A Trajectory Planning Method for Unpaved Roads Based on Dynamics Response Prediction

Abstract

Unpaved roads are marked by irregular undulations and more pronounced geometric curvature than paved roads. For autonomous vehicles, trajectory planning that ignores vehicle dynamic response characteristics easily gives rise to safety risks in such complex scenarios. This paper proposes a trajectory planning method that takes vehicle-road cooperative information as input and incorporates vehicle dynamic response prediction. First, local road reference lines, elevation models and risk models are constructed with sensing data from on-board and roadside devices; a quintic polynomial model is adopted to pre-plan candidate trajectory clusters, and a vehicle dynamics model is introduced for virtual pre-tracking to predict vehicle dynamic response states during tracking. A multi-objective cost function integrating vehicle safety, path smoothness, trajectory deviation and predicted dynamic response is designed to evaluate and optimize candidate trajectories under road spatial constraints and vehicle dynamic constraints. Performance tests are conducted on 84,000 road models with diverse convex-concave features generated via Latin Hypercube Sampling based on real unpaved roads. Results show that compared with traditional dynamic programming and collision detection methods, the proposed method improves the average planning success rate by 23.8% and 55.99%, increases the average planned path length by 15.4% and 47.2%, reduces the rollover index by 39.0% and 13.6%, and cuts the yaw index by 58.1% and 34.9%, respectively, which significantly enhances vehicle safety and traffic efficiency in complex road scenarios and demonstrates application potential in vehicle-to-infrastructure systems.