each point in the graph represents the average for 3 participants. The graph represents the average sensitivity to the haptic sensations presented at the different frequencies. The horizontal axis is the frequency (scaled in logio) being tested and the vertical axis is the force output required for detection. The error bars are the standard deviation for the samples taken. Note that the vertical scale of the graph is different from that of Figure 7. 27 Figure 9 shows the average results of the Detection Threshold for the triangle waveform. The error bars represent the standard deviation of the 3 samples taken. This graph is similar in shape and magnitude to that of the sine waveform. There is a noticeable increase in sensitivity for the higher frequencies. Note that the scale of the vertical axis in this graph is different from the one used in Figure 8. DT for Triangle Waveform o a 1 p o 7 / < fc fi Q O n \ 9- — • 0.1 1 10 100 1000 Frequency (Iog10 Hz) Figure 9. Detection Threshold results for the triangle waveform. Each point in the graph represents the average for 3 participants. The graph represents the average sensitivity to the haptic sensations presented at the different frequencies. The horizontal axis is the frequency (scaled in logio) being tested and the vertical axis is the force output required for detection. The error bars are the standard deviation for the samples taken. Note that the vertical scale for the graph is different from that of Figure 8. 28 Figure 10 shows the average results of the Detection Threshold for the sawtooth waveform. The error bars represent the standard deviation of the 3 samples taken. This graph is similar in magnitude to that of the square waveform. There is a subtle increase in sensitivity for the higher frequencies. However, the change is smaller than for the sine and triangle waveforms (Figure 7 and 9). Note that the scale of the vertical axis in this graph is different from those used in Figures 7 and 9. DT for Sawtooth Waveform CL F sm-i ' 1 nn I .ou 1 60 1 40 \ 1 °0 \ 1 nn • \ I -UU \ . n n n * ff\ R n U.Du • u._u 9r©9-0.1 1 10 100 1000 F r e q u e n c y (Iog10 Hz) Figure 10. Detection Threshold results for the sawtooth waveform. Each point in the graph represents the average for 3 participants. The graph represents the average sensitivity to the haptic sensations presented at the different frequencies. The horizontal axis is the frequency (scaled in logio) being tested and the vertical axis is the force output required for detection. The error bars are the standard deviation for the samples taken. Note that the vertical scale of the graph is different from that of Figures 7 and 9. Figure 11 shows a comparison of the results obtained for the Detection Threshold experiment with all four waveforms. In this graph, we can see two trends for the results: • The results for the sine and triangle waveforms both show a noticeable increase in sensitivity with the increasing frequency. 29 • The graph for the square and sawtooth waveforms show an increase in sensitivity at the higher frequencies as well; however, this increment is subtle. Overall, the participants have a higher sensitivity at the lower end of the frequency range for the square and sawtooth waveforms. At the higher end of the frequency range, the sensitivity levels are similar for all four waveforms. There is an apparent increase in the force required for detection for all waveforms at 1 Hz. We have no clear explanation for this phenomenon. DT Results E z E cr CD E < ?-i L 6 ' < s A ^ 1—V—4— —• 2— ——-— * - * 1. I , 0.1 10 F r e q u e n c y (Hz) 100 1000 Figure 11. Comparative Detection Threshold for all tested waveforms. The participants have a higher sensitivity at the lower end of the frequency range for the square and sawtooth waveforms than for the sine and triangle waveforms. At the higher frequencies tested, the sensitivity levels are similar for all four waveforms. 4.2.7 Conclusions Overall, the detection threshold for the square and sawtooth waveforms is lower than that of the sine and triangle waveforms particularly at the low end of the frequency range. We believe this is due to the abrupt changes in force magnitude when displaying the square and sawtooth waveforms. The sine waveform force graph has a smooth 1st derivative and the 30 triangle a continuous albeit sharp-edged derivative. On the other hand, the square and sawtooth have impulse functions as 1st derivatives. This high frequency component is more easily perceived when the stimulus presented has a low fundamental frequency. There still is a lower detection threshold with the increasing frequencies for these discontinuous waveforms. The sine and triangle waveforms were perceptually very similar to each other. This prompted us to avoid using the triangle waveform in our icon set for the formal MDS analysis. As expected, participants were more sensitive to the icons with higher frequencies for our chosen stimulus set. Due to hardware constraints we were unable to generate frequencies beyond 200Hz. Previous work on vibro-tactile perception [18] suggests that the frequency response graph for vibro-tactile stimuli should begin to drop-off beyond 800 Hz. We were unable to corroborate these results due to the limitations in our hardware. At the frequencies allowed by our existing setup (0-200Hz), the results obtained match previous work in the area of vibro-tactile perception quite accurately. The participants were most sensitive to the highest frequency used for our experiments, 200 Hz. 4.3 Frequency Differentiation Experiment Frequency is a key ingredient of simple haptic sensations. In order to determine the correct frequency spacing for the different haptic icons, we must first determine which frequencies humans are able to differentiate from one another. Therefore, after determining the Detection Threshold (Chapter 4.2) for the waveforms to be used in the haptic sensations, we selected an easily detectable amplitude to obtain the Difference Threshold for frequency. 31 The Difference Threshold (or "Just Noticeable Difference") is the minimum amount by which a stimulus characteristic must be changed in order to produce a noticeable variation in sensation. The purpose of the Frequency Differentiation experiment is to determine the just noticeable difference (JND) in frequency between haptic sensations with a fixed amplitude and waveform. We have completed experiments to collect data from participants to determine their perceptual limits with a set of haptic sensations with differences in frequency. 4.3.1 Methodology Participants are presented with haptic stimuli using a 2 Alternative Forced Choice (2AFC) paradigm similar to the one used for the Detection Threshold (DT) experiment (Chapter 4.2). The stimuli are presented as four intervals divided into two interval pairs. Three of the intervals hold identical stimuli (Control Stimuli). One of the intervals holds a stimulus with a higher frequency than the other 3. All stimuli have the same amplitude and waveform. The order of the intervals is randomized. The participant is asked to identify which of the interval pairs contains the different haptic sensation. 4.3.2 Stimuli Selection The haptic sensations used for this experiment consist of four waveforms: sine, square, triangle and sawtooth presented at 6 base frequencies: 0.5, 1, 5, 10, 50 and 100 Hz at a fixed amplitude of 24.53 mNm. These frequencies were selected to spread over the frequency range allowed by our existing setup. The amplitude selected is twice the Detection Threshold for the least sensitive waveform-frequency combination as determined by the DT experiment (Chapter 4.2). 32 Each base frequency was presented in the control stimuli (the 3 identical stimuli). The initial frequency for the stimulus containing a difference (Test Stimulus) was set to be 150% of the base frequency being tested. For example, when testing a sine waveform at 50Hz., the control stimuli all present the same sine waveform at 50Hz. The test stimulus initially presents a sine waveform with a frequency of 75F£z. The experiment software dynamically adjusts the frequency for the test stimulus thereafter. All force magnitudes presented through our haptic knob are expressed as peak-to-peak torque values in mNm (Newton/millimeters). The duration for all the stimuli to be used was set at 2 seconds per interval as to allow a single complete cycle for the lowest frequency to be tested (0.5 Hz). 4.3.3 Considerations In order to properly test the participant's ability to distinguish differences in frequency we had to address several issues: 4.3.3.1 Completeness of Cycles When presenting a stimulus of fixed duration but changing frequency, the resulting output of the haptic device could be introducing un-wanted noise when the waveform is cut to fit the desired timeframe (Figures 12 and 13). If for example we wish to present a sine waveform with a frequency of 0.6 Hz. and duration of 2 seconds, we end up presenting 1.2 cycles of said waveform. In this case, the force being displayed at the end of the stimulus ends abruptly (Figure 13). This presents the participant with an unintentional difference in feel between the stimuli. Our goal is to test the differentiability of the haptic sensations based on frequency differences alone, so this must be avoided. 33 Figure 12. Haptic sensation with complete cycles. Both the starting and ending forces displayed for this haptic sensation are neutral. Force Amplitude 13 D. E < Time Abrupt Change in Force Figure 13. Haptic sensation with incomplete cycles. This graph presents an incomplete cycle; the problem with this is that the force ends in an abrupt manner. This could allow the participants to differentiate the haptic sensations based on other aspects than frequency differences. We solved this problem by presenting the participant with complete cycles for the haptic sensations while keeping the intervals at 2 seconds. For a stimulus with a frequency of 2.1 Hz, 2 cycles were presented. This gave the stimulus a duration of 1.90 seconds. A "blank" sensation of 0.1 seconds was added to the stimulus to keep the interval at 2 seconds. 34 4.3.3.2 Counting of Cycles When a participant is presented with a stimulus lasting 2 seconds, a 2 Hz stimulus consists of 2 complete cycles while a 4Hz stimulus would be 4 complete cycles. At low frequencies, this could allow the participant to distinguish between the stimuli by "counting" the complete cycles. This can be avoided using one of the following methods: • Present only frequencies higher than 5 Hz. This prevents the participant from counting by keeping the speed of the cycles above what can be easily counted. • Present the participant with long stimuli. This method prevents the counting by making it hard to keep track of the cycles presented. When a long stimulus is presented, the participant can no longer count the cycles effectively. • Present the participant with stimuli that contain the same number of cycles independent of the frequency being tested. This is the chosen method for our experiments. We present stimuli with different frequencies but we keep the same number of cycles presented so counting cannot be used to distinguish between stimuli. This could have allowed the participants to discriminate between the stimuli based on differences in length, however we believe this not to be the case given that we used this method only for frequencies 5 Hz and below. This way we kept the changes in stimulus length to a minimum. 35 Presenting the participant with the same number of cycles in each stimulus allowed us to test very low frequencies (0.5 Hz) and still be able to keep the stimulus short (2 seconds) by not having to present numerous cycles of the stimulus. 4.3.4 Experimental Procedure A complete session yielded the Frequency Differentiation Threshold (FD) for a specific waveform at the 6 selected frequencies. Each session was divided into 6 blocks of trials. The FD Threshold for a specific frequency was determined in each block. We used Parameter Estimation by Sequential Testing (PEST) to determine the FD Threshold. PEST decided the number of trials required for each frequency being tested. All trials were self-paced with breaks allowed between blocks or when the participant felt the need for relaxing the grip on the knob. The participants were told that the experiment was not timed. 