Grasping Objects of Various Shapes by Calibration-Free Vision-Guided Robots

An approach to vision-based control of calibrationfree robots is introduced. Similarly to living organisms, it uses an uncalibrated stereo vision system, but no world coordinates and no inverse pcrspective or kinematic transformations, and it enables an automatic adaptation to changing optical, mechanical and control parameters of the robot. We implemented the approach on a vision-guided calibration-free articulated arm manipulator with five degrees of freedom. In real-word grasping experiments we showed that robot control without calibration is, indeed, feasible and that it leads to robots with a rcmarkable degree of robustness against disturbances of the sensors and unexpected changcs of the robot's parameters.