Variable learning rate neuromorphic guidance controller for automated transit vehicles

This paper presents the development and the performance of a guidance controller for automated transit vehicles operating at high speeds. The controller is based on a feedforward neural network with the back propagation algorithm for learning. Traditional back-propagation neural controllers make use of a fixed learning factor. Herein, a controller with variable learning rate, whose value depends on the operating parameters of the vehicle is described. The operating parameters considered are the linear speed of the vehicle, the instantaneous position and the orientation offsets of the longitudinal axis of the vehicle with respect to the track. Empirical relationships are derived to compute the suitable learning rates in real-time. Simulation studies illustrate that the vehicle recovers from initial offsets and follows the track within few seconds for vehicle speeds less than 4.0 m/s (14 km/hr).