Combining fuzzy logic and neural networks to control an autonomous vehicle

The authors present an approach to a controller design that enables a simulated car to drive autonomously around a race track. The input to the controller is the current speed of the car and several sensor signals indicating the properties of the race track, and as its output the controller determines the car´s change of direction and speed in response to the information received. The basic idea is to let a fuzzy controller supply the training data for a backpropagation neural network and use the trained network to drive the car on an unknown race track. The implementation of the proposal is described, and the driving performance of the car is evaluated. The results indicate that the combined neural/fuzzy approach is superior to solutions in which either the fuzzy controller or the neural network alone is used to drive the car