| The dynamics of multi-link robot manipulators are highly nonlinear and tightly coupled and derivation of such dynamics is not easy. Moreover, the parameters of the system such as variable payload may not be known exactly or may change. Consequently, the design of controllers for robot manipulators is not simple especially when a complex maneuver such as collision avoidance is considered. This controller design becomes even more challenging when an on-line real-time implementation is required.; Recently, attempts have been made in designing robotic controllers by using global linearization or adaptive approaches. Since these methods deal with the entire dynamics, the methodology is still difficult to implement due to the large scale dynamics.; In this dissertation, a two-level controller scheme is applied, i.e., local and global control to simplify the design problems. The objective of the global controller is to perform real-time trajectory planning and to handle complex tasks such as obstacle avoidance. The dynamics of the robot must be considered so that errors are minimized. The purpose of the local controller is to handle all the unpredictable disturbances caused by nonlinearities, parameter variations, or disturbances from the environment.; This proposed scheme is accomplished by applying a new control scheme called decentralized model reference control at the local control level. The entire dynamics of a multi-link arm are geometrically decomposed into multi-subsystems which represent each link dynamics. The coupling effects among links are modelled by the dynamics of a waveform disturbance. Only the states of an individual link are used in the decentralized local control to make the link behave as its model, in spite of coupling effects and disturbances. The global control, which is at a higher control level, utilizes the proposed trajectory decomposition technique for a real-time trajectory planning to circumvent the complexities of design and to reduce the computational load of implementation. By decomposing the entire trajectory into many small segments, where each one is controlled by an optimal constant input, a nonlinear optimal control problem can be reduced to a quadratic optimization problem which can be simplistically solved analytically or numerically. The entire control scheme is implemented on an IBM-AT to drive a three-degree-of freedom cylindrical arm. The main results of the experiments are presented and discussed. |
