A Fast Approach in Generating High Quality Grasps Watcharapol - - PowerPoint PPT Presentation

▶
a fast approach in generating high quality grasps
SMART_READER_LITE
LIVE PREVIEW

A Fast Approach in Generating High Quality Grasps Watcharapol - - PowerPoint PPT Presentation

Oct 11, 2022 112 likes •700 views

A Fast Approach in Generating High Quality Grasps Watcharapol Watcharawisetkul Adviser: Asst. Prof. Nattee Niparnan, Ph.D. 1 Outline Introduction Background Knowledge Our Approach Evaluations Conclusion 2 Outline

slide-1
SLIDE 1

A Fast Approach in Generating High Quality Grasps

Watcharapol Watcharawisetkul Adviser: Asst. Prof. Nattee Niparnan, Ph.D.

1

slide-2
SLIDE 2

2

Outline

  • Introduction
  • Background Knowledge
  • Our Approach
  • Evaluations
  • Conclusion
slide-3
SLIDE 3

3

Outline

  • Introduction
  • Background Knowledge
  • Our Approach
  • Evaluations
  • Conclusion
slide-4
SLIDE 4

4

Robotic grasping

dvice.com brown.edu darpa.mil

slide-5
SLIDE 5

5

suin.mx bloggaida.blogspot.com itcentralstation.com updatedtrends.com

slide-6
SLIDE 6

6

Grasping pipeline

Object perception Grasp synthesis Grasp planning Grasp execution

1 2 3 4

slide-7
SLIDE 7

7

Focus on

Object perception Grasp synthesis Grasp planning Grasp execution

1 2 3 4

slide-8
SLIDE 8

Contribution

  • Propose grasp synthesis algorithm

○ Calculate a large number of high quality secure grasps in short time

8

Input

  • Position and normal of
  • bject surface points

Output

  • List of secure grasps
slide-9
SLIDE 9

9

Outline

  • Introduction
  • Background Knowledge
  • Our Approach
  • Evaluations
  • Conclusion
slide-10
SLIDE 10

10

Finger model

  • Hard finger with friction

○ Can exerts only pure force lie in a circular friction cone

slide-11
SLIDE 11

Wrench

11

  • Force ( )

○ 3D

  • Torque (τ

)

○ 3D

  • Wrench [

T τT ]T

○ Concatenation of force and torque ○ 6D

·

slide-12
SLIDE 12

Secure grasp

12

  • Force closure

○ Ability to resist any external disturbance ○ Contact wrenches generated by fingers positively span entire wrench space

✘

✔

slide-13
SLIDE 13

Number of fingers

  • X. Markenscoff et al. , “The Geometry of Grasping,”

1990. ○ Sufficient no. of fingers to archived force closure

13

without friction with friction 2D 4 3 3D 7 4

slide-14
SLIDE 14

14

Outline

  • Introduction
  • Background Knowledge
  • Our Approach
  • Evaluations
  • Conclusion
slide-15
SLIDE 15

Our Approach

  • Objective

○ Propose grasp synthesis algorithm that can calculate a large number of high quality grasps in short time

  • Randomized algorithm bound with time limit
  • Concurrent grasp

15

slide-16
SLIDE 16

Concurrent grasp

16

  • C1. There exist lines in each

friction cone that intersect in a single point

  • C2. The vectors parallel to

these lines and pointing inward positively span force space

Concurrent point

Concurrent grasps ⊂ Force closure grasps

slide-17
SLIDE 17

Finding concurrent grasp algorithm

17

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

slide-18
SLIDE 18

Finding concurrent grasp algorithm

18

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

slide-19
SLIDE 19

Finding concurrent grasp algorithm

19

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

slide-20
SLIDE 20

Finding concurrent grasp algorithm

20

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

slide-21
SLIDE 21

Finding concurrent grasp algorithm

21

  • C1. There exist lines in each

friction cone that intersect in a single point

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

slide-22
SLIDE 22

Finding concurrent grasp algorithm

22

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

slide-23
SLIDE 23

Finding concurrent grasp algorithm

23

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

  • C2. The vectors parallel to

these lines and pointing inward positively span force space

slide-24
SLIDE 24

Finding concurrent grasp algorithm

24

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

  • C2. The vectors parallel to

these lines and pointing inward positively span force space

slide-25
SLIDE 25

Finding concurrent grasp algorithm

25

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

slide-26
SLIDE 26

Finding concurrent grasp algorithm

26

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

slide-27
SLIDE 27

Time complexity

27

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

O(|V|3 log|V|+K)

  • N. Niparnan et al. , 2006

○ Force-closure grasps ○ 2D objects ○ Frictionless ○ 4 fingers

slide-28
SLIDE 28

Concurrent points

28

Input : S = object surface points frictionHalfAngle timeLimit Output : Gans = list of concurrent grasp 1:while usedTime < timeLimit 2: randomly pick concurrent point p 3: M ← {s | s∈S, p in DFCs} 4: V ← {vm | vm=inward(p-m) ∀m∈M} 5: G ← {(a,b,c,d)|va,vb,vc,vd∈V, va,vb,vc,vd positively span ℝ3} 6: Gans ← Gans ∪ G

Different point lead to difference result

slide-29
SLIDE 29

Concurrent points (1)

  • "Uniform"

