symbolic search and abstraction heuristics for cost
play

Symbolic Search and Abstraction Heuristics for Cost-Optimal Planning - PowerPoint PPT Presentation

Feb 09, 2024 •206 likes •1.35k views

Symbolic Search and Abstraction Heuristics for Cost-Optimal Planning Alvaro Torralba Advisors: Daniel Borrajo and Carlos Linares L opez Universidad Carlos III de Madrid June 2, 2015 Alvaro Torralba PhD Defense June 2, 2015 1 /

  1. Optimizing Image Computation Image computation is the main bottleneck in symbolic search How to represent the Transition Relation? ◮ Monolithic relation ⇒ may use exponential memory ◮ Solution in G AMER ⇒ One TR for each action Idea 1: Separate preconditions and effects → avoid using auxiliary variables! move-T (A, B) pre: at-T A eff: at-T B load-P (A) pre: at-T A, at-P A eff: at-P T move-T (B, A) pre: at-T B eff: at-T A . . . ´ Alvaro Torralba PhD Defense June 2, 2015 13 / 54

  2. Optimizing Image Computation Image computation is the main bottleneck in symbolic search How to represent the Transition Relation? ◮ Monolithic relation ⇒ may use exponential memory ◮ Solution in G AMER ⇒ One TR for each action Idea 1: Separate preconditions and effects → avoid using auxiliary variables! Idea 2: Conjunction Tree → check preconditions of all operators simultaneously at-T move-T (A, B) A B ⋆ load-P (A) at-P eff: at-T A . . . move-T (B, A) A B T ⋆ . . . eff: at-P T eff: at-P A eff: at-T B ∅ ´ Alvaro Torralba PhD Defense June 2, 2015 13 / 54

  3. Optimizing Image Computation Image computation is the main bottleneck in symbolic search How to represent the Transition Relation? ◮ Monolithic relation ⇒ may use exponential memory ◮ Solution in G AMER ⇒ One TR for each action Idea 1: Separate preconditions and effects → avoid using auxiliary variables! Idea 2: Conjunction Tree → check preconditions of all operators simultaneously Idea 3: Aggregate TRs → different strategies to group the actions move-T (A, B) move-T (A, B) load-P (A) load-P (A) move-T (B, A) move-T (B, A) . . . . . . ´ Alvaro Torralba PhD Defense June 2, 2015 13 / 54

  4. Empirical Results Compare image computation methods: TR 1 : baseline approach 1 TR 1 + : avoid using x ′ variables 2 CT L 20 : conjunction tree 3 T DT 100 k : aggregate TRs 4 Total coverage of symbolic search algorithms over 1375 instances: TR 1 TR 1 + CT L T DT 20 100 k Forward uniform-cost search 699 676 724 742 Backward uniform-cost search 444 525 529 532 Bidirectional uniform-cost search 729 763 769 793 BDDA ∗ with SPDBs 705 717 724 764 TR 1 ≤ TR 1 + ≤ CT L 20 ≤ T DT 100 k (across all domains) ´ Alvaro Torralba PhD Defense June 2, 2015 14 / 54

  5. Time of Bidirectional Search Solving time of TR 1 (seconds) 10 3 10 2 10 1 10 0 10 0 10 1 10 2 10 3 Solving time of T DT 100 (seconds) ´ Alvaro Torralba PhD Defense June 2, 2015 15 / 54

  6. Outline Introduction 1 Cost-Optimal Planning Symbolic Search 2 (Background) Symbolic Search Image Computation State Invariants Abstraction Heuristics 3 (Background) Abstractions Merge-and-Shrink for Symbolic Search Symbolic Perimeter Merge-and-Shrink Symbolic Bidirectional Heuristic Search 4 5 Conclusions Final Results: IPC14 Conclusions ´ Alvaro Torralba PhD Defense June 2, 2015 16 / 54

  7. Motivation: State Invariants in Symbolic Search Invariant: holds in all states that may belong to a solution path Mutex: pair of facts that cannot be true in the same state 1 → a truck cannot be simultaneously at two locations Invariant group: Set of facts such that exactly one is true 2 → a truck must be somewhere Generated computing h 2 in both directions Useful for: Removing operators from the planning task 1 Pruning invalid states during the search 2 ´ Alvaro Torralba PhD Defense June 2, 2015 17 / 54

  8. Encoding Constraints with cBDD cBDD : BDD that describes invalid states Mutex: f i ∧ f j 1 “At-least-1” invariant: ¬ ( f 1 ∨ f 2 ∨ . . . ∨ f n ) 2 Remove invalid states from S g : S g \ cBDD v 1 v 1 v 1 v 2 v 2 v 2 v 2 v 2 v 3 v 3 v 3 v 3 T F T F T F (c) cBDD (d) S g (e) S g \ cBDD e-deletion : encode invariants in the TRs → no invalid states are generated ´ Alvaro Torralba PhD Defense June 2, 2015 18 / 54

  9. Experimental Results Constraints found in 35 out of 43 domains 10%-74% invalid operators found in 17 out of 43 domains Mutex types: ◮ Baseline (B) ◮ Not pruning invalid states: M ∅ ◮ Pruning invalid states: cBDD or e-deletion ( e - del ) Remove invalid ops B M ∅ cBDD e - del Forward uniform-cost search 699 742 745 750 Backward uniform-cost search 509 532 677 696 Bidirectional uniform-cost search 765 793 836 841 BDDA ∗ with SPDBs 736 764 777 781 ´ Alvaro Torralba PhD Defense June 2, 2015 19 / 54

