基于深度学习的自动驾驶车辆模型迁移路径规划研究

Research on Migration Path Planning of Autonomous Vehicle Model Based on Deep Learning

为解决传统路径规划方法中无车辆动力学约束和消除车辆模型跟踪误差的问题,本文提出一种基于深度学习的自动驾驶车辆模型迁移路径规划方法。首先,根据真实环境建立虚拟行车环境模型,该模型应用经深度学习训练后的最优自动驾驶策略;其次,通过MATLAB自动驾驶场景设计器将实际场景问题迁移至虚拟抽象模型中;最后,经过仿真实验验证该方法。实验结果表明,所提方法能够减少横向跟踪误差和提高模型的泛化性能,减小过度依赖问题。

In order to solve the problem of no vehicle dynamics constraints in the traditional path planning method and eliminate the tracking error of the vehicle model,a deep learning-based autonomous vehicle model migration path planning method is proposed.First establish a virtual driving environment model based on the real environment,which applies the optimal autonomous driving strategy after deep learning training;then transfer the actual scene problem to the virtual abstract model through the MATLAB automatic driving scene designer,and finally verify it by simulation experiment this method.Experimental results show that the proposed method can reduce the lateral tracking error,improve the generalization performance of the model,and reduce the problem of excessive dependence.