Repetitive Learning Control For Tracking Of Robot Manipulators

Robotic manipulators are highly nonlinear with strong couplings between joints.High-precisiton trajectory tracking control of robotic manipulators is especiallychallenging and has been widely studied in the past years. Many industrial applicationsrequire robots to perform repetitious tasks. Given the myriad of industrial applicationsthat require a robot to move in repetitive manner, researchers have been motivated toinvestigate control methods that exploit the periodic nature of the the robot dynamics,and hence, increase link position tracking performance. As a result of this work, apromising control approach is internal model-based repetitive learning control. Thisapproach does not utilize the modeling information in the control formulation, and thuspermits easy implementation. In this thesis, some new classes of repetitive learningcontrollers are proposed. The main research works can be described as follows.1. A new class of nonlinear decentralized repetitive learning control for robotmanipulators is proposed to give faster response and higher tracking precision over thecommonly used repetitive learning control without increased torque. The proposednonlinear decentralized repetitive learning control is formulated with a calss ofnonlinear saturated function with the characteristics of ‘enlargement of small error andsaturated in large error’.2. As a matter of fact, in practice, velocity measurement are prone to beingcontaminated by noise and degrading the performance of the controller. Moreover,measurement devices add extra weight and cost to the system. To overcome thesedifficulties, we propose in this thesis an output feedback repetitive learning controllerfor trajectory tracking control of robot manipulators with model uncertainty. A nonlinearfilter is utilized in the controller development to remove the requirement of link velocitymeasurement. Despite the fact that only link position is available, the proposedcontroller obtains favorable performance. The repetitive control strategy ensures that thelink position globally asymptotically tracks the desired periodic reference signals.3. Taking into account actuator saturation, a sliding mode based repetitive learningcontrol method is proposed for high-precision tracking of robot manipulators. It isproven that robot systems subject to bounded inputs can be asymptotically stabilized.Advantages of the proposed control include the absence of model parameter in thecontrol law formulation and an ability to ensure actuator constraints are not breached.This is accomplished by selecting control gains a priori, removing the possibility ofactuator failure due to excessive torque input levels. 4. A terminal sliding mode based repetitive learning control method is developed byincorporating characteristics of terminal sliding mode control into repetitive learningcontrol. The proposed hybrid control scheme utilizes learning-based feedforward termsto compensate for periodic dynamics and terminal sliding mode-based feedback termsto compensate for nonperiodic dynamics. Advantages of the proposed control includethe absence of model parameter in the control law formulation and improved robustnessand tracking performance in comparison with the conventional approaches.5. Lyapunov’s direct method and LaSalle’s invariance principle are employed toprove (semi-global) global asymptotic tracking. Simulation results on roboticmonipulators demonstrate the effectiveness and improved performances of the proposedschemes.