Object-action abstraction for teleoperation

In telerobotic systems human actions are mapped to robot actions. In an illustrative object manipulation experiment various human grasps were translated to configurationally similar robotic grasps. The experiment´s results highlight the problems and suboptimal performance incurred when such a resemblance is maintained. A new approach to telerobotics based on the construction of object-action pairs is presented. Actions are identified in the context of the object they are being performed on according to features extracted from the human grasp and transport motion. A priori knowledge is introduced to the robot controller using object centered programming and a relational database.