prm and multi-space planning problems : how to handle many motion planning queries?

28
PRM and Multi-Space Planning Problems: How to handle many motion planning queries? Jean-Claude Latombe Computer Science Department Stanford University 1 (based on discussions with Tim Bretl and Kris Hauser)

Upload: trent

Post on 24-Feb-2016

37 views

Category:

Documents


0 download

DESCRIPTION

PRM and Multi-Space Planning Problems : How to handle many motion planning queries?. Jean-Claude Latombe Computer Science Department Stanford University. (based on discussions with Tim Bretl and Kris Hauser). PRM Planning in Single Space. Applicable to robots with many dofs - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

1

PRM and Multi-Space Planning Problems:

How to handle many motion planning queries?

Jean-Claude LatombeComputer Science Department

Stanford University

(based on discussions with Tim Bretl and Kris Hauser)

Page 2: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

2

PRM Planning in Single Space

Applicable to robots with many dofs In expansive configuration spaces:

Probabilistically complete + fast convergence

But unable to detect that no solution exists Cutoff on running time

Page 3: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

3

Convergence of a PRM Planner

???What should be the cutoff time?

Page 4: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

4

Planning in Multiple Spaces

Example 1: Climbing Robot

4-contact move

3-contact move

Page 5: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

5

Climbing Robot Dilemma[Bretl, 2005]

Thousands of spaces many PRM queries Most queries have no solution Running times for feasible queries are highly variable

Large time cutoff Prohibitive time is wasted on infeasible queries Small time cutoff Critical queries might not be solved

difficult queriesor bad luck?

Page 6: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

6

Other Examples Navigation on irregular terrain [Hauser,

2008]

Page 7: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

7

Other Examples Dexterous manipulation

Page 8: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

8

Other Examples Mechanical assembly

Page 9: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

9

Other Examples Spatial re-arrangements of movable

objects

[Stillman and Kuffner, 2007]

Page 10: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

Modular reconfigurable robotsOther Examples

[Yim]

Page 11: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

11

Other Examples Integration of task and motion planning

Change battery

Go to toolboxGrasp screwdriver

Go to old batteryUnscrew screws

Grasp old batteryUngrasp screwdriver

Remove old battery

Page 12: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

12

Basic Architecture

High-level Planner

(graph searching)

Motion Planner(PRM)

query resultMany queries are infeasible “climbing-robot” dilemma

Each query involves a distinct configuration space, with its own dimensionality, parameterization, and/or

constraints. queries cannot be processed using one single

precomputed roadmap

Page 13: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

13

Possible Approaches

Estimating query feasibility

Lazy PRM planning

High-level Planner

(graph searching)

Motion Planner(PRM)

query result

Page 14: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

Learning Transition Feasibility[Hauser, 2008] Create a large dataset of labeled transitions

Train a classifier Q : transition {feasible, non-feasible} Use classifier to select sequences of spaces with

likely feasible transitions between them But no work yet on learning feasibility of entire

queries (that require connecting two transitions)14

4 contacts 3 contacts

Non-feasible if empty

Page 15: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

15

Possible Approaches

Estimating query feasibility

Lazy PRM planning

High-level Planner

(graph searching)

Motion Planner(PRM)

query result

Page 16: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

16

Lazy PRM Planning[Bohlin & Kavraki, 2000; Sanchez-Ante, 2001]

Observation: PRM planning wastes much time testing that sampled configurations and connections are valid (e.g., free of collision).

Idea: Perform a computation only when there is enough evidence that it may be useful.

Page 17: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

17

Lazy Collision Checking of Connections [Sanchez-Ante, 2001]

sg

X

Page 18: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

18

Lazy Collision Checking of Connections [Sanchez-Ante, 2001]

sg

Page 19: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

Rationale Configuration spaces are rarely chaotic: so, the

connection between close valid configurations has high probability of being valid

Most of the time spent by a PRM planner is in testing connections

Most valid connections will not be part of the final solution

Testing connections is more expensive for valid connections than for invalid ones

Postpone testing a connection until the test is likely to be useful

19

Page 20: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

20

Extending Lazy PRM Planning

Create a bag of fine-grain computational probes:

Nodesampling

NodeConnection

Page 21: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

21

Extending Lazy PRM Planning

Sample a node and partially test if it is valid

p1 p8

p7p6p5

p4

p3p2

r d d > r+r’ p1 = 1d ≤ r+r’ p1 ~ d/r+r’

r’

Page 22: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

22

Extending Lazy PRM Planning

Create connection and partially test if it is valid

p1 p8

p7p6p5

p4

p3p2

p12

p23

p24

p45

p38

p46

p47

Page 23: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

23

Extending Lazy PRM Planning

Test further that a node is valid

p1 p12

p23

p24

p45

p38

p46

p47

p8

p7p6p5

p4’

p3p2

Page 24: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

24

Extending Lazy PRM Planning

Test further that a connection is valid

p1 p8

p7p6p5

p4’

p3p2

p12

p23

p24

p45

p38

p46

p47’

Page 25: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

25

Potential Advantages More choices opportunity for much smarter,

more efficient strategies

More flexibility in distributing computation over several spaces, e.g., focus on queries that have the highest probability of being feasible

Compatibility with probabilistic modeling of uncertainty, e.g., probabilistic distribution of obstacles

Page 26: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

26

Conclusion We will have to live with imperfect motion

planners like PRM planners Important problems require handling many

motion planning queries in distinct spaces “climbing-robot” dilemma

Possible approaches to address this dilemma:—Fast and reliable evaluation of query feasibility

(e.g., using trained classifiers)—Extended lazy PRM planning

Page 27: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

27

Page 28: PRM and  Multi-Space Planning Problems : How to handle many motion planning queries?

28

Narrow Passages I don’t think they are the main issue

in PRM planning. They are unlikely to occur by chance. Intentionally creating

complex narrow passages is not easy.

Alpha puzzle