1. Coping with Time & Continuous Quantities David E. Smith desmith@arc.nasa.gov Collaborators: Jeremy Frank, Ari Jónsson

2. The Ambitious Spacecraft Observation-1 priority time window target instruments duration Observation-2 Observation-3 Observation-4 … Objective: maximize science return

3. Complications Observation-1 priority time window target instruments duration Observation-2 Observation-3 Observation-4 … calibration target1 target2 … consumables: fuel power data storage cryogen angle between targets  turn duration Objective: maximize science return linked

4. The Spacecraft’s Choices Observations Instrument Calibration target Ordering Communication Instrument Status

5. The Spacecraft’s Choices Observations Instrument Calibration target Ordering Communication Instrument status continuous time time constraints concurrent actions metric resources optimization

6. The Choice Spectrum planning job-shop scheduling

7. The Choice Spectrum planning job-shop scheduling Job1 task1 < task2 < task3 < … Job2 Job3 … R3R7R1 Ordering choices only

8. The Choice Spectrum cascading levels of choice planning job-shop scheduling … ………… ……

9. The Choice Spectrum job-shop scheduling planning resource choices (RCSP) umfagoggin clavitracle fernambulator 5 11 Task1 Task2 Task3 Task4 Task5 Task6 Task8 Task7 [8,17] Ordering choices Resource choices

10. The Choice Spectrum job-shop scheduling planning resource choices (RCSP) alternative processes process3process7 process8 Task1 Task2 Task3 Task4 Task5 Task6 Task8 Task7 Ordering choices Resource choices Process choices

11. The Choice Spectrum planning ambitious spacecraft job-shop scheduling resource choices (RCSP) alternative processes Observation choices Instrument choices Calibration target choices Ordering choices Communication choices Instrument status choices

12. The Choice Spectrum planning Subset Selection ambitious spacecraft observation scheduling process planning job-shop scheduling resource choices (RCSP) alternative processes

13. Strips Assumptions Instantaneous actions No time constraints No concurrent actions No continuous quantities Bleak?

14. Some STRIPS Operators TakeImage (?target, ?instr): Pre: Status(?instr, Calibrated), Pointing(?target) Eff: Image(?target) Calibrate (?instrument): Pre: Status(?instr, On), Calibration-Target(?target), Pointing(?target) Eff: ¬Status(?inst, On), Status(?instr, Calibrated) Turn (?target): Pre: Pointing(?direction), ?direction ≠ ?target Eff: ¬Pointing(?direction), Pointing(?target) no time, no resources

15. Outline Resources Reusable Consumable Time Uncertainty?

16. Resources DiscreteContinuous single/multiple (metric) Reusable instruments bandwidth Consumable solid rockets fuel

17. Reusable Resources (single-capacity) TakeImage (?target, ?instr): Res: ?instr Pre:Status(?instr, Calibrated), Pointing(?target) Eff:Image(?target) Prevent concurrent actions Mutual exclusion between actions: Graphplan trivial Satplan ¬A 1  ¬A 2 POCLA 1 <A 2  A 2 <A 1

18. Reusable Resources (multiple-capacity) TakeImage (?target, ?instr): Res: ?instr Pre:Status(?instr, Calibrated), Pointing(?target) Eff:Image(?target) Suppose: 2 cameras Simple approach: name the cameras Mutual exclusion between actions: Graphplan trivial Satplan ¬A 1  ¬A 2 POCL ?instr 1 ≠?instr 2  A 1 <A 2  A 2 <A 1

19. Reusable Resources (multiple-capacity) TakeImage (?target): Res: Camera(1) Pre:Pointing(?target) Eff:Image(?target) 2 Cameras: Not named Explicit checking of concurrent actions to ensure:  usage ≤ capacity Capacity(Camera)=2

20. Reusable Resources (multiple-capacity)  usage > capacity Mutual exclusion between sets of actions: Graphplan n-ary mutex Satplan ¬(A 1  A 2  A 3  …) POCLA 1 <A 2  A 2 <A 1  A 1 <A 3  A 3 <A 1  … Graph Techniques IxTeT Profiling heuristics Sadeh, Fox, Beck O-Plan

21. Summary Resources Reusable Single-capacity  binary mutexeasy Multiple-capacity  n-ary mutextricky specialized algorithms search heuristics Consumable Time Uncertainty?

22. Consumable Resources Turn (?target): Pre:Pointing(?direction), ?direction ≠ ?target Fuel ≥ Angle(?direction,?target) *ConsumptionRate Eff:¬Pointing(?direction), Pointing(?target) Fuel –= Angle(?direction,?target) *ConsumptionRate Simple approach: wait until all variables are instantiated

23. LP Approaches Zeno: open conditions: P 1, P 2, P 3, P 4, P 5, E 1, E 2, E 3 constraints: Fuel ≤ 10 Simplex

24. LP Approaches: LPSAT Turn (?target): Pre:Pointing(?direction), ?direction ≠ ?target Fuel ≥ Angle(?direction,?target) *ConsumptionRate Eff:¬Pointing(?direction), Pointing(?target) Fuel –= Angle(?direction,?target) *ConsumptionRate ¬Turn-T3-T7 1 v {Pointing-T3 1  C1 1  ¬Pointing-T3 2  Pointing-T7 2  C2 2 } C1 1  Fuel 1 ≥ Angle(T3,T7) *ConsumptionRate C2 2  Fuel 2 = Fuel 1 – Angle(T3,T7) *ConsumptionRate Simplex SAT engine

