A Novel Approach for Integrated Vehicle Dynamics Management Using Improved Adaptive Neural Network

The present work has proposed a novel approach for autonomous vehicle path tracking, as well as integrated vehicle dynamics management. Our approach adopts the nonlinear model-based control (NMPC) model to account for the unknown nonlinearity in vehicle dynamics, together with a cost function to enable minimum input parameters for our NMPC model. An improved Random Projection Neural Network (RPNN) is then proposed, featuring a novel weight adaptation method utilizing gradient descent from the input to the hidden layer, and an adaptive law grounded in Lyapunov stability theory from the hidden to the output layer. This improvement aims to address computational burdens and the challenges posed by unknown nonlinear uncertainties in lateral tyre forces, which are not adequately managed by Nonlinear Model Predictive Control (NMPC). Experimental results demonstrate that the developed approach, integrating the improved RPNN with NMPC, achieves accurate path-tracking performance with a minimal amount of data. This is attributed to the high adaptability of the enhanced RPNN algorithm, which updates weights in real time using incoming data.