CS 730/730W/830: Intro AI

Size: px

Start display at page:

Download "CS 730/730W/830: Intro AI"

Bernadette Felicia Potter
5 years ago
Views:

1 CS 730/730W/830: Intro AI 1 handout: slides Wheeler Ruml (UNH) Lecture 18, CS / 15

2 Comparison Extensions Setting Break Wheeler Ruml (UNH) Lecture 18, CS / 15

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

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

5 Setting Comparison Extensions Setting Break STRIPS assumes static, deterministic world, discrete time, single discrete actions. 1. time, resources 2. concurrent actions 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 / 15

6 Break Comparison Extensions Setting Break asst 8 asst 9 full project proposals Wheeler Ruml (UNH) Lecture 18, CS / 15

7 Wheeler Ruml (UNH) Lecture 18, CS / 15

8 task decomposition actions = goals for lower level actions = restrictions for lower level actions = hueristic for lower level Wheeler Ruml (UNH) Lecture 18, CS / 15

9 Hierarchical Task Networks states and tasks actions: preconditions, effects, primitive vs high-level method: preconditions, tasks goal : complete decomposition into primitive actions downward refinement: high-level guaranteed to refine into legal primitives SHOP2 planner Wheeler Ruml (UNH) Lecture 18, CS / 15

10 HTN Example: Logistics actions: Drive, Load, Unload method: MovePackageByTruck(p,s,d, t) pre: At(p,s) post: At(p,d) subtasks: Drive(t, s), Load(p,t), Drive(t,d), At(p,d) Wheeler Ruml (UNH) Lecture 18, CS / 15

11 Hierarchical Goal Networks (IJCAI, 2013) operators as in STRIPS goal network: partially-ordered set of DNF formulas over literals method: preconditions and subgoals. postconditions are last subgoal. subgoal: conjunction of literals 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 / 15

12 HGN Example: Logistics actions: Drive, Load, Unload method: MovePackageByTruck(p,s,d, t) pre: At(p,s) subgoals: At(t, s), In(p,t), At(t,d), At(p,d) Wheeler Ruml (UNH) Lecture 18, CS / 15

13 Example: Dragon Age: Origins Wheeler Ruml (UNH) Lecture 18, CS / 15

14 Class Outline 1. search: heuristics, CSPs, games 2. knowledge representation: FOL, resolution 3. planning: STRIPS, MDPs 4. learning: supervised, unsupervised 5. uncertainty: particle filters, HMMs Wheeler Ruml (UNH) Lecture 18, CS / 15

15 EOLQs What question didn t you get to ask today? What s still confusing? What would you like to hear more about? Please write down your most pressing question about AI and put it in the box on your way out. Thanks! Wheeler Ruml (UNH) Lecture 18, CS / 15

Lecture 2 Agents & Environments (Chap. 2) Outline

Lecture 2 Agents & Environments (Chap. 2) Based on slides by UW CSE AI faculty, Dan Klein, Stuart Russell, Andrew Moore Outline Agents and environments Rationality PEAS specification Environment types