Decentralized Communication-less Multi-Agent Task Assignment with Cooperative Monte-Carlo Tree Search

Abstract

Cooperative task assignment is an important subject in multi-agent systems with a wide range of applications. These systems are usually designed with massive communication among the agents to minimize the error in pursuit of the general goal of the entire system. The problem is often formulated as finding the best routing configuration to capture $N_{A}$ goals by $N_{A}$ agents in an $N\times N$ 2-Dim grid with no collision. In this work, we propose an approach for Decentralized Cooperative Communication-less Multi-Agent Task Assignment employing Monte-Carlo Tree Search (MCTS). We design a Multi-Agent MCTS with a high success rate where each agent is moving toward the collective goal effectively by knowing the current location of other agents, with no additional communication overhead. We show that by employing separated MCTS on each agent armed with a collective reward value, the total accuracy could be maximized compared to the solutions where a single MCTS executed for all the agents. As an evaluation and comparison, in the proposed MA-MCTS, agents accomplish a high success-rate by capturing all 20 random positioned goals, in a 20 by 20 2-D grid in 9.9s process-time.