This thesis addresses the problem of positioning robots relative to objects in the surrounding work space. The object's position and orientation (posture) are measured using a non-contact sensor on-board the robot. For gripping/docking tasks the goal posture is specified relative to the object and not in a fixed coordinate system of the robot. Emphasis is on cases when the initial posture uncertainty is in the same order of magnitude as the size of the object. This should be contrasted with the current use of industrial robots. The task of the robot may even be to move so as to reduce the uncertainty of an essentially unstructured work space.

Several cases of position and motion control are studied using non-contact sensing. The approach is model-based including dynamics, sensor uncertainty and task tolerances. The results include:

• Algorithms and experiments for docking a mobile robot to a rectangular object, within prespecified tolerances.

• Control algorithms using measurements of the distance between the arm-tip and the object in order to compensate for deflection and vibration.

• For relative positioning the uncertainty in a range camera is essentially proportional to the distance-to-go. To give insight into optimal approach velocities, a 1-dimensional problem has been solved in the distance-to-go variable.

Feedback from non-contact sensors offers a large potential for a variety of novel applications, like handling of non man made objects and for robots working in hazardous environments. In a longer perspective the model-based approach is well suited considering the fast development in computing and laser based sensing.