In the last few years, numerous complex dextrous robotic hands have been developed. They are now employed as manipulators in larger robotic systems in various fields of activity as for example tele-operation and service robotics. So far, research on these robotic hands was solely occupied with isolated laboratory systems. Only recently, triggered by the increasing capabilities of the hardware, real world tasks have been implemented. In doing so, it was found that most algorithms for planning and smooth control of robotic grasping require more information than was generally available from actual measurements: The exact point of contact between an object to be grasped and the surface of a finger, and the position of the object. Due to the lack of small but nevertheless high resolution tactile sensors, the contact point is not directly measurable in the vast majority of robotic grippers. The information can nevertheless be retrieved, using and interpreting other sensors. Two new approaches to this interpretation are presented in this work. The first method evaluates position sensors. It exploits kinematic velocity constraints imposed on the motion of fingers and an object through rigid or quasi-rigid contact. It is able to obtain the contact points on the surface of the finger and the relative motion between the fingers and the object. The second method estimates the position of contact and the respective forces at this point from the force / torque measurements available at a finger. Based on the contact information, this work strives to determine the position of the object itself: The thesis at hand proposes an optimised algorithm to compare the derived information about the point of contact with a geometric model of the object. Thereby, the object can be located. This is especially required if other, e.g. visual information, is either not available or unreliable as may be the case due to obstruction of vision by other objects or infeasible lighting conditions. This algorithm implements a ``blind man's'' approach to grasping. The proposed methods are evaluated in computer simulations and experimentally verified using the DLR Hand II.
«
In the last few years, numerous complex dextrous robotic hands have been developed. They are now employed as manipulators in larger robotic systems in various fields of activity as for example tele-operation and service robotics. So far, research on these robotic hands was solely occupied with isolated laboratory systems. Only recently, triggered by the increasing capabilities of the hardware, real world tasks have been implemented. In doing so, it was found that most algorithms for planning and...
»