Neural network path following for differential drive vehicle

In this paper, a neural network path following controller for a differential drive vehicle is proposed. A differential drive vehicle has nonlinear characteristics and it is difficult to know its position exactly without using global/external position sensors. The neural network controller which has the characteristic of the PID controller is used in this study. We propose 1. using the additional/redundant derivative information in the input layer to improve learning efficacy, 2. applying a sigmoid function as the activation function in the output layer to remove difficulties on input saturation of actuators, and 3. novel cost function to improve the learing efficacy. Simulation and experimental results show the effectiveness of the proposed method.