Neural network approaches for lateral control of autonomous highway vehicles

Abstract

The research reported in this paper focuses on the automated steering aspects of intelligent highway vehicles. Proposed is a machine vision system for capturing driver views of the oncoming highway environment. The objective is to investigate various designs of artificial neural networks for processing the resulting images and generating acceptable steering commands for the vehicle. The research effort has involved the construction of a computer graphical simulation system, called the Road Machine, which is used as the experimental environment for analyzing, through simulation, alternative neural network approaches for controlling autonomous highway vehicles. The Road Machine serves as both the training environment and the experimental testing environment for the autonomous highway vehicle. It is composed of five (5) major modules: Highway design, Driver view simulation, Image processing, Neural network design and training, and Autonomous driving simulation. Two types of neural network control structures are under active research, Back-propagation and Adaptive Resonance. The Road Machine is written in C and operates on Silicon Graphics workstations using Unix and the SGI graphics language.