dezhen song, ken goldberg uc berkeley, united states anatoly pashkevich

30
1 ShareCam Part II: Approximate and Distributed Algorithms for a Collaboratively Controlled Robotic Webcam Supported in part by the National Science Foundation Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich State University of Informatics and Radioelectronics, Belarus

Upload: bono

Post on 15-Jan-2016

30 views

Category:

Documents


1 download

DESCRIPTION

ShareCam Part II: Approximate and Distributed Algorithms for a Collaboratively Controlled Robotic Webcam. Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich State University of Informatics and Radioelectronics, Belarus. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

1

ShareCam Part II: Approximate and Distributed Algorithms for a Collaboratively Controlled

Robotic Webcam

Supported in part by the National Science Foundation

Dezhen Song, Ken GoldbergUC Berkeley, United States

Anatoly PashkevichState University of Informatics and

Radioelectronics, Belarus

Page 2: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

2

Robot System Taxonomy (Tanie, Matsuhira, Chong 00)

Single Operator, Single Robot (SOSR):

Single Operator, Multiple Robot (SOMR):

Multiple Operator, Multiple Robot (MOMR):

Multiple Operator, Single Robot (MOSR):

Page 3: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

3

Page 4: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

4

Contents

• Related work

• Problem definition

• Algorithm– Approximation bound– Distributed algorithm

• Results

• Future work

Page 5: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

5

Related Work• Facilities Location Problems

– Megiddo and Supowit [84]

– Eppstein [97]

– Halperin et al. [02]

• Rectangle Fitting – Grossi and Italiano [99,00]

– Agarwal and Erickson [99]

– Mount et al [96]

– Kapelio et al [95]

Page 6: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

6

Related Work

• Similarity Measures – Kavraki [98]

– Broder et al [98, 00]

– Veltkamp and Hagedoorn [00]

• Distributed robot algorithms – Sagawa et al [01], Safaric[01]

– Parker[02], Bulter et al. [01]

– Mumolo et al [00], Hayes et al [01]

– Agassounon et al [01], Chen [99]

Page 7: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

7

Related Work• Existing algorithms for ShareCam

– Song, van der Stappen, Goldberg [02] O(n2)

– Har-Peled, Koltun, Song, Goldberg [03] O(n log n)

Page 8: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

8

One Optimal Frame

ShareCam Problem: Given n requests, find optimal frame

Page 9: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

9

Problem Definition• Assumptions

– Camera has fixed aspect ratio: 4 x 3– Candidate frame c = [x, y, z] t

– (x, y) R2 (continuous set)– z Z (continuous set)

(x, y)3z

4z

Page 10: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

10

Problem Definition

Requested frames: ri=[xi, yi, zi], i=1,…,n

Page 11: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

11

Problem Definition• “Satisfaction” for user i: 0 Si 1

Si = 0 Si = 1

= c ri c = ri

Page 12: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

12

• Measure user i’s satisfaction:

)1),/min(()/(

1,)(

)(min

)(

)(),(

zzap

csize

rsize

rArea

rcAreacrs

iii

i

i

ii

Satisfaction Metrics

Requested frame ri Area= ai

Candidate frame c

Area = api

Page 13: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

13

Optimization Problem

n

iii

crcs

1

),(max

Page 14: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

14

Algorithm Overview

• Grid based approach• Derive approximation bound

– Price to pay for enlarging a candidate frame– Optimal frame must be enclosed by a large

frame on the sampling lattice. The size difference depends on lattice resolution

– Bound depends on inputs and lattice resolution

• Distributed algorithm

Page 15: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

15

Approximation Algorithm

n)dd

whgO(

spacing resolution :

spacing lattice :

z2

zd

d x

y

d

Compute S(x,y) at lattice of sample points:

w, h : width and height,g: Resolution range

Page 16: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

16

Approximation Bound

Requested frames

Page 17: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

17

Approximation Bound

c

Requested frames

Candidate frame

z

ncrscs

z

zzap

crs

n

iii

iii

ii

1

),()(

/1

)/)(/(

),(

Page 18: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

18

Approximation Bound

ca

cb

Requested frames

Candidate frames

bb

aa

zncs

zncs

/)(

/)(

b

a

a

b

z

z

cs

cs

)(

)(

Page 19: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

19

Approximation Bound

b

a

a

b

z

z

cs

cs

)(

)(

ca

cb

Requested regions

Candidate frames

Page 20: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

20

Approximation Algorithm

ca

cb

za

a

b

a

z

dz

z

z

z

dd

2

3set

Page 21: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

21

Approximation Algorithm

– Run Time: – O(n / 3)

c* : Optimal frame

: Optimal at lattice (Algorithm output)

c~

: Smallest frame at lattice that encloses c*

c

)ˆ()~()( * cscscs

)(

)ˆ(

)(

)~(1

** cs

cs

cs

cs

zdz

z

2...

min

min

1

Page 22: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

22

Distributed Algorithms

•Server O(n+1/3)•Client O(1/3)•Robustness to dropouts…

Page 23: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

23

Distributed Lattice• Define Final Lattice (Define d)

dd

Page 24: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

24

Distributed Lattice• Divide Lattice point based on n (Assume n=4)

Page 25: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

25

Distributed Lattice• Sub lattice for each user

Page 26: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

26

Robustness to client failures

Page 27: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

27

Results

• A demo with 6 inputs

t

Page 28: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

28

Current & Future Work - Satellite Application

Page 29: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

29

Current & future work - Functional Box Sums

• Efficient reporting of

n

i i

ii

n

ii φArea

φAreaωsS

11 )()Φ(

)Φ()Φ(

Φ

kφ[Zhang et al 2002]

)log)/1(( 23 nnO

Page 30: Dezhen Song, Ken Goldberg UC Berkeley, United States Anatoly Pashkevich

30

www.co-opticon.net