Compliant Control of a Transfemoral Prosthesis Combining Predictive Learning and Primitive-Based Reference Trajectories

Abstract

This paper reports the development of a novel compliant controller for a transfemoral prosthesis that combines a feed-forward prediction torque component with a feedback error correction. The controller architecture aims to track primitive-based reference trajectories by the prosthetic joints. It relies on Locally Weighted Projection Regression, a function approximator that acts as an inverse internal model of the prosthesis. The proposed strategy is validated in a simulation environment.