Deep Reinforcement Learning for Physics-Based Musculoskeletal Simulations of Healthy Subjects and Transfemoral Prostheses’ Users During Normal Walking

This paper proposes to use deep reinforcement learning for the simulation of physics-based musculoskeletal models both healthy subjects and transfemoral prostheses' users during normal level-ground walking. The algorithm is based on proximal policy optimization approach in combination with imitation guarantee a natural walking gait while reducing computational time training. Firstly, implemented OpenSim model subject validated experimental data from public data-set. Afterwards, generic prosthesis' user, which has been obtained by number muscles around knee ankle joints and, specifically, keeping only uniarticular ones. user shows stable gait, forward dynamic comparable subject's, yet using higher muscles' forces. Even though computed forces could not be directly used as control inputs muscle-like linear actuators due their pattern, this study paves way design architecture prostheses.