Grasp recognition and mapping on humanoid robots

摘要

In this paper, we present a system for vision-based grasp recognition, mapping and execution on a humanoid robot to provide an intuitive and natural communication channel between humans and humanoids. This channel enables a human user to teach a robot how to grasp an object. The system comprises three components: human upper body motion capture system which provides the approaching direction towards an object, hand pose estimation and grasp recognition system, which provides the grasp type performed by the human as well as a grasp mapping and execution system for grasp reproduction on a humanoid robot with five-fingered hands. All three components are real-time and markerless. Once an object is reached, the hand posture is estimated, including hand orientation and grasp type. For the execution on a robot, hand posture and approach movement are mapped and optimized according to the kinematic limitations of the robot. Experimental results are performed on the humanoid robot ARMAR-IIIb.