基于高级运动模型轨迹预测的不确定性分析

Uncertainty analysis of trajectory prediction based on advanced motion model

为研究车辆模型不确定性与环境感知不确定性对车辆轨迹预测精度的影响,利用Matlab建立恒横摆率和恒速度(constant turn rate and velocity,CTRV)轨迹预测模型,结合不同工况下采集的全球定位系统(global positioning system,GPS)与惯性传感器(inertial measurement unit,IMU)融合定位数据集,运用扩展卡尔曼滤波(extended Kalman filters,EKF)算法处理车辆模型的过程噪声与传感器的测量噪声。在此基础上进行仿真实验,分析不同行驶工况下车辆轨迹预测误差及对轨迹预测模型5个状态量的影响。结果表明:EKF算法能够很好地处理车辆模型过程噪声和传感器测量噪声,直线行驶工况下的轨迹预测误差控制在0.3 m以内,小曲率弯道行驶工况下的轨迹预测误差范围为1~9 m,大曲率弯道行驶工况下的轨迹预测误差范围为2~38 m;利用基于CTRV的轨迹预测模型结合EKF算法处理不确定性时,道路曲率的大小会直接影响车辆偏航角的滤波,甚至会使车辆偏航角滤波轨迹发散,导致大曲率弯道工况下的轨迹预测误差较大。

In order to study the influence of vehicle model uncertainty and environmental perception uncertainty on the vehicle trajectory prediction accuracy,a model of constant yaw rate and constant turn rate and velocity(CTRV)trajectory prediction was established by using Matlab.Combined with the global positioning system(GPS)and inertial measurement unit(IMU)fused positioning datasets under different working conditions,the extended Kalman filters(EKF)algorithm was used to process the vehicle model process noise and sensor measurement noise.On this basis,simulation experiments were carried out to analyze the vehicle trajectory prediction error under different driving conditions and its influence on the five state variables of the trajectory prediction model.The results show that the EKF algorithm can handle the vehicle model process noise and sensor measurement noise well,and the trajectory prediction error under straight-line driving conditions is controlled within 0.3 m,the trajectory prediction error range under the small curvature curve driving conditions is 1~9 m,and the trajectory prediction error range under large curvature curve driving conditions is 2~38 m.When the trajectory prediction model based on CTRV is combined with the EKF algorithm to deal with the uncertainty,the road curvature will directly affect the filtering of the vehicle yaw angle,and even cause the vehicle yaw angle filtering trajectory to diverge,resulting in a large trajectory prediction error under the condition of large curvature curves.