Recent advances of reinforcement learning in multi-robot systems:A survey

Abstract

Multi-robot optimization control based on reinforcement learning is a research frontier of robotics and distributed artificial intelligence in recent years.Some characteristics in multi-robot systems,such as distribution,heterogeneity and high-dimensional continuity,lead to a series of challenges in theoretical and methodological research for multi-robot reinforcement learning.Therefore,recent advances of multi-robot reinforcement learning are systematically surveyed.Firstly,the fundamental theoretical models and optimization objectives are analyzed.Based on a contrastive analysis for existing algorithms,the difficulties in theoretical research and implementations are discussed,and the possible solutions are summarized in detail.Several benchmark problems and applications are listed.Finally,current work and future research directions are concluded.