Evaluation of fuzzy and neural vehicle control

Abstract

The authors present a neural and fuzzy solution to the collision avoidance problem of an automated guided vehicle (AGV). They describe the AGV and its sensor characteristics. Two methods based on neural networks and fuzzy logic, respectively, have been developed. The advantages and problems of each approach are evaluated. Experiments showed that the collision avoidance problem can be successfully tackled by both neural networks and fuzzy logic. Both approaches have the advantage that almost no control-specific knowledge is needed. Neural network controllers are easier to design, whereas the operation of the fuzzy logic controller is more understandable, i.e., individual rules can be adjusted to optimize certain parts of the controller behavior.<>