○ Randomly select concurrent points from a uniform distribution within axis- aligned minimum bounding box ○ Use as baseline

29

slide-30
SLIDE 30

Concurrent points (2)

  • "NearCM"

○ Randomly select concurrent points near

  • bject's center of mass

○ Normal distribution

  • 30
  • Grasp quality

○ Ponce et al. 1997, Ding et al. 2001 ○ Distance between the centroid and the center of mass

slide-31
SLIDE 31

Concurrent points (3)

  • "MedialPoint"

○ Select concurrent points from object's medial axis

31

  • Medial axis

○ Centers of spheres which touch the object's surface at two or more points

slide-32
SLIDE 32

3D medial axis approximation

  • Ma et al. 2011

○ Input

■ Position and normal of

  • bject surface points

○ Output

■ Medial point

corresponding to each

  • bject surface point

32

slide-33
SLIDE 33

33

Outline

  • Introduction
  • Background Knowledge
  • Our Approach
  • Evaluations
  • Conclusion
slide-34
SLIDE 34

34

Evaluation objective

  • Propose grasp synthesis algorithm

○ No. generated grasps ○ Grasp quality

slide-35
SLIDE 35

Competitor

Our algorithm

  • "Uniform"

○ Select concurrent points from a uniform distribution

  • "NearCM"

○ Select concurrent points near cm of the object

  • "MedialPoint"

○ Select concurrent points from object's medial axis

35

slide-36
SLIDE 36

Competitor (2)

  • "Random4SP"

○ C. Borst et al. , “Grasping the dice by dicing the grasp,” 2003. ○ Find force closure grasp

36

(N. Niparnan et al. 2007) (Y. Zheng et al. 2009)

1:while usedTime < timeLimit 2: Random 4 surface points 3: Apply heuristic filter 4: Perform force closure test

slide-37
SLIDE 37

Experiments

  • Set time limit to 0.5 seconds for all

experiments

  • Perform 30 tests for each object and each

method

  • Set half-angle of friction cone as 10 degree

37

slide-38
SLIDE 38

Object

  • Each object has ~800 facets.
  • Use center of each facet as surface points.

38

KIT Object Database

slide-39
SLIDE 39

Result (no. generated grasps)

39

  • "Uniform"
  • "NearCM"
  • "MedialPoint"
  • "Random4SP"
slide-40
SLIDE 40

Group A : graphs (no. grasps)

40

slide-41
SLIDE 41

Group A : object models

41

slide-42
SLIDE 42

Group B : graphs (no. grasps)

42

slide-43
SLIDE 43

Group B : object models

43

slide-44
SLIDE 44

Group C : graphs (no. grasps)

44

slide-45
SLIDE 45

Group C : object models

45

slide-46
SLIDE 46

Moon : graphs (no. grasps)

46

slide-47
SLIDE 47

Grasp quality measure

  • C. Ferrari and J. Canny, “Planning optimal grasps,”

1992 ○ ɛ-metric ○ Smallest wrench that can break the grasp

47

slide-48
SLIDE 48

Result (grasp quality)

48

➢ ɛ-metric

  • "Uniform"
  • "NearCM"
  • "MedialPoint"
  • "Random4SP"
slide-49
SLIDE 49

Group A : graphs (quality)

49

slide-50
SLIDE 50

Group B : graphs (quality)

50

slide-51
SLIDE 51

Group C : graphs (quality)

51

slide-52
SLIDE 52

Moon : graphs (quality)

52

slide-53
SLIDE 53

53

Outline

  • Introduction
  • Background Knowledge
  • Our Approach
  • Evaluations
  • Conclusion
slide-54
SLIDE 54

Conclusion

  • We propose a fast algorithm to find a large

number of high quality force closure grasps.

○ Use the condition for concurrent grasp ○ Use only position and normal of object surface points as inputs

54

slide-55
SLIDE 55

Conclusion

  • Sampling concurrent points from object's

medial axis lead to high number of grasps

  • Sampling concurrent points near object's

center of mass lead to high grasp quality

55

slide-56
SLIDE 56

Paper (1)

1.

  • W. Watcharawisetkul, M. Borwornpadungkitti, N. Niparnan, and A.

Sudsang, “A Randomized Approach in Identifying High Quality Force Closure Grasp from Contact Points in Real Time,” in Applied Mechanics and Materials, 2015, vol. 781, pp. 483–486. 2.

  • W. Watcharawisetkul, M. Borwornpadungkitti, N. Niparnan, and A.

Sudsang, “The Quickgrasp Algorithm for Grasp Synthesis,” in 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2015.

56

slide-57
SLIDE 57

Paper (2)

3.

  • M. Borwornpadungkitti, W. Watcharawisetkul, N. Niparnan, and A.

Sudsang, “Improved Method for Computation of Grasp Quality Metric Using Minimal Breaking Force on Objects,” in 2014 IEEE International Conference on Robotics and Biomimetics (ROBIO), 2014,

  • pp. 2197–2202.

4.

  • M. Borwornpadungkitti, W. Watcharawisetkul, N. Niparnan, and A.

Sudsang, “Exact Calculation for Disturbance Force Rejection Grasp Quality Measure,” in 2015 IEEE/RSJ International Conference

  • n Intelligent Robots and Systems (IROS), 2015.

57

slide-58
SLIDE 58

Q & A

58