COMPUTATIONAL ALGORITHMS FOR ADAPTIVE ROBOT CONTROL by Guinter

Effective adaptive controller designs potentially combine high-speed and high-precision in robot manipulation. Furthermore, they can considerably simplify high-level programming by providing consistent performance in the face of large variations in loads and tasks. Globally convergent adaptive manipulator controllers have recently been developed for this purpose. However, computational complexity has so far restricted their applicability to manipulators with few degrees-of-freedom. This work presents a new recursive algorithm, which effectively implements the simple, yet globally tracking-convergent, adaptive sliding manipulator controller. To further improve efficiency, a procedure is derived to select and evaluate a minimal set of inertial parameters of the manipulator. Since many tasks require direct cartesian control, the algorithm is subsequently expanded to handle cartesian data hi real-time and possibly exploit kinematic redundancies. Finally, the application to constrained motion is discussed, introducing impedance control features to guarantee stable contacts to arbitrary passive environments. The complete system allows the effective use of adaptive strategies on multi degree-offreedom manipulators for a variety of tasks. The developments are implemented and illustrated experimentally on a four degree-offreedom articulated robot arm and suggest a wide range of application beyond adaptation to grasped loads. Thesis Supervisor: Professor Jean-Jacques E. Slotine Title: Associate Professor of Mechanical Engineering