Development of Adaptive Model Predictive Control Using Radial Basis Function Neural Network Approximator for Autonomous Vehicles

This paper proposes an adaptive model predictive control algorithm for autonomous lateral control using a neural network approximator. The model predictive control algorithm can ensure reasonable control performance by reflecting the physical constraints of the system and deriving the optimal control input. However, disturbances and model uncertainties during the prediction process of the model predictive controller can have a negative influence on the control performance. To overcome these limitations, this study proposes the estimation of the weight matrix of the model predictive controller using a neural network approximator. A prediction model based on the lateral error dynamics model is designed to formulate the model predictive control, where the designed prediction model is reflected in the objective function to minimize it and derive optimal inputs satisfying the constraints. The estimated weighting matrices using a radial basis function neural network, are then incorporated into the model predictive controller. The performance evaluation of the proposed control algorithm was conducted using a double lane change scenario. The proposed control algorithm was designed in the Matlab/Simulink environment, and CarMaker simulator-based performance evaluation was performed.