1. Hand Posture Subspaces for Dexterous Robotic Grasping Dipartimento di Ingegneria dell’Informazione Università degli Studi di Siena IIT- Genova 24 January 2011 A review of Columbia University’s work

2. Targets -Outline affinities with Hands.dvi project -Underline Ciocarlie and Allen’s results -Find possible suggestions and define a different way for our project

3. Eigengrasps (1/2) -Low-dimensional hand posture subspaces to express coordination patterns between multiple DOFs for robotic hands -Based on Santello’s results -Same meaning of the synergies vectors -Defined on a 20 DOFs human hand model, the concept has been extended to different robotic hands - M.T. Ciocarlie and P.K. Allen, “Hand Posture Subspaces for Dexterous Robotic Grasping,” The International Journal of Robotics Research, vol. 28, Jun. 2009, pp. 851-867. - M.T. Ciocarlie, C. Goldfeder, and P.K. Allen, “Dexterous Grasping via Eigengrasps : A Low-dimensional Approach to a High-complexity Problem,” Proceedings of the Robotics: Science & Systems, 2007. - C.Goldfeder, M.T. Ciocarlie, and P.K. Allen, “Dimensionality reduction for hand- independent dexterous robotic grasping,” IROS 07, Citeseer, 2007.

4. Eigengrasps (2/2) -Empirical mapping on non-human hands -Use similarities with human hands -For Barreth Hand, spread angle DOF mapped into human finger abduction

5. Grasp Synthesis through Low- dimensional Posture Optimization (1/5) -Control algorithms operate on eigengrasp directions and they do not need to be customized for low-level operations -All of the results presented were obtained by treating all hand models identically, without the need for any hand- specific tuning or change in parameters -Form closure -Maximization of a high-dimensional quality function p hand posture, w wrist position and orientation, d number of hand DOFs

6. Grasp Synthesis through Low- dimensional Posture Optimization (2/5) -If d=20 then 26-dimensional optimization domain -Proposed solution -New problem -Only 8 parameters to compute when b=2

7. Grasp Synthesis through Low- dimensional Posture Optimization (3/5) - Quality function formulation - Simulated annealing used for optimization

8. Grasp Synthesis through Low- dimensional Posture Optimization (4/5) -Obtained grasps -Solution: use this result as pre-grasp position and complete the grasping by closing fingers

9. Grasp Synthesis through Low- dimensional Posture Optimization (5/5) -Results -Number of form-closed grasps obtained from 20 pre-grasps found in a two-dimensional eigengrasp space

10. On-line Interactive Dexterous Grasping (1/2) -Remove computation of wrist position through a human operator that move the hand -Quality Function Formulation using Scaled Contact Wrench Spaces

11. On-line Interactive Dexterous Grasping (2/2) M.T. Ciocarlie and P.K. Allen, “On-Line Interactive Dexterous Grasping”

12. Conclusion - Definition of pre-grasp position obtained in synergy subspaces - Definition of eigengrasps for different hand models - Form closure grasp obtained from pre-grasp position - On-line interactive grasping