Safe path planning for fixed‑wing UAVs in dynamic urban wind fields

This paper addresses safe path planning for fixed-wing UAVs in dynamic urban wind fields. To handle the coupled challenges of wind-induced risk and fixed-wing kinematic constraints, a Two-Layer Forward-Looking Risk-Aware Reinforcement Learning method for Fixed-Wing UAVs (2LRRL-FW) is proposed. The method first constructs a forward conical feasible action space under the maneuverability constraints of fixed-wing UAVs to ensure maneuvering feasibility in the decision-making process. It then introduces a Markov-process-based multi-scenario wind-field switching model to characterize the temporal dynamics of the wind field. On this basis, a two-layer forward-looking risk assessment is further carried out to simultaneously consider the current wind risk and the potential future wind risk within the UAV’s forward reachable region. Results in simulated environments and data-driven reconstructed urban environments show that 2LRRL-FW has significant advantages in safety. Compared with the benchmark methods, the proposed method reduces the average trajectory risk level and the proportion of high-risk path segments by 13%–25%, while increasing the path length by only 3%–7%. At the same time, the method maintains good stability and reliability under dynamically switching wind directions.