Neural-network-based vision processing for autonomous robot guidance

摘要

The autonomous land vehicle in a neural network (ALVINN) project addresses the problem of training artificial neural networks in real time to perform difficult perception tasks. ALVINN is a modular connectionist system that uses inputs from a video camera and an imaging laser rangefinder to guide the CMU Navlab, a modified Chevy van. This paper describes a technique for rapidly training expert networks for new driving circumstances. A rule-based integration scheme that uses a symbolic planning system to combine multiple experts is also presented.