关键词: 飞行车辆, 多模态路径规划, 博弈学习, 运行距离, 路径能耗

Abstract: With the air-ground multi-domain cooperative operation, the path planning problem of flying vehicles is expanded to three-dimensional space, which effectively improves the task efficiency. Traditional path planning algorithms no longer meet the planning requirements of multi-modal three-dimensional paths for the flying vehicle. A multi-modal path planning strategy based on game learning is proposed, which considers running distance and energy consumption. Firstly, a new reward function of the Q-learning algorithm, considering the influence of flight obstacle crossing capability, is designed to guide the vehicle to explore the short two-dimensional ground distance by mode switching. Secondly, the two-layer pursuit-evasion game is presented for the decision sequence of switching timing and location, and the Nash equilibrium solution is obtained by constructing a “separation distance-energy consumption” compound benefit table. In the process of interactive learning between the vehicle and environment, this table is constantly updated and a series of mode switching decisions considering running distance and energy consumption is given. Finally, the proposed strategy is verified on a 40×40 km map. The multi-modal path of the proposed strategy is shortened by 10.785 km compared with the ground travel path of the traditional reinforcement learning algorithm. And the energy consumption is 16.38% lower compared to that of the multi-modal path with minimum running distance. When the task is more urgent, the flying vehicle focuses more on the shortening of the running distance. The multi-modal path is 11.933 km shorter than the ground travel path, and the energy consumption is 0.04% lower than that of the multi-modal path with minimum running distance.

Key words: flying vehicles, multi-modal path planning, game learning, running distance, energy consumption