Reinforcement Learning for Multi-Robot System: A Review

摘要

The optimization control of multi-robot systems based on Reinforcement Learning is the frontier field of Robotics and distributed Artificial Intelligence in recent years. Multi-robot systems have the characteristics of distribution, heterogeneity, and high-dimensional spatial continuity, which makes the research of reinforcement learning for multi-robot systems face a series of challenges. This paper reviews the challenges in four practical problems of the multi-robot system which are distributed collaborative driving of multiple vehicles, mobile sensing robot team, multi-robot collaborative monitoring, and multi-UAV cooperative task planning and the latest solutions of them. Methods based on Deep Reinforcement Learning and Multi-Agent Reinforcement Learning are also described. This review may be useful to guide researchers and technologists from the industry in their choice of better cope with the multi-robot system's problems.