Journal of South China University of Technology(Natural Science Edition)

• Academic Achievement Album of Young Editorial Committee • Previous Articles     Next Articles

Three-Dimensional Path Planning for UAV in Complex Wind Fields Based on an Improved Q-Learning Algorithm

WANG Haibo1  WANG Zihao1  CHENG Huazhen2*  LIN Ke2  CHEN Ningning1，2  ZENG Weiliang3    

  1. 1. School of Civil and Transportation Engineering, Guangdong University of Technology, Guangzhou 510006, Guangdong, China;

    2. Zhenye Uctrl Technology Corp. Ltd., Zhongshan 528400, Guangdong, China;

    3. School of Automation, Guangdong University of Technology, Guangzhou 510006, Guangdong, China

  • Published:2026-02-11

Abstract:

This study proposes an enhanced Q-Learning algorithm for UAV path planning in complex 3D wind fields. By designing a hierarchical composite reward function that integrates wind vectors, path smoothness, and spatial constraints, the algorithm optimizes the trade-off between flight distance and environmental costs. Validated across six diverse wind field scenarios, the results indicate that the proposed method significantly minimizes cumulative wind speed consumption and headwind exposure compared to conventional algorithms. While a slight increase in geometric length is observed, the trajectories feature significantly improved smoothness and reduced inflection points, aligning better with UAV kinematic constraints. The algorithm demonstrates stable convergence within 2,000 iterations. The results confirm the robustness and adaptability of this method for safe and efficient UAV autonomous navigation in dynamic wind environments.

Key words: 3D path planning, Q-Learning algorithm, complex wind fields, hierarchical composite mechanism, multi-objective optimization