Representing the robot's workspace through constrained manipulability analysis

Quantifying the robot's performance in terms of dexterity and maneuverability is essential for the analysis and design of novel robot mechanisms and for the selection of appropriate robot con gurations in the context of grasping and manipulation. It can also be used for monitoring and evaluating the current robot state and support planning and decision making tasks, such as grasp selection or inverse kinematics (IK) computation. To this end, we propose an extension to the well-known Yoshikawa manipulability ellipsoid measure [40], which incorporates constraining factors, such as joint limits or the self-distance between manipulator and other parts of the robot. Based on this measure we show how an extended capability representation of the robot's workspace can be built in order to support online queries like grasp selection or inverse kinematics solving. In addition to single handed grasping tasks, we discuss how the approach can be extended to bimanual grasping tasks. The proposed approaches are evaluated in simulation and we show how the extended manipulability measure is used within the grasping and manipulation pipeline of the humanoid robot ARMAR-III.