  10. Time of Bidirectional Uniform-Cost Search 2 , 000 Solving time of O h2 (seconds) 2 , 000 Solving time of O (seconds) 1 , 500 1 , 500 1 , 000 1 , 000 500 500 0 0 0 500 1 , 000 1 , 500 2 , 000 0 500 1 , 000 1 , 500 2 , 000 Solving time of O h2 (seconds) Solving time of e - del (seconds) (f) remove operators (g) prune invalid states Solving time of M 100 k (seconds) 2 , 000 1 , 500 1 , 000 500 0 0 500 1 , 000 1 , 500 2 , 000 Solving time of e - del (seconds) (h) e-deletion vs cBDD ´ Alvaro Torralba PhD Defense June 2, 2015 20 / 54

  11. Comparison with State-of-the-Art Planners 800 600 Coverage 400 200 C G AMER -BD G AMER - BD 0 10 0 10 1 10 2 10 3 Time (seconds) ´ Alvaro Torralba PhD Defense June 2, 2015 21 / 54

  12. Comparison with State-of-the-Art Planners 800 600 Coverage 400 200 C G AMER - FW E XPLICIT -FW 0 10 0 10 1 10 2 10 3 Time (seconds) ´ Alvaro Torralba PhD Defense June 2, 2015 21 / 54

  13. Comparison with State-of-the-Art Planners 800 600 Coverage 400 200 A ∗ + LM- CUT BDDA ∗ + SPDBs 0 10 0 10 1 10 2 10 3 Time (seconds) ´ Alvaro Torralba PhD Defense June 2, 2015 21 / 54

  14. Comparison with State-of-the-Art Planners 800 600 Coverage 400 C G AMER -BD G AMER - BD C G AMER - FW 200 E XPLICIT -FW A ∗ + LM- CUT BDDA ∗ + SPDBs 0 10 0 10 1 10 2 10 3 Time (seconds) ´ Alvaro Torralba PhD Defense June 2, 2015 21 / 54

  15. Summary Image computation 1 ◮ Analyzed different methods for image computation ◮ Best method: aggregate TRs State invariants 2 ◮ Pruning invalid states (specially useful in bw search) ◮ Best encoding for symbolic search: e-edeletion These significantly improved performance of symbolic planning → Symbolic bidirectional blind search is the current state-of-the-art for cost-optimal planning ´ Alvaro Torralba PhD Defense June 2, 2015 22 / 54

  16. Outline Introduction 1 Cost-Optimal Planning Symbolic Search 2 (Background) Symbolic Search Image Computation State Invariants Abstraction Heuristics 3 (Background) Abstractions Merge-and-Shrink for Symbolic Search Symbolic Perimeter Merge-and-Shrink Symbolic Bidirectional Heuristic Search 4 5 Conclusions Final Results: IPC14 Conclusions ´ Alvaro Torralba PhD Defense June 2, 2015 23 / 54

  17. Motivation: Heuristics in Symbolic Search Heuristics Delete-relaxation: h max , h + Landmarks: h LA , LM-cut Abstractions: PDBs, M&S, CEGAR, Fork Critical paths: h m Flow max add LP ´ Alvaro Torralba PhD Defense June 2, 2015 24 / 54

  18. Motivation: Heuristics in Symbolic Search Heuristics Delete-relaxation: h max , h + Landmarks: h LA , LM-cut Abstractions: PDBs, M&S, CEGAR, Fork Critical paths: h m Flow max add LP Challenge: How to evaluate h ( s ) on a set of states? ´ Alvaro Torralba PhD Defense June 2, 2015 24 / 54

  19. Motivation: Heuristics in Symbolic Search Heuristics Delete-relaxation: h max , h + Landmarks: h LA , LM-cut Abstractions: PDBs, M&S, CEGAR, Fork Critical paths: h m Flow max add LP Challenge: How to evaluate h ( s ) on a set of states? ´ Alvaro Torralba PhD Defense June 2, 2015 24 / 54

  20. Abstraction Heuristics Abstraction: Mapping from states to abstract states ◮ Smaller abstract state space → easier to search ◮ Use optimal distances in abstract state space as heuristic ◮ Preserve transitions → admissible estimation 4 5 6 7 start 00 10 20 30 0 3 2 1 01 11 21 31 3 2 1 0 02 12 22 32 ´ Alvaro Torralba PhD Defense June 2, 2015 25 / 54

  21. Abstraction Heuristics Abstraction: Mapping from states to abstract states ◮ Smaller abstract state space → easier to search ◮ Use optimal distances in abstract state space as heuristic ◮ Preserve transitions → admissible estimation 4 5 6 7 start 00 10 20 30 0 3 2 1 01 11 21 31 3 2 1 0 02 12 22 32 h = 3 2 1 0 ´ Alvaro Torralba PhD Defense June 2, 2015 25 / 54