4.3.4.1 Participants For this experiment, five participants were recruited through a posting in a bulletin board in the computer science building. The participants included 2 females and 3 males, all right-handed and between 20 and 29 years old. None of the participants had any disabilities or limitations in the sight or touch senses. All participants were recruited from the Computer Science Building and were paid cash for their involvement. Figure 14 shows details of the participants' involvement in the experiments. Each waveform was completed by 3 participants; some participants completed more than one session. 36 Part ic ipat ion in Detect ion Thresho ld Exper iments Part ic ipant Sine Tr iangle Square Sawtoo th W a v e f o r m W a v e f o r m W a v e f o r m W a v e f o r m 1 X X X X 2 X X 3 X X 4 X X 5 X X Figure 14. Participant involvement in Frequency Differentiation experiment. The X denotes participation in the experiment for the matching column. Each experiment was completed by 3 participants. 4.3.4.2 Stimuli Presentation For each trial, two pairs of 2-second intervals were presented. All intervals contained stimuli with the same waveform at a fixed amplitude. Three of the intervals contained identical haptic sensations at a fixed base frequency; the fourth interval presented a haptic sensation with a different frequency as specified by the experiment software. The order for these intervals was randomized. During the stimuli presentation, the screen was cleared and the participants were required to wear noise-canceling headphones to ensure the haptic stimulus was the only factor determining their responses. 4.3.4.3 Instruction of Participants We followed the same procedure as for the DT experiment (Chapter 4.2.4.3). 37 434.4 Experiment Session We used the same procedure and methodology followed for the DT experiment (Chapter 4.2.4.4); however, the participants were asked to determine the location of one different stimulus amongst two interval pairs for this experiment. 4.3.5 Data Collection and Experiment Output We used the same procedure and methodology followed for the DT experiment (Chapter 4.2.5); however, the results were stored as 6 values that represent the Just Noticeable Difference (JND) in frequency for each base frequency with the waveform being tested. 4.3.6 Results for the Frequency Differentiation Experiment The results for the Frequency Differentiation experiment are displayed in Figures 15 to 19. Each waveform tested (sine, square, triangle and sawtooth) is shown in a separate graph; a fifth graph presents a comparative of the results. The plots for these results present the Just Noticeable Difference in frequency in a logio scale. 38 Frequency Differentiation Results for Sine Waveform 100.0 1 • o 0.1 J : Base Frequency (loglOHz) Figure 15. Frequency Differentiation experiment results for the sine waveform. The graph represents sensitivity to changes in frequency of the haptic sensations. The horizontal axis is the base frequency being tested and the vertical axis is the frequency change required for detection. The error bars are the standard deviation of the 3 samples taken. Both the axes in this graph are in logio scale. Frequency Differentiation Results for Square Waveform 100.0 -| 3 o.o J Base Frequency (Iog10 Hz) Figure 16. Frequency Differentiation experiment results for the square waveform. The graph represents the average sensitivity to changes in frequency of the haptic sensations. The horizontal axis is the base frequency being tested and the vertical axis is the frequency change required for detection. The error bars are the standard deviation of the 3 samples taken. Both the axes in this graph are in logio scale. 39 Frequency Differentiation Results for Triangle Waveform 100.00 -i 0.01 J Base Frequency (Iog10 Hz) Figure 17. Frequency Differentiation experiment results for the triangle waveform. The graph represents the average sensitivity to changes in frequency of the haptic sensations. The horizontal axis is the base frequency being tested and the vertical axis is the frequency change required for detection. The error bars are the standard deviation of the 3 samples taken. Both the axes in this graph are in logio scale. Frequency Differentiation Results for Sawtooth waveform 100.0 -i o 0.1 J Base Frequency (loglOHz) Figure 18. Frequency Differentiation experiment results for the sawtooth waveform. The graph represents the average sensitivity to changes in frequency of the haptic sensations. The horizontal axis is the base frequency being tested and the vertical axis is the frequency change required for detection. The error bars are the standard deviation of the 3 samples taken. Both the axes in this graph are in logio scale. 40 Figure 19 presents the Frequency Differentiation experiment results for all four waveforms in a graph. Both axes on the graph are in logio scale. The graph shows the average results for the participants tested. Frequency Differentiation Results 100.0 T . o.o J Base Frequency (log Hz) Figure 19. Frequency Differentiation experiment result summary. The horizontal axis in this graph represents the base frequency being tested. The vertical axis represents the frequency change required for detection. Both axes in this graph are in logio scale. 4.3.7 Conclusions for F D Experiment From the graph in Figure 19, it becomes apparent that the JND for frequency is approximately 1/5 of the base frequency for those frequencies 1 Hz and above. These results seem to follow Weber's law quite well. Weber's law [19] states that the size of the just noticeable difference (JND) of a stimulus is a constant proportion of the original stimulus value. 41 For our case, the minimum detectable frequency change A / is a fixed proportion k of the base frequency being tested / . All waveforms tested exhibit this phenomenon clearly at frequencies beyond 1Hz (Figure 19). 4.4 Conclusions The results obtained for the Detection Threshold and Frequency Differentiation experiments were used to select appropriate amplitudes and frequencies to be used as stimuli for our formal MDS analysis. The stimuli selected for MDS Experiment 1 consists of 36 different haptic sensations. These were created from the combination of 3 waveforms (sine, square and sawtooth), 4 frequencies (0.5, 5, 20 and 100 Hz) and 3 amplitudes (12.3, 19.6 and 29.4 mNm). The minimum amplitude (12.3 mNm) is twice the value of the detection threshold level for the waveform-frequency combination for which participants were the least sensitive in the DT experiment (sine waveform at 0.1 Hz). The frequencies were chosen to be easily differentiable from one another. The triangle waveform was not used due to its perceptual similarity to the sine waveform. 42 Chapter 5 Mapping the Perceptual Space In this chapter we present a novel approach to studying the perception of haptic icons using an exploratory statistical method called Multidimensional Scaling (MDS) [20,21]. We begin by presenting some of the methods and tools utilized in our work (Chapter 5.1) followed by a description of the experiments carried out and the results obtained. The study is divided in 2 experiments: • Experiment 1 (Chapter 5.3) We use MDS to analyze a set of haptic sensations created based on the results for the Detection Threshold (Chapter 4.2) and Frequency Differentiation (Chapter 4.3) experiments. • Experiment 2 (Chapter 5.4) The results obtained in Experiment 1 (Chapter 5.3) prompted the design of a second haptic icon set to do further testing in order to determine a specific frequency which might posses the most information carrying capabilities. Chapters 5.4 and 5.6 present Experiment 2 and the results obtained from testing this second icon set. Chapter 5.5 presents individual results for both experiments. Chapter 5.7 presents a discussion of the results obtained for both experiments. 43 5.1 Multidimensional Scaling Analysis Techniques 5.1.1 Introduction to Multidimensional Scaling Analysis (MDS) Before a researcher can understand why an organism reacts to a stimulus in a specific manner, the researcher must first understand what aspects of the stimulus are attended by such organism. Identifying such aspects directly for complex stimuli can be hard to do. In order to simplify this identification, we use an exploratory statistical method known as Multidimensional Scaling (MDS). MDS is a set of mathematical techniques that enable a researcher to uncover the "hidden structure" behind data. MDS is similar to principal component analysis (PCA), a mathematical procedure that transforms a number of (possibly) correlated variables into a (smaller) number of uncorrelated variables called principal components. However, PCA cannot take into account nonlinear structures, structures consisting of arbitrarily shaped clusters or curved manifolds, since it describes the data in terms of a linear subspace. PCA is a linear projection data reduction method while MDS is a non-linear projection method. An example illustrating a Multidimensional Scaling application can be seen in a 1968 election study conducted by the Survey Research Center of the University of Michigan [22]. For this experiment, the participants were asked to evaluate 12 actual or possible candidates for President of the United States. The idea was to determine how similarly the candidates were viewed by the public and what identifiable features the public discerned in the candidates that could help us understand what led individual citizens to their decisions. MDS is useful in answering such questions by locating the political candidates in a spatial configuration or "map". Once we have located the candidates (or points) on a 44 multidimensional space, we seek to determine the hidden structure, or meaningful representation of this map of candidates. Applying an MDS procedure to this data set provided a way to reduce the information from a collection of perceived distances between the 12 candidates to a two-dimensional map representing the hidden structure of the data (in this case, partisanship and ideology). By finding key differences between political candidates at opposite ends of each dimension, we can attempt to develop indicators of variables that can be measured in future elections. MDS allows you to analyze N objects (in our case, haptic icons) according to their measured dissimilarity. A dissimilarity matrix is a set of values representing the perceived distances between each object in a set. MDS takes as input a dissimilarity matrix and generates a multidimensional configuration of the objects in an N dimensional space such that the distances in the Euclidean space approximate the dissimilarities specified by the matrix. MDS produces results in several representations ranging from one to N dimensions. According to the literature reviewed [22,23] most object sets can be satisfactorily represented with 2 to 3 dimensions. Each dimensional representation obtained from an MDS experiment has a "fit to data" measure. There are several different methods for measuring this fitness value: Stress, S-Stress, F-stress and 1-R2. All these methods obtain a measure for goodness of fit to the data tested. For our tests we chose to use Stress as a fit to data measure. 45 The formula for stress is: V scale In the equation, dtj refers to the Euclidean distance, across all dimensions, between points i and j on the map, f(xy) is some function of the input data, and scale refers to a constant scaling factor, used to keep stress values between 0 and 1. When the MDS map perfectly reproduces the input data, (/(•*,•,•)- dtj) is zero for all i and j, so stress is zero. Thus, the smaller the stress, the better the representation. This measure of fitness can be used as a rough guideline to determine the number of dimensions appropriate for representing the data. A greater number of dimensions in the representation generally produce a lower stress value. In order to determine how many dimensions are appropriate for representing the dissimilarities/configuration, we must look at a plot of the stress values for the different representations (dimensionalities); this is called a Scree Plot [24]. According to the Scree criteria, the particular dimensionality at which a sharp "elbow" is seen in the stress curve indicates an appropriate choice of solution dimensionality. For the sample graph shown in Figure 20, the correct number of dimensions to use for the configuration is 2. In our experiments, we used this measure as a preliminary indication of the algorithm's performance; we followed it by visually inspecting the MDS solutions for neighboring dimensions and confirming that the variability was being absorbed in meaningful way. 46 Stress (AtSCfll) 0.5. 