驾驶机器人车辆动态制动力矩补偿

为了减小长期自动驾驶过程中制动性能下降带来的影响,提出了一种驾驶机器人车辆动态制动力矩补偿方法。首先建立了以车速和制动踏板力为输入,制动力矩为输出的驾驶机器人车辆制动性能离线自学习模型。然后考虑到驾驶机器人车辆长期自动驾驶导致离线自学习模型可靠性下降,建立了以车速和制动踏板力为输入,制动力矩为输出的扩展自回归在线辨识模型,并采用模糊变遗忘因子递推最小二乘法进行参数辨识。模糊变遗忘因子递推最小二乘法通过引入遗忘因子的方式,对数据施加时变加权系数,以避免出现数据增长导致的数据饱和现象。模糊变遗忘因子控制器以制动力矩辨识误差为输入,经模糊规则推理实时输出合适的遗忘因子进行参数辨识,能够有效均衡驾驶机器人车辆制动性能参数辨识的稳定性与收敛速度。驾驶机器人车辆自动驾驶过程中,根据当前车速与目标车速的大小计算出所需的制动力矩,加上反馈回来的制动力矩误差,并结合当前时刻的车速,利用制动性能离线自学习模型与机械腿逆向运动学模型实时计算出制动电机输出位移量,实现对驾驶机器人车辆制动力矩的在线补偿。仿真与试验结果表明:利用所提出的方法对车辆动态制动力矩进行辨识时,通过调节遗忘因子,辨识结果能够快速收敛且辨识误差较小。在此基础上,控制驾驶机器人车辆进行纵向车速跟踪时,能够有效减小制动性能下降造成的影响,保证控制车速跟踪误差在±1km·h-1之内。

Dynamic Braking Torque Compensation for a Driving Robot Vehicle

To reduce the impact of the braking performance of a driving robot vehicle during long-term automatic driving, a dynamic braking torque compensation method was proposed. First, an off-line self-learning model for the braking performance was established, which used the vehicle speed and braking pedal force as input and the braking moment as output. Then, considering the fact that the reliability of the off-line self-learning model decreased because of the long-term automatic driving of driving robotic vehicles, an extended autoregressive on-line identification model with the vehicle speed and brake pedal force used as input and the brake moment used as output was established. In addition, the parameters were identified using the fuzzy variable forgetting factor recursive least squares method. This method imposed time-varying weighting coefficients on data by introducing a forgetting factor to avoid data saturation caused by data growth. The fuzzy variable forgetting factor controller used the braking moment identification error as input, and it output the appropriate forgetting factor in real time through fuzzy rule reasoning. This could effectively balance the stability and convergence speed of the braking performance parameter identification of the driving robot vehicle. During automatic driving, the required braking moment was calculated based on the current and target speeds. On this basis, the output displacement of the brake motor was calculated in real time using the off-line self-learning model of braking performance and the inverse kinematics model of the mechanical leg combined with the feedback error of braking moment and the current vehicle speed. In this manner, on-line compensation of the braking moment of the driving robot vehicle could be realized. Simulation and experimental results show that when the dynamic braking torque of the vehicle is identified by the proposed method, the identification result can be quickly converged, and the identification error is small when the forgetting factor is adjusted. On this basis, when the driving robot vehicle is controlled to track the longitudinal speed, the impact of braking performance degradation can be effectively reduced and vehicle speed tracking error is within ±1 km·h^-1.