25. ILP planning ¬Turn-T3-T7 1 v {Pointing-T3 1  C1 1  ¬Pointing-T3 2  Pointing-T7 2  C2 2 } C1 1  Fuel 1 ≥ Angle(T3,T7) *ConsumptionRate C2 2  Fuel 2 = Fuel 1 – Angle(T3,T7) *ConsumptionRate ¬Turn-T3-T7 1 v Pointing-T3 1 ¬Turn-T3-T7 1 v C1 1 ¬Turn-T3-T7 1 v ¬ Pointing-T3 2 ¬Turn-T3-T7 1 v Pointing-T7 2 ¬Turn-T3-T7 1 v C2 2 C1 1  Fuel 1 ≥ Angle(T3,T7) *ConsumptionRate C2 2  Fuel 2 = Fuel 1 – Angle(T3,T7) *ConsumptionRate Kautz & Walser Vossen et al.

26. ILP planning (1 - Turn-T3-T7 1 ) + Pointing-T3 1 ≥ 1 M(1 - Turn-T3-T7 1 ) + (1 + D1 1 ) ≥ 1 (1 - Turn-T3-T7 1 ) + (1 - Pointing-T3 2 ) ≥ 1 (1 - Turn-T3-T7 1 ) + Pointing-T7 2 ≥ 1 M(1 - Turn-T3-T7 1 ) + (1 ± D2 1 ) ≥ 1 D1 1 = Fuel 1 – Angle(T3,T7) *ConsumptionRate D2 2 = Fuel 2 – (Fuel 1 – Angle(T3,T7) *ConsumptionRate) ¬Turn-T3-T7 1 v Pointing-T3 1 ¬Turn-T3-T7 1 v C1 1 ¬Turn-T3-T7 1 v ¬ Pointing-T3 2 ¬Turn-T3-T7 1 v Pointing-T7 2 ¬Turn-T3-T7 1 v C2 2 C1 1  Fuel 1 ≥ Angle(T3,T7) *ConsumptionRate C2 2  Fuel 2 = Fuel 1 – Angle(T3,T7) *ConsumptionRate

27. Summary Resources Reusable Consumable Equality and Inequality conditions POCL + LPslow SAT + LPfast ILP medium Graphplan? Concurrent continuous changeugly! Time Uncertainty?

28. Concurrent Continuous Change Turn (?target): Pre:Pointing(?direction), ?direction ≠ ?target Fuel ≥ Angle(?direction,?target) *ConsumptionRate Eff:¬Pointing(?direction), Pointing(?target) Fuel –= Angle(?direction,?target) *ConsumptionRate Fuel = F(t) Fuel Time Fuel Time Fuel Time

29. Outline Resources Time Uncertainty?

30. World Description State-centric: for each time describe propositions that are true Proposition-centric: for each proposion describe times it is true Pointing(A7) Status(Cam2, Calibrated) Turn(A7) Pointing(Earth) Status(Cam2, Calibrated) ¬ Image(A7) Turn(A7) Pointing(A7) Status(Cam2, Calibrated) ¬ Image(A7) Pointing(Earth)

31. Allen’s Ordering Relationships A B A before B A B A meets B A B A overlaps B A contains B A B A = B A B A B A starts B A B A ends B

32. TakeImage Schema TakeImage(?target, ?instr) A  P {Status(?instr, Calibrated) P  Contains( P, A )}   Q {Pointing(?target) Q  Contains( Q, A )}   R {Image(?target) R  Meets( A, R )} TakeImage (?target, ?instr): Pre: Status(?instr, Calibrated), Pointing(?target) Eff: Image(?target)

33. Shorthand Pointing(?target) Status(?instr, Calibrated) TakeImage(?target, ?instr) Image(?target) meets contains TakeImage (?target, ?instr) contained-byStatus(?instr, Calibrated) contained-byPointing(?target) meetsImage(?target)

34. Turn Turn (?target) met-byPointing(?direction) meetsPointing(?target) Pointing(?target) Pointing(?direction) Turn(?target) meets

35. Calibrate Status(?instr, Calibrated) Pointing(?target) CalibrationTarget(?target) Calibrate(?instr) meets contains Status(?instr, On) Calibrate (?instr) met-byStatus(?instr, On) contained-byCalibrationTarget(?target) contained-byPointing(?target) meetsStatus(?instr, Calibrated)

36. Algorithm Choose: introduce an action & instantiate constraints coalesce propositions Propagate constraints

37. Initial Plan PastImage(?target) meets Pointing(Earth) Status(Cam1, Off) Status(Cam2, On) CalibrationTarget(T17) Future meets -∞ ∞

