无人驾驶汽车路径跟踪控制方法拟人程度研究

无人驾驶汽车路径跟踪控制是无人驾驶汽车运动控制的核心所在，目前常用的路径跟踪模型主要以路径跟踪精度为主要控制目标，在很大程度上忽略了无人驾驶汽车的乘坐舒适性和控制的拟人程度。为了研究无人驾驶汽车路径跟踪控制算法的拟人程度并提高乘坐舒适性，基于转向几何学、汽车运动学和汽车动力学理论建立实车中常用的4种路径跟踪模型，提出以路径跟踪过程中的最大横向加速度a_ymax和方向盘转角平方和δ_w²共同表征路径跟踪模型的拟人程度和横向乘坐舒适性。基于驾驶人实车换道试验数据，建立多项式拟人换道参考路径，搭建CarSim/Simulink联合仿真模型，并对其进行不同车速下的车辆换道试验。研究结果表明：路径跟踪模型的横向循迹偏差均会随着车速的提高而增加，但都能较好实现路径跟踪；带预瞄路径跟踪模型和动力学前馈最优LQR路径跟踪模型拟人程度较好；汽车运动学路径跟踪模型的乘坐舒适性最差，方向盘修正激烈；在100 km·h^-1，a_ymax>0.7 m·s^-2，δ_w²>2.7×10³时，拟人程度最差；不带预瞄路径跟踪模型循迹精度最高，且拟人程度最高，乘坐舒适性最好，120 km·h^-1时，a_ymax ≤ 0.5 m·s^-2。

Investigation of Humanoid Level of Path Tracking Methods Based on Autonomous Vehicles

Autonomous vehicle path tracking has an important role in the motion control of unmanned vehicles. Presently, commonly used path tracking models consider path tracking precision as the main control goal, and largely ignore the unmanned vehicle ride comfort and humanoid level. To investigate the anthropomorphic degrees of unmanned vehicle path tracking control methods, and to improve ride comfort, four typical path tracking models were established based on the steering geometry, vehicle kinematics, and dynamics theories. The humanoid level and ride comfort of different path tracking methods were characterized by the maximum lateral acceleration a_ymax and the sum of squares of the steering wheel angles (δ_w²) during path tracking. A polynomial lane change path was established based on the actual experimental vehicle data. Additionally, a CarSim/Simulink co-simulation model was established. The results of lane changing under different vehicle speeds revealed that the tracking error of the above models increased with the increase of vehicle speed, and satisfactory path tracking results were achieved. The steering geometry method with preview and optimal dynamic control with feed forward term path tracking models had a good humanoid level degree. The ride comfort of the kinematics path tracking model was the least satisfactory with a_ymax being over 0.7 m·s^-2 and δ_w²>2.7×10³ being under 100 km·h^-1. The steering geometry path tracking model without preview had the highest tracking accuracy and highest degree of humanoid level, with a maximum lateral acceleration below 0.5 m·s^-2 under 120 km·h^-1.