Reward Shaping in Reinforcement Learning for Prosthetic Knees

Abstract

In this paper we have looked into Reinforcement Learning Algorithm for Prosthetic Knee Joint. Model Based Aggregate Policy Optimization (MBAPO).Model Free Q Learning with Model Based Learning,It has been found out that with Reward Shaping is an important aspect of Reinforment Learning and can improve the learning in many ways.We have compared the different reward functions and compared the performance on each learning.This approach by far shows the better learning reward and learn faster.This is done in OpenSim and Prosthetics Environment was used.