0.4 0.3 0.2 0.1 h 0 1 2 3 4 Number of Dimensions Figure 20. Scree plot of stress values for MDS solutions in several dimensions. The horizontal axis represents the number of dimensions for the MDS representation. The vertical axis is the stress (fit to data) measure. We can observe a sharp "bend" in the graph at 2 dimensions; this suggests that a two-dimensional representation of the solution for this particular example is appropriate. The dissimilarity (distance) of the object pairs is estimated by averaging the judgments of several participants. The number of participants needed is dependent on the estimated number of dimensions for the representation and the number of stimuli to be used in the experiment. [25,26] The higher dimensionality expected, the greater the number of stimuli needed - usually the number of stimuli should be at least 4 to 5 times the expected number of dimensions, preferably more. Also, the dissimilarity data should be as stable as possible - the more observations per stimulus pair, the more stable, but usually 5-10 observations per pair is enough to get reasonable stability. This means that 5-10 participants are required to complete the experiment. 47 For most MDS tests, the number of dimensions is not known a priori. The researcher must choose the dimensionality based on the interpretability of the results and the fit to the data values. The interpretation of the solutions must rely on meaningful stimulus features, orderings or groupings of stimuli that correspond to meaningful stimulus attributes. A pilot study can be carried out to get a better idea of the representation dimensionality to be used for an experiment. Like most user studies, MDS requires that a standard set of instructions be given to the participants prior to participating in the experiment. The directions should standardize the participants' expectations about the stimuli to be judged. For example, these instructions can be used to inform the participants if they are expected to differentiate frequency, amplitude or any other specific characteristic of the stimulus; or conversely, they might (consistently) give the participant no such guidance and leave the criteria for discrimination up to the participant. As part of the pre-test preparations, the participants typically are required to review all of the stimuli to be judged after getting the directions but before making any judgments to allow them to calibrate their judgments. 5.1.2 Types of MDS There are three types of MDS analyses; each addresses a specific type of problem to be solved. • Dimensional MDS. The dimensions recovered by MDS are taken to be the salient aspects for the organism and the stimulus coordinates recovered in the scaling are interpreted as the location of the objects along these salient aspects. 48 • Data Reduction Applications [27,28] No interpretation is given to the dimensions recovered in MDS; the analysis focuses on clusters of stimuli found within the plot of the MDS dimensions. The scaling only serves as a device for reducing the data to a form in which the stimulus clusters can be readily displayed graphically. • Configural Verification Type This method is used to confirm a theory. The experimenter begins from a hypothesis that specifies the number of dimensions that should be obtained if the proximity measure is submitted to a non-metric MDS. The results are used to confirm or reject an existing theory of behavior. The final goal for this study is the discovery of the underlying psychophysics for the perception of haptic icons. We decided to use Dimensional MDS because it allows the extraction of the perceptual axes we are trying to discover. For these experiments, the pre-test instructions will not advise the participants what aspects of the stimuli to differentiate. This gives them creative freedom when rating the stimuli, however, it could allow for more strategic influences on the results. Different participants might use different strategies in their analysis of the haptic sensations. 5.1.3 Data Gathering Techniques for MDS Tests 5.1.3.1 Traditional Direct Comparison Method In the Direct Comparison method, the participant is presented with a pair of stimuli and asked to give a measure of similarity for it. All possible pair combinations have to be 49 presented at least one time. For our scenario, this means presenting a pair of haptic icons and asking the participant to rate them on a scale from Same to Different. This method presents all possible combinations of icon pairs in random order and asks the participant to give a measure of similarity between them. A screenshot of the interface used to perform this test is shown in Figure 21. This method is straightforward and simple to implement, however, the tests are long and their duration increases geometrically with an increase in the number of icons to be tested. Figure 21. Interface used to gather similarity data for a test set using direct comparison. 50 For an experiment consisting of some 12 icons, this is not a problem. This test set requires 72 comparisons. A larger test set of say 20 icons would require 200 comparisons. To calculate the required number of comparisons we use: Comparisons = ^ Where n is the number of icons in the test set. For our formal testing we decided to use an icon set composed of 36 icons. If we chose to use this methodology we would need a total of 648 comparisons for each participant. This is infeasible since such an experiment would take approximately 3 hours to complete and would undoubtedly be too tedious to perform reliably. Direct comparison for a large number of stimuli becomes un-desirable method not only due to the duration of the tests, but also for repeatability and accuracy of the results. After consulting with a couple of experts on the field of perception and psychology and studying previous work presenting similar experiments in the field of vision [11,12], we decided to develop a different testing method for our haptic icon set. 5.1.3.2 Icon Grouping Test Method Another option for performing these tests was developed based on the work by Lawrence Ward on the visual perception of spaces [11,12]. In this work, the participants were asked to group pictures of different natural and artificial (human created) places. An experiment session consists of 5 blocks. For each block, the participants are asked to sort items into categories. For the first block, the participants are allowed to select the 51 number of categories that they wish to use. For the remaining blocks in the experiment, the participants place the items into a predetermined number-of-categories as to obtain a complete categorization into 3, 6, 9, 12, and 15 groups. From this set of number-of-categories (3, 6, 9, 12 and 15), the one closest to that selected freely by the participant in the first block is eliminated. Thus, if a participant selects 5 categories for the first block, the following blocks ask the participant to sort the items into 3, 9, 12 and 15 categories, substituting the sorting into 6 categories with the participant's chosen 5. The order of number-of-categories was randomized among the last 4 blocks. The similarity data for the set of items is calculated as follows: For each block in an experiment session, if two particular items coincide in the same category, a similarity value equal to the number of categories for that block is given to the particular item pair. These values are added for all blocks in the session. For example, if items 1 and 3 are grouped together for the blocks with 3, 9 and 15 categories but not for the blocks with 6 and 12 categories, the similarity measure for this pair is calculated as 3 + 9 + 15 = 27. The maximum similarity value obtainable with this methodology is 45, when a pair of items is grouped together for all blocks in a session (3 + 6 + 9 + 12+ 15 = 45). This grouping methodology was adopted for all our MDS experiments (Chapters 5.3 and 5.4). 5.2 MDS Pilot Test Before appropriately defining the haptic icons to use for our formal MDS analysis, we performed a pilot study using a set of 9 different very crudely prepared haptic icons. The 52 pilot was completed by 10 participants recruited informally in the lab. The icons for the test were formed by combining 3 frequencies (100, 300 and 500 Hz) and 3 amplitudes (6.1, 18.3 and 30.5 mNm) of a sine waveform. Figure 22 shows the output obtained for this pilot test. In this case, a two-dimensional representation was chosen based on the scree criteria. The plot for the results shows a strong grouping for the smallest magnitude. The participants seem able to differentiate icon amplitude easily but do not differentiate frequency. We can see that for the specific frequencies and amplitudes selected, there is no orientation for the graph that can be mapped directly to scales of frequency and amplitude. The 3 haptic icons on the upper right corner of the graph represent the 3 frequencies at the lowest amplitude. The fact that they are placed so close together in the graph indicates that they are perceptually similar. 2.5 Differences in perception of haptic Icons '•J V -1.5 A A -0.5 -=6r <%5---2--vAf-0 0.5 "I A 1 /' \ Figure 22. Two-dimensional MDS representation for the simple haptic icons used in the pilot study. A small graphical icon identifies the locations of the test stimuli in the MDS solution space. This icon's size represents stimulus amplitude and its period, the frequency used. 53 Although this was designed to be a simple pilot test, we obtained useful information for our later experiments. As we later discovered in the Detection Threshold experiment, our selection for the lowest amplitude to use in the pilot test was too low (6.1 raNrn). Another problem with the selected test set were the frequencies used; the frequencies selected for this test (100, 300 and 500 Hz) were above what the display could accurately reproduce (200Hz). We also learned that the use of the direct comparison data gathering method (Chapter 5.1.3.1), proved to be too tedious to perform accurately. For this reason, we selected to use the icon grouping methodology (Chapter 5.1.3.2) for our formal MDS analyses. 5.3 Mult id imensional Scaling Experiment 1 Having performed the Detection Threshold and Frequency Differentiation tests, we designed a haptic icon set to be tested using MDS techniques. 5.3.1 Methodology Participants are presented with the haptic stimuli using the icon grouping MDS test method. A complete session with the test set consisting of 36 haptic icons required between 45 minutes and 1 hour to complete. 5.3.2 St imul i Selection The stimuli selected for this experiment consists of 36 different haptic sensations. These are created from the combination of 3 waveforms (sine, square and sawtooth), 4 frequencies (0.5, 5, 20 and 100 Hz) and 3 amplitudes (12.3, 19.6 and 29.4 mNm). We selected the amplitudes and frequencies based on the results for the Detection Threshold and Frequency Differentiation tests. The minimum amplitude (12.3 mNm) is 54 twice the value of the detection threshold level for the waveform-frequency combination for which participants were the least sensitive in the DT experiment (sine waveform at 0.1 Hz). The frequencies were chosen to be easily differentiable from one another. Every stimulus used in this experiment has 2-second duration. All force magnitudes presented through our haptic knob are scaled as peak-to-peak torque values in mNm (Newton/millimeters). 5.3.3 Experimental Procedure Participants are presented with the haptic stimuli using the icon grouping MDS test method (Chapter 5.1.3.2). This method is interactive and self paced. A complete session yields a dissimilarity matrix containing the calculated perceived distances between the icons in the set. Each session is divided into 5 blocks of stimuli sorting. 