Fingertips motion planning and force computation for dextrous - PowerPoint PPT Presentation

IROS 2010 IROS 2010 Workshop On Workshop On Grasp Planning and Task Grasp Planning and Task Learning by Imitation Learning by Imitation Fingertips motion planning and force computation for dextrous manipulation N.Daoud, J.P.Gazeau, S.Zeghloul, M.Arsicault

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation Outline • Context of research • Introduction & Prior Art • Motion strategy • Simulation of manipulation tasks • Grasp stability • Experimental results • Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation 2006 - 1 dof underactuated hand Context 1991 - 1 dof 2009 underactuated of research gripper 2002 - 16 dof ABILIS project exosqueleton Introduction & Prior Art Motion strategy Simulation of manipulation tasks Grasp stability Experimental results 1996 – A mechanical hand with 16 dof fully actuated Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation Context The experimental site : of research • Hand embedded controller Introduction • Hand position & force control & Prior Art • Kuka KR16 6 axis robot Electronic Interface Motion strategy Multi Axis Controller Mechanical Hand A general strategy for Simulation of object manipulation POWER SUPPLY manipulation including : tasks • Planning method for object Grasp manipulation; stability • Efficient algorithm to compute fingertip forces Experimental • PC with SMAR Grasp synthesis results Software Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation A geometrical reasoning Context The manipulation task is defined by the object trajectory P d (t)  6 of research such as :   ( ) r t Introduction  ( ) P t    d & Prior Art  ( )  t Motion   3 : position of the center of mass of the grasped object r strategy    3 : pitch, yaw and roll angles of the object, and t the time. Simulation of The object trajectory is divided into a succession of small manipulation tasks displacements T i-1,i : T 0,f =T 0,1 .T 1,2 ..T n-1,n .T n,f Grasp stability Where T 0,f is a homogeneous transformation that represents Experimental of the object from its initial to its final configuration. results Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation Assumptions Z d R d Context n • Objects are rigid of research X d • Manipulation is only carried out E Y d Introduction Q with fingertips P & Prior Art • Fingertip is hemispherical Z ob R ob • Shape of object, dimensions Motion G strategy Y ob and mass are known X ob Simulation of For each small displacement T i-1,i , 2 contact modes between object manipulation tasks and finger : • Fixed point mode Grasp • Rolling without sliding mode stability Experimental Sliding mode is not considered results Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation 01 Initial grasp choice 02 (Definition of contact points Pi object- finger) Context 03 Decomposition of the trajectory of the Strategy for manipulation 04 object in N small displacements dP with of research 05 dP= d  1 ,d  2 , d  3 , dX G , dY G , dZ G 06 FOR i=1 TO N Introduction Find initial grasp with 3 fingers 07 (For each small object displacement N ° i) & Prior Art 08 Update rotation of the object:  i =  i + d  i with i=i+1 09 Object motion Motion 10 Update translation of the object: (fingers rolling without sliding on object surface) 11 X G = X G + dX G ; strategy 12 Y G = Y G + dY G ; Manipulation 13 Z G = Z G + dZ G Check : No Simulation of achieved with - Collisions 14 Compute small joint displacements dq i manipulation success - Joint limit 15 IF Solution is OK THEN tasks 16 Update joint parameters q=q+dq Yes 17 ELSE Find a new optimal 3 fingers grasp Grasp 18 Solution out of range: with genetic algorithms 19 Repositioning of the fingers on the object stability 20 Computation of the new contact point P Use finger gaiting to reach the new grasp 21 END (the fourth finger will be used to keep Experimental the object position and orientation unchanged) results Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation Strategy for manipulation Context Fingers roll without sliding : 01 Initial grasp choice of research 02 (Definition of contact points Pi object- finger)     03 Decomposition of the trajectory of the