1 / 21

Exploiting Domain Knowledge with a Concurrent Hierarchical Planner

Exploiting Domain Knowledge with a Concurrent Hierarchical Planner. Brad Clement and Ed Durfee University of Michigan Artificial Intelligence Laboratory. Exploit Domain Knowledge by Structuring Hierarchical Plans. Domain expert specifies how goals/subgoals can be achieved by providing

melora
Télécharger la présentation

Exploiting Domain Knowledge with a Concurrent Hierarchical Planner

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. Exploiting Domain Knowledge with a Concurrent Hierarchical Planner Brad Clement and Ed Durfee University of Michigan Artificial Intelligence Laboratory

  2. Exploit Domain Knowledge by Structuring Hierarchical Plans • Domain expert specifies how goals/subgoals can be achieved by providing • choices of methods that can achieve higher level goal/task under different contexts, • task networks that describe a partial ordering of a group of tasks representing higher level method, and • primitive actions that are directly executable • Robust execution systems (PRS, RAPS, JAM, etc.) • HTN Planners (NOAH, NONLIN, etc.)

  3. HTN Planners A C B B A C • Top-down • Resolve conflicts by • choosing decomposition methods • adding ordering constraints • binding variables • Efficiency depends on these decisions and • choosing tasks to expand • selecting conflicts to resolve • Solution is a consistent primitive plan on A B on B C on A B on B C

  4. Reason at Abstract Levels to Make Better Decisions • Until recently, decisions were blind to conditions underneath.FAF (“fewest alternatives first”) is a most-constrained-variable heuristic. • expand the task with the least number of decomposition methods • bind the variable with the least number of values • ExCon (external condition) heuristic reasons about conditions at abstract level. (Tsuneto, Hendler, & Nau 1998) • expand tasks that can achieve or threaten an external condition • Summarize conditions of potential refinements at abstract levels to reason about task interactions. • resolve all threats at abstract level • prune inconsistent choices at abstract levels • make refinement choices based on threats among abstract plans

  5. Concurrent Hierarchical Planning evacuate evacuate evacuate evacuate noswitch oneswitch twoswitches noswitch oneswitch twoswitches cw ccw go tofarthest switch & goto farthest go to safe loc move move move move move move move move

  6. Concurrent Hierarchical Planning evacuate evacuate evacuate evacuate noswitch oneswitch twoswitches noswitch oneswitch twoswitches A C B cw ccw B A C on A B go tofarthest switch & goto farthest go to safe loc on B C on A B on B C move move move move move move move move

  7. Planning at Abstract Levels • Resolve conflicts at high level to minimize search time • Better solutions may exist at lower levels crisper solutions lowerplanningcost planning levels

  8. Concurrent Hierarchical Plans (CHiPs) • pre, in, & postconditions - sets of literals for a set of propositions • type - and, or, primitive • subplans - execute all for and, one for or; empty for primitive • order - conjunction of point or interval relations B - before DA A B B DB B B B B B B

  9. Summary Information pre: at(A,1,3)in: at(A,1,3), at(B,1,3), at(B,0,3), at(A,0,3), at(B,0,4), at(A,1,3) post: at(A,1,3), at(B,1,3),at(B,0,3), at(A,0,4), at(A,0,3),at(B,0,3), at(B,0,4) 1,3->0,4HI • external preconditions • external postconditions • internal conditions • must, may • always, sometimes 1,3->0,3 0,3->0,4 pre: at(A,1,3)in: at(A,1,3), at(B,1,3), at(B,0,3)post: at(A,0,3), at(A,1,3), at(B,1,3), at(B,0,3) pre: at(A,0,3)in: at(A,0,3), at(B,0,3), at(B,0,4)post: at(A,0,4), at(A,0,3), at(B,0,3), at(B,0,4) pre: at(A,1,3)in: at(A,1,3), at(B,1,3), at(B,0,3), at(B,1,4), at(A,0,3), at(A,1,4), at(B,0,4), at(A,1,3)post: at(A,1,3), at(B,1,3),at(B,0,3), at(A,0,4), at(A,0,3), at(A,1,4), at(B,0,3),at(B,1,4), at(B,0,4) 0 1 2 3 4 1,3->0,4 DA 0 A 1,3->0,4HI 1,3->0,4LO 1 DB 2 B

