Researchers at Wuhan University of Technology have developed a hierarchical path planning framework for unmanned autonomous vehicles operating in highly dynamic, obstacle-dense unstructured environments, such as off-road terrain, campus roads, and mixed-traffic zones. The approach decouples global and local planning to balance efficiency, smoothness, and kinematic feasibility.
Key takeaways
- Global layer: An RRT* variant enforces maximum steering-angle constraints, uses a dynamic KD-tree for faster nearest-neighbor searches, and applies B-spline smoothing to produce a curvature-continuous reference path.
- Local layer: An adaptive weighted artificial potential field dynamically adjusts goal-directed and environmental weights, enabling rapid response to static and dynamic obstacles while respecting curvature limits.
- Performance: In grid-map and ROS simulations across off-road, mixed-obstacle, and campus scenarios, the method reduced average planning time by 18–20%, node count by 30–36%, and path length by 8–12% versus existing sampling-based planners, with significantly higher obstacle avoidance success.
By separating global connectivity from local reactivity, the framework addresses common pitfalls of single-layer planners—such as poor real-time performance and infeasible trajectories—offering a practical solution for autonomous navigation in cluttered, unstructured settings. This could benefit applications like园区物流, security patrol, and mining transport.
Source: 《机器人》期刊 (robot.sia.cn) · Published 2026-07-15 · “非结构化场景中无人自主车辆分层路径规划”