  22. Abstraction Heuristics Abstraction: Mapping from states to abstract states ◮ Smaller abstract state space → easier to search ◮ Use optimal distances in abstract state space as heuristic ◮ Preserve transitions → admissible estimation 4 5 6 7 start 00 10 20 30 0 3 2 1 01 11 21 31 3 2 1 0 02 12 22 32 h = 3 2 1 0 Pattern Databases (PDBs) ◮ Ignore some variables in the problem ◮ Limitation: ignoring a single variable may relax too much ´ Alvaro Torralba PhD Defense June 2, 2015 25 / 54

  23. Abstraction Heuristics Abstraction: Mapping from states to abstract states ◮ Smaller abstract state space → easier to search ◮ Use optimal distances in abstract state space as heuristic ◮ Preserve transitions → admissible estimation 2 2 1 0 start 00 10 20 30 0 1 1 1 01 11 21 31 0 0 0 0 02 12 22 32 h = 0 0 0 0 Pattern Databases (PDBs) ◮ Ignore some variables in the problem ◮ Limitation: ignoring a single variable may relax too much ´ Alvaro Torralba PhD Defense June 2, 2015 25 / 54

  24. Merge-and-Shrink Algorithm (M&S) Algorithm 1: M&S α 1 ← Π υ 1 foreach υ ∈ { υ 2 . . . υ n } : if | α | > N : shrink ( α i − 1 ) ⊗ Π i α i ← α i − 1 ⊗ Π i return α Merge strategy: Linear → variable ordering Shrink strategy → reduce abstraction size ´ Alvaro Torralba PhD Defense June 2, 2015 26 / 54

  25. Merge-and-Shrink Algorithm (M&S) α 1 = T A move A , B Algorithm 1: M&S A B α 1 ← Π υ 1 foreach υ ∈ { υ 2 . . . υ n } : move B , A if | α | > N : shrink ( α i − 1 ) ⊗ Π i α i ← α i − 1 ⊗ Π i return α Merge strategy: Linear → variable ordering Shrink strategy → reduce abstraction size ´ Alvaro Torralba PhD Defense June 2, 2015 26 / 54

  26. Merge-and-Shrink Algorithm (M&S) α 1 = T A move A , B Algorithm 1: M&S A B α 1 ← Π υ 1 foreach υ ∈ { υ 2 . . . υ n } : move B , A if | α | > N : α 3 = T A , P T , P A shrink ( α i − 1 ) ⊗ Π i move A , B α i ← α i − 1 ⊗ Π i return α AA BA move B , A load/unload A Merge strategy: Linear move A , B → variable ordering Ac Bc Shrink strategy move B , A load/unload B → reduce abstraction move A , B size AB BB move B , A ´ Alvaro Torralba PhD Defense June 2, 2015 26 / 54

  27. Merge-and-Shrink Algorithm (M&S) α 1 = T A move A , B Algorithm 1: M&S A B α 1 ← Π υ 1 foreach υ ∈ { υ 2 . . . υ n } : move B , A if | α | > N : α 3 = T A , P T , P A shrink ( α i − 1 ) ⊗ Π i move A , B α i ← α i − 1 ⊗ Π i return α AA BA move B , A load/unload A Merge strategy: Linear move A , B → variable ordering Ac Bc Shrink strategy move B , A load/unload B → reduce abstraction move A , B size AB BB move B , A ´ Alvaro Torralba PhD Defense June 2, 2015 26 / 54

  28. Outline Introduction 1 Cost-Optimal Planning Symbolic Search 2 (Background) Symbolic Search Image Computation State Invariants Abstraction Heuristics 3 (Background) Abstractions Merge-and-Shrink for Symbolic Search Symbolic Perimeter Merge-and-Shrink Symbolic Bidirectional Heuristic Search 4 5 Conclusions Final Results: IPC14 Conclusions ´ Alvaro Torralba PhD Defense June 2, 2015 27 / 54

  29. Merge-and-Shrink for Symbolic Search Hypothesis: BDDA ∗ lacks good heuristics → BDDA ∗ + M&S can improve results How to use M&S in symbolic search: M&S algorithm M&S heuristic to ADD ADD to BDDs BDDs to use in symbolic search ´ Alvaro Torralba PhD Defense June 2, 2015 28 / 54

  30. Merge-and-Shrink as ADDs α 1 = T A move A , B A B move B , A α 3 = T A , P T , P A move A , B AA BA move B , A load/unload A move A , B Ac Bc move B , A load/unload B move A , B AB BB move B , A ´ Alvaro Torralba PhD Defense June 2, 2015 29 / 54

  31. Merge-and-Shrink as ADDs α 1 = T A move A , B A B move B , A T A α 3 = T A , P T , P A P T move A , B AA BA P A move B , A load/unload A move A , B Ac Bc move B , A load/unload B move A , B AB BB move B , A ´ Alvaro Torralba PhD Defense June 2, 2015 29 / 54

  32. Merge-and-Shrink as ADDs α 1 = T A move A , B A B move B , A T A α 3 = T A , P T , P A P T move A , B AA BA P A move B , A load/unload A move A , B Ac Bc move B , A load/unload B move A , B AB BB move B , A ´ Alvaro Torralba PhD Defense June 2, 2015 29 / 54

  33. Merge-and-Shrink as ADDs α 1 = T A move A , B A B move B , A T A α 3 = T A , P T , P A P T move A , B AA BA P A move B , A load/unload A move A , B AA Ac AB ✗ BA Bc BB ✗ Ac Bc move B , A load/unload B move A , B AB BB move B , A ´ Alvaro Torralba PhD Defense June 2, 2015 29 / 54

