基于深度学习辅助修正的履带车辆路径跟随控制研究

To enhance the path-following accuracy of unmanned tracked platforms in urban and factory area applications, and to reduce the impact of skid-steering slip on the vehicle’s path, a deep learning assisted path-following control method for tracked vehicles was proposed. A vehicle slip ratio prediction model was established based on a convolutional neural network-radial basis function (CNN-RBF) network, which keeps root mean square error(RMSE) lower than 0.101 recognition and prediction accuracy for the slip ratio based on rotational speeds of the tracks on both sides of the vehicle during urban road driving. A vehicle path-following control algorithm was developed using linear time varying-model predictive control (LTV-MPC) technology, and the predicted slip ratio was used for auxiliary correction to improve path-following control accuracy of tracked vehicles in urban application scenarios. The Recurdyn-Simulink co-simulation results show that, compared to path-following control without correction assisted, the deep learning assisted path-following control algorithm for tracked vehicles optimizes the following error by an average of 45.5%, with a maximum optimization of 67%.