基于MPC的自动驾驶车辆横向路径跟踪控制

Lateral path tracking control of self-driving vehicle based on MPC

由于驾驶技能、生理极限等原因,人工驾驶员在转向操作中存在响应迟滞、动作超调等问题,控制性能优良的自动驾驶车辆可以改善上述问题。设计了一种基于模型预测控制(MPC)的自动驾驶车辆横向路径跟踪控制器。基于预瞄跟随理论建立了最优侧向加速度的驾驶员转向模型,以分析驾驶员方向盘操作中预瞄时间和车速对车辆跟踪参考路径的影响。基于模型预测控制算法设计了车辆横向路径跟踪控制器,利用反馈校正机制改进车辆预测模型,以处理参数不完全确定和外部干扰对模型精度带来的影响;采用松弛因子对目标函数进行处理,以保证目标函数具有可行解;进一步地,将所设计的模型预测控制器每一步的优化求解转化为带约束的二次规划问题,利用模型预测控制滚动优化的特点,求解跟踪参考路径所需的方向盘转角,作用于自动驾驶车辆。实验结果表明:预瞄时间和车速对驾驶员操控车辆跟踪参考轨迹的影响较大,MPC控制器下的车辆实际行驶轨迹与参考轨迹之间的最大横向偏差为0.085 m,小于熟练驾驶员操控的车辆,同时,MPC控制器下的车辆转向起始时刻相对于熟练驾驶员操控的车辆提前0.89 s。

Due to driving skills,physiological limits and other reasons,manual drivers have some problems such as response lag and action overshoot in steering operation.Self-driving vehicles with excellent control performance can improve the above problems.In this paper,a lateral path tracking controller of self-driving vehicle based on model predictive control(MPC)was designed.First,based on the theory of preview and following,a driver steering model with optimal lateral acceleration was established to analyze the influence of the driver preview time and vehicle speed on the path tracking performance.Second,a vehicle lateral path tracking controller was designed based on the MPC method,in which the predictive vehicle model was corrected by using relaxation factors to ensure that the objective function had feasible solutions.Further,the optimization solutions of the proposed MPC controller was considered as a constrained quadratic programming problem in each step.The desired steering angle for path tracking was applied to the autonomous driving vehicle by using the rolling optimization of MPC.The experimental results showed that the preview time and speed have great influence on the vehicle tracking reference track controlled by the driver.The maximum lateral deviation between the actual trajectory and the reference track of the vehicle under the MPC controller was 0.085 m,which was smaller than that of the vehicle controlled by the skilled driver.At the same time,the steering start time of the vehicle under the MPC controller was 0.89 s earlier than that of the vehicle controlled by the skilled driver.