  34. Merge-and-Shrink as ADDs α 1 = T A move A , B A B move B , A T A α 3 = T A , P T , P A P T move A , B AA BA P A move B , A load/unload A move A , B AA Ac AB ✗ BA Bc BB ✗ Ac Bc move B , A load/unload B move A , B AB BB move B , A ´ Alvaro Torralba PhD Defense June 2, 2015 29 / 54

  35. Merge-and-Shrink as ADDs α 1 = T A move A , B A B move B , A T A α 3 = T A , P T , P A P T move A , B AA BA P A move B , A load/unload A move A , B AA Ac AB ✗ BA Bc Ac Bc move B , A load/unload B move A , B AB BB move B , A ´ Alvaro Torralba PhD Defense June 2, 2015 29 / 54

  36. Theoretical Results M&S to ADDs/BDDs in polynomial time Related empirical results: ◮ ADD representation of heuristics reduces memory ◮ Variable ordering has a huge impact ADD/BDD reduction rules may achieve exponential gain in memory with respect to shrinking perfect strategies → shows potential of improvement for M&S strategies ´ Alvaro Torralba PhD Defense June 2, 2015 30 / 54

  37. Empirical Results Used M&S in symbolic search → Worse than symbolic PDBs 657 M&S g 749 638 M&S b 739 568 Blind ( ∅ ) 756 550 600 650 700 750 Coverage Symbolic-A ∗ Explicit-A ∗ ◮ Contradicts our hypothesis ´ Alvaro Torralba PhD Defense June 2, 2015 31 / 54

  38. Outline Introduction 1 Cost-Optimal Planning Symbolic Search 2 (Background) Symbolic Search Image Computation State Invariants Abstraction Heuristics 3 (Background) Abstractions Merge-and-Shrink for Symbolic Search Symbolic Perimeter Merge-and-Shrink Symbolic Bidirectional Heuristic Search 4 5 Conclusions Final Results: IPC14 Conclusions ´ Alvaro Torralba PhD Defense June 2, 2015 32 / 54

  39. Motivation: Combine Symbolic Search and M&S Symbolic PDBs: larger abstract state spaces 1 M&S: flexible abstractions 2 Can we get the best of both worlds? ´ Alvaro Torralba PhD Defense June 2, 2015 33 / 54

  40. Motivation: Combine Symbolic Search and M&S Symbolic PDBs: larger abstract state spaces 1 M&S: flexible abstractions 2 Can we get the best of both worlds? → Use symbolic search to search M&S abstractions! Symbolic Perimeter M&S: Symbolic M&S abstractions: larger M&S abstract state spaces 1 Perimeter abstractions 2 ´ Alvaro Torralba PhD Defense June 2, 2015 33 / 54

  41. SM&S Hierarchy Enlarged M&S abstractions: to perform symbolic search v 1 v 2 v 3 v 4 v 5 α SM&S (original problem) 0 α SM&S α 1 1 α SM&S α 2 2 α SM&S α 3 3 α SM&S α 4 (M&S abstraction) 4 ´ Alvaro Torralba PhD Defense June 2, 2015 34 / 54

  42. SM&S Hierarchy Enlarged M&S abstractions: to perform symbolic search v 1 v 2 v 3 v 4 v 5 α SM&S (original problem) 0 α SM&S α 1 1 α SM&S α 2 2 α SM&S α 3 3 α SM&S α 4 (M&S abstraction) 4 ´ Alvaro Torralba PhD Defense June 2, 2015 34 / 54

  43. SM&S Hierarchy Enlarged M&S abstractions: to perform symbolic search v 1 v 2 v 3 v 4 v 5 α SM&S (original problem) 0 α SM&S α 1 1 α SM&S α 2 2 α SM&S α 3 3 α SM&S α 4 (M&S abstraction) 4 ´ Alvaro Torralba PhD Defense June 2, 2015 34 / 54

  44. SM&S Hierarchy Enlarged M&S abstractions: to perform symbolic search v 1 v 2 v 3 v 4 v 5 α SM&S (original problem) 0 α SM&S α 1 1 α SM&S α 2 2 α SM&S α 3 3 α SM&S α 4 (M&S abstraction) 4 ´ Alvaro Torralba PhD Defense June 2, 2015 34 / 54

  45. SM&S Hierarchy Enlarged M&S abstractions: to perform symbolic search v 1 v 2 v 3 v 4 v 5 α SM&S (original problem) 0 α SM&S α 1 1 α SM&S α 2 2 α SM&S α 3 3 α SM&S α 4 (M&S abstraction) 4 ´ Alvaro Torralba PhD Defense June 2, 2015 34 / 54

  46. SM&S Hierarchy Enlarged M&S abstractions: to perform symbolic search v 1 v 2 v 3 v 4 v 5 α SM&S (original problem) 0 α SM&S α 1 1 α SM&S α 2 2 α SM&S α 3 3 α SM&S α 4 (M&S abstraction) 4 ´ Alvaro Torralba PhD Defense June 2, 2015 34 / 54

