A dual-layer hybrid path planning method for autonomous vehicles via multi-strategy enhanced GWO and adaptive DWA

To address the limitations of traditional path planning algorithms-such as insufficient initial exploration, a tendency to get stuck in local optima, and weak dynamic obstacle avoidance capabilities-this paper proposes a two-layer hybrid path planning framework consisting of an upper-layer global planning component and a lower-layer local obstacle avoidance component. The upper-layer algorithm incorporates multiple strategic improvements to the traditional Grey Wolf optimization algorithm. It introduces the Sobol sequence and a lens-based reverse learning mechanism to optimize the initial population distribution. It combines annealed Softmax dynamic weights with a spiral approximation mechanism to enhance optimization accuracy and robustness. The underlying algorithm employs an adaptive prediction time-domain strategy tailored for the dynamic window method, enabling it to dynamically adjust the prediction horizon based on the complexity of the environment. Furthermore, by reformulating the evaluation function to incorporate dynamic safety constraints and motion smoothing factors, we significantly enhanced vehicle safety during high-speed operation. Experimental results show that, in a static grid environment, the improved global algorithm reduced path lengths by approximately 17.92% and 53.20% on maps of different scales compared to the original algorithm. In a high-fidelity Gazebo physics simulation containing multiple dynamic obstacles, the hybrid framework demonstrated exceptional real-time re-planning capabilities. Compared to existing fusion approaches, it generated shorter planning paths, achieved lower average runtime, and significantly reduced the number of turns.