5.3.3.1 Participants Due to the nature of this experiment, which focuses on determining basic parameters for perception of haptic icons, we recruited participants without regard to background or abilities on the assumption that basic tactile perceptual abilities would be consistent across the general population. For this experiment, eight participants were recruited through a posting in a bulletin board in the computer science building. The participants included 4 females and 4 males, all right-handed and between 20 and 29 years old. None of the participants had any disabilities or limitations in the sight or touch senses. All participants were recruited from the Computer Science Building and were paid cash for their involvement. 55 5.3.3.2 Stimuli Presentation The graphic icon sorting method is thoroughly interactive. The participants decide which haptic icons are to be displayed by clicking on a graphic representation of the haptic icons presented on the screen. Each icon can be displayed as many times as the participant requires. 5.3.3.3 Instruction of Participants The session began by presenting a set of simple instructions on the screen. The instructions were then read to the participant by the experimenter. These instructions informed the participant of the tasks to be completed. The participants were told to sort the sensations based on perceived similarity using whatever criteria they choose. Copies of these instructions are included in Appendix D. After making sure that the participant was seated comfortably and the haptic display was adjusted to be within easy reach, we observed the participant interact with the system and answered any questions. Since the icon sorting method is an interactive procedure, the participants can learn the basics of operation by exploring the interface without affecting the end results of the experiment. Participants were instructed to perform the first block under our supervision. After this, the participants were again asked if they had any questions. Having answered any further questions the participants were required to wear the noise canceling headphones and continue with the experiment. After each block the participants were reminded by the experiment software to take a break. 56 5.3.3.4 The Experiment The participant is presented with a set of 36 graphical buttons. Each of these buttons represents a different haptic icon. When a button is pressed, the haptic icon is displayed on the knob. The buttons can be moved on the screen by right clicking and dragging them. A screenshot of the program used for the experiment is presented in Figure 23. Figure 23. Interface used to obtain the dissimilarity data for the haptic icons. This is the screen after a few icons have been placed in groups. Each group has a space for the participant to type a description and must contain at least 1 haptic icon. The participants are asked to move the buttons into boxes in the screen, sorting them into groups. Following the methodology of previous work explained in Chapter 5.1.3.2 we ask the participants to perform 5 sorting tasks, one for each block of the session. The number of categories used is the only change between blocks. 57 For each block in the session, the participants can feel the haptic icons as many times as they require. They can also move the icons between the groups as they see fit. When the participants are comfortable with the sorting, they can proceed to the next block by clicking on the appropriate button on the screen. At this time, the experiment software reminds the participants to take a break. Figure 23 shows a screenshot of a typical session in progress. Some of the haptic icons have been put into groups (the blue buttons have been placed in the boxes on the top of the screen). The groups (represented by boxes on the screen) have a space provided for the participant to type a description. This description not only allows the participant to identify the group, but also provides us with additional information as to what characteristics were considered when sorting the haptic icons. For this example, the participant chose 6 categories for the first block in the session. 5.3.4 Data Collection and Experiment Output After a participant completes the 5 blocks in a session, having sorted the haptic icons into 3, 6, 9, 12 and 15 categories, the program calculates the similarity between the haptic icons based on the groupings as explained in Chapter 5.1.3.2. The data gathered from each session was stored into individual files. The name of these files was formed by combining the date and time of the session with the participant's name. The file contained the participant's name, age and gender and the dissimilarity matrix obtained. The dissimilarity matrix was processed using A L S C A L PC [29,30] to obtain the MDS results. The results are presented using graphing software developed by the author. 58 5.3.5 Results for MDS Experiment 1 The stress measure (fit to data) we obtained for the average results for MDS Experiment 1 (Figure 24), suggests that a two-dimensional representation of the output data is appropriate. Some individuals make more complex distinctions than others, but on average, 2 dimensions seem to produce a good overall representation. Stress (Fit to Data) Measure 1 2 3 4 5 Dimensions Figure 24. Stress values for the average results of MDS Experiment 1. The marked "elbow" at 2 dimensions suggests that a two-dimensional representation of the results is appropriate for analysis of the data. Figure 25 presents the two-dimensional representation for the average results of MDS Experiment 1. We observed higher dimensional representations for these results, but they provided no additional information. Graphs of the individual results for this experiment can be found in appendix A. In addition to this graph, we present a partial view of the dissimilarity matrix obtained for this experiment (Figure 27) and a graph showing the mapping of the haptic icon input parameters onto the MDS solution space (Figure 26), to aid in the interpretation of the results. In order to aid in the interpretation of the MDS solutions presented beforehand, we devised a procedure to visualize the relation of the stimulus input parameters to the MDS 59 representation. Figure 26 shows essentially a projection of the original input parameter space (i.e., frequency, amplitude and waveform) onto the Experiment 1 MDS solution. For example, each point in the set of points corresponding to frequency is the average of all MDS solution coordinates for a single frequency (including every shape, magnitude and participant); and likewise for the shape and magnitude sets. This view suggests a possible interpretation of the accommodation of stimulus variability among the two solution axes, wherein it can be seen that frequency dominates both dimensions. Note that the plot is presented at a larger scale than the MDS solution to aid in interpretability. From Figure 25, we can observe that frequency differences dominate both axes in the plot. The tight frequency grouping indicates a strong tendency to discriminate haptic icons based on frequency. Changes in frequency amongst the haptic icons overwhelm the other changes in the parameters that make the icons feel different. A closer look at the graph reveals that waveform shape is of secondary importance followed by amplitude. When looking only at a specific frequency, we can still see a clear grouping based on waveform and also some discrimination for amplitude changes. This means that the participants are still able to differentiate changes in waveform and amplitudes, but these differences are not as perceptually salient as are the changes in frequency. Another interesting feature illustrated in the plot is the tight grouping for the lowest (0.5 Hz) and highest (100 Hz) frequencies. The groupings for 5 Hz and 20Hz are not as tight as the 0.5 and 100 Hz frequency groups. Within a single frequency, the most spread in the plot is noticeable at 20 Hz. This leads us to believe that there could be a specific frequency 60 at which the other characteristics for the haptic icons are most noticeable, making this frequency the one with the most information carrying capability. This phenomenon prompted a new set of experiments. These experiments, designed to find that specific frequency with the highest expressive capability (the most spread in the graph), are presented in Chapter 5.4 and 5.6. 61 :T5 o Freq (Hz): • 0.5 5 20 100 Mag (mNm): 12.3 | | 19.6 • 29.4 Waveform: Osine ~J Square y\ Sawtooth Figure 25. M D S Experiment 1 results. The haptic icons are represented by figures in the plot with sizes representing frequencies, tones representing magnitudes and shapes representing the different waveforms used to create the haptic icons. The smallest black circle represents the haptic icon with a sine waveform, a magnitude of 29.4 m N m and a frequency of 0.5 H z . Conversely, the largest light gray triangle represents a haptic icon generated with a sawtooth waveform with magnitude of 12.3 m N m and a frequency of 100 Hz . This Graph shows a tight grouping based on frequency. The tightest groupings are for the 0.5 and 100 H z frequencies. Waveform is of second importance in the grouping, followed by amplitude. The frequency with the most spread in the graph is 20 Hz . 62 Input Parameter Mapping to MDS space X Ampli tude Figure 26. Projection of design parameters onto the Experiment 1 MDS solution. Each marker represents the average MDS coordinates for all stimuli containing the parameter level indicated: for example, the Frequency marker labeled "0.5 Hz" is the average of coordinates for 6 stimuli (36/6); whereas the Magnitude marker labeled "12.2 mNm" is the average of 12 stimuli (36/3). Markers of a given parameter family are linked to illustrate the overall trend for direction for that family. Note that the scale for this plot is larger than the one used to present the MDS results. Figure 27 shows a partial view of the dissimilarity matrix obtained in Experiment 1 (the complete matrix can be found in Appendix C). The left column and bottom row represent the icon. The perceived distance between these icons is shown in the corresponding intersection in the matrix. 63 g Amplitude (mNm) o Frequency (Hz) o U l 12.3 0 U l 12.3 825 0 ro o 12.3 983 983 0 o o 12.3 1000 1000 925 0 o U l 19.6 158 800 983 1000 0 U l 19.6 867 333 983 1000 875 0 ro o 19.6 992 992 533 942 992 992 0 o o 19.6 1000 1000 925 317 1000 1000 942 0 o U l 29.4 283 833 983 1000 192 867 992 1000 0 01 29.4 950 467 992 1000 958 475 983 1000 950 0 ro 0 29.4 1000 967 642 942 1000 1000 417 942 1000 958 0 0 0 29.4 1000 1000 942 550 1000 1000 917 333 1000 992 933 0 Icon 0.5 5 20 100 0.5 5 20 100 0.5 5 20 100 Frequency (Hz) Ampl i tude ( m N m ) 12.3 12.3 12.3 12.3 19.6 19.6 19.6 19.6 29.4 29.4 29.4 29.4 Figure 27. Portion of Dissimilarity Matrix for sine waveform obtained in MDS Experiment 1. Shown here is part of the dissimilarity matrix obtained for Experiment 1. This section of the dissimilarity matrix represents the distances between the icons within the sine waveform. The complete matrix is shown in Appendix C. 64 5.4 Multidimensional Scaling Experiment 2 The results obtained in MDS Experiment 1 (Chapter 5.2) suggested the existence of a specific frequency that would allow the maximum spread in the MDS graph. For the results presented in Figure 25, we can observe that within a specific frequency, 20 Ftz has the most spread in the graph. We speculate that there is a specific frequency that allows the most spread in the MDS graph and that it should lie somewhere between 5 and 20 Hz. This frequency would have the largest bandwidth and thus the greatest Expressive Capability (EC). We designed a second haptic icon set with a smaller frequency spread to test our theory and performed an MDS analysis on the data obtained for it. 5.4.1 Methodology The same procedures were used as in Experiment 1 (Chapter 5.2.1). 