Iterative learning control for industrial robots with end effector sensing

This dissertation considers precise Tool Center Point (TCP) tracking for industrial robots, i.e. tracking of reference trajectories in the Cartesian-space by the center point of a tool at the end effector of a robot. The difficulty of the control are due to disturbances and uncertainties in the reducers, such as non-linear friction, backlash, transmission error, and flexibility of the reducers. Under an assumption that a robot repeat the same tracking task repeatedly, we consider Iterative Learning Control (ILC) to accomplish precise TPC tracking.;We focus on the frequency domain design of ILC and we develop a systematic design method for ILC based on Hinfinity synthesis. First we present the LFT (Linear Fraction Transformation)-based ILC design method, which was originally proposed for control of wafer scanners. We extend the original design to be able to apply non-causal learning as well as causal learning. We also propose an LMI (Linear Matrix Inequality)-based ILC design method. This method utilizes zero-phase weighting functions and realizes low order controllers which ensure robustness to disturbances and uncertainties. The controller performance is verified by experiments on an industrial robot.;We introduce link-side ILC for industrial robots, i.e. ILC utilizing the link-side measurement for tracking in the Cartesian-space by the LMI-based ILC. ILC may be applied to each joint of a robot separately: i.e. Single-Input Single-Output (SISO) ILC design. We also consider Multiple-Input Multiple-Output (MIMO) ILC design, since Cartesian-space motion is a composite of multiple-joint motions. We divide the discussion of the link-side MIMO ILC in the Cartesian-space into three parts. First is the investigation of link-side ILC for a single-joint model. Secondly, we introduce MIMO ILC in the joint-space to compare the performance of SISO ILC and MIMO ILC. The last part is the investigation of ILC in the Cartesian-space which involves inverse kinematics of robots. All the ILC designs are evaluated by simulations.;The third part of this dissertation is on vision-based ILC, which utilizes vision data as link-side measurement. The vision sensors provide the error between the TCP position and the target reference, but, they do not provide the TCP position itself. In this part, we estimate the desired link position to cancel the error in Cartesian-space by utilizing the kinematic relationship between the joint angle and the vision data. The controller performance is verified on a industrial robot.