Calibration problems in robotics.

Abstract

It is expected that the integration of a vision system with a robot will increase a manipulation system's capability and overall flexibility. This thesis explores certain problems of a robot-vision integrated system arising from parametric uncertainty. Specifically, the thesis explores the effective implementation of adaptive coordinated manipulator controller, off-line calibration of the extrinsic camera parameters, and robust vision based controller design. A computationally efficient adaptive controller has been developed for two coordinated manipulators handling an unknown object, based on the recursive Newton-Euler dynamics formulation. The resulting control algorithm achieves global convergence in trajectory tracking of the desired motion of the object while maintaining the desired internal forces. Kinematic calibration arising from a wrist mounted camera is considered next. Dual quaternions are used for the parameterization of the rigid transformations. A closed form solution is found in the case of perfect data. A numerical off-line calibration technique using gradient descent has been implemented to deal with the presence of measurement noise. Finally, simple robust visual servo control algorithms are presented using stereo vision. The algorithms arise from the structure of the camera model associated with the visual servo system. The significant feature of the control strategies is that they use fixed gain laws based only on the desired set point in the image plane. Nevertheless, they result in global convergence. These controllers do not require estimates of the gripper's spatial position, a significant source of calibration sensitivity. Robustness against rotational miscalibration is suggested by a series of simulations and experiments.