Multi-Agent Cooperative Search in Multi-Object Uncertain Environment

Multiagent multi-object search (MAMOS) in uncertain rescue environment has always been considered a challenge, which can be modeled as an Object Oriented-Multiagent Partially Observable Markov Decision Process (OO-MPOMDPs). This paper proposes a Guidance and Belief Synchronization-Object Oriented-Multi-agent Partially Observable Upper Confidence Bound Apply to Tree (GBS-OO-MPOUCT) algorithm which contains two parts. One is belief decomposition, update and synchronization to achieve environment representation, update and multi-agent internal communication, another is deadlock recognition and resolution when robots are getting trapped due to insufficient depth of Monte Carlo tree. We construct many examples to test the performance of the algorithm. The results show that our algorithm can find all objects in fewer steps and observation noise has little impact on algorithm performance.