Abstract:

The fuzzy neural network (FNN) controller was formed by combining the fuzzy control and the neural networks. The fuzzy rules were collected automatically by reinforcement Q-Learning (QL) on-line beforehand. The rules were  implicitly stored in the FNN after the off-line training. In control applications, the complex search and inference of the rules were unnecessary, and the best adaptive behavior could be produced in the output without looking up the table. The simulation results show that because all training samples are from the trained fuzzy rules, the output is almost the same as the result of the training rules.
 

Key words: mobile robot, fuzzy neural network, adaptive behavior; , fuzzy control rules