5.4.2 Stimuli Selection In order to better determine the frequency with the greatest Expressive Capability, we devised a new haptic icon set consisting of 30 icons created from the combination of 2 amplitudes, 5 frequencies and 3 waveforms. We selected to use 2 amplitudes (unlike the icon set for Experiment 1 which uses 3) based on the results obtained for MDS Experiment 1 that suggest that differences in amplitude of the haptic icons are not as perceptually salient as are the differences in frequency or waveform. We speculate that by using only two amplitudes for the second icon set we can make the now more obvious differences in amplitude become more perceptually salient. The new torque amplitudes chosen for this set were 12.265 and 24.53 mNm. The largest amplitude is smaller in magnitude than the 65 largest amplitude for Experiment 1 (29.4 mNm). We chose these values based on the words used by the participants to describe the categories for MDS Experiment 1. Amongst these comments were: "bad", "violent-unpleasant", "big vibration" and "forceful". It's a possible issue that the stimuli selected for MDS Experiment 1 were subjectively unpleasant at the larger amplitudes. Our objective is to determine the frequency with the most Expressive Capabilities so we decided to try 5 different frequencies for this second haptic icon set. We use the same 3 waveforms as with Experiment 1: sine, square and sawtooth. Having 2 amplitudes (12.265 and 24.53 mNm), 3 waveforms (sine, square and sawtooth) and 5 frequencies (3, 7, 10, 16 and 25 Hz), our new test set consists of 30 different icons ( 2 x 3 x 5 = 30). Every stimulus used in this experiment has a 2-second duration. All force magnitudes presented through our haptic knob are scaled as peak-to-peak torque values in mNm (Newton/millimeters). 5.4.3 Experimental Procedure The same procedures were used as in MDS Experiment 1 (Chapter 5.2.3). 5.4.3.1 Participants For this experiment, nine participants were recruited through a posting in a bulletin board in the computer science building. The participants included 5 females and 4 males, all right-handed and between 20 and 29 years old. None of the participants had any disabilities or limitations in the sight or touch senses. Three of the participants had completed Experiment 1. All participants were recruited from the Computer Science Building and were paid cash for their involvement. 66 5.4.3.2 Stimuli Presentation The same procedures were used as for MDS Experiment 1 (Chapter 5.2.3.2). 5.4.3.3 Instruction of Participants The same procedures were used as in MDS Experiment 1 (Chapter 5.2.3.3). 5.4.3.4 The Experiment The same procedures were used as in MDS Experiment 1 (Chapter 5.2.3.4); however, the participant is presented with a set of 30 graphical buttons representing each different haptic icon in Experiment 2. 5.4.4 Data Collection and Experiment Output The same procedures were used as in MDS Experiment 1 (Chapter 5.2.4). 5.4.5 MDS Results for Experiment 2 The stress measure (fit to data) we obtained for MDS Experiment 2 (Figure 28), suggests that a two-dimensional representation of the output data is appropriate. However, the bend in the graph at 2 dimensions is not as clear as for Experiment 1 so we present the results for a three-dimensional representation as well. The two-dimensional representation for the solution is shown in Figure 29. The three-dimensional representation is presented in 3 graphs showing the different views of the solution (Figures 31, 33 and 35). Figures 30, 32, 34 and 36 present a projection of the stimulus input parameters to the MDS representation plots. In addition to these graphs, we present a partial view of the dissimilarity matrix (Figure 37) obtained for this experiment to aid in the interpretation of the results. 67 Stress (Fit to Data) Measure o -I , , , , 1 1 2 3 4 5 Dimensions Figure 28. Stress values for the average results for MDS Experiment 2. The "elbow" at 2 dimensions suggests that a two-dimensional representation of the results is appropriate for analysis of the data. Figure 29 shows the two-dimensional average results for MDS Experiment 2. Individual results for this experiment are included in Appendix B. The experiment was designed to find the frequency with the most discriminability (spread in the graph). Having a reduced frequency range for the second icon set produced a graph where the influence of frequency is not as overwhelming as with the icon set used in Experiment 1. Several interesting features can be distinguished in this graph. Frequency is mapped on the horizontal axis, increasing from right to left (Figure 29). Frequency increases from right to left for all waveforms, however specific frequencies for the different waveforms do not align. Another noticeable aspect of this graph is the clear separation between the sine waveform and the square and sawtooth waveforms. Both the square and sawtooth waveforms contain sharp changes in force (jerk), while the sine waveform is smooth. We can also notice that, as frequencies increase, the separation between smooth (sine) and sharp changing (square 68 and sawtooth) waveforms diminishes. As the frequency of a sine waveform increases, it appears to become perceptually closer to the square and sawtooth waveforms. This means that at higher frequencies, the perceptual differences for the different waveforms are less apparent. z • Waveform: OSine • Square A Sawtooth o ° a • A Mag (mNm): 1 12.3 I 24.5 Figure 29. Two-dimensional representation of the results for MDS Experiment 2. The icon set was selected to find the frequency with most spread in the graph. Reducing the frequency range for the haptic icon set produces a graph less influenced by changes in frequency. This has allowed a different mapping of the axes within the graph; frequency is represented in the horizontal axis (increasing to the left). Another noticeable feature is the separation of the sine waveform (smooth) and the square and sawtooth waveforms (with discontinuities). 69 Figure 30 illustrates the mapping of the input parameters to the MDS space. Frequency maps closely to the horizontal axis while waveform dominates the vertical axis. Amplitude shares both axes and is not as perceptually salient as the other factors, but is more salient relative to other factors than it was for Experiment 1. Input Parameter Mapping to MDS Space •Frequency •Waveform •amplitude Figure 30. Projection of design parameters onto the two-dimensional Experiment 2 MDS solution. Each marker represents the average MDS coordinates for all stimuli containing the parameter level indicated: for example, the Frequency marker labeled "3 Hz" is the average of coordinates for 6 stimuli (30/5); whereas the Magnitude marker labeled "12.2 mNm" is the average of 15 stimuli (30/2). Markers of a given parameter family are linked to illustrate the overall trend for direction for that family. Note that the scale for this plot is larger than the one used to present the MDS results. 70 Figure 31 shows the first of 3 views of the three-dimensional interpretation of the MDS solution for Experiment 2. Dimensions 1 and 2 from the solution map to the horizontal and vertical axes respectively. In this graph, we can see a mapping of the frequency to the horizontal axis (Dimension 1), increasing from right to left. The vertical axis represents the different waveforms (Dimension 2). This is presented more clearly in Figure 32. Figure 31. View 1 of the three-dimensional representation of the results for MDS Experiment 2. Dimensions 1 and 2 from the solution map to the horizontal and vertical axes respectively. Frequency is represented in the horizontal axis (increasing to the left). The vertical axis represents the different waveforms (Dimension 2). 71 In Figure 32, dimensions 1 and 2 from the solution map to the horizontal and vertical axes respectively. A mapping of the frequency to the horizontal axis (Dimension 1), increasing from right to left is clear. The vertical axis represents the different waveforms (Dimension — • — F r e q u e n c y 1 1 - • - W a v e f o r m -A -Amp l i t ude Figure 32. Projection of design parameters for view 1 of the three-dimensional representation of Experiment 2 MDS solution. Each marker represents the average MDS coordinates for all stimuli containing the parameter level indicated: for example, the Frequency marker labeled "3 Hz" is the average of coordinates for 6 stimuli (30/5); whereas the Magnitude marker labeled "12.2 mNm" is the average of 15 stimuli (30/2). Markers of a given parameter family are linked to illustrate the overall trend for direction for that family. Note that the scale for this plot is larger than the one used to present the MDS results. Input Parameter Mapping to MDS Space 3 Hz 72 For Figure 33, dimensions 1 and 3 from the three-dimensional representation of the MDS Experiment 2 solution map to the horizontal and vertical axes respectively. The horizontal axis (Dimension 1) maps loosely to frequency, increasing from right to left. The vertical axis (Dimension 3) maps loosely to amplitude, from small (at the top) to large (at the bottom). — O Freq (Hz] • 3 • 7 | 1 0 16 25 o Waveform: Osine • Square A Sawtooth |Mag (mNm): I 12.3 I 24.5 • o A • Figure 33. View 2 of the three-dimensional representation of the results for MDS Experiment 2. Dimensions 1 and 3 from the solution map to the horizontal and vertical axes respectively. Frequency is represented in the horizontal axis (increasing to the left). The vertical axis maps to differences in amplitude, from small to large (top to bottom). 73 Figure 34 shows dimensions 1 and 3 from the three-dimensional representation of the MDS Experiment 2 solution mapped to the horizontal and vertical axes respectively. The horizontal axis (Dimension 1) maps loosely to frequency, increasing from right to left. The vertical axis (Dimension 3) maps loosely to amplitude, from small (at the top) to large (at the bottom). Input Parameter Mapping to MDS Space —•—Frequency - • - W a v e f o r m -A- Amplitude Figure 34. Projection of design parameters for view 2 of the three-dimensional representation of Experiment 2 MDS solution. Each marker represents the average MDS coordinates for all stimuli containing the parameter level indicated: for example, the Frequency marker labeled "3 Hz" is the average of coordinates for 6 stimuli (30/5); whereas the Magnitude marker labeled "12.2 mNm" is the average of 15 stimuli (30/2). Markers of a given parameter family are linked to illustrate the overall trend for direction for that family. Note that the scale is larger than the one used to present the MDS results. 74 For Figure 35, dimensions 2 and 3 from the three-dimensional representation of the M D S Experiment 2 solution map to the horizontal and vertical axes respectively. The horizontal axis (Dimension 2) maps loosely to waveform. The vertical axis (Dimension 3) maps loosely to amplitude, from small (at the top) to large (at the bottom). A */ A a CP CP Freq(Hz): • 3 • ' • • A A ^ K, •* O O Osine Mag (mNm): • Square • 12.3 A Sawtooth • 24.5 Figure 35. View 3 of the three-dimensional representation of the results for M D S Experiment 2. Dimensions 2 and 3 from the solution map to the horizontal and vertical axes respectively. The horizontal axis (Dimension 1) maps loosely to waveform. The vertical axis (Dimension 3) maps loosely to amplitude, from small (at the top) to large (at the bottom). 75 Figure 36 maps dimensions 2 and 3 from the three-dimensional representation of the MDS Experiment 2 solution to the horizontal and vertical axes respectively. The horizontal axis (Dimension 2) maps loosely to waveform. The vertical axis (Dimension 3) maps loosely to amplitude, from small (at the top) to large (at the bottom). Input Parameter Mapping to MDS Space Sawtooth — • — Frequency - • - W a v e f o r m —ir- Amplitude Figure 36. Projection of design parameters for view 3 of the three-dimensional representation of Experiment 2 MDS solution. Each marker represents the average MDS coordinates for all stimuli containing the parameter level indicated: for example, the Frequency marker labeled "3 Hz" is the average of coordinates for 6 stimuli (30/5); whereas the Magnitude marker labeled "12.2 mNm" is the average of 15 stimuli (30/2). Markers of a given parameter family are linked to illustrate the overall trend for direction for that family. Note that the scale for this plot is larger than the one used to present the MDS results. 