A Deep Reinforcement Learning Approach to Multi-agent Search and Rescue in Unknown Environments

The increasingly frequent occurrence of natural disasters and industrial accidents has heightened the demand for efficient and dependable rescue operations. In this paper, we introduce map-sharing multi-agent proximal policy optimization (MS-MAPPO), a novel multi-agent reinforcement learning algorithm, to address the challenge of multi-agent search and rescue in completely unknown environments. Our algorithm empowers agents to independently explore the environment and exchange the generated maps with other agents, thereby enhancing the efficiency of rescue operations. Through evaluations in a simulated multi-agent rescue environment, we demonstrate the efficacy of our approach.