Distributed Reinforcement Learning for Robot Teams: a Review

Recent advances in sensing, actuation, and computation have opened the door to multi-robot systems consisting of hundreds/thousands robots, with promising applications automated manufacturing, disaster relief, harvesting, last-mile delivery, port/airport operations, or search rescue. The community has leveraged model-free multi-agent reinforcement learning (MARL) devise efficient, scalable controllers for (MRS). This review aims provide an analysis state-of-the-art distributed MARL cooperation. Decentralized MRS face fundamental challenges, such as non-stationarity partial observability. Building upon “centralized training, decentralized execution” paradigm, recent approaches include independent learning, centralized critic, value decomposition, communication approaches. Cooperative behaviors are demonstrated through AI benchmarks real-world robotic capabilities motion/path planning. survey reports challenges surrounding cooperation existing classes We present along a discussion on current open avenues research.