27
The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL Fast C-obstacle Query Computation for Motion Planning Liang-Jun Zhang 12/13/2005 Liang-Jun Zhang 1 Young J. Kim 2 Gokul Varadhan 1 Dinesh Manocha 1 1: University of North Carolina - Chapel Hill, USA 2: Ewha Womans University, Korea, http://gamma.cs.unc.edu/cobstacle

Fast C-obstacle Query Computation for Motion Planning

  • Upload
    elita

  • View
    37

  • Download
    0

Embed Size (px)

DESCRIPTION

Fast C-obstacle Query Computation for Motion Planning. Liang-Jun Zhang 1 Young J. Kim 2 Gokul Varadhan 1 Dinesh Manocha 1 1: University of North Carolina - Chapel Hill, USA 2: Ewha Womans University, Korea, http://gamma.cs.unc.edu/cobstacle. Liang-Jun Zhang 12/13/2005. - PowerPoint PPT Presentation

Citation preview

Page 1: Fast C-obstacle Query Computation for Motion Planning

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Fast C-obstacle Query Computationfor Motion Planning

Liang-Jun Zhang12/13/2005

Liang-Jun Zhang1 Young J. Kim2 Gokul Varadhan1 Dinesh Manocha1

1: University of North Carolina - Chapel Hill, USA2: Ewha Womans University, Korea,

http://gamma.cs.unc.edu/cobstacle

Page 2: Fast C-obstacle Query Computation for Motion Planning

2

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Configuration space

Free space

C-obstacle

Do they intersect? Is its configuration in C-obstacle or free space?

• Free space and C-obstacle

Robot

Obstacle

Page 3: Fast C-obstacle Query Computation for Motion Planning

3

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

C-obstacle query in C-space

• C-obstacle query♦ Does a primitive lie completely

inside C-obstacle?♦ The primitive usually is a cell.

• Goal♦ Design an efficient C-obstacle query algorithm

Free spaceC-obstacle

Page 4: Fast C-obstacle Query Computation for Motion Planning

4

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Motivation- An important query for Motion Planning

• Cell Decomposition Method ♦ Label Cells as FULL and EMPTY

• Complete motion planning♦ Able to find a path or report path non-existence

EMPTY

init

goal

FULL

Page 5: Fast C-obstacle Query Computation for Motion Planning

5

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Previous work• Computing the boundary of C-

obstacle♦ Exponential complexity [Sacks 99, Sharir

97]♦ Degeneracy and floating point error

• Contact surface constraints♦ [Latombe 91, Zhu 91]♦ Complexity of contact surface enumeration♦ To deal with non-linear contact surfaces

Page 6: Fast C-obstacle Query Computation for Motion Planning

6

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Interpretation of C-obstacle query

C-obstacle

Free space

•Does the cell lie inside C-obstacle?

• Do the robot and obstacle intersect at all configurations?

•Can the robot ‘escape’ from the obstacle at some moment?

Obstacle

Page 7: Fast C-obstacle Query Computation for Motion Planning

7

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Algorithm overview• Penetration Depth

♦ How much does the robot penetrate into the obstacle at a configuration q ?

• Bounding Motion ♦ How much motion the robot can undergo,

when its configuration changes from q but within the query primitive?

• Query criterion♦ If Penetration Depth > Bounding Motion

the robot can not escape♦ the query primitive lies inside C-

obstacle

q

A(q)

Page 8: Fast C-obstacle Query Computation for Motion Planning

8

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Translational Penetration Depth: PDt

• Minimum translation to separate A, B ♦ [Dobkin 93, Agarwal 00,

Bergen 01, Kim 02]• PDt: not applicable♦ The robot is allowed to

both translate and rotate. ♦ Undergoing rotation, A

may ‘escape’ from B easier.

B

A

A’

A

B

Page 9: Fast C-obstacle Query Computation for Motion Planning

9

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Generalized Penetration Depth: PDg

• Consider both translation and rotation♦ [Zhang, Kim, Varadhan, Manocha: UNC-CS

TR05]

• Difficult for non-convex objects

• Convex A, B PDg(A,B)=PDt(A,B)

Page 10: Fast C-obstacle Query Computation for Motion Planning

10

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Algorithm-Lower bound on PDg

1. Convex decomposition 2. Eliminate non-overlapping pairs3. PDt over overlapping pairs4. LB(PDg) = Max over all PDts

A

B

A2

A1

B1

B2

A2

A1

B2

B1

Page 11: Fast C-obstacle Query Computation for Motion Planning

11

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

C-obstacle Query Criterion

If Lower Bound (PDg(A(qa), B))> Bounding Motion,

the cell C is in C-obstacle.