  47. Perimeter Abstractions Challenges addressed with symbolic search Regression 1 Expensive operations: 2 ⋆ membership in perimeter ⋆ frontier mapping S ⋆ Set perimeter radius 3 Exp ( α 0 ) Contributions Exp ( α 1 ) Multiple abstraction levels 1 Improved initialization of 2 abstract searches ´ Alvaro Torralba PhD Defense June 2, 2015 35 / 54

  48. Perimeter Abstractions Challenges addressed with symbolic search Regression 1 Expensive operations: 2 ⋆ membership in perimeter ⋆ frontier mapping S ⋆ Set perimeter radius 3 Exp ( α 0 ) Contributions Exp ( α 1 ) Multiple abstraction levels 1 Exp ( α 2 ) Improved initialization of 2 abstract searches ´ Alvaro Torralba PhD Defense June 2, 2015 35 / 54

  49. Symbolic Perimeter Merge-and-Shrink V α 0 = ∅ Exp ( α 0 ) υ 0 υ 1 υ 2 S 1 S 2 ∅ S ⋆ υ 3 . . . h 0 1 2 truncated h Exp ( α 0 ) h SPM&S heuristic is admissible and consistent ´ Alvaro Torralba PhD Defense June 2, 2015 36 / 54

  50. Symbolic Perimeter Merge-and-Shrink V α 1 υ 0 V α 0 = ∅ υ 1 M&S e 0 e 1 e 2 Exp ( α 0 ) υ 0 υ 1 υ 2 ∅ S ⋆ S 1 S 2 υ 3 . . . h 0 1 2 truncated h Exp ( α 0 ) h SPM&S heuristic is admissible and consistent ´ Alvaro Torralba PhD Defense June 2, 2015 36 / 54

  51. Symbolic Perimeter Merge-and-Shrink V α 1 υ 0 V α 0 = ∅ υ 1 M&S e 0 e 1 e 2 Exp ( α 0 ) Exp ( α 1 ) υ 0 υ 1 α 1 α 1 υ 2 S 3 ∅ S ⋆ S 1 S 2 S 2 υ 3 . . . h 0 1 2 2 3 truncated truncated h Exp ( α 0 ) h Exp ( α 1 ) h SPM&S heuristic is admissible and consistent ´ Alvaro Torralba PhD Defense June 2, 2015 36 / 54

  52. Symbolic Perimeter Merge-and-Shrink V α 1 υ 0 υ 0 V α 0 = ∅ V α 2 υ 1 υ 1 M&S M&S υ 2 e 0 e 1 e 2 e 0 e 1 e 2 Exp ( α 0 ) Exp ( α 1 ) Exp ( α 2 ) υ 0 υ 1 α 1 α 1 α 2 α 2 υ 2 S 3 S 3 ∅ S ⋆ S 1 S 2 S 2 S 4 υ 3 . . . h 0 1 2 2 3 3 4 truncated truncated h Exp ( α 0 ) h Exp ( α 1 ) h Exp ( α 2 ) h SPM&S heuristic is admissible and consistent ´ Alvaro Torralba PhD Defense June 2, 2015 36 / 54

  53. Empirical Results LM- CUT 796 M&S 650 809 SPM&S SPPDB 822 SP 800 650 700 750 800 850 Coverage ´ Alvaro Torralba PhD Defense June 2, 2015 37 / 54

  54. Empirical Results: Expanded Nodes 10 8 10 8 Expanded nodes M&S bop 10k Expanded nodes LM- CUT 10 6 10 6 10 4 10 4 10 2 10 2 10 0 10 0 10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 7 10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 7 Expanded nodes SPM&S bop 10k Expanded nodes SPM&S bop 10k ´ Alvaro Torralba PhD Defense June 2, 2015 38 / 54

  55. Empirical Results: Expanded Nodes 10 8 Expanded nodes SPPDB 10 6 Expanded nodes SP 10 6 10 4 10 4 10 2 10 2 10 0 10 0 10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 7 10 8 10 0 10 1 10 2 10 3 10 4 10 5 10 6 10 7 10 8 Expanded nodes SPM&S bop 10k Expanded nodes SPM&S bop 10k ´ Alvaro Torralba PhD Defense June 2, 2015 39 / 54

  56. Summary Symbolic Perimeter M&S Combines M&S, perimeter abstractions and symbolic search Improvements to perimeter abstractions Synergy between symbolic search and perimeter abstractions More accurate heuristic than both! But... Results still worse than symbolic bidirectional uniform-cost search ´ Alvaro Torralba PhD Defense June 2, 2015 40 / 54

  57. Outline Introduction 1 Cost-Optimal Planning Symbolic Search 2 (Background) Symbolic Search Image Computation State Invariants Abstraction Heuristics 3 (Background) Abstractions Merge-and-Shrink for Symbolic Search Symbolic Perimeter Merge-and-Shrink Symbolic Bidirectional Heuristic Search 4 5 Conclusions Final Results: IPC14 Conclusions ´ Alvaro Torralba PhD Defense June 2, 2015 41 / 54

  58. Motivation: Heuristics in Symbolic Bidirectional Search Observations Bidirectional brute-force search is a state-of-the-art technique 1 Good symbolic abstraction heuristics 2 ´ Alvaro Torralba PhD Defense June 2, 2015 42 / 54

  59. Motivation: Heuristics in Symbolic Bidirectional Search Observations Bidirectional brute-force search is a state-of-the-art technique 1 Good symbolic abstraction heuristics 2 Use abstraction heuristics in symbolic bidirectional search! ´ Alvaro Torralba PhD Defense June 2, 2015 42 / 54

