Dynamic Path Planning of Low-altitude Aircraft Based on TCP-DQN Algorithm in Complex Environment

To address the issues of inefficient training, slow convergence, and poor path feasibility encountered by deep reinforcement learning algorithms in solving dynamic path planning for low-altitude aircraft, a TCP-DQN (target-guided curriculum learning and prioritized replay deep Q-network) based dynamic path planning algorithm is proposed. Firstly, a curriculum learning mechanism is introduced into the framework of reinforcement learning algorithms, where target-guided maneuver strategies are set to improve the training speed of the algorithm while optimizing the feasibility of the planned paths. Secondly, a combined reward function for training is constructed to resolve the sparsity problem of DQN reward values, and obstacle avoidance experiences of low-altitude aircraft are prioritized for replay to enhance the learning performance of the algorithm. Finally, simulation results of the TCP-DQN algorithm for path planning in 3D low-altitude dynamic environment are presented. The simulation results demonstrate that the algorithm can quickly plan the safe and efficient paths for low-altitude aircraft in dynamic and unknown threat environments.