A Target-coupled Multiagent Reinforcement Learning Approach for Teams of Mobile Sensing Robots

Abstract

A mobile sensing robot team (MSRT) is a typical application of multiagent system, which faces the huge challenge of environment dynamism that the change of action selection of one robot may influence behaviors of the other robots. In this paper, we investigate reinforcement learning methods for MSRT problem. A target-coupled multiagent reinforcement learning approach is proposed to solve the MSRT problem by taking advantage of both the knowledge of each agent and the local environment information sensed by the agent for achieving a shared goal in a common environment. We show the strength of our approach compared to the existed decentralized Q-learning in MSRT problem, where sensing robots are able to meet targets coverage requirement by discovering a coordinated strategy.