38. Expansion 1 Image(A7)Future meets Past meets Pointing(Earth) Status(Cam1, Off) Status(Cam2, On) CalibrationTarget(T17) -∞ ∞ Pointing(A7) Status(?instr, Calibrated) TakeImage(A7, ?instr) meets contains before

39. Expansion 2 Pointing(Earth) Status(Cam1, Off) Status(Cam2, On) CalibrationTarget(T17) Image(A7) Pointing(A7) Status(?instr, Calibrated) TakeImage(A7, ?instr) meets contains Pointing(?direction) Turn(A7) Pointing(?caltarget) CalibrationTarget(?caltarget) Calibrate(?instr) meets contains Status(?instr, On) before

40. Coalescing Pointing(Earth) Status(Cam1, Off) Status(Cam2, On) CalibrationTarget(T17) Image(A7) Pointing(A7) Status(Cam2, Calibrated) TakeImage(A7, Cam2) meets contains Pointing(?direction) Turn(A7) Pointing(T17) Calibrate(Cam2) meets contains before

41. Coalescing Pointing(Earth) Status(Cam1, Off) Status(Cam2, On) CalibrationTarget(T17) Image(A7) Pointing(A7) Status(Cam2, Calibrated) TakeImage(A7, Cam2) meets contains Turn(A7) Pointing(T17) Calibrate(Cam2) meets contains before

42. Expansion 3 Pointing(Earth) Status(Cam1, Off) Status(Cam2, On) CalibrationTarget(T17) Image(A7) Pointing(A7) Status(Cam2, Calibrated) TakeImage(A7, Cam2) meets contains Turn(A7) Pointing(T17) Calibrate(Cam2) meets contains Turn(T17) meets Pointing(?direction)

43. Coalescing Pointing(Earth) Status(Cam1, Off) Status(Cam2, On) CalibrationTarget(T17) Image(A7) Pointing(A7) Status(Cam2, Calibrated) TakeImage(A7, Cam2) meets contains Turn(A7) Pointing(T17) Calibrate(Cam2) meets contains Turn(T17) meets

44. POCL vs CBI POCL CBI search goal-directed goal-directed strategy temporal before/aftercomplex reasoning protection causal links meets conflicts threats mutex conflict proactive lazy resolution

45. Causal Links & Threats proposition action meets action proposition action proposition action threatens proposition action proposition mutex POCLCBI Causal links: Threats :

46. CBI Planners Zeno (Penberthy) intervals, no CSP Trains (Allen) Descartes (Joslin) extreme least commitment IxTeT (Ghallab) functional rep. HSTS (Muscettola) functional rep., activities

47. Timelines Pointing=Turn (?target) met-byPointing=?direction meetsPointing=?target EarthA7Turn(A7) T17 ?direction ? Pointing: Turn (?target) met-byPointing(?direction) meetsPointing(?target)

48. Explanatory Frame Axioms Pointing(?target) met-byPast or met-byTurn (?target) Turn (?target) met-byPointing(?direction) meetsPointing(?target) Pointing(?target) Pointing(?direction) Turn(?target) meets

49. A Stupid Question Why should CBI perform any better than POCL? less commitment better propagation timelines

50. A Stupid Question Why should CBI perform any better than POCL? less commitment better propagation timelines reachability analysis heuristic estimators

51. Reachability Analysis Graphplan: optimistic projection of achievability … … … level 0level 1level 2level 3 level 0level 1level 2

52. Reachability Analysis Pointing(Earth) Pointing(A7) Calibrate(Cam1,T5) Status(Cam1, Calibrated) meets contains Turn(A7) meets [0, ∞] [5, ∞] meets [5, ∞] Pointing(T5) Turn(T5) [3, ∞] meets [3, ∞] meets [0, ∞] [3, ∞] [8, ∞] [0, 13] mutex rules more complex (Smith & Weld - IJCAI99)

53. Benefits? 1. Pruning during plan construction 2. Basis for heuristic distance measures McDermott (UNPOP) Geffner (HSP-R) RAO

54. Extracting Heuristic Estimates Pointing(Earth) Pointing(A7) Calibrate(Cam1,T5) Status(Cam1, Calibrated) meets contains Turn(A7) meets [0, ∞] [5, ∞] meets [5, ∞] Pointing(T5) Turn(T5) [3, ∞] meets [3, ∞] meets [0, ∞] [3, ∞] [8, ∞] [0, 13] Distance metrics time required fuel required …

55. Summary Multi-capacity MetricOptimizationContinuousSpeed ResourcesQuantities Time Graphplan hard? hard hard? hardgood IPP TGP SAT moderate?moderatemoderate?very hard?good planning LPSAT ILP easy? easy easyvery hard? fair? planning ILP-plan ILP-plan CBImoderate moderatemoderate? easy fair? planning IxTeT Zeno HSTS

56. Critical Issues Multi-capacity MetricOptimizationContinuousSpeed ResourcesQuantities Time Graphplan hard? hard hard? hardgood IPP TGP SAT moderate?moderatemoderate?very hard?good planning LPSAT ILP easy? easy easyvery hard? fair? planning ILP-plan ILP-plan CBImoderate moderatemoderate? easy fair? planning IxTeT Zeno HSTS