Abstract: To address the limitations of traditional dynamic window approach (DWA) in global path tracking accuracy, delayed dynamic obstacle avoidance, and inadequate adaptation to vehicle kinematic constraints, a multi-dimensional improved DWA algorithm was proposed. Methodologically, the study: 1) constructed a turning radius constraint cost function based on the Ackerman steering model, employing exponential penalty terms to suppress out-of-limit path generation; 2) designed a spatiotemporal trajectory prediction mechanism that utilized models to forecast dynamic obstacle trajectories and calculate collision risks, thereby enhancing dynamic obstacle avoidance capability; 3) proposed a hierarchical dynamic weight adjustment strategy that optimized weight allocation in real time based on target distance and obstacle proximity; 4) introduced a fractional-order PID controller for trajectory tracking optimization, employing Grünwald-Letnikov discretization to reduce overshoot and steady-state errors. Simulation and physical vehicle experiments demonstrate that in static obstacle scenarios, the improved algorithm reduced motion time by 9.5% and shortened path length by 6.8%; in dynamic obstacle environments, it achieved an additional 5.9% path length reduction with 9.9% fewer iterations. The fractional-order PID controller exhibited a maximum tracking error of merely 0.2 m, significantly outperforming the 0.6 m error of traditional PID control. The conclusion verifies that the improved algorithm effectively improves path planning feasibility, dynamic obstacle avoidance efficiency, and trajectory tracking precision, making it suitable for unmanned vehicle local path planning in complex dynamic environments.