Approximate Model Predictive Control with Recurrent Neural Network for Autonomous Driving Vehicles

Abstract

This paper addresses an approximate model predictive control (MPC) with recurrent neural network. It has been reported in the literature that MPC is an effective method in vehicle lateral control and applied to lane keeping system. It was also shown that MPC improves tracking performance even in the presence of irregularity of waypoints. However, applying the standard MPC control law is computationally demanding in real-time control with an ECU having limited computing power. To cope with this problem, in this paper we developed a recurrent neural network to provide the approximated output of the standard MPC with off-line trained weighting matrix. For training the RNN, standard MPC is used to provide the training data set. The performance of the proposed RNN-MPC for waypoints tracking is validated through computational experiments. We conclude that the trained network shows the potential to implement the waypoint tracking system even in the presence of irregularity in waypoints.