基于模糊双曲正切模型的车辆路径跟踪算法

为了实现铰接式车辆无人驾驶技术。针对路径跟踪问题,该文提出了基于模糊双曲正切模型的铰接式车辆路径跟踪控制算法。首先根据实地试验测得铰接式车辆的横向偏差、横向偏差变化率、航向角偏差、航向角偏差变化率和转向角的样本数据,建立其模糊双曲正切模型。在此基础上,采用改进的自适应反向传播(back propagation,BP)神经网络对模型进行参数辨识,并推导了基于Cauchy鲁棒误差估计器的权系数调解率公式。然后设计基于极点配置方法的控制器,得到转角的反馈控制率。从试验数据可以看出:车辆横向位置偏差、航向角偏差、转角控制量分别控制在0.008 m、0.07 rad(0.5°)、0.21 rad(12°)附近,各向偏差均稳定,误差控制在1%以内。该种路径跟踪控制算法的研究可为铰接式车辆无人驾驶提供参考。

Path tracking algorithm of vehicles based on fuzzy hyperbolic tangent model

Abstract: To solve the autonomous driving problems of underground mining truck, especially the path tracking problem, predecessors have researched articulated vehicle kinematic and control through the analysis of the kinematics modeling. Although articulated vehicle dynamic was taken into consideration in path tracking control fields, there would be also lack of accurate system modeling. So considering this multi variable, strong coupling, highly complex nonlinear dynamic system of articulated vehicles, it''s difficult not only for establishment of an accurate model, but also for design of a precise control algorithm. Fuzzy hyperbolic model and the method of pole placement controller design in this article could provide a better way to solve these problems above. In this method, information with driver driving the truck included the vehicle kinematics relations, and by repeat of the same process, the vehicle dynamics relationship would be clearer. This method reduced the complexity in process modeling. The fuzzy hyperbolic model could cleverly convert the coefficient matrix of the nonlinear system to a constant matrix, then neural networks as a supervised learning method could be used to identify the parameters, which made the model more closed to the true model and it would be convenient to design a control algorithm. Pole assignment as a classical control algorithm was simple and effective and it could be appropriately applied into this kind of model. Following steps were needed to establish the fuzzy tangent model and the pole assignment method. Initially, the sample data, including the lateral displacement error and orientation error, were collected through the driver controlling articulated vehicle at the speed of 3 m/s. During the multiple times of driving in the same roadway, little change could be achieved for the lateral displacement error and the orientation error. After that, by using the improved adaptive BP neural network model and the mediation rate of error estimator based on the method of Cauchy robust, the system error of neural network learning was effectively reduced, and the fitting error and relative error of data were decreased. So the system was better matched, and the weights were well identified. Finally, the pole assignment method with choosing the appropriate poles was designed to control articulate angle. The Hardware-In-the-Loop (HIL) simulation platform was set up on the basis of PXI and C-RIO as a host computer. The kinematic model established in the Adams platform was downloaded into the PXI as the simulation plant, and the path tracking algorithm compiled by Simulink was embedded to the C-RIO as the real electronic control unit. The host computer coupled the vehicle model and the path tracking algorithm via the Labview platform and displayed the simulation status in the upper monitor. The results suggested in the process of driving this method could describe both the quantitative relation of the horizontal position deviation and course angle deviation with their rates of change, and could provide an accurate control. HIL results showed that the lateral displacement error, orientation error and the articulate angle of the vehicle were respectively controlled at 0.008 m, 0.07 rad (0.5°), 0.21 rad (12°). All the deviations were asymptotically stable. Overshoot and the response time were less than expectations and eventually stabilize. Articulated vehicles were maintained to the low level deviation and reference trajectory coincides with the track and also the simulation time and calculating time were all 80 s in the hardware in the loop simulation process, so the controller met the requirement of the real-time control performance. The method is demonstrated to be effective and reliable in path-tracking for the underground vehicles.