Multi-UAV Collaborative Search and Strike based on Reinforcement Learning

Abstract

Abstract The problem of multi-UAV collaboration is one of the important research contents of multi-UAV systems. Aiming at the optimal strategy of multi-UAV collaborative search and strike against learning targets for moving targets in an unknown environment, a multi-UAV collaborative search and strike algorithm based on multi-team reinforcement learning is proposed. First, we establish a search probability map for the environment to search for moving targets After finding the target, the moving target is allocated by auction to form an independent investigation team for different goals. Then each inspection team learns its optimal strategy in cycles at the same time. At the same time, by changing the learning speed and supervising the learning results during the learning process, the convergence and effectiveness of the learning results are guaranteed. The simulation results show that the UAV can better adapt to the dynamic environment through training. Under the condition of no prior information, it can effectively execute the search and strike task against the moving target with learning ability. And compared with the learning method of Markov decision process, the convergence speed and learning speed are faster.