Advances in architectures and algorithms for high-performance robot control

This thesis addresses two practical problems in robot control. The first problem is how, in order to implement control algorithms whose computational need exceeds the most powerful available computer, to harness multiple computers to do the job. In particular performance measures of update and latency are defined for distributed digital controllers. These quantities are measured in an actual working system, and experimentally demonstrated (i) to depend intimately on both the computation and communication structure of the implementation vary dramatically with changes in the system structure, a significant point which has been largely ignored in the robotics literature, and (ii) to vary dramatically with changes in the system communication protocol. The effect controller latency and update rate on stability and performance of closed loop robot tracking is experimentally demonstrated.;The second problem addressed in this thesis is the mathematical development and comparative experimental testing of a class of globally stable adaptive control laws for robot reference trajectory tracking. This class of control laws is the only reported class of controllers whose stability can be globally demonstrated with respect to the commonly accepted rigid-body robot dynamical model.;A new member of this class, an adaptive controller for reference trajectory tracking of robot arms, is presented with a rigorous proof of global stability.;Finally, this thesis presents the first comparative experimental study of the new and several established globally stable adaptive controllers. The study, obtained from both a GMFanuc A-500 industrial arm and the new "Yale Buhgler" three degree of freedom direct drive juggling robot (i) reconciles several previous contrasting empirical studies, (ii) demonstrates and compares the superior tracking performance of the adaptive algorithms, and (iii) examines contexts which compromise their advantage.;As a testbed for the experiments reported herein, a distributed robot control hardware and software architecture is developed for the Yale-GMFanuc A-500 robot arm. This implementation (i) demonstrates the advantages of message passing architectures over conventional shared-memory bus-based architectures, (ii) utilizes a modular architecture comprised of (at present) control, actuator, and host subsystems and (iii) supports the utility medium-grained real-time distributed control architectures.