  60. Motivation: Heuristics in Symbolic Bidirectional Search Observations Bidirectional brute-force search is a state-of-the-art technique 1 Good symbolic abstraction heuristics 2 Use abstraction heuristics in symbolic bidirectional search! However, bidirectional heuristic search is not so easy: ◮ Very promising since years ago ◮ Never really able to outperform A ∗ or bidirectional uniform-cost search ´ Alvaro Torralba PhD Defense June 2, 2015 42 / 54

  61. Algorithm Main idea: Start symbolic bidirectional uniform-cost search 1 ⋆ If it succeeds → done! Detect when it is going to fail and activate heuristics 2 Abstraction heuristics: Bidirectional, Partial, Perimeter s 0 s ⋆ ´ Alvaro Torralba PhD Defense June 2, 2015 43 / 54

  62. Algorithm Main idea: Start symbolic bidirectional uniform-cost search 1 ⋆ If it succeeds → done! Detect when it is going to fail and activate heuristics 2 Abstraction heuristics: Bidirectional, Partial, Perimeter Decide which search advance: useful and feasible s 0 s ⋆ ´ Alvaro Torralba PhD Defense June 2, 2015 43 / 54

  63. Algorithm Main idea: Start symbolic bidirectional uniform-cost search 1 ⋆ If it succeeds → done! Detect when it is going to fail and activate heuristics 2 Abstraction heuristics: Bidirectional, Partial, Perimeter Decide which search advance: useful and feasible s 0 s ⋆ ´ Alvaro Torralba PhD Defense June 2, 2015 43 / 54

  64. Algorithm Main idea: Start symbolic bidirectional uniform-cost search 1 ⋆ If it succeeds → done! Detect when it is going to fail and activate heuristics 2 Abstraction heuristics: Bidirectional, Partial, Perimeter Decide which search advance: useful and feasible s 0 s ⋆ ´ Alvaro Torralba PhD Defense June 2, 2015 43 / 54

  65. Algorithm Main idea: Start symbolic bidirectional uniform-cost search 1 ⋆ If it succeeds → done! Detect when it is going to fail and activate heuristics 2 Abstraction heuristics: Bidirectional, Partial, Perimeter Decide which search advance: useful and feasible s 0 s ⋆ ´ Alvaro Torralba PhD Defense June 2, 2015 43 / 54

  66. Empirical Results Best 873 SM&S 840 PDB ran Full SymBA ∗ 844 PDB cgl 842 BD ( ∅ ) 842 820 840 860 880 Coverage ´ Alvaro Torralba PhD Defense June 2, 2015 44 / 54

  67. Empirical Results 870 Best 873 842 SM&S 840 Full SymBA ∗ 836 PDB ran No perimeter 844 abstraction 844 PDB cgl 842 842 BD ( ∅ ) 842 820 840 860 880 Coverage ´ Alvaro Torralba PhD Defense June 2, 2015 44 / 54

  68. Empirical Results 868 Best 870 873 842 842 SM&S Full SymBA ∗ 840 No perimeter 836 PDB ran 836 abstraction 844 No bidir 837 abstraction PDB cgl 844 842 842 BD ( ∅ ) 842 842 820 840 860 880 Coverage ´ Alvaro Torralba PhD Defense June 2, 2015 44 / 54

  69. Summary Contributions: ◮ SymBA ∗ : a symbolic bidirectional heuristic search algorithm ◮ Bidirectional search in abstract state spaces ◮ Synergy: Symbolic search + Bidirectional search + Perimeter abstractions Symbolic Bidirectional A ∗ is possible ◮ Future work: domain-independent abstraction strategies (better than a random selection) ´ Alvaro Torralba PhD Defense June 2, 2015 45 / 54

  70. Outline Introduction 1 Cost-Optimal Planning Symbolic Search 2 (Background) Symbolic Search Image Computation State Invariants Abstraction Heuristics 3 (Background) Abstractions Merge-and-Shrink for Symbolic Search Symbolic Perimeter Merge-and-Shrink Symbolic Bidirectional Heuristic Search 4 5 Conclusions Final Results: IPC14 Conclusions ´ Alvaro Torralba PhD Defense June 2, 2015 46 / 54

  71. Final Results 800 600 Coverage 400 200 Sym-BD G AMER -BD 0 10 0 10 1 10 2 10 3 Time (seconds) ´ Alvaro Torralba PhD Defense June 2, 2015 47 / 54

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Symbolic execution as search, and the rise of solvers Search and SMT Symbolic execution is

Symbolic execution as search, and the rise of solvers Search and SMT Symbolic execution is

Symbolic execution as search, and the rise of solvers Search and SMT Symbolic execution is appealingly simple and useful , but computationally expensive ! We will see how the effective use of symbolic execution boils down to a kind of

494 views • 24 slides
Construction Heuristics Iterated Greedy GRASP Adaptive Iterated Construction Search Multilevel

Construction Heuristics Iterated Greedy GRASP Adaptive Iterated Construction Search Multilevel

Outline DM811 1. Complete Search Methods A best-first search HEURISTICS AND LOCAL SEARCH ALGORITHMS FOR COMBINATORIAL OPTIMZATION 2. Incomplete Search Methods 3. Other Tree Search Based Incomplete Methods Rollout/Pilot Method Lecture 5

