基于相关向量机和样条曲线的换道轨迹规划

Programming of Lane Change Trajectory Based on RVM and Sample Curve

针对现有跟踪算法不能很好地适应强非线性动态跟踪问题,提出利用稀疏特性更强的相关向量机(RVM)对换道参数进行估计,并采用样条曲线对估计值进行规划。将车辆纵向车速和横向车速以及横摆角速度作为表征参数,依据真车试验数据建立相应的运动方程、RVM输出方程以及条件概率方程,选取具有良好非线性特性的高斯函数作为核函数,依据不同带宽下模型性能的优劣确定最佳的带宽。对较为敏感的横摆角速度进行了滤波处理,通过对边缘似然函数进行迭代求解确定出权重分布和估计噪声,并采用RVM模型对换道参数进行估计,利用B样条曲线规划换道轨迹。测试结果表明:RVM具有良好的估计特性,相比SVM而言,其对核函数的敏感度较低,测试时间短,对大样本数据具有良好的适应特性,经过B样条曲线规划后,估计值的连续性和尖峰特性得到最大限度地改善。

Because the existing track method is not appropriate for the strong non-linear problems, the more sparse relevance vector machine （RVM） is used to predict trajectory changing parameters, and sample curve is used to plan the estimation value. Longitudinal speed, lateral speed and yaw rate are taken as the characteristic parameters, corresponding movement equation, the output equation of RVM and condition probability equation are established on the basis of the real vehicles test data. Then well non- linear gauss function is to be as kernel function in the built model. According to the model performance in different band width, the best bandwidth can be determined. The more sensitive yaw rate data are filtered by Kalman. The distribution of weight and noise estimation can be determined by the iteration of log marginal likelihood. Moreover RVM model is adopted to estimate the parameters of lane change and the change trajectory programming can be finished by sample curve B. The test results show that RVM haswell estimation performance. By comparison with SVM, RVM is less sensitive to kernel function and has shorter test time. Moreover, RVM has good adaptability for large samples. After programming of lane change by sample curve B, the continuity and peak performance of estimation value can be improved significantly.