76 Figure 37 shows a partial view of the dissimilarity matrix obtained in Experiment 2. The matrix shows the measures for the sine waveform only. The left column and bottom row represent the icon. The perceived distance between these icons is shown in the corresponding intersection in the matrix. g Amplitude (mNm) o Frequency (Hz) co 12.3 0 - j 12.3 489 0 o 12.3 733 556 0 cn 12.3 926 926 785 0 ro cn 12.3 941 919 852 541 0 co ro -fc. cn 452 615 866 985 956 0 ro cn 830 637 770 978 985 689 0 o ro cn 911 881 719 963 970 815 563 0 CD ro cn 978 955 933 644 785 941 926 793 0 ro cn ho -t^  cn 993 993 963 800 630 956 926 867 563 0 Icon 3 7 10 16 25 3 7 10 16 25 Frequency (Hz) Ampl i tude ( m N m ) 12.3 12.3 12.3 12.3 12.3 24.5 24.5 24.5 24.5 24.5 Figure 37. Portion of Dissimilarity Matrix for sine waveform obtained in MDS Experiment 2. This section of the dissimilarity matrix represents the distances between the icons within the sine waveform. The complete matrix is shown in Appendix C. 77 5.5 Individual Results for Experiments 1 and 2 The results presented for both MDS experiments (Figures 25, 29,31, 33, and 35) are the average results for all participants. It is useful to observe the results for individual participants in order to better understand differences in perceptual abilities and uncover the diverse strategies for differentiating the haptic icons, as well as to appreciate the "error bars" in the group MDS solution. We will present 2 cases whose solution plots represent the extremes in the results for the participants tested. Figure 38 shows the two-dimensional MDS solution obtained for Participant 1 in Experiment 1. Participant 1 is a left-handed female between the ages of 20-29. The plot reveals an acute ability to identify the different frequencies used in the experiment. The sharp frequency-based groupings indicate that this person utilized frequency as the primary characteristic when evaluating the haptic icons. A closer look at the plot reveals that the participant is able to distinguish the icons based on waveform differences and amplitude changes as well. Within each frequency cluster, we observe tight groupings based on waveform and an ordering based on increasing amplitude as well. Each frequency cluster contains 3 groups, one for each waveform. Within this waveform sub-cluster, we see an ordering based on increasing amplitude towards the center of the plot. Figure 39 shows the mapping of the input parameters onto the two-dimensional MDS solution. The data presented in Figure 39 differs from the average results (Figure 26) in the perceptual distance between 0.5 and 20 Hz as well as the perceptual salience of changes in waveform and amplitude. For Participant 1, the icons with 0.5 Hz are perceptually equidistant from the icons with 5 and 20 Hz. The differences in amplitude and waveform are perceptually 78 small in comparison with the changes in frequency. The stress plot for Participant 1 is shown in Figure 40. Figure 38. MDS Experiment 1 results for Participant 1. The plot shows an acute ability to identify the different frequencies used in the experiment. The sharp frequency-based groupings indicate that this person utilized frequency as primary characteristic when evaluating the haptic icons. A closer look at the plot reveals that the participant is able to distinguish the icons based on waveform differences and amplitude changes as well. Within each frequency cluster, we observe tight groupings based on waveform differences and an ordering based on increasing amplitude as well. Each frequency cluster contains 3 groups, one for each waveform. Within this waveform sub-cluster, we see an ordering based on increasing amplitude towards the center of the plot. 79 Input Parameter Mapping to MDS Space )< Ampli tude Figure 39. Projection of design parameters onto the Experiment 1 MDS solution for Participant 1. Each marker represents the average MDS coordinates for all stimuli containing the parameter level indicated: for example, the Frequency marker labeled "0.5 Hz" is the average of coordinates for 6 stimuli (36/6); whereas the Magnitude marker labeled "12.2 mNm" is the average of 12 stimuli (36/3). Markers of a given parameter family are linked to illustrate the overall trend for direction for that family. Note that the scale for this plot is larger than the one used to present the MDS results. 80 Experiment 1 Stress Measure for Participant 1 0.6 0.5 0.4 (A in £ 0.3 55 0.2 0.1 0 1 2 3 4 5 Dimensionality Figure 40. Experiment 1 stress values for participant 1. The marked "elbow" at 2 dimensions suggests that a two-dimensional representation of the results is appropriate for analysis of the data. Figure 41 shows the results obtained for Participant 2 in Experiment 1. Participant 2 is a right-handed male between the ages of 20-29. The graph shows a good spread in the solution for all but the lowest (0.5 Hz) frequency. For the higher frequencies (20 and 100 Hz) the representation places the haptic icons separate from one another in waveform and increasing in size from the top to the bottom of the plot. Figure 42 shows the mapping of the input parameters onto the MDS solution. The stress values for Participant 2 are shown in Figure 43. The plot suggests that 2 dimensions are appropriate to represent the solution. Figure 42 illustrates an increased sensitivity for changes in amplitude and waveform when compared to the average results (Figure 26). This plot shows frequency as the most salient aspect for the stimuli, followed by amplitude and waveform. 81 Freq (Hz): • 0.5 5 20 Mag (mNm): V^©Wbrm: 1 1 2 3 Osine g 19.6 ~J Square | 29.4 / [ Sawtooth Figure 41. MDS Experiment 1 results for Participant 2. This participant was able to distinguish nearly every haptic icon from each other. The graph shows a good spread in the solution for all but the lowest (0.5 Hz) frequency. 82 Input Parameter Mapping to MDS Space M Wave fo rm >< Ampli tude Figure 42. Projection of design parameters onto the Experiment 1 MDS solution for Participant 2. Each marker represents the average MDS coordinates for all stimuli containing the parameter level indicated: for example, the Frequency marker labeled "0.5 Hz" is the average of coordinates for 6 stimuli (36/6); whereas the Magnitude marker labeled "12.2 mNm" is the average of 12 stimuli (36/3). Markers of a given parameter family are linked to illustrate the overall trend for direction for that family. Note that the scale for this plot is larger than the one used to present the MDS results. 83 Experiment 1 Stress Measure for Participant 2 0.7 -r 0.6 -0.5 • in 0.4 -in o w 0.3 -0.2 -0.1 -0 -1 2 3 4 5 Dimensionality Figure 43. Experiment 1 stress values for participant 2. The marked "elbow" at 2 dimensions suggests that a two-dimensional representation of the results is appropriate for analysis of the data. Participant 1 completed both Experiment 1 and 2. We present the results for Experiment 2 (Figure 44) as a two-dimensional representation for comparability purposes. The plot for these results does not show a grouping for frequency like for Experiment 1 (Figure 38). For Experiment 2, Participant 1 took more account of parameters other than frequency for her judgments. Figure 45 shows the mapping of the input parameters onto the MDS solution. The results for Experiment 2 for Participant 1 are very similar to the average results obtained (Figure 30). 84 Figure 44. MDS Experiment 2 results for Participant 1. The plot for these results does not show a grouping for frequency like the results for Experiment 1 (Figure 38). For the icon set in Experiment 2, participant 1 was not as keen in identifying the differences in frequency. 85 Input Parameter Mapping to MDS Space Sine 12.2 m N m 3 Hz 7 Hz 25 Hz 24.5 mr *Jm 10 Hz 16 Hz Square Frequency Waveform Amplitude Sawtooth Figure 45. Projection of design parameters onto the Experiment 2 MDS solution for Participant 1. Each marker represents the average MDS coordinates for all stimuli containing the parameter level indicated: for example, the Frequency marker labeled "3 Hz" is the average of coordinates for 6 stimuli (30/5); whereas the Magnitude marker labeled "12.2 mNm" is the average of 15 stimuli (30/2). Markers of a given parameter family are linked to illustrate the overall trend for direction for that family. Note that the scale for this plot is larger than the one used to present the MDS results. 86 5.6 Finding the Frequency w i th most Expressive Capabilities To more precisely ascertain the frequency allowing the greatest spread in other parameters in the M D S solution for a particular dimensionality, we defined a measure of the "Discriminability" amongst icons that share a specific frequency while varying in the other parameters. Df is the sum of the squared distances between solution coordinates for all stimuli sharing that frequency. A larger value for Df means more variability due to non-frequency parameters at that frequency, d represents the perceived dissimilarity measure between the haptic icon n and m within a specific frequency. ./V is the number of stimuli at a given frequency. The measure for each frequency is obtained by adding the square of the distances between solution coordinates for all icons within the frequency. A larger value for Df means more variability due to non-frequency parameters at that frequency. Figure 46 shows the average results obtained from our Discriminability Analysis. The data for this graph was taken from the data gathered for M D S Experiment 1 and 2. According to this measure, 10 H z permits the greatest spread, followed closely by 7 Hz . N-l N 87 25 Discriminability - • Experiment 1 • Experiment 2 40 60 Frequency [Hz] 100 Figure 46. Average results for the Discriminability Analysis. Shown here are the results for the Discriminability Analysis for the data from Experiment 1 and 2. The frequency with the most spread in the MDS graph is 10 Hz, followed closely by 7Hz. This means that amongst the frequencies tested, 10 Hz has the most information carrying capabilities for the waveforms selected. 5.7 Discussion of MDS Experiments 5.7.1 Experiment 2 Dimensionality The results for Experiment 2 were presented as two and three-dimensional maps. We did this due to the lack of a clear "elbow" in the Scree plot for the solution (Figure 28). The two-dimensional representation for Experiment 2 (Figure 29) shows the horizontal axis mainly representing frequency and the vertical axis expressing changes in waveform. The differences in amplitude are absorbed by the two axes. When the results are represented in three dimensions, changes in amplitude are closely represented by the third dimension. However, this dimension also represents the differences between the square waveform and the other waveforms used (sine and sawtooth). This suggests that changes in waveform 88 (between square and the rest) are perceived in the same manner as changes in amplitude. The icons generated with the square waveform seem perceptually louder than those generated with the sine and sawtooth waveforms. This holds true for the icon set selected. 5.7.2 Participant Extremes The individual results presented in Chapter 5.5 provide an insight into the different strategies and individual differences amongst the participants. Participant 1 had particularly interesting results in Experiment 1. She was able to perfectly identify the different frequencies in the stimulus set giving little to no attention to the other parameters. After analyzing the results obtained, we inquired on her particular abilities and background to try and explain the results. Participant 1 is musically trained, which leads us to believe some sort of cross-modality training is responsible for the results presented here. A second musically trained (as we later found out) participant obtained strikingly similar results for this experiment as well, however, further testing is required to be able to determine if this phenomenon is indeed caused by cross-modality training. Participant 1 completed both Experiment 1 and 2. Figure 44 presented the results for Experiment 2 as a two-dimensional representation for comparability purposes. The plot for these results does not show a grouping for frequency as do the results for Experiment 1 (Figure 38). For Experiment 2, Participant 1 appears to take more account of parameters other than frequency for her judgments; however, this icon set has a much smaller frequency range and perhaps the differences in frequency for this icon set are perceived in a different manner than for Experiment 1. Another possible explanation for these results is the fact that the participant had a better knowledge of the experiment design and operation when she performed Experiment 2. All participants' questions regarding the experiments 89 were answered after every session. In fact, all participants took more account of the amplitude and waveform parameters with this new icon set; the results for Participant 1 may simply reflect the general trend of the results. Figure 41 shows the results obtained for Participant 2 in Experiment 1. Participant 2 is a 26-year-old right-handed male. This participant was able to distinguish nearly every haptic icon from each other. The plot shows an acute ability to separate the different sensations from one another. This was the other extreme in the results obtained. Figure 42 shows an increased sensitivity for changes in amplitude and waveform when compared to the average results (Figure 26). Still, this plot shows frequency as the most salient aspect for the stimuli, followed by amplitude and waveform. For Experiment 1, participants in general focused on the changes in frequency for differentiating the haptic icons. Participant 1 shows the extreme of relying on frequency to differentiate them. Participant 2 shows the other extreme, considering amplitude and waveform, while still mainly relying on frequency for the differentiation. Even the extremes presented here, still show the same general trend as the group; they differentiate the haptic icons mainly based on frequency. For Experiment 2, the individual results showed even less variability. Participants in this experiment still relied on frequency as the main parameter to differentiate haptic icons, however, amplitude and waveform now played an important role in this differentiation. 