Intelligent control of the Intelledex 605T robot manipulator

The authors present the results of the experiments which indicate how controlled robotic motion might be achieved through pattern-based paradigms, implemented for real-time operation on the Intelledex 605T robot manipulator with artificial neural nets (ANN). Previous attempts at pattern-based control have often failed, primarily because of the need for storage of an enormous number of training-set patterns and the long times required for pattern processing. It is demonstrated that these problems can be overcome through use of artificial neural networks implemented by parallel distributed processing. The feedforward Rumelhart net is investigated for the constrained robot manipulator. The robot arm, with two degrees of freedom, must move its end effector toward an observed target in the presence of disturbances. Such a control action need not be programmed in detail. Presented with a small number of training situations, the ANN can generalize and perform in many different situations. Preliminary results obtained using an Intelledex 605T and an IBM PC/AT are described.<>