Soft computing paradigms for learning fuzzy controllers with applications to robotics

Three soft computing paradigms for automated learning in robotic systems are briefly described. The first employs genetic programming to evolve rules for fuzzy behaviors to be used in mobile robot control. The second paradigm develops a two-level hierarchical fuzzy control structure for flexible manipulators. It incorporates genetic algorithms in a learning scheme to adapt to various environmental conditions. The third paradigm concentrates on a methodology that uses a neural network to adapt a fuzzy logic controller in manipulator control tasks. Simulation results of fuzzy controllers learned with the aid of these soft computing paradigms are presented.