Neural network based hybrid adaptive controller for an autonomously driving car using thin plate spline radial basis activation function

This paper presents a hybrid lateral and longitudinal controller for a self-driving passenger car. The controller comprises a Proportional Derivative (PD) controller as a closed loop controller and Neural Network (NN) based adaptive compensator as a feed forward controller. The activation function of the NN adaptive stage is based on a poly-harmonic Thin Plate Spline (TPS) Radial Basis Function (RBF), which promises better accuracy, smoother interpolation and closed form solutions. The controller development and testing has been performed using a non-linear vehicle dynamics model, which has been developed using the Matlab/Simulink tool. The Controller performance in terms of vehicle lane following (lateral deviation control) and safe cruising control (longitudinal spacing error control) have been verified through simulations. Reductions of lateral deviation error by 15% and longitudinal spacing error by 7% have been achieved. © (2014) Trans Tech Publications, Switzerland.