A Novel UAV Path Planning Method Based on Layered PER-DDQN

Path planning is a key technology for Unmanned Aerial Vehicles (UAVs) to complete the operational mission in a complex battlefield environment. A step-by-step path planning method based on the Layered Double Deep Q-Network with Prioritized Experience Replay (Layered PER-DDQN) is proposed in this paper. The novel method is constructed by combining the threat avoidance network and collision-free network based on the PER-DDQN framework. By analyzing the current environment of the UAV, the networks output threat avoidance action vector and obstacle avoidance action vector, and the method does a weighted summation of the action vectors according to the weight of the subproblems to obtain the final action. The simulation experiment verifies that the Layered PER-DDQN path planning method has better convergence and practicability than the Deep Q-Network and A* algorithm.