Developments in autonomous vehicle navigation

An approach to autonomous vehicle navigation using neural networks to imitate human driving behavior is presented. The results are based on the measured actions of a real human driver in a real car. The car environment and the information that is recorded during driving are described. The problem that should be solved by the neural network is discussed. The actual learning of the driving task is presented. The quality of the leaned driving behavior is reviewed. The experiments showed that medium-sized neural networks were able to approximate the driving behavior within a maximum error of 5%.<>