Connectionist dynamic control of robotic manipulators

Abstract

Motion control of robotic manipulators has traditionally consisted of following preprogrammed sequences of movement. In order to take full advantage of the inherent flexibility and versatility of these manipulators, there is a need for correspondingly more flexible and robust control schemes. This thesis finds a solution to the problem of coordinated motion control and trajectory synthesis for robotic manipulators, by combining the learning and adaptive techniques of a connectionist paradigm with the proved and sound methodology of control theory. The development of this thesis includes the use of the functional decomposition principle that simplifies the connectionist models, the dynamic optimality principle that gives a convenient Cartesian feedback, a theoretical analysis of stability and performance of optimal trajectory control and the differentiated on-line learning for adjustments of the connectionist feedforward during control. Extensive simulation experiments have provided examples of the capabilities of these models, and possible applications to real systems. Formal theory is used to demonstrate the soundness of the developed control schemes. The conclusions of this work yield some insight on how to proceed towards an integrated architecture for dynamic control in robotics. (Copies available exclusively from Micrographics Department, Doheny Library, USC, Los Angeles, CA 90089-0182.)