Survey of Multi-Agent Reinforcement Learning to Solve Inverse Kinematic Problems of Redundant Robotic Manipulators

Abstract

During the past decades, robots have become an important part of human life. However, due to algorithmic and hardware limitations, they have been unable to achieve high intelligence. Flexible manipulators with redundant robot technology provide a noticable degrees of freedom. Such systems offer enormous promise in a variety of sectors, including assistive and surgical, transportation, industry, and agriculture. However, to enable practical application in these environments, intelligent control frameworks and AI are required. This paper presents a survey on various MARL algorithms and models to solve inverse kinematics for robotics manipulators. We first introduce MARL from basic concepts to advanced methods and highlight inherent benefits in dealing with the issues posed by robotics research. Finally, this paper focuses on selecting model-based or model-free, including different usage of value-function-based or policy-search methods.