ISSN 0280-5316 ISRN LUTFD2/TFRT--5690--SE 6-DOF Visual Servoing Using the Lie Group of Affine Transformations

In this thesis a visual servoing approach for uncalibrated systems with 6 degrees of freedom (DOF) is evaluated. Assuming weak perspective, the observed motions of the robot end-effector are constrained to affine trans-formations in the image space. This allows us to express the control error directly in terms of an affine transformation in the camera images. By this approach, visual servoing is realized without previous knowledge of the camera parameters or the kinematic properties of the robot. Similarly to conventional image based visual servoing, but without the usual drawbacks, such as suboptimal trajectories in cartesian space. The system was evaluated for different camera configurations, regarding stability and precision, using both a single camera and a stereo rig config-uration. Also, a motion primitive to define complex pick and place tasks is proposed. Applications of such an approach are in an industrial environment where a flexible behavior of the robot is expected, e.g. handling slightly changing parts in a manufacturing line. Additional applications are in harsh envi-ronments where conventional sensors are not suitable for some reasons or applications where the kinematics of the robot are changing, resp. unknown, for example in the growing field of reconfigurable robots. Being an image based visual servoing approach, our method is more robust with respect to maintenance and repair then position based visual servoing. The entity of sensor, actor and controller breaks up and every system can be regarded in particular. A replacement of the camera, the end-effector or even the whole kinematic setup does not involve modifications on the other parts of the system. Some practical considerations were examined using a virtual robot envi-ronment and validated using a ABB 6DOF industrial robot and a stereo camera system.