A(qa): set A’s config as qa

aq

C

Page 12: Fast C-obstacle Query Computation for Motion Planning

12

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Upper bound of Motion for line segment

♦ Configurations qa and qb • [Schwarzer,Saha,Latombe 04]

Max trajectory length over points on the moving

robot

bqaq,( , )a bq qA

( )aA q

( )bA q

Page 13: Fast C-obstacle Query Computation for Motion Planning

13

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Upper bound of Motion for cell

,( , ) max ( , )a bq qA C A

Any diagonal line segment yields maximum bounding motion.

aq

cqbqqa is the center of the cell Cqb is any point on the boundary of the cell.

C

,( , ) max ( , )a bq qA C A

,( , )a cq qA

Page 14: Fast C-obstacle Query Computation for Motion Planning

14

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

C-obstacle Query Criterion

If Lower Bound (PDg(A(qa), B))>

the cell C is in C-obstacle.

A(qa): set A’s config as qa

aqcqbq

C,( , )a cq qA

Page 15: Fast C-obstacle Query Computation for Motion Planning

15

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Application• Star-shaped roadmap: a

complete motion planning approach♦ [Varadhan and Manocha 05]

• To identify cells which lie inside C-obstacle♦ No subdivisions are applied for them

Page 16: Fast C-obstacle Query Computation for Motion Planning

16

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Results-`gear’: 2T+1R

video

Page 17: Fast C-obstacle Query Computation for Motion Planning

17

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Results

‘Piano’ ‘World map’

Page 18: Fast C-obstacle Query Computation for Motion Planning

18

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Effective and Performance

• Query timing: 0.04ms to 0.12 ms

Culled C-obstacle Cells• Cell Culling Ratio =

All C-obstacle Cells

Page 19: Fast C-obstacle Query Computation for Motion Planning

19

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Speedup For Star-shaped roadmap method

Page 20: Fast C-obstacle Query Computation for Motion Planning

20

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Conclusion• A fast C-obstacle query algorithm for

rigid robots• Based on generalized penetration

depth and bounding motion computation. ♦ Need not explicit computation of the boundary of

free space.♦ Robust and efficient

• Applied for accelerating a complete motion planning approach for 2D rigid robot.

Page 21: Fast C-obstacle Query Computation for Motion Planning

21

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Ongoing and Future work

• A Simple Path Non-Existence Algorithm for low DOF robots ♦ [L. Zhang, Y.J. Kim, D. Manocha]

WAFR2006

• Apply for 3D rigid robots• Handle articulated robots

Page 22: Fast C-obstacle Query Computation for Motion Planning

22

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Acknowledgements• Army Research Office• DARPA/REDCOM• NSF• ONR• Intel Corporation• KRF, STAR program of MOST,

Ewha SMBA consortium, the ITRC program (Korea)

Page 23: Fast C-obstacle Query Computation for Motion Planning

23

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Thanks

Any Questions?

Page 24: Fast C-obstacle Query Computation for Motion Planning

24

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Appendix

Page 25: Fast C-obstacle Query Computation for Motion Planning

25

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Generalized Penetration Depth: PDg

• Consider both translation and rotation♦ [Zhang, Kim, Varadhan, Manocha et al. 05]

• Trajectory length Separating path

Robot

Obstacle

Robot

Page 26: Fast C-obstacle Query Computation for Motion Planning

26

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

• PDg

• Difficult for non-convex objects• Convex A, B

PDg(A,B)=PDt(A,B)

MIN over all possible separating paths MAX of the trajectory length over all on the moving robot

Generalized Penetration Depth: PDg

Page 27: Fast C-obstacle Query Computation for Motion Planning

27

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Upper bound of Motion for line segment

• [Schwarzer,Saha,Latombe 04] ♦ Max trajectory length

over points on the moving robot

• ♦ The weighted sum of difference for x, y,

components between qa and qb

,

, , , , , ,

( , )a bq q

a x b x a y b y a r b

A

q q q q R q q

x

ybq

aq

R: ‘radius’ of the object

,( , )a bq qA