Adaptive control of free-floating and free-flying robotic manipulators

New control algorithms are required which are capable of adapting to the unstructured and highly nonlinear uncertainties inherent in space manipulator operations. In this thesis a neural network approach is adopted to address the problem. The resulting controller is capable of “learning” the necessary uncertainties while controlling the manipulator in a stable and convergent manner. The neurocontrol algorithms presented in this thesis represent a new and important enabling technology for space manipulator control.; The joint space dynamics of fixed-base manipulators, spacecraft, and space manipulators are extensively detailed. Viewed at a high level of mathematical abstraction, space manipulator systems possess remarkable similarity to their well-studied terrestrial counterparts. The various dynamic descriptions are unified into a single high level mathematical set of equations. This unified dynamic description facilitates an investigation into the applicability of fixed-base control strategies for use with space manipulators in joint space.; Although many conventional fixed-base control strategies are inadequate for use with space manipulators, a neural network based control scheme shows great promise in solving the difficult problem of controlling a space manipulator in the presence of uncertainty. An established fixed-base neurocontrol algorithm is extended and modified for use with space manipulators. The stability and convergence of the algorithm are analyzed, and computer simulation results are presented which verify its effectiveness for both space manipulators and fixed-base manipulators.; A detailed dynamic development is also presented for both fixed-base manipulators and space manipulators in task space. Fixed-base control strategies in task space are investigated, again yielding the conclusion that many are inadequate for use with space manipulators, especially in the presence of uncertainty. The joint space neurocontrol algorithm is significantly modified and extended, generating a new space manipulator neurocontrol algorithm for use in task space. The assumptions of a known Jacobian and an uncertain Jacobian axe each considered. In both cases, the stability and convergence properties of the algorithm are analyzed, and computer simulation results axe presented which demonstrate the effectiveness of the task space neurocontroller.