  10. e' e' e e' Asserting, Clobbering, Achieving, Undoing e l l e • easserts in/postcondition l at t • eachieves precondition l of e' • eclobbers pre/in/postcondition l of e' • eundoes postcondition l of e' • must - for all executions in all histories • may - for some execution in some history l l e l l e e e' l l e e' l l l l e' l l

  11. Deriving Summary Information • Recursive procedure bottoming out at primitives • Derived from those of immediate subplans • O(n2c2) for n non-primitive plans in hierarchy and c conditions in each set of pre, in, and postconditions • Proven procedures for determining must/may - achieve/undo/clobber • Properties of summary conditions are proven based on procedure

  12. Determining Temporal Relations • CanAnyWay({relation}, {psum, qsum}) - relationcan hold for any wayp and q can be executed • MightSomeWay({relation}, {psum, qsum}) - relationmight hold for some wayp and q can be executed B - before O - overlaps CanAnyWay({before}, {psum, qsum}) ØCanAnyWay({overlaps}, {psum, qsum}) MightSomeWay({overlaps}, {psum, qsum})

  13. Concurrent Hierarchical Planner • Derive summary information for hierarchy expanded to primitive level (iteratively expand for infinite recursion of methods) • Expand hierarchy from the top down • After each expansion, try to resolve threats • add ordering constraints • bind variables • check CAW and MSW

  14. Searching for CHiPs • search state • set of expanded plans • set of temporal constraints • set of blocked subplans • set of variable bindings • search operators • expand, block, constrain, bind blocked temporal constraints

  15. DA A DB B Reasoning at Abstract Levels Can Improve Performance Total Cost top-level best mid-level best primitive-level best Computation CostExecution Cost

  16. level #plans #conds /plan #operations toderive summ. info. #test operations /solution candidate solutionspace 0 1 O(bdc') O(1) 1 O(b2(bd-1c')2) = O(b2dc'2) ... 1 b O(bd-1c') O(b2(b(d-1)c')2) = O(b2dc'2) O(b!) O(bb2(bd-2c')2) = O(b2d-1c'2) 1 2 b ...... 2 b2 O(bd-2c') O(b4(b(d-2)c')2) = O(b2dc'2) O(b2!) O(b2b2(bd-3c')2) = O(b2d-2c'2) ................................. d-2 .......... bd-2 O(b2c') O(b2(d-2)(b2c')2) = O(b2dc'2) O(bd-2!) O(bd-2b2(bc')2)= O(bd+2c'2) ............ d-1 bd-1 3c'+b3c'= O(bc') O(b2(d-1)(bc')2)= O(b2dc'2) O(bd-1!) O(bd-1b2c'2) = O(bd+1c'2) ... ... ....... d bd 3c' O(b2dc'2) O(bd!) O(1) 1 2 b 1 2 b O(b2d-ic'2) i bi O(bd-ic') O(b2dc'2) O(bi!) • Summary information for each plan grows exponentially up the hierarchy • Number of plan steps per level grows exponentially down the hierarchy • Complexity of testing an ordering of plans is constant throughout hierarchy • Number of orderings of plans grows exponentially down hierarchy • Resolving threats is NP-complete (reduced from Hamiltonian Path)

  17. Search Heuristics • Prune inconsistent global plans (MightSomeWay) • “Expand most threats first” (EMTF) • expand subplan involved in most threats • focuses search on driving down to source of conflict • “Fewest threats first” (FTF) • search plan states with fewest threats first • or subplans involved in most threats are blocked first • Branch & bound - abstract solutions help prune space where cost is higher

  18. NEO Domain Experiments • Compare different strategies of ordering search states and ordering expansions • FAF-FAF • DFS-ExCon • FTF-EMTF • FTF-ExCon • 4 - 12 locations • 2 - 4 transports • no, partial, & complete overlap in locations visited

  19. FTF-EMTF found solutions for 23 problems, 14 optimal • FTF-ExCon found solutions for 19 problems, 12 optimal • FAF-FAF found solutions for 22 problems, 14 optimal • DFS-ExCon found solutions for 6 problems, 3 optimal (not shown)

  20. Future Work • What properties of plan hierarchies benefit which heuristics? • For different domains, how can the hierarchies be restructured to take advantage of different heuristics? • How beneficial is it to prune inconsistent and more costly choices at abstract levels?

  21. Contributions • Complexity analysis showing that resolving conflicts at higher levels is much easier than at lower levels • FTF and EMTF heuristics that take advantage of summary information • Sound and complete concurrent hierarchical planner • Preliminary experiments showing that these techniques can greatly improve the search for optimal plans

More Related