Multiagent Cooperative Search Learning With Intermittent Communication

Communication plays a crucial role in coordinating the behavior of multiple agents. However, unstable communication connections in complex environments may lead to intermittent communication, information delays, and control strategy failures. This study proposes the Multi-Agent Cooperative Search Learning (MACSL) algorithm to achieve efficient search tracking in dynamic partially observable environments with intermittent communication. First, to enhance search efficiency when global communication links are unreachable, we propose a cooperative search strategy based on reinforcement learning from the perspective of teammate strategy learning. By designing an environment aware map to guide agent exploration and learning, effective distributed coverage search is realized. Second, to mitigate the impact of communication interruptions on shared information loss, we investigate target information prediction based on recurrent neural network. The update rule of the target probability map and the cooperative model are optimized. Experimental results validate the effectiveness of the MACSL algorithm for cooperative search with intermittent communication.