Research on Cooperative Collision Avoidance Problem of Multiple UAV Based on Reinforcement Learning

Abstract

To solve the problem of multi-UAV collision avoidance, propose the solution framework of multi-UAV collision avoidance based on the reinforcement learning. Firstly, four elements of state space, action space, the environment model and return value in the multi UAV cooperative collision avoidance problem were analyzed in depth. For the environment model, modeling it from two aspects of the UAV dynamic model and multi UAV relative distance calculation model; for the reward calculation, analyzing it from three aspects of the effect of collision avoidance, pre route deviation and maneuver execution cost. Then, proposing the multi UAV cooperative collision avoidance control algorithm based on the reinforcement learning. The experiment results show that the algorithm can effectively realize multi UAV coordination collision avoidance, control decision time is about 100 milliseconds, and the algorithm has practical feasibility.