Abstracts

This research aims to develop a personalized powered ankle-foot prosthesis. Unilateral below-knee amputees generally have walking gait patterns with temporal, kinematic, and kinetic asymmetries resulting from neuromuscular adaptations required to compensate for the lack of ankle function leading to a host of secondary musculoskeletal impairments. Powered ankle-foot prostheses can restore some ankle function by active compensation at the prosthetic joint. However, current methods are limited by: (i) reliance on able-bodied gait data for trajectory synthesis, (ii) expert tuning of subject specific parameters, (iii) an inability to adapt with the user in the long term, and (iv) a limited amount of mechanical personalization. This work seeks to directly address the first two limitations with a symmetry learning controller that automatically tunes control trajectories by iteratively adjusting the prosthetic ankle torque to match the intact ankle torque in the frequency domain. The challenge is to avoid divergence caused by the time-vary human-robot dynamics. Towards this, a rule based iterative learning algorithm is introduced that adjusts the frequency dependent learning rate based on the changes in error. The symmetry controller could conceivably be adapted for indefinite use e.g., by caching, storing, and rewriting training instances, and thus address the third limitation. To reducing the active requirement of the device, a nonlinear cam-based spring that can be customized to the user and/or activity is introduced, which provides some level of passive personalization (the fourth limitation). The new method enables symmetric ankle moment control and is tailored to the individual. The method was implemented and experimentally verified. Results indicated that symmetry of the ankle moment was significantly increased and unbounded growth of the control signal was avoided.Advisors/Committee Members: Devasia, Santosh (advisor), Klute, Glenn (advisor).