Open Collections

Chen, Guanxiong

University of British Columbia

Accurate estimation of mass properties of the human musculoskeletal system is of great interest to many tasks, from gait analysis in biomechanics to motion tracking and control in computer animation. Previous work typically simplified the human musculoskeletal structure as a chain of rigid capsules, with muscle mass lumped with body segments. Such simplifications lead to errors in the system’s inertia matrix, and the error propagates to torque and pose estimates. In this study, we show that we can estimate the generalized joint-space inertia matrix of a human in motion, using a deep neural network or with a simple statistical model. The models do not make any assumptions other than that effective inertia matrices must be symmetric and positive definite. The models are trained and tested with real-world human data that includes synchronized motion and ground reaction forces. We show that a joint-space inertia matrix estimated from data can be physically plausible by revealing inertial coupling which a rigid, lumped inertia matrix fails to entail, and that effective inertia estimates are motion-type dependent. Moreover, we show that our neural inertia model SPDNet can predict inertia matrices parametrized by pose, body mass and height, that its predicted matrices are physically plausible, and that it generalizes well to unseen poses and mass distributions when used to reconstruct motion.

Text

2024-04-29

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/88062

Computer Science

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/