Neuro Fuzzy Control of Robotic Arm Movement

Abstract

This thesis first outlines the theory, historical background, and application of neural networks and fuzzy logic. The review of neural networks and fuzzy logic is followed by a discussion of the combination of the two technologies -- neuro-fuzzy techniques. The two tools have been successfully combined to maximize their individual strengths and compensate for shortcomings. A survey is given of previous work done in applying these technologies to control systems. The problem of moving a robotic arm in the presence of an obstacle is discussed. In particular, trajectory planning of a planar, redundant manipulator is studied. The primary weakness of previous methods for determining acceptable trajectories is the massive amount of computer time needed to obtain a solution. Neuro-fuzzy systems offer not only the benefit of the parallel nature of its computations, but also the ability to learn the control of an arm by following a human's example. Several neuro-fuzzy controllers are trained using sample data obtained from a human's control of a robotic arm. Their performance is quantified and compared. It is shown that the definition of the fuzzy membership functions plays a significant role in the ability of the neuro-fuzzy controller to learn and generalize. Possible directions for future work are suggested.