考虑时变参考的L2+车辆避障规控方法研究OA

Current emergency obstacle avoidance functions achieve limited adaptability in complex and dynamic traffic environments.Although high-level automated avoidance systems improve environmental interaction and decision-making capability,they usually depend on high-precision perception and strong onboard computing resources,which restrict efficient deployment on L2+vehicles.This challenge is more prominent on low-cost platforms that mainly rely on millimeter-wave radar,because such platforms face both limited environmental representation and strict real-time constraints in planning and control.To address these issues,this paper develops an integrated obstacle avoidance planning and control method with time-varying references for L2+vehicles.It aims to improve the coordination between trajectory generation and tracking under constrained perception and computation.Thus,the vehicle maintains safe and smooth real-time performance in dynamic scenarios. At the planning level,the method builds a local planning framework in the Frenet coordinate system to handle complex road conditions with clear geometric structure and manageable computational complexity.The framework decouples longitudinal and lateral motion and generates candidate trajectories with quintic polynomials,it also evaluates each candidate under road boundary constraints,obstacle constraints,and vehicle dynamic feasibility.This design allows the planner to screen infeasible trajectories early and retain trajectories that satisfy both collision avoidance requirements and motion continuity requirements.On this basis,it further optimizes trajectory shape and speed distribution,so that the final local trajectory maintains continuity,smooth curvature evolution,and executable motion characteristics while still preserving traffic efficiency.Therefore,the planning strategy seeks a collision-free path and explicitly balances obstacle avoidance safety,trajectory smoothness,and driving efficiency in a unified framework. At the control level,the method addresses a key gap between planned trajectories and actual vehicle execution.In real driving scenarios,road curvature changes continuously,while steering actuation exhibits non-negligible delay.These factors often degrade tracking accuracy if the controller fails to consider them in prediction and optimization.To address the problem,it builds a predictive tracking model that incorporates time-varying road curvature parameters and explicitly describes steering delay characteristics.The model places path geometry variation and vehicle steering dynamics in a unified prediction framework,improving the controller's ability to follow planned trajectories under rapidly changing road geometry.To satisfy the real-time requirement of resource-limited onboard platforms,the method further introduces a lookup-mapping mechanism based on the inverse matrix derived from the KKT conditions.This design avoids heavy online iterative computation and enables fast analytical calculation of control parameters,reducing online computation and improving real-time control capability without sacrificing tracking precision. Simulation and vehicle tests verify the effectiveness of the proposed method.Under the 64 km/h condition,it reduces lateral displacement error by 35.8％and constrains lateral velocity error and heading angle error within 0.05 m/s and 0.072° respectively.Under the double-lane-change condition,it reduces lateral error and heading angle error by 60.8％and 61.1％respectively,compared with a conventional MPC controller.Real vehicle tests further show the maximum lateral error remains within 0.1 m and the heading angle error stays below 0.02°.All these demonstrate the proposed method achieves dynamic obstacle avoidance with strong safety,good smoothness,and high real-time performance under limited perception and computing resources.It may provide a practical planning and control solution for engineering deployment of L2+intelligent vehicles.