Intelligent self-organising controllers for autonomous guided vehicles: comparative aspects of fuzzy logic and neural nets

Concerns vehicle-dependent motion control of AGVs. Model-based methods are restricted to constant velocity and environmental conditions and small perturbations. Humans are able to generate driving algorithms with little physical insight, but with a great deal of experimental knowledge, that is flexible, robust, sufficiently precise for proper functioning, and intelligent, in that they adapt to differing environmental and payload conditions. This paper reviews two approaches to AGV intelligent control; self-organising fuzzy logic controllers, and associative-memory-type neural nets. Both approaches have the potential to provide real time, adaptive, convergent, robust decision strategies with little a priori knowledge. There are many similarities between the two approaches: a transformed input space is required in both cases: initial, approximate plant models are required; both adopt local weight/rule adaptation schemes; there is an equivalence of learning through least mean squares based adaptation rules based upon B-splines; both methods are extremely robust and fault tolerant