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

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. © 2023 South China University of Technology. All rights reserved.