Comparison of benchmarking algorithms in Prosthesis

Abstract

Due to the remarkable performance of deep reinforcement algorithms in numerous benchmark tests and current environmental setups, Deep Reinforcement Learning has gained a lot of attention in recent years. In such methods, the combination of Deep Learning, which is already well established and strong, and Reinforcement Learning, which is unique, provides amazing results. To enable valuable technical and practical insights into these methods and their results, it is deemed necessary that this paper provides a comprehensive look at these methods and their results, which is concise, accurate, and comparable. Passive prostheses don't fully restore lost functions like robot prostheses do for amputees with transfemoral amputations. There are, however, several issues that must be resolved before an automatic control system could be developed for prosthetic devices. It is a natural tool to use that is based on reinforcement learning (RL). The controlled prosthesis should be able to adapt to different task environments as quickly and easily as possible when it is controlling the prosthesis with the user in the loop. When an environment changes during a runtime, most real-time agents have to re-learn all the rules.