1. Department of Computer Science, Iowa State University Robot Grasping of Deformable Planar Objects Yan-Bin Jia (with Ph.D. students Feng Guo and Huan Lin ) Department of Computer Science Iowa State University Ames, IA 50010, USA

2. Department of Computer Science, Iowa State University Rigid Body Grasping – Form Closure The object has no degree of freedom (Reuleaux, 1875). frictionless contacts What cannot be generated by the contact force?

3. Department of Computer Science, Iowa State University Rigid Body Grasping – Force Closure The contacts can apply an arbitrary wrench (force + torque) to the object (Nguyen 1988). contact friction cones  These wrench vectors positively span the 3D wrench space W.  Equivalently, their convex hull contains the origin in the interior.  Each force (normal or tangential) at a contact generates a vector in the 3D wrench space W (6D for a 3D object). Not form closure. Form closure does not imply force closure.  They can resist an arbitrary external wrench.

4. Department of Computer Science, Iowa State University Related Work on Rigid Body Grasping  Form closure grasps  Bounds on # contact points: Mishra et al. (1986); Markenscoff et al. (1987)  Synthesis: Brost & Goldberg (1994); van der Stapper et al. (2000)  Force closure grasps  Testing & synthesis: Nguyen (1988); Trinkle (1988); Ponce et al. (1993); Ponce et al. (1997)  Caging: Rimon & Blake (1999); Rodriguez et al. (2012)  Grasp metrics: Kerr & Roth (1986); Li & Sastry (1988); Markenscoff & Papadimitriou (1989); Mirtich & Canny (1994); Mishra (1995); Buss et al. (1988); Boyd & Wegbreit (2007)

5. Department of Computer Science, Iowa State University Barrett Hand Grasping a Foam Object

6. Department of Computer Science, Iowa State University Deformable Body Grasping Is Difficult  Form closure impossible (infinite degrees of freedom)  Force closure inapplicable (changing geometry, growing contacts)  High computation cost of deformable modeling using the finite element methods (FEM) Very little research done in robotics (most limited to linear objects) Wakamatsu et al. (1996); Hirai et al. (2001); Gopalakrishnan & Goldberg (2005); Wakamatsu & Hirai (2004); Saha & Isto (2006); Ladd & Kavraki (2004)  Contact constraints needed for modeling do not exist at the start of a grasp operation.

7. Department of Computer Science, Iowa State University Displacement-Based Deformable Grasping A change of paradigm from rigid body grasping.  Specified forces cannot guarantee equilibrium after deformation.  Deformation computed under geometric constraints ensures force and torque equilibrium.  Easier to command a finger to move to a place than to exert a prescribed grasping force. Specify finger displacements rather than forces.

8. Department of Computer Science, Iowa State University Positional Constraints & Contact Analysis Deformation update during a grasp needs positional constraints. Resort to varying finger contacts  They are maintained by friction.  Contact regions grow or shrink.  Individual contact points slide or stick. Incrementally track contact configuration! Instantaneous deformation is assumed in classical elasticity theory. How can we predict the final contact configuration from the start of a grasp operation?

9. Department of Computer Science, Iowa State University Assumptions  Deformable, isotropic, planar or thin 2-1/2 D object.  Two rigid grasping fingers coplanar with the object.  Frictional point or area contacts.  Gravity ignored.  Small deformation (linear elasticity).

10. Department of Computer Science, Iowa State University Linear Plane Elasticity Displacement field:

11. Department of Computer Science, Iowa State University Strains Extensional strain – relative change in length before after Shear strain – rotation of perpendicular lines toward (or away) from each other.

12. Department of Computer Science, Iowa State University Strain Energy translationrotation

13. Department of Computer Science, Iowa State University Finite Element Method (FEM) a) Discretize the object into a triangular mesh. b) Obtain the strain energy of each triangular element in terms of the displacements of its three vertices. c) Sum up the strain energies of all elements. stiffness matrix (symmetric & positive semidefinite)

14. Department of Computer Science, Iowa State University Energy Minimization Total potential energy : load potential vector of all nodal forces

15. Department of Computer Science, Iowa State University Stiffness Matrix translations of all nodesrotation of all nodes Spectral decomposition: orthogonal matrix

16. Department of Computer Science, Iowa State University Deformation from Contact Displacements Forces at nodes not in contact: known

17. Department of Computer Science, Iowa State University Submatrices from Stiffness Matrix null space contact node indices

18. Department of Computer Science, Iowa State University Solution Steps

19. Department of Computer Science, Iowa State University Matrix for Solution of Deformation

20. Department of Computer Science, Iowa State University Uniqueness of Deformation Computational complexity m is small

21. Department of Computer Science, Iowa State University Reduced Stiffness Matrix Forces at m contact nodes: Strain energy:

22. Department of Computer Science, Iowa State University Squeeze by Two Point Fingers Minimizing potential energy is equivalent to minimizing strain energy. Solution: squeeze depth

23. Department of Computer Science, Iowa State University Pure Squeeze  object translation or rotation during deformation. where squeeze depth

24. Department of Computer Science, Iowa State University Example for Comparison (stable squeeze) (pure squeeze)

25. Department of Computer Science, Iowa State University Squeeze Grasp with Rounded Fingers Translate the fingers to squeeze the object.  Contact friction.  Contacts growing into segments.

