基于预瞄时间的斯坦利控制算法

The Stanley Control Algorithm Based on Preview Time

斯坦利算法以前轮中心为误差基准,缺乏人类前视驾驶特性,导致方向频繁修正。同时,当前自动驾驶仿真方法与平台在拟真度上的局限,难以有效验证算法改进的实际效果。文章提出了一种最优预瞄理论与斯坦利控制算法相结合的“预瞄-斯坦利”控制算法,首先基于Carla仿真平台,创建了虚拟组合导航模块和车辆控制器局域网(CAN)通信模块;然后,建立了自动驾驶域控制器在环的仿真平台,并将“预瞄-斯坦利”算法部署在自动驾驶域控制器中进行了在环测试;最后进行了真实物理场景下的实车测试验证。结果可知,一方面通过对比硬件在环仿真和实车测试结果中的“斯坦利-实车”与“预瞄-斯坦利-实车”横向误差,表明了“预瞄-斯坦利”算法整体误差小于斯坦利算法;另一方面对比“预瞄-斯坦利-仿真”与“预瞄-斯坦利-实车”横向误差,验证了所创建的硬件在环系统的准确性和拟真度。

The Stanley algorithm,which utilizes the front wheel center as the error reference point,lacks the human-like forward-looking driving characteristic,resulting in frequent directional corrections.Furthermore,the current autonomous driving simulation methodologies and platforms are limited in their fidelity,making it difficult to effectively validate the practical impact of algorithm improvements.In this paper,a"preview-Stanley"control algorithm combining the optimal preview theory and Stanley control algorithm is proposed.Firstly,a virtual integrated navigation module and a vehicle controller area network(CAN)communication module are created based on Carla simulation platform.Then,the simulation platform of the autopilot domain controller-in-the-loop is built,and the"preview-Stanley"algorithm is deployed in the autopilot domain controller and tested-in-the-loop.Finally,the real vehicle test under the real physical scene is verified.The results show that,on the one hand,by comparing the lateral errors of"Stanley-real vehicle"and"preview-Stanley-real vehicle"in hardware simulation and real vehicle test results,the overall error of the"preview-Stanley"algorithm is smaller than that of the Stanley algorithm;On the other hand,by comparing the lateral errors of"preview-Stanley-simulation"and"preview-Stanley-real vehicle",the accuracy and fidelity of the created hardware-in-the-loop system are verified.