Enhancing Search for Satisficing Temporal Planning with Objective-driven Decisions

18
Enhancing Search for Satisficing Temporal Planning with Objective- driven Decisions J. Benton Patrick Eyerich Subbarao Kambhampati

description

Enhancing Search for Satisficing Temporal Planning with Objective-driven Decisions. Patrick Eyerich. Subbarao Kambhampati. J. Benton. g-value plateaus in Temporal Planning. Common temporal planning objective function (:metric (minimize (total-time ))) - PowerPoint PPT Presentation

Transcript of Enhancing Search for Satisficing Temporal Planning with Objective-driven Decisions

Page 1: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

Enhancing Search for Satisficing Temporal Planning

with Objective-driven DecisionsJ. Benton Patrick

EyerichSubbarao

Kambhampati

Page 2: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

2

g-value plateaus in Temporal Planning

Common temporal planning objective function (:metric (minimize (total-time)))

Makespan as the evaluation function is inefficient for satisificing search g-value plateaus Leads to worst case cost-variance between search

operations The usual approach: Use a Surrogate Search

Choose a surrogate evaluation function that allows for scalability, improving the cost-variance between search states

Objective Function ≠ Evaluation Function We want to improve “keeping track” of objective

function

Page 3: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

3

Temporal Fast Downward Temporal Fast Downward (TFD)

Objective Function

CorrespondingEvaluation Function

SurrogateEvaluation Function

Page 4: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

4

Temporal Fast Downward Search5

@ end eff @ end eff

4

6

3@ end eff

@ start @ end eff

@ start @ end eff

2

2

Page 5: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

5

Temporal Fast Downward Search5

@ end eff @ end eff

4

6

3@ end eff

@ start @ end eff

@ start @ end eff

2

2…

Page 6: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

6

Find the Better Path Force consideration of better-makespan

path Should maintain surrogate evaluation

function’s scalability

Our idea: Determine whether operators are useful according to makespan and force their expansion

Page 7: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

7

Useful Operators Related to Wehrle et al.’s useless

actions At parent state s

Remove operator o from the domain Find heuristic value for , Apply operator o to generate Find heuristic value for , If then operator is possibly

useful Its degree of usefulness is

Page 8: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

8

Makespan-Usefulness Example

Get all trucks to An optimal plan

Page 9: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

Makespan-Usefulness Example

9

Page 10: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

10

Lookahead on Useful Operators Force expansion of most makespan-

useful state before other states Remove ‘best’ node from queue Analyze for child states for makespan-

usefulness Expand state given by most useful

operator Evaluate each resulting grandchild state

according to the surrogate evaluation function and push into queue

Page 11: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

11

Useful Operator Lookahead5

@ end eff @ end eff

4

6

3@ end eff

@ start @ end eff

@ start @ end eff

2

2

Page 12: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

12

Empirical Evaluation 4 Anytime search variations

TFD TFD with useful lookahead, TFD with lazy evaluation followed by TFD

with useful lookahead (and without lazy evaluation),

TFD with lazy evaluation followed by TFD without lazy evaluation,

Makespan heuristic using STN 30 minute timeout Compared on IPC score

Page 13: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

13

Results Over Time

Page 14: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

14

Results Over Time

Page 15: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

15

At 30 Minutes

Page 16: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

16

Quality Change

Page 17: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

17

Summary Used notion of operator usefulness

Lookahead on most useful operator Use in combination with surrogate

search

Shown to improve plan quality in some domains

Continues to help when combined with a portfolio-like approach

Page 18: Enhancing Search for  Satisficing Temporal Planning with Objective-driven Decisions

18

Future Work Lookahead more than one step

k-deep local lookaheads on most useful operators combined with best-first search

Use relaxed solutions YAHSP-style lookahead but stop when no

makespan-useful operators are applicable