Cooperative Search Strategy of Multi-UAVs Based on Reinforcement Learning

Aiming at the cooperative search problem of multiple unmanned aerial vehicles for moving targets in unknown areas in a wide area, a cooperative search strategy training method based on reinforcement learning is proposed. This method is based on Markov decision model, centralized training and distributed execution. Through strategy sharing, the dimension of reinforcement learning training process is reduced, and the dynamic environment problem is processed by fusing pheromone model. The simulation shows that the cooperative search training method based on reinforcement learning is not affected by the target movement, and can adapt to different search areas, so that the UAV cluster has the functions of cooperative search in unknown areas and online obstacle avoidance and re planning. Compared with the traditional push sweep search method, the search algorithm based on reinforcement learning training performs better when the search area is wide and the target moves fast.