V (P finger/Rd) = V (P object/Rd) Introduction 04 object in N small displacements dP with & Prior Art 05 dP= d  1 ,d  2 , d  3 , dX G , dY G , dZ G 06 FOR i=1 TO N We write the small 07 (For each small object displacement N ° i) Motion displacement model : 08 Update rotation of the object: strategy  i =  i + d  i with i=i+1 09 10 Update translation of the object:  finger object Simulation of dP dP 11 X G = X G + dX G ; manipulation 12 Y G = Y G + dY G ; tasks 13 Z G = Z G + dZ G 14 Compute small joint displacements dq i 15 IF Solution is OK THEN Grasp 16 Update joint parameters q=q+dq stability 17 ELSE 18 Solution out of range: Experimental 19 Repositioning of the fingers on the object results 20 Computation of the new contact point P 21 END Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation doigt objet  dP dP   objet     J ( q , q , q ) dq + R n J (q , q , q ) dq = dP  v 0 1 2 0 1 2 Context dP object = dx + PG ^ d  of research Introduction A linear system with 3 equations and 3 unknown factors dq j (j=1..3) & Prior Art q j = q j + dq j (j=1..3) Motion strategy The new contact point P’ : R d Z d Simulation of     manipulation E' P ' R. n ' X d E tasks Y d n E’ Q P with R radius of hemispherical n’ Grasp fingertip. Q’ P’ stability Experimental Evolution of the point of contact results for a small object displacement Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation Manipulation tasks... Context of research Introduction & Prior Art Motion strategy Simulation of ... for a cylinder manipulation tasks Grasp stability ... for a prism Experimental results Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation Problem : Context Apply to the object the global external force Fe, necessary of research to ensure its stability. W. F g = F e Introduction with F g = ( F g 1 T , F g 2 T , F g 3 T ) T & Prior Art The 6x9 grasping matrix W is defined as follows : Motion strategy    0 z y   i i I I I        W 0 R z x Simulation of where   i i i  R R R    manipulation 1 2 3   0  y x i i tasks The contact force must be applied to the object without sliding or Grasp breaking contact. We write the static friction constraints : stability        2 2  . 1 . 0 F F F n i i i i   Experimental   . 0 F i n 1 , 2 , 3 i results i Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation We formulate the fingertip force F g that can be obtained from F e with : Context with Fg (  ) = W T [W.W T ] -1 . Fe + N.  of research - - - - - - 0 0 0 x x y y z z x x y y z z Introduction 2 1 2 1 2 1 1 2 1 2 1 2 N T & Prior Art = - - - 0 0 0 - - - x x y y z z x x y y z z 3 1 3 1 3 1 1 3 1 3 1 3 - - - - - - 0 0 0 x x y y z z x x y y z z 3 2 3 2 3 2 2 3 2 3 2 3 Motion strategy An optimization problem Simulation of   T      manipulation Find the vector 1 2 3 tasks 1   T that minimizes the following quadratic function ( ) F F F 2 Grasp with the static friction constraints. stability Experimental results Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation Problem resolution : Context Use of sequential unconstrained minimization techniques (SUMT). of research Advantages : Initialization of the process doesn’t require any feasible solution. Introduction & Prior Art Time cost 1.4ms / C++ Time comparison in Matlab (0.34s vs 1.4s for fmincon) Motion strategy Simulation of manipulation tasks Grasp stability Finger F x F y F z |F| [N] 1.1582 -0.864 0.5895 1 1.5606 Experimental -0.234 0.8617 0.5981 2 1.0748 results -0.923 0.0025 0.7743 3 1.2052 Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation Force Control Context A learning based method of research for force evaluation Introduction Contact Measure & Prior Art 80.00 60.00 Motion strategy 40.00 20.00 Simulation of 0.00 manipulation tasks -20.00 0.00 0.40 0.80 1.20 1.60 Grasp Time (s) stability Normal force evaluation N  1 NEURAL Contact on distal phalanx Experimental  2 NETWORK Contact on intermediate phalanx FINGER N°i results  3 Contact on proximal phalanx Conclusion

IROS 2010 Workshop On Grasp Planning and Task Learning by Imitation Position Control Context …Manipulation of research Introduction & Prior Art Motion strategy Simulation of manipulation tasks … Reach Grasp and Grasp stability Experimental results Conclusion