521 views • 14 slides
Set 3: Informed Heuristic Search ICS 271 Fall 2014 Kalev Kask Overview Heuristics and

Set 3: Informed Heuristic Search ICS 271 Fall 2014 Kalev Kask Overview Heuristics and

Set 3: Informed Heuristic Search ICS 271 Fall 2014 Kalev Kask Overview Heuristics and Optimal search strategies heuristics hill-climbing algorithms Best-First search A*: optimal search using heuristics Properties of A*

1.08k views • 78 slides
5.1 Seminar [08.11] Domain Studies [15.11] Abstraction Heuristics I [22.11] Abstraction

5.1 Seminar [08.11] Domain Studies [15.11] Abstraction Heuristics I [22.11] Abstraction

Seminar: Search and Optimization October 4, 2012 5. Schedule and Topics Seminar: Search and Optimization 5. Schedule and Topics 5.1 Seminar Schedule and Topics Next steps Malte Helmert Universit at Basel 5.2 Software project October

314 views • 5 slides
Outline 1. (Meta-) Heuristics for Local Search Local Search (Meta-) Heuristics Heuristics for

Outline 1. (Meta-) Heuristics for Local Search Local Search (Meta-) Heuristics Heuristics for

Topic 17: Constraint-Based Local Search 1 (Version of 26th November 2020) Pierre Flener Optimisation Group Department of Information Technology Uppsala University Sweden Course 1DL441: Combinatorial Optimisation and Constraint Programming,

1.18k views • 114 slides
Online Saturated Cost Partitioning for Classical Planning Jendrik Seipp October 21, 2020

Online Saturated Cost Partitioning for Classical Planning Jendrik Seipp October 21, 2020

Online Saturated Cost Partitioning for Classical Planning Jendrik Seipp October 21, 2020 University of Basel 1/17 Setting optimal classical planning multiple abstraction heuristics cost partitioning 2/17 A search +

574 views • 38 slides
Constraint Programming - Heuristics in Search Variable and Value selection Static and

Constraint Programming - Heuristics in Search Variable and Value selection Static and

Constraint Programming - Heuristics in Search Variable and Value selection Static and Dynamic Heuristics Incomplete Search Strategies Symmetry Breaking (introduction) 10 November 2011 Constraint Programming 1 Heuristic Search -

746 views • 59 slides
Outline 1. Construction Heuristics DMP204 General Principles SCHEDULING, Metaheuristics A

Outline 1. Construction Heuristics DMP204 General Principles SCHEDULING, Metaheuristics A

Construction Heuristics Local Search Software Tools Outline 1. Construction Heuristics DMP204 General Principles SCHEDULING, Metaheuristics A search TIMETABLING AND ROUTING Rollout Beam Search Iterated Greedy GRASP Lecture 9 2.

701 views • 21 slides
Automatically Deriving Abstraction Heuristics PDB Abstractions Explicit-State Abstractions

Automatically Deriving Abstraction Heuristics PDB Abstractions Explicit-State Abstractions

Transition Systems and Abstractions Automatically Deriving Abstraction Heuristics PDB Abstractions Explicit-State Abstractions Conclusion Malte Helmert Albert-Ludwigs-Universit at Freiburg, Germany STAIR 2008 About This Talk

870 views • 43 slides
Foundations of Artificial Intelligence 14. State-Space Search: Analysis of Heuristics Malte

Foundations of Artificial Intelligence 14. State-Space Search: Analysis of Heuristics Malte

Foundations of Artificial Intelligence 14. State-Space Search: Analysis of Heuristics Malte Helmert and Thomas Keller University of Basel March 23, 2020 Properties of Heuristics Examples Connections Summary State-Space Search: Overview

480 views • 15 slides
Living on the Edge: Safe Search with Unsafe Heuristics Erez Karpas Carmel Domshlak Faculty of

Living on the Edge: Safe Search with Unsafe Heuristics Erez Karpas Carmel Domshlak Faculty of

Safeness of Heuristics Path Dependent Heuristics Unjustified Actions The Connection More than a Pruning Mechanism Experimental Evaluation Living on the Edge: Safe Search with Unsafe Heuristics Erez Karpas Carmel Domshlak Faculty of

706 views • 37 slides
Foundations of Artificial Intelligence 4. Informed Search Methods Heuristics, Local Search

Foundations of Artificial Intelligence 4. Informed Search Methods Heuristics, Local Search

Foundations of Artificial Intelligence 4. Informed Search Methods Heuristics, Local Search Methods, Genetic Algorithms Wolfram Burgard, Bernhard Nebel, and Martin Riedmiller Albert-Ludwigs-Universit at Freiburg Mai 17, 2011 Contents

346 views • 33 slides
Outline DM811 Fall 2009 Heuristics for Combinatorial Optimization 1. Complete Search Methods

Outline DM811 Fall 2009 Heuristics for Combinatorial Optimization 1. Complete Search Methods

Complete Search Incomplete Search General Search Methods A search Metaheuristics Heuristics for TSP Outline DM811 Fall 2009 Heuristics for Combinatorial Optimization 1. Complete Search Methods General Search Methods A search

432 views • 14 slides
1 Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level

1 Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level

Hyper-heuristics: Raising the Level of Generality of Search Hyper-heuristics: Raising the Level of Generality of Search Methodologies Methodologies Graham Kendall Contents The University of Nottingham What is a hyper-heuristic?

