Learning Walking on a Musculoskeletal Human System with a Prosthesis

Abstract

Incompetent design of prosthesis for amputees inflict pain in muscles and bones contingent to the prosthesis. Simulation models mimicking human movement promise a prosthesis with improved movement capability for amputees. Musculoskeletal models enable better anticipation of prosthesis contributions to the human musculoskeletal system during walking movement. In this paper, we apply a simulation of musculoskeletal model on an amputated human model with a prosthesis using Gaussian Process Regression Machine Learning Predictor and deep reinforcement learning. The performance of two versions of a prosthesis, one being a simpler version (passive prosthesis) and one being relatively better version (active prosthesis) are evaluated and compared to that of a healthy human model.