Inhalt des Dokuments
Grasping has been an active area of research for several decades. In spite of many efforts, a robust, general approach to grasping in unstructured environments does not exist. Most research makes the strong assumption that perfect knowledge of the robot and the environment are accessible. This assumption does not hold in the scenarios we are interested in. Only very recently have innovative approaches to grasping come under investigation. These approaches also broaden the view of grasping and include perception and mechanism design as part of the problem. Nevertheless, the grasping problem, in spite of its centrality in robotics, remains to large extents unsolved.
- A parallel jaw gripper by Meka, the BarrettHand, and the Meka H2 compliant hand with five degrees-of-freedom.
- © Robotics
We believe the grasping problem can only be solved by investigating the interactions between its seemingly separate but in fact highly connected
sub-fields: mechanism design, perception, manipulation, planning, and control. We therefore seek to devise novel grasping methods by exploring the best division between sensing, modeling, manipulation.
Consequently, the concept of interactive perception - tight coupling of perception and action - will play an important role in our grasping research.
Our approach to grasping is motivated by the "mitten thought experiment". This experiment illustrates that a sensory information-deprived subject (blindfolded, wearing a thick mitten to eliminate tactile feedback) is able to grasp a large variety of objects reliably by simply closing the hand, provided that a second experimenter appropriately positioned the object relative to the hand.
This thought experiments illustrates that an appropriate perceptual strategy (the experimenter) in conjunction with a simple compliance-based control strategy (the mitten hand) can lead to outstanding grasping performance.
We are developing perceptual primitives to directly perceive grasping affordances, sidestepping the difficult issue of explicitly modeling object geometry to then determine those affordances. A grasping affordance is associated with a geometric shape in the environment that matches the geometry of the hand/gripper during a grasp. We can therefore devise perceptual primitives based on knowledge of the grasping capabilities of the hand.
Contact: Clemens Eppner
Flexible Skill Acquisitionen and Intuitive Robot Tasking for Mobile Manipulation in the Real World (First MM) - funded by European Commision, in the program Cognitive Systems and Robotics,
award number FP7-ICT-248258,
February 2010 - July 2013