Multi-agent Cooperative Search based on Reinforcement Learning

Regional cooperative search is an important scenario for Unmanned Aerial Vehicle (UAV) fleet. It is difficult for the UAV fleet to adapt to the dynamic environment by using traditional search method, which needs pre-planning. In this paper, a reinforcement learning approach is employed to train the UAV agents to search without pre-determined strategies. By designing reward function, state and action spaces, agents can learn to make decisions all by themselves based on their own observations and cooperate with each other effectively. The proposed method is experimentally validated by numerical simulations, and the result show that the UAV fleet can complete task with a pretty high coverage and few repetitions at the same location.