5.7.3 Interpretation of Dimensions A principal goal of our experiments was to relate our three stimulus design parameters to human perceptual axes. Perhaps the most important finding of this study is that while both 90 sets of stimuli (those used for Experiment 1 and Experiment 2) are perceived in two dimensions, the interpretation of those dimensions is markedly different for the two sets. For Experiment 1 (Figure 25), whose stimulus set exhibited wide variation in the most salient design parameter, that parameter dominated both perceived dimensions. That is, participants sorted stimuli based on frequency and little else - yet found two dimensions out of one control parameter. In Experiment 2, all three parameters were evident in the MDS solution - frequency and shape the most strongly and most nearly orthogonal to one another, while magnitude shared shape's dimension as illustrated by Figure 30. The fact that frequency axes for different wave shapes do not precisely align (Figure 29) could mean that participants perceive frequency differently for each wave shape. This and similar observations for the other parameters imply the effect of moderate to strong individual differences, which will need to be accommodated in future icon design. The thee-dimensional representations for the MDS solution provide little extra information about the icon set tested. The third dimension maps to changes in amplitude only weakly. 5.7.4 Max imiz ing Expressive Capacity through Frequency The inconsistent domination by frequency in the results for MDS Experiment 1 (least evident for 5 and 20 Hz), suggested that there might be a "sweet spot" for frequency where the information carrying capability (expressive capacity) of the signal was maximized. Thus, choosing a frequency range for the 2nd stimulus set which optimized discriminability permitted more stimuli to be perceived as markedly different from one another. Among the 5 frequencies considered in Experiment 2, 7 and 10 Hz had the maximum expressive capability according to our simple measure; and the entire range (3-25 Hz) permitted a 91 much greater expressive capacity of the stimulus set as a whole than did that of Experiment 1 (Figure 46). Thus, to maximize information delivery in haptic icons which employ frequency as a design parameter, frequency range should be carefully considered; and at least for the set of design parameters and stimulus duration used here, a range between about 5 and 20 Hz is optimal. Frequencies outside this range may also be used, but the contribution from other design parameters will not be perceived. 5.7.5 Effect of Shape: Smooth vs. Jerky As observed in Figure 29, there is a clear separation between the sine and the square / sawtooth wave shapes. Both the square and sawtooth waves are discontinuous while the sine wave has smooth derivatives. Further, the MDS solution's separations between smooth (sine) and sharply changing (square and sawtooth) shapes diminish; suggesting that at higher frequencies, shape differences become less perceptible. 5.7.6 The Range Effect - A Possible Explanation for Our Results When a set of stimuli has variations on a specific characteristic, the results of a perceptual test for said stimuli might be affected by a phenomenon known as the Range Effect. The results affected might show strong grouping at the extremes of the range for the specific characteristic being changed in the stimuli. The effect is present when participants are able to clearly identify one or both of the extremes in a range (for our case 0.5 and 100 Hz) and not so capable at identifying steps within this range. The participants then use these extremes as anchors from which to base their perceptual judgments. Participants have a tendency to group their responses close to the perceived anchors; anything not lying near 92 the anchors is not as precisely identified in the scale. This could explain the grouping found in the results for MDS Experiment 1 (Figure 25). 93 Chapter 6 Conclusions and Future Work 6.1 Conclusions The stimulus design, data collection and analysis techniques presented here together comprise an innovative approach to the design of haptic icons, which in turn have the potential of becoming a new communication medium for many kinds of interactive devices. The current results support the promise of the overall method. Specific contributions of the present study include observations that for synthetic haptic icons, (a) frequency plays a dominant perceptual role; (b) the meaning (in terms of design parameters) of users' perceptual dimensions across an entire set of icons is strongly affected by the range of frequencies employed in that icon set; (c) the perceptual variability between users across an entire set of icons is dependant on the icon set selected; (d) to maximize expressive capability of other design parameters, frequency should be varied around a relatively narrow range - probably 5-20 Hz; and (e) beyond frequency, wave shape and finally force magnitude appear to be most important perceptually. There are important limitations both to the scope of the present study (which covered only a few design parameters, and confined itself to passively felt stimuli) and to the MDS technique in general - which tells us about differentiability, but not about salience. It will provide more useful results, for example, if supplied with data from novel wave shapes that are judged as "somewhat similar" and "fairly similar" rather than at the two extremes of the similarity-dissimilarity scale, regardless of frequency. We consider this work an important 94 first step, and are already widening its impact through modifications of the current technique, and combining it with other forms of analysis that will broaden its applicability. 6.2 Future Work The work done in this thesis is only the first step in defining haptic icons. The second important step towards usable haptic icons is to create icons that convey complex meanings using simple haptic feedback devices (i.e., how to make a knob feel like it is controlling water flow or fan speed). Another possibility is the designing of "themes" for the haptic icons (i.e., make a set of haptic icons that seem happy, serious or festive). The work done in this thesis answered a few questions but it also raised many new ones about the perception of haptic icons. Amongst the questions raised by this work is the possible existence of some sort of cross-modality training. Some of the participants tested obtained significantly different results from the rest of the tested population. These participants were musically trained, suggesting that their improved performance for determining differences in frequency for the haptic icons might be due to this strong musical background. Further testing is required to be able to examine what caused this phenomenon. Our work concentrated on the use of passive haptic icons, where the participant does not need to explore the environment to obtain information about it. Another follow up experiment could be to carry on similar testing methods for exploratory haptic icons, those where the participant needs to move the haptic display to obtain information about it. This could allow us to better understand the perception of simple textures when displayed through a low degree of freedom (DOF) haptic display. 9 5 REFERENCES: [1] Westendorp, P. & Waarde, K. (2001). "Icons: Support or Substitute?" Information Design Journal 10(2): 91-94. [2] Taylor, M . M . & Creelman, C. D. (1967). "PEST: Efficient estimates on probability functions." Journal of the Acoustical Society of America 41: 782-787. [3] Huggins, W. H. & Entwisle, D. R. (1974). Iconic Communication: an annotated biography. Baltimore, John Hopkins University Press. [4] Yazdani, M . & Goring, D. (1990). "Iconic Communication." Technical Report, Dept. of Computer Science, Exeter University. [5] Gaver, W. W. (1988). "Everyday Listening and Auditory Icons." Ph.D. Dissertation, Dept. of Computer Science, University of California, San Diego. [6] Gaver, W. W. (1989). "The SonicFinder: An Interface that Uses Auditory Icons." In Human Computer Interaction 4(1): 67-94. [7] Brewster, S., Wright, P. & Edwards, A. (1992). "A detailed investigation into the effectiveness of earcons" In G. Kramer (Ed.), Auditory Display: Sonification, Audification and Auditory Interfaces. Addison-Wesley : 471-498. [8] Brewster, S., Wright, P. & Edwards, A. (1993). "An evaluation of earcons for use in auditory human-computer interfaces." Proc. InterCHJ93, Amsterdam, Netherlands 5: 222-227. [9] Hollins, M . , Faldowski, R., Rao, S. & Young, F. (1993). "Perceptual Dimensions of Tactile Surface Texture: A Multidimensional Scaling Analysis" Perception and Psvchophvsics 54: 697-705 [10] Mojsilovic, A., Kovacevic, J., Kail, D., Safranek, R. J. & Ganapathy, K. (1999). "The Vocabulary and Grammar of Color Patterns" The IEEE Image Processing special issue on image and video processing for digital libraries, December 1999. [11] Ward L. , (1977). "Multidimensional Scaling of the Molar Physical Environment", Multivariate Behavioral Research 12: 23-42. 96 [12] Ward L. , Russell J. (1981). "Cognitive Set and the Perception of Place", Environment and Behavior 13: 610-632. [13] Bonebright, T. L . (2001). "Perceptual Structure of Everyday Sounds: A Multidimensional Scaling Approach" Proceedings of the 2001 International Conference on Auditory Display, Espoo, Finland, 2001. [14] Rinker, M . A., Craig, J. C. & Bernstein, L . E. (1998). "Amplitude and Period discrimination of haptic stimuli" Journal of the Acoustic Society of America 104(1): 453-464. [15] Tan, H. Z., Srinivasan, M . A., Eberman, B., & Cheng, B. (1994). "Human Factors for the Design of Force-Reflecting Haptic Interfaces" Proceedings of the 3rd Annual Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems 55(1) [16] Heeger, D (1997). "Signal Detection Theory" Department of Psychology, Stanford University. [17] De Valois, R.L., De Valois, K.K. (1988). "Spatial Vision" Oxford: Oxford University Press 147-168. [18] Coren, S., Ward, L . M . & Enns, J.T. (1999). "Sensation And Perception" 5 m edition, Harcourt Brace 226-237. [19] Weber, E. H. (1995). "De Tactu" and "Der Tastsinn und das Gemeingefuhl" Handworterbuch der Physiologie 3: 481-588. [20] Kruskal J. (1964). "Multidimensional Scaling by Optimizing Goodness of Fit to a Non-metric Hypothesis" Psychometrika 29: 1-27. [21] Kruskal J. (1964). "Non-metric Multidimensional Scaling" Psychometrika 29: 115-129. [22] Kruskal, J. & Wish, M . (1978). "Multidimensional Scaling" Beverly Hills: Sage Publications. 