307 views • 14 slides
4 Heuristic Search 4.0 Introduction 4.3 Using Heuristics in Games 4.1 An Algorithm for

4 Heuristic Search 4.0 Introduction 4.3 Using Heuristics in Games 4.1 An Algorithm for

4 Heuristic Search 4.0 Introduction 4.3 Using Heuristics in Games 4.1 An Algorithm for Heuristic Search 4.4 Complexity Issues 4.2 Admissibility, 4.5 Epilogue and Monotonicity, and References Informedness 4.6 Exercises Additional

1.04k views • 78 slides
4 Heuristic Search 4.0 Introduction 4.3 Using Heuristics in Games 4.1 An Algorithm for

4 Heuristic Search 4.0 Introduction 4.3 Using Heuristics in Games 4.1 An Algorithm for

4 Heuristic Search 4.0 Introduction 4.3 Using Heuristics in Games 4.1 An Algorithm for Heuristic Search 4.4 Complexity Issues 4.2 Admissibility, 4.5 Epilogue and Monotonicity, and References Informedness 4.6 Exercises Additional

536 views • 53 slides
Introduction to Human Computer Interaction Course on NPTEL, Spring 2018 Week 11 GOMS, Others

Introduction to Human Computer Interaction Course on NPTEL, Spring 2018 Week 11 GOMS, Others

Introduction to Human Computer Interaction Course on NPTEL, Spring 2018 Week 11 GOMS, Others Ponnurangam Kumaraguru (PK) Associate Professor ACM Distinguished & TEDx Speaker Linkedin/in/ponguru/ 1 fb/ponnurangam.kumaraguru,

1.03k views • 19 slides
Adversarial Search Chapter 6 Section 1 4 Outline Optimal decisions in games Which

Adversarial Search Chapter 6 Section 1 4 Outline Optimal decisions in games Which

Adversarial Search Chapter 6 Section 1 4 Outline Optimal decisions in games Which strategy leads to success? Perfect play minimax decisions - pruning Resource limits and approximation evaluation

626 views • 51 slides
Heuristic Search Heuristic Search Best-First A * Heuristic Functions Some material

Heuristic Search Heuristic Search Best-First A * Heuristic Functions Some material

RN, Chapter 4.1 4.2 Heuristic Search Heuristic Search Best-First A * Heuristic Functions Some material from: D Lin, J You, JC Latombe 1 Search Overview Introduction to Search Blind Search Techniques Heuristic

807 views • 53 slides
AEAC U S & RHADAM AN THUS MC4BSM FNAL May 18-20, 2015 Joel W. Walker SHSU

AEAC U S & RHADAM AN THUS MC4BSM FNAL May 18-20, 2015 Joel W. Walker SHSU

AEAC U S & RHADAM AN THUS MC4BSM FNAL May 18-20, 2015 Joel W. Walker SHSU AEAC U S & RHADAM AN THUS MC4BSM FNAL May 18-20, 2015 Joel W. Walker SHSU Cutting with AEAC U S (Algorithmic Event Arbiter and

629 views • 34 slides
Addition about Prototypes http://www.useit.com/papers/guerrilla_hci.html 29/01/04 LMU

Addition about Prototypes http://www.useit.com/papers/guerrilla_hci.html 29/01/04 LMU

Vorlesung Mensch-Maschine-Interaktion Evaluation Ludwig-Maximilians-Universitt Mnchen LFE Medieninformatik Heinrich Humann & Albrecht Schmidt WS2003/2004 http://www.medien.informatik.uni-muenchen.de/ 29/01/04 LMU Mnchen

428 views • 12 slides
1 Optimality Summary What is Search For? Tree search: Models of the world: single agents,

1 Optimality Summary What is Search For? Tree search: Models of the world: single agents,

Announcements Introduction to Artificial Intelligence Please ask for help on assignment V22.0472-001 Fall 2009 Lecture 4: Constraint Lecture 4: Constraint Satisfaction Problems Rob Fergus Dept of Computer Science, Courant Institute,

333 views • 8 slides
Heuristics, Approximation Algorithms Lecture 24 Nov 18, 2016 Chandra & Ruta (UIUC) CS473

Heuristics, Approximation Algorithms Lecture 24 Nov 18, 2016 Chandra & Ruta (UIUC) CS473

CS 473: Algorithms, Fall 2016 Heuristics, Approximation Algorithms Lecture 24 Nov 18, 2016 Chandra & Ruta (UIUC) CS473 1 Fall 2016 1 / 34 Part I Heuristics Chandra & Ruta (UIUC) CS473 2 Fall 2016 2 / 34 Coping with

716 views • 57 slides
TCP ECN Experience with enabling ECN on the Internet Padma Bhooma Apple 1 ECN deployment

TCP ECN Experience with enabling ECN on the Internet Padma Bhooma Apple 1 ECN deployment

TCP ECN Experience with enabling ECN on the Internet Padma Bhooma Apple 1 ECN deployment Padma Bhooma MAPRG 98th IETF Chicago March 2017 Using ECN from client side Apple enabled negotiation of TCP ECN (RFC 3168) from the client-side

467 views • 20 slides

More recommend