1 / 26

Temporal Planning: Part 2

This article explores the use of a model-based execution approach with a planner for temporal planning. It covers task decomposition, execution histories, goals, projective task expansion, scheduler, and temporal planner.

medinac
Télécharger la présentation

Temporal Planning: Part 2

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Temporal Planning: Part 2 Brian C. Williams 16.412J/6.834J Oct 3rdst, 2001 1

  2. Model-based Execution (w planner) Task-Decomposition Execution Histories(?) Goals Projective Task Expansion Scheduler Temporal Planner Flexible Sequence (Plans) Task Dispatch Plan Runner Goals Modes Reactive Task Expansion Commands Observations

  3. Outline • Representing Plans • Representing Time • Plan Consistency and Scheduling

  4. Cassini Saturn Orbital Insertion courtesy JPL

  5. Planning in a continuous world Attitude Control System

  6. Planning in a continuous world • Turn (a, b) Attitude Control System

  7. Planning in a continuous world • Turn (a, b) Attitude Control System

  8. Planning in a continuous world • Turn (a, b) Attitude Control System

  9. Planning in a continuous world • Turn (a, b) Attitude Control System

  10. Planning in a continuous world • Turn (a, b) Attitude Control System

  11. Planning in a continuous world • Turn (a, b) Attitude Control System

  12. Planning in a continuous world: • Point (b) Attitude Control System

  13. Planning in a continuous world: • Point (b) Attitude Control System

  14. Action/State dichotomy in a continuous world • Turn is an action, Point is a state • Both Turn and Point are actions • Both Turn and Point are state Point (b) Turn (a, b) Attitude Control System

  15. Tokens represent procedure invocations Delta_V(direction=b, magnitude=200) Point(a) Point(b) Turn(b,a) Turn(a,b) Off Both Actions and States are Tokens Thrust Goals Power Attitude Thrust (b, 200) Engine Off Warm Up

  16. Delta_V(?x, ?y) Off f Planning Constraints are Specified by Compatibilities Thrust Goals Power equals contained_by Point(?x) Attitude contained_by meets met_by Thrust (?x, ?y) Engine Warm Up

  17. Delta_V(direction=b, magnitude=200) Example of Compatibility Instantiation Thrust Goals Power contains Attitude Thrust (b, 200) Engine

  18. Delta_V(direction=b, magnitude=200) Off Example of Compatibility Instantiation Thrust Goals Power equals contained_by Point(b) Attitude contained_by meets met_by Thrust (b, 200) Engine Warm Up

  19. Planning Experts Search Control Search engine Model (DDL) Planner/Scheduler Architecture Engine Domain Knowledge Goals Plan Plan Database Initial state to EXEC from EXEC

  20. Plan has flaws Plan is consistent Planner Resolves Flaws PLAN NO Uninstantiated compatibility . . . Instantiate compatibility . . . Backtrack Schedule token NO YES

  21. Types of Flaws • No Disjunct of Compatibility Selected. • Unsatisfied Subgoal • There is a hole in the timeline immediately before Thrust (b, 200) • Floating Token, Needs to be Placed on Timeline. • Uninstantiated Variable in a Token • Thrust (?x = {a, b, c}, ?y = [190, 300] The Plan Database can be modified only by primitives that resolve each of these flaws.

  22. Planning Experts Search engine Model (DDL) Planner/Scheduler Architecture Engine Domain Knowledge Search Control Goals Plan Plan Database Initial state to EXEC from EXEC

  23. Search Control Determines Order and Type of Flaw Resolution PLAN T f2I(P): Plan has no subgoal flaws Unsatisfied subgoal Search Control . . . Instantiate subgoal . . . . . . Backtrack Schedule token f1I(P): Plan is consistent F {T, ?}

  24. Ready( ) Search Control Language met_by Turning_on( ) Camera: (:subgoal (:master-match (Camera = Ready)) (:slave-match (Camera = Turning_on))  (:priority 50) (:method-priority ((:method :add) (:sort :asap)) ((:method :connect)) ((:method :defer))))

  25. DS1 Planner/Scheduler • DS1 PS is a constraint-based, backtrack search, generative planner operating on a fully temporal domain model • Model size (Remote Agent Experiment) • state variables 18 • procedure types 42 • Plan size • tokens 154 • variables 288 (81 time points) • constraints 232 (114 distance bounds) • Performance • search nodes 649 • search efficiency 64 %

  26. Model-based Execution (w planner) Task-Decomposition Execution Histories(?) Goals Projective Task Expansion Scheduler Temporal Planner Flexible Sequence (Plans) Task Dispatch Plan Runner Goals Modes Reactive Task Expansion Commands Observations

More Related