97 [23] Cox,T & Cox, M . (1988). "Multidimensional Scaling" London: Chapman & Hall. [24] Cattel, R. B. (1978). "The scientific use of factor analysis in behavioral and life sciences." New York: Plenum Press. [25] MacCallum, R. C. (1979). "3-mode-MDS: A computer program to execute Tucker's three-mode multidimensional scaling" Applied Psychological Measurement 3(24). [26] Spence, I. and Domoney, D.W. (1974). "Single subject incomplete designs for nonmetric multidimensional scaling" Psychometrika 39: 460-490. [27] Shepherd R., (1963). "Analysis of Proximities as a Technique for the Study of Information Processing in Man" Human Factors 5: 33-48. [28] Shepherd R., (1964). "Attention and the Metric Structure of the Stimulus Space" Journal of Mathematical Psychology 1: 54-87. [29] Young, F. & Lewyckyj, R. (1990). "ALSCAL, ALTERNATING L E A S T SQUARES SCALING" [30] Young, F. & Lewyckyj, R. (1996). "Alscal User's Guide" L. L. Thurstone Psychometric Laboratory, University of North Carolina, Chapel Hill, North Carolina. 98 Appendix A Individual Results for MDS Experiment 1 Freq (Hz): 0.5 5 20 Mag (mNm): 12.3 | 19.6 • 29.4 Waveform: OSine "] Square / \ Sawtooth MDS Experiment 1 results for 26 year old, right-handed female. 99 n • A I Freq (Hz): 0.5 5 20 100 Mag (mNm): §f12.3 H 19.6 • 29.4 Waveform: Osine ~J Square /] Sawtooth MDS Experiment 1 results for 29 year old, left-handed female. 100 cP i t | Freq (Hz): 0.5 5 I 20 100 Mag (mNm): 12.3 • 19.6 • 29.4 Waveform: Osine J Square y\ Sawtooth MDS Experiment lresults for 20 year old, right-handed male. 10 MDS Experiment lresults for 23 year old, left-handed female. 102 Freq (Hz): 0.5 20 100 CO LJ O A Mag (mNm): 12.3 • 19.6 • 29.4 ZI 1 z ZI Waveform: Osine ] Square £ Sawtooth MDS Experiment 1 results for 26 year old, right-handed male. 103 I Freq (Hz): 0.5 5 I 20 100 Mag (mNm): 12.3 §§ 19.6 • 29.4 Waveform: Osine J Square ^ Sawtooth MDS Experiment lresults for 25 year old, right-handed female. 104 MDS Experiment 1 results for 22 year old, right-handed male. 105 Freq (Hz): 0.5 5 20 100 Mag (mNm): I 1 2 3 | 19.6 • 29.4 Waveform: Osine ] Square /] Sawtooth • o • o MDS Experiment 1 results for 24 year old, right-handed male. 106 Appendix B Individual Results for MDS Experiment 2 A ^ A o Freq (Hz): • 3 • 7 • 9 o waveiorm: Osine Mag (mNm): • Square 8 12.3 A Sawtooth • 24.5 MDS Experiment 2 results for 22 year old, right-handed male. 107 o O | Freq (Hz): 3 I 7 • 10 16 25 • Waveform: Osine D Square A Sawtooth • A V Mag (mNm): I 12.3 I 24.5 MDS Experiment 2 results for 29 year old, left-handed female. 108 ' Q ) | Freq (Hz): 3 7 I 1 6 25 Waveform: Osine • Square A Sawtooth Mag (mNm): t 12.3 124.5 • A -rtj MDS Experiment 2 results for 23 year old, left-handed female. 109 o © A O | Freq (Hz): 3 7 I 16 25 A Waveform: Osine D Square A Sawtooth A [Mag (mNm): S 12.3 I 24.5 MDS Experiment 2 results for 24 year old, right-handed male. © LJ | Freq (Hz): 3 7 I 16 25 O o ...J Waveform: o Osine Mag (mNm): D Square 12.3 A Sawtooth • 24.5 MDS Experiment 2 results for 24 year old, right-handed female. I l l I Freq (Hz): 3 I 7 • 10 16 25 Waveform: Osine • Square A Sawtooth Mag (mNm): • 12.3 • 24.5 MDS Experiment 2 results for 22 year old, right-handed male. 112 o • MDS Experiment 2 results for 24 year old, right-handed female. 113 Waveform: Osine • Square A Sawtooth |Mag (mNm): 1 12.3 I 24.5 MDS Experiment 2 results for 20 year old, right-handed male. 114 Appendix C Dissimilarity Matrices for MDS Experiment 1 and 2 Experiment 1 < 3 J Amplitude 1 o g g o g o g S § = o g § g i IS o g a i 8 | o s g 3 | o 8 8 ft •X-IS ° o s I | 1 o O CO o s g 8 8 8 g is s I o s g I a a 8 § a ° o e 8 g ft ft | i ft s a o s § g g ft s | 1 8 CO o ro ° • s s o a 8 | I O a o o o - — — • g Di a i s i § j is 1 1 ! f 1 I | • g 8 • g I 1 • s K § 1 § § 1 1 a | I o • £ u, Di a is 8 i a 8 § 1 |s a I i S cS O • .3 i S ' £ • s r i ' i § ! a 8 a 1 1 IS s a • I ro o § 550 I 992 110001 967 I 559 i 1000s 1000 ; 9421 | § | a a I a il | crt I g co il CO ro IO <£• OO • g 550 I 992 110001 967 I 559 i 1000s 1000 ; 9421 • § a 550 I 992 110001 967 I 559 i 1000s 1000 ; 9421 • s § 550 I 992 110001 967 I 559 i 1000s 1000 ; 9421 CO o cS 5 a g 1 O s o | § CO § ft a 5 ! Sn O o --• £ 1 § 550 I 992 110001 967 I 559 i 1000s 1000 ; 9421 § i i§ i § i § 1 § to i ca s i o g CO o g CO i o -1 i l -g g 550 I 992 110001 967 I 559 i 1000s 1000 ; 9421 cn C« ; g ! g ! O ^} g o, -J 550 I 992 110001 967 I 559 i 1000s 1000 ; 9421 ro 8 i | i ft CO [ <0 ro [ o g | o a |s o a g s is s 550 I 992 110001 967 I 559 i 1000s 1000 ; 9421 1 I g I i g a g § ft 1 O 1 * 1 • 1 -ft S l S3 g a g 1 g ° s co § ca a is 1 a n | § ] s a o i § Di 1 g s § 11 s o ro O o *4 s a a a if a § § CO o 8 ! 992 1 917 1 208 f 1000; 1000 1 958 ! 192 | I s is I § g o a 8 I r6 a 8 ! 992 1 917 1 208 f 1000; 1000 1 958 ! 192 | •n fo o ro a is CO 1 is 0* •g Di jg o> ro o 8 Di 1 3 s _i_ Ul to ro co ° s 1 ! 992 1 917 1 208 f 1000; 1000 1 958 ! 192 | ro | g __ £ g .!° 1 § s i 8 a ! 992 1 917 1 208 f 1000; 1000 1 958 ! 192 | •» 1 g r o o is 8 i rS ^ rS 8 O o 1 ft Di o o o O 1 L Z L I , £ is i o ! 992 1 917 1 208 f 1000; 1000 1 958 ! 192 | s , a t rS is a S § ° is § ft 8 is a a ! 992 1 917 1 208 f 1000; 1000 1 958 ! 192 | § I IS a a s a g g $ o a ft is !S a i ro a s g a 8 O o c i i ! 992 1 917 1 208 f 1000; 1000 1 958 ! 192 | I ^ ^ 1 | i § 1 Frequency p u, 8 § g s o g o, ro 8 s 8 8 KA ro I g 8 § g ro § Ul g u. 8 8 A m p l i t u d e ro ro ro £ £ £ P> *& ro 8 £ £ £ S ro ro t £ >o 8 8 W a v e f o r m o o O O o O o o o O O • • • • • • • • • • • • ^) T h e k ft c o l u m n a n d b o t t o m r o w represent the haptic i con . T h e p e r c e i v e d distance b e t w e e n these i cons is s h o w n in the c o r r e s p o n d i n g intersect ion in the matrix. 1 Experiment 2 < 3 .o > 3 o. o 0 ro co o o -j ro co 0 •0 o o o ro co - J CO CO Ul Ul 47* o 0 — — — o cn ro co •0 ro on 40 ro cn -o CO Ul o ro i n ro co •£> -*">• 40 40 CO Ul ro u i o o 0 ro Ul Ul ro 0 ui CO 0 40 CO Ul 40 Ul 47* o o - j ro •F--In oo CO o Cr. CO - J - j -j 0 40 -vl CO 40 CO Ul 47* oo 40 o o 0 ro * -Ul <0 OO CO - J 40 40 47* CO 40 -vl 0 oo Ul Ul 47* CO 0 o 0 ro Ul 40 CO 40 Ul Ul 40 CO CO 47* *-• - J CO Ul 40 *»• 40 ro 47* -si 40 CO 0 o ro ro Ul vO CO 40 40 CO 40 40 CT. CO 40 OO 0 0 47* CO 0 40 Ul 47* 40 ro 47* oo 47* -M Ul 47* CO 0 • 0 ro co OC" ro ro 40 40 CO o 0 o CO 40 47* 40 47* CO 40 oo Ul 40 CO Ul 0 0 o o • - j ro co vO Ul Ul 40 00 CO 4O 40 40 CO o 0 o 40 47* CO 40 ro 47* 40 Ul 47* 40 CO Ul 0 0 o -sl CO 0 • o ro co 10 40 CO 40 40 CO CO Cn -si 40 oo 40 40 40 CO 40 0 Ul 40 0 4*t 40 0 40 Cr-CO CO 40 47* 0* CO 0 • o* ro co "0 - J OO o o 0 40 -sl o "Sl 0 •»«• 0 o 0 40 40 CO eo 40 47* 0 - J •£»• CO CO o 40 - J CO 40 ro 47* -vl -sl o 0 • ro ro co o o o o o o 40 CO Ul -vl -vl o Ul -si o O o o 40 40 CO 40 -vl o oo ro ro Ul *»• oo 40 -vl OO 40 47* CO CO Ul 40 -sl o • 0 ro *»• Ul \o -vl o 40 CO Ul •0 -g o 40 40 CO 0 o 0 CO Ul 40 40 o 40 Ul Ul 40 - j 0 40 40 CO 47* Ul CO Ul ro 40 CO CO 40 -vl 0 40 -vJ 00 o • —j ro Ul oo Ul 40 -vl o •0 Ul Cr. 0 0 o 0 0 o 40 CO CO Ul ro -sl 40 CO 40 -vl 0 40 -vl 0 CO -sl 4*--sl ro 47* CO ro ro 40 -vl 0 40 -vl o -j o • o ro Ul 40 oo Ul 40 oo Ul 40 Ul Cr-40 40 CO 40 00 Ul 40 oo Ul CO CO oo -f». •F* 40 0 CO 40 ro cr-40 OO oo 40 47* oo ro ro 40 O **• 40 ro c* CO Ul ro Ul CO CO o • 0 ro Ul o o o o 0 0 40 CO Ul 40 CO CO 40 47* 40 -vl oo 40 47* CO 40 47* CO -vl ~S| CO -sl 47* CO 40 -vl CO 40 -vl CO oo -sl *»• -sl 40 CO oo CO 40 -vl 0 CO oo 40 -vl o u- O • ro Ul ro U l o o o 0 o 0 40 CO Ul 40 CO ui o o 0 40 40 CO 40 -sJ 0 -vl 40 ro 47* Ul ro 40 Ul 47* 40 Ul 47* CO CO 40 CO Ul ro -sl - J oo 40 "Nl o CO oo 0 o 0 0 Ul 0 0 0 ro CO o» -si 40 -vl 0 40 ~vl 0 40 40 CO o o o 40 Ul 47* o o 0 0 o 0 40 40 CO o 0 o Ul Ul 47* 0 CO 40 40 47* CO 40 -sl CO 40 -sJ OO 0 -vl * -40 -sl 0 40 -vl 0 40 -sl 0 40 -sl 0 0 -<j ro cc-•0 40 -o o 40 ^ 1 o o 0 0 40 40 CO 40 -vl 0 40 oo Ul 40 oo Ul 40 40 CO 40 40 CO 47* Ul CO CO -vl 40 o 4* 40 -sl 0 40 -vl 0 40 0 40 47* CO 40 0 0 40 -vl o 40 ~vj CO Ul o o ro CO 40 ro cn 40 00 oo oo 40 40 40 CO o o 0 40 Ul Ul 40 CO Ul 40 oo Ul 40 CO U l O 0 0 47* 47* -sl Ul 47* CO -sJ CO Ul 40 47* CO 40 47* CO 0 Ul 40 oo 47* -vJ 40 ro 0 40 0 40 -sJ 0 -sl 0 0 Ul ro O ^ d ro CO 40 CO Ul 40 CO CO CO - J 4W CO 40 Cr. CO Ul 40 40 40 CO o 0 0 40 -vl o 40 Ul 47* 40 -vl CO 40 CO Ul 40 Ul -vl CO 00 CO -sl oo 47* "Sl 40 V-CO 40 0 40 40 Ul 0 40 Ul 0 0 40 0 0 0 -vl -vl Ul Ul O ro Ul ro CO <0 40 CO 40 40 CO 40 cr> CO -sl 40 47* CO 40 40 CO 0 o 0 40 47* CO oo 0 -vl CO CO 40 40 00 40 47* CO -vl oo Ul CT. CO -sJ Ul -vl O 40 47* CO 40 0 40 0 Ul 0 0 40 ro 0 40 0 0 40 0 Ul 40 o co ro •*>> Ul 40 Ul cr. 40 ro on 40 Cr. CO 40 40 CO 0 o 0 CO CO 0 0 o 0 40 - J oo 40 40 CO o 0 o -vl CO CO 40 40 ro 47* 40 -sl OO 40 - J 0 47* OO 40 40 0 40 -vl 0 40 -vl 0 40 - J 0 •*»• 40 0 - J 4 > 0 0 OO 40 0 0 40 0 o -sl ro Ul 40 40 CO 40 47* CO 40 Cr. CO 40 40 CO 40 CO Ul 40 CO Ul 40 Ul 47* 40 0 40 40 CO 40 40 CO oo oo 40 CO -vl *»• 40 4 > 40 47* CO 40 ro 47* -s) - J 00 40 0 * -40 0 0 40 -vl o 40 -sl 0 -vl -vl 0 0 CO -sl 0 40 ro 0 40 0 0 0 -sl fm. 0 o ro Ul 40 40 CO 40 - J oo 40 ro 47* 40 40 CO 40 CO Ul 40 40 CO 40 o f-CO 40 47* 40 -sl CO 40 -sl CO 40 +*• -sl CO CO CO -vl 4 > 40 OO 40 CO CO OO Ul 40 - J ro 0 0 0 40 40 4*-0 40 Ul Ul 40 0 0 40 0 40 Ul Ul 0 40 0 40 -sl o 0 Ul -sl O o ro Ul o o o 40 -vl CO 40 CO -vJ 4 * 40 o o o 0 0 0 40 Ul 47* OO 40 en 40 Ul 47* 40 Ul 47* -sl 40 CO 47* Ul ro 00 CO o CO ro ro OO Ul 40 - j ro 47* -sl 0 Ul 0 0 0 40 0 40 0 0 40 - J o 0 0 40 -sl 0 0 0 40 40 ro 0 0 0 -vJ 0 ro ro o ro Ul ro Ul 40 -vl oo 0 0 0 40 CO Ul 40 ro 0 CO ro 40 o o 0 40 CO Ul 40 47* CO CO Ul - J CO 40 -vl CO 40 -vl CO 40 *- Ul 40 CO 47* -sl *»• 40 *» CO 40 ro 47* -sl o 0 -sl 0 O -sl 40 -sl 0 40 -sl o 40 -vl 0 40 0 0 ro ro 40 -sl 0 40 Ul 0 40 0 Ul ro o Frequency CO -sl 0 47* ro Ul CO -sl 0 47* ro Ul CO -sl 0 47* ro Ul 0 -sl O 0 ro Ul 0 -vl o 0 ro UI 0 -sl 0 0 ro Ul Amplitude ro CO ro CO ro CO ro CO ro CO ro Ul ro Ul ro *-Ul ro Ul ro •*>» Ul ro CO ro CO ro CO ro CO ro CO ro ro «>•. In ro UT ro •**>• Ul ro In ro CO ro CO ro CO ro CO ro CO ro Ul ro *«• Ul ro *-in ro •*>> Ul ro Ul Waveform O o O O o o o o O o • • • • • • • • • • ^3 ^ d ^ d ^ d 1 L The left column and bottom r o w represent the haptic icon. The perceived distance between these icons is shown in the corresponding intersection in the matrix. 116 Appendix D Instructions for Multidimensional Scaling Experiments Haptic Icon MDS Test v 2.0 The following test is intended to gather information as to how humans perceive simple forces applied through a knob. There is a set of different force feelings that will be tested. You can perceive each different force by left clicking on one of the buttons labeled with numbers on the bottom right of the screen. They are graphic representations of the force sensations. You can click on these buttons as many times as you need. These buttons can be moved to another location of the screen individually by right-clicking on the button (picking it up) and right clicking again on the on the screen in the place you want to drop the button. The buttons can only be dropped in the marked rectangular boxes in the screen. The test consists of 5 similar tasks. For the first step you will be asked to sort the buttons representing the forces into as many groups as you want (from 2 to 15 groups can be selected). By default you will get 2 groups for sorting. You can add more by clicking on the + button on the bottom of the screen. You can remove boxes by clicking on the - button on the bottom of the screen. For the first part of the test, you are required to sort the buttons into the number of boxes you selected based on perceived similarity. You can use any criteria you Want. To aid in the sorting, you can label the boxes by clicking on the label located at the top left part of each box and typing a description. This description is for your use only, but try to write words that are understandable and may aid you in identifying the box later on. The other 4 steps will be quite similar with the exception that you will not get to choose the number of boxes on the screen but rather you will be asked to sort the buttons on the given number of boxes. Remember: -You can take as much time as you need for each sort. -You can feel the buttons as many times as you need. -You can re-position the buttons as you wish. -There are no wrong answers! If you feel tired, Take a break! If you require any assistance, don't hesitate to ask! Now...to carry on to the test, Click on the button below. 117 

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.831.1-0051677/manifest

Comment

Related Items