Online Learning of the Dynamical Internal Model of Transfemoral Prosthesis for Enhancing Compliance

Powered prostheses hold promises to restore adaptive and robust locomotion lower-limb amputees. However, their daily use is still challenged by several shortcomings, on top of which those related control methods. This letter reports the development an controller for a transfemoral prosthesis that combines predictive torque component with feedback error-correction mechanism. The module based Locally Weighted Projection Regression (LWPR) algorithm achieves nonlinear function approximation dynamical model joints. performance proposed strategy are assessed simulated biped walker unilateral amputation. Results show LWPR-based provides accurate predictions ankle knee torques, resulting in precise position tracking. allows reducing gains mechanism one order magnitude, leading contribution total joint lower than 3% 8% knee, respectively. Compliance both joints enhanced accordingly. In addition, architecture speed changes while internal continuously learned. approach thus promising controllers prostheses.