Online planning for multi-agent systems with consensus protocol

This article is concerned with decision-making and coordination among agents under uncertain conditions. We propose a novel algorithm for on-line planning of a multi-agent system based on Decentralized Partially Observable Markov Decision Processes (DEC-POMDP). This algorithm helps the multi-agent team to make a prudent decision in uncertain environment as prompt as possible. When communication is permitted, the agents exchange their partial information using a consensus protocol, such that the agents can approach a unique belief space. This guarantees that every agent makes a distributed decision which is the most beneficial to the team, unnecessary or false decisions are avoided. Furthermore, the algorithm can effectively reduce the agents´ dependency on observation and historical information, computational complexity is also decreased. Simulation results show feasibility and validity of the proposed algorithm.