Research on UAV Path Planning Algorithm Based on Improved DQN
Xiaobo Wang, Shan Yang, Haina Ye, Ti Wang, Zhongyan Du, Dongchun Wu, Xinwei Wang · 2024
To address the issues of insufficient exploration capability and long training times in the classic DQN algorithm for UAV path planning, an improved DQN path planning algorithm is proposed. This algorithm optimizes the reward function by redefining the reward signals, enabling the UAV to more intelligently perceive the environment and task requirements. To enhance the UAV's exploration ability, the algorithm also introduces an adaptive epsilon adjustment mechanism, allowing the UAV to automatically adjust its exploration strategy according to the environment and task. In simulation experiments, the average path planning time was only 0.09s, and the path length was only 50m. The training time of the algorithm was reduced from 12 hours to 4 hours.