26. Department of Computer Science, Iowa State University Contact Configuration  Which nodes are in contact.  Which of them are sticking and which are sliding. slidingsticking Deformation update based on FEM: Maintain two sets:

27. Department of Computer Science, Iowa State University Overview of Squeeze Algorithm

28. Department of Computer Science, Iowa State University Squeeze Grasp Algorithm success Contact Event Analysis min extra squeeze Either finger slips? failure yes no yes no

29. Department of Computer Science, Iowa State University Movement of a Contact Node  A sticking node moves with its contacting finger.  A sliding node also slides on its contacting finger.

30. Department of Computer Science, Iowa State University Deformation under Extra Squeeze

31. Department of Computer Science, Iowa State University Contact Events Check for all values of extra squeeze depth at which a event could happen, and select the minimum.  Event A – New Contact  Event B – Contact Break

32. Department of Computer Science, Iowa State University More Contact Events  Event C – Stick to Slip  Event D – Slip to Stick

33. Department of Computer Science, Iowa State University Termination of Squeeze  A grasping finger starts to slip. At either one of the following situations:  Strain at some node exceeds the material’s proportional limit.  The object can be picked up against its weight vertically. All contact nodes with the finger are slipping in the same direction.

34. Department of Computer Science, Iowa State University Experiment slip stick

35. Department of Computer Science, Iowa State University Stick to Slip

36. Department of Computer Science, Iowa State University Stick to Slip back to Stick Second (convex) shape

37. Department of Computer Science, Iowa State University Experiment with Ring-like Objects Degenerate shells.

38. Department of Computer Science, Iowa State University Adversary Finger Resistance

39. Department of Computer Science, Iowa State University What Optimality? Rigid body grasping Total force/wrench to resist unit adversary force/wrench. Deformable body grasping Work to resist unit translation by adversary finger. Optimality criterion should reflect the effort of resistance.

40. Department of Computer Science, Iowa State University Work Minimization Finger contact sets change during resistance: 3) Change of contacts under Coulomb friction (general case). Solution steps:

41. Department of Computer Science, Iowa State University The Case of Fixed Point Contacts Work done by grasping fingers: Stable squeeze: Minimization is subject to Pure squeeze:

42. Department of Computer Science, Iowa State University An Example Resistance by a stable squeeze.

43. Department of Computer Science, Iowa State University The Case of Fixed Segment Contacts Generalize over the case of fixed point contacts.

44. Department of Computer Science, Iowa State University General Case of Frictional Segment Contacts  Between two events the contact configuration does not change.

45. Department of Computer Science, Iowa State University Outcomes of Resistance

46. Department of Computer Science, Iowa State University Simulation

47. Department of Computer Science, Iowa State University Resistance Trajectories

48. Department of Computer Science, Iowa State University Resistance Experiment force meter

49. Department of Computer Science, Iowa State University Simulation vs Experiment 2.1 0.82 8.0 1.53 0.0027 0.008 6.57 0.018  Measurement errors. Reasons for discrepancies: Experiment (“Optimal”)Simulation

50. Department of Computer Science, Iowa State University “Optimal” vs “Arbitrary” Resistances stable resistance

51. Department of Computer Science, Iowa State University Summary  Displacement-based grasping strategy for deformable objects.  Stable and pure squeezes.  Event-driven algorithm combined with contact mode analysis.  Energy-based grasp optimality.  Computational efficiency from one-time matrix decomposition.

52. Department of Computer Science, Iowa State University Future Work  Computationally efficient grasp outcome prediction.  Rigid body grasping vs. deformable body grasping.  Why deformable objects are often easier to grasp?  Soft fingers on a rigid body vs. Hard fingers on a deformable body  Design of grasping algorithms for 3D deformable objects.  Energy-based grasping metrics.

53. Department of Computer Science, Iowa State University Acknowledgement US National Science Foundation IIS-0915876

54. Department of Computer Science, Iowa State University Online Papers IEEE International Conference on Robotics and Automation (2013, published) http://www.cs.iastate.edu/~jia/papers/ICRA13.pdf Extended version (submitted to International Journal of Robotics Research) http://www.cs.iastate.edu/~jia/papers/IJRR13-submit.pdf IEEE/RSJ International Conference on Intelligent Robots and Systems (2013, accepted) http://www.cs.iastate.edu/~jia/papers/IROS13-submit.pdf