Emergency Steering Modelling of Vehicles via Ensemble Radial Basis Function Neural Network(RBFNN)

Road traffic injuries from rear-end collisions are one of the leading causes of death from vehicular collisions. An accurate driver model must be developed for autonomous vehicles to prevent such accidents. Existing models such as the conventional car-following models have been created for obstacle avoidance adopted by the parametric modeling techniques. Mentioned technique falls short due to the modeling relationship between the vehicle and the human’s driver reaction. Therefore, this study aims to develop an ensemble radial basis function neural network (RBFNN) to produce the predicted steering angle for the driver model in emergency collision avoidance. The data is collected from the ground truth rear-end collision experiment where professional drivers are involved. The variables used during modeling are relative velocity, relative distance, yaw rate, and steering angle. The performance of the developed models is determined by accessing the value of the root mean square error (RMSE) and coefficient of determination, \(R^{2}\) from the generalisation test. In addition, the comparison analysis between the proposed model and standalone RBFNN as well as feed-forward artificial neural network (FFANN) are conducted. Based on the results, ensemble RBFNN produced an outstanding performance in predicting the steering angle.