cs 730 730w 830 intro ai
play

CS 730/730W/830: Intro AI Beyond STRIPS Hierarchy Wheeler Ruml - PowerPoint PPT Presentation

CS 730/730W/830: Intro AI Beyond STRIPS Hierarchy Wheeler Ruml (UNH) Lecture 18, CS 730 1 / 15 Beyond STRIPS Comparison Extensions Setting Break Hierarchy Beyond STRIPS Wheeler Ruml (UNH) Lecture 18, CS 730 2 / 15


  1. CS 730/730W/830: Intro AI Beyond STRIPS Hierarchy Wheeler Ruml (UNH) Lecture 18, CS 730 – 1 / 15

  2. Beyond STRIPS ■ Comparison ■ Extensions ■ Setting ■ Break Hierarchy Beyond STRIPS Wheeler Ruml (UNH) Lecture 18, CS 730 – 2 / 15

  3. Comparison Forward: states Beyond STRIPS ■ Comparison + state known: strong heuristic, expressivity ■ ■ Extensions ■ Setting − irrelevant states ■ ■ Break Backward: sets of states Hierarchy + relevant states ■ − partial states: larger space, weaker heuristic, expressivity ■ Partial-order: plans + small space ■ + / − least commitment ■ − poor heuristics ■ Wheeler Ruml (UNH) Lecture 18, CS 730 – 3 / 15

  4. STRIPS Extensions negated goals: no problem with CWA Beyond STRIPS disjunctive precondition: for regression, just branch ■ Comparison ■ Extensions conditional effects: for regression, if we need the effect, plan ■ Setting ■ Break for the condition Hierarchy universal preconditions and effects: just ground goals and preconditions Wheeler Ruml (UNH) Lecture 18, CS 730 – 4 / 15

  5. Setting STRIPS assumes static, deterministic world, discrete time, single Beyond STRIPS discrete actions. ■ Comparison ■ Extensions ■ Setting 1. time, resources ■ Break 2. concurrent actions Hierarchy 3. abstraction: hierarchical planning 4. uncertainty: eg, disjunctive effects 5. temporally extended goals 6. execution monitoring, replanning 7. continuous state 8. multiple (self-interested) agents Wheeler Ruml (UNH) Lecture 18, CS 730 – 5 / 15

  6. Break asst 8 ■ Beyond STRIPS asst 9 Tue Apr 7 ■ Comparison ■ ■ Extensions wildcard vote Thu Apr 2 ■ ■ Setting ■ Break Hierarchy Wheeler Ruml (UNH) Lecture 18, CS 730 – 6 / 15

  7. Beyond STRIPS Hierarchy ■ Hierarchy ■ HTNs ■ HTN Example ■ HGNs ■ HGN Example ■ DAO ■ Class Outline Hierarchy ■ EOLQs Wheeler Ruml (UNH) Lecture 18, CS 730 – 7 / 15

  8. The Many Forms of Hierarchy task decomposition/refinement ■ Beyond STRIPS actions = goals for lower level ■ Hierarchy ■ Hierarchy actions = restrictions for lower level ■ ■ HTNs actions = hueristic for lower level ■ ■ HTN Example ■ HGNs ■ HGN Example ■ DAO ■ Class Outline ■ EOLQs Wheeler Ruml (UNH) Lecture 18, CS 730 – 8 / 15

  9. Hierarchical Task Networks states, tasks, methods, actions ■ Beyond STRIPS actions: preconditions, effects ■ Hierarchy ■ Hierarchy methods: preconditions, subtasks ■ ■ HTNs ‘goal’: complete decomposition into primitive actions ■ ■ HTN Example ■ HGNs ■ HGN Example downward refinement: high-level guaranteed to refine into legal ■ DAO primitives ■ Class Outline ■ EOLQs planning is semi-decidable, plan verification is NP-hard SHOP2 planner Wheeler Ruml (UNH) Lecture 18, CS 730 – 9 / 15

  10. HTN Example: Logistics actions: Drive, Load, Unload Beyond STRIPS method: Hierarchy ■ Hierarchy MovePackageByTruck(p,s,d, t) ■ HTNs ■ HTN Example pre: At(p,s) ■ HGNs post: At(p,d) ■ HGN Example ■ DAO subtasks: Drive(t, s), Load(p,t,s), Drive(t,d), Unload(p,t,d) ■ Class Outline ■ EOLQs Wheeler Ruml (UNH) Lecture 18, CS 730 – 10 / 15

  11. Hierarchical Goal Networks (IJCAI, 2013) operators as in STRIPS ■ Beyond STRIPS goal network: partially-ordered set of DNF formulas over ■ Hierarchy ■ Hierarchy literals ■ HTNs method: preconditions and subgoals. postconditions are last ■ ■ HTN Example ■ HGNs subgoal. ■ HGN Example subgoal: conjunction of literals ■ DAO ■ ■ Class Outline ■ EOLQs planner branches on: progressing state using applicable actions ■ ‘decomposing’ problem using applicable methods ■ applicable in state and relevant to goal methods are only for search guidance! Wheeler Ruml (UNH) Lecture 18, CS 730 – 11 / 15

  12. HGN Example: Logistics actions: Drive, Load, Unload Beyond STRIPS method: Hierarchy ■ Hierarchy MovePackageByTruck(p,s,d, t) ■ HTNs ■ HTN Example pre: At(p,s) ■ HGNs subgoals: At(t, s), In(p,t), At(t,d), At(p,d) ■ HGN Example ■ DAO ■ Class Outline ■ EOLQs Wheeler Ruml (UNH) Lecture 18, CS 730 – 12 / 15

  13. Example: Dragon Age: Origins Beyond STRIPS Hierarchy ■ Hierarchy ■ HTNs ■ HTN Example ■ HGNs ■ HGN Example ■ DAO ■ Class Outline ■ EOLQs Wheeler Ruml (UNH) Lecture 18, CS 730 – 13 / 15

  14. Class Outline 1. search: heuristics, CSPs, games Beyond STRIPS 2. knowledge representation: FOL, resolution Hierarchy ■ Hierarchy 3. planning: STRIPS, MDPs ■ HTNs 4. learning: supervised, unsupervised ■ HTN Example ■ HGNs 5. uncertainty: particle filters, HMMs ■ HGN Example ■ DAO ■ Class Outline ■ EOLQs Wheeler Ruml (UNH) Lecture 18, CS 730 – 14 / 15

  15. EOLQs What question didn’t you get to ask today? ■ Beyond STRIPS What’s still confusing? ■ Hierarchy ■ Hierarchy What would you like to hear more about? ■ ■ HTNs ■ HTN Example Please write down your most pressing question about AI and put ■ HGNs ■ HGN Example it in the box on your way out. ■ DAO Thanks! ■ Class Outline ■ EOLQs Wheeler Ruml (UNH) Lecture 18, CS 730 – 15 / 15

Recommend


More recommend