基于经验迁移的赛车学习预测控制研究

Research on learning predictive control based on experience transfer for racing car

为提高车辆控制算法对不同道路的适应能力,在原有学习预测控制算法的基础上,本文提出一种基于经验迁移的赛车学习预测控制策略.基于所建立的赛车曲线坐标系模型,记录小车在历史赛道上的行驶轨迹,将其作为采样安全集.采样安全集蕴含了车辆行驶的经验信息.在新赛道上,通过与采样安全集内曲率相近的轨迹进行特征匹配,找出新赛道的虚拟路径跟踪轨迹.然后,对虚拟路径跟踪轨迹附近的采样点进行坐标变换,将历史轨迹转换为新赛道的虚拟采样轨迹,实现对历史赛道上的行驶经验的迁移.构造了迁移学习预测控制(TLMPC),使小车在新的赛道上能够通过学习预测控制器以更快的速度行驶.本文在4个典型赛道上进行了仿真,结果表明所设计的控制策略控制效果有明显提升.与LMPC相比,10次迭代结果中单圈耗时至少减少了1.2 s.

To improve the adaptability of the racing car control algorithm to different roads,a learning predictive control strategy based on the experience transfer is proposed.Based on the established racing car model in curve-coordinate,the driving trajectory of the car on the historical track is recorded and used as the sampled safety set.The sampled safety set contains the driving experience information of the racing car.On the new track,the virtual path tracking trajectory can be obtained by feature matching,which is carried out by comparing the current trajectory curvature with that in the sampled safety set.Then,the coordinate transformation is performed on the sampling points near the virtual path tracking trajectory,and the historical trajectory is converted into the virtual sampling trajectory of the new track,so as to realize the transfer of driving experience on the history track.Then the transfer learning model predictive control(TLMPC)is constructed,the car can travel at a faster speed with the learning predictive controller on the new track.Simulations were carried out on four typical tracks,and the results show that the control effect of the designed control strategy is significantly improved.Compared with LMPC,the time per lap in 10 iterations is reduced by 1.2 s at least.