A new approach to robust control of robot manipulators based on decompositions of dynamics model and uncertainty is proposed in this thesis. The fundamental idea of this approach is to distinguish between different uncertain parameters and variables based on their physical features, and design a separate compensator for each type of uncertainty. The overall controller is generated by the combination of these compensators.

Two robust control schemes are derived using the proposed approach. The first scheme is based on robot dynamics decomposition. The dynamic model of the robot manipulators is re-formulated in such a way that model uncertainties of different physical types are treated distinctly from each other. Based on this decomposition, separate robust compensators are introduced corresponding to each type of model uncertainty. The combination of these robust compensators yields the overall control law. The control parameters of the controller can be directly related to the dynamic behavior of the robot and adjusted accordingly.

The second scheme is based on model uncertainty decomposition. Model uncertainty in a robot manipulator system is categorized into 'parameterized uncertainty' and 'unparameterized uncertainty', and a specific compensator is designed for each uncertainty group. The overall robust control is formulated by combining the two compensators. Such an approach allows the use of integral control to compensate for the parameterized uncertainty, and the requirement for high feedback gains is dramatically reduced.

The stability of the proposed methods is established in the sense of uniform ultimate boundedness using Lyapunov theory. Design examples and computer simulations are used for illustration of the theoretical results.

Extensive experiments on a direct drive robot arm were carried out to evaluate the proposed methods and compare them with published control schemes. The results of these experiments have confirmed the desirable features and efficiency of the proposed methods. Useful insights gained during the experiments are discussed.

A related issue, that the same controller would regulate both free and constraint motions in the presence of model uncertainty, was also addressed in this research. A robust hybrid impedance control method is developed based on a new tracking control scheme. This method has been tested experimentally, and the results are presented in this thesis.