Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Lecture 20 of 42 Wednesday, 15 October 2008 William H. Hsu Department of Computing and Information Sciences, KSU KSOL course page: Course web site: Instructor home page: Reading for Next Class: Section 12.1 – 12.4, Russell & Norvig 2 nd edition HTN Planning and Robust Planning Discussion: CSP & Game Trees Review

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Constraint satisfaction problems (CSPs): Review Standard search problem:  state is a "black box“ – any data structure that supports successor function, heuristic function, and goal test CSP:  state is defined by variables X i with values from domain D i  goal test is a set of constraints specifying allowable combinations of values for subsets of variables Simple example of a formal representation language Allows useful general-purpose algorithms with more power than standard search algorithms © 2004 S. J. Russell From: Reused with permission.

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Constraint graph: Review Binary CSP: each constraint relates two variables Constraint graph: nodes are variables, arcs are constraints © 2004 S. J. Russell From: Reused with permission.

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Standard search formulation: Review Let's start with the straightforward approach, then fix it States are defined by the values assigned so far Initial state: the empty assignment { } Successor function: assign a value to an unassigned variable that does not conflict with current assignment  fail if no legal assignments Goal test: the current assignment is complete 1.This is the same for all CSPs 2.Every solution appears at depth n with n variables  use depth-first search 3.Path is irrelevant, so can also use complete-state formulation 4.b = (n - l )d at depth l, hence n! · d n leaves © 2004 S. J. Russell From: Reused with permission.

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Arc consistency algorithm AC-3: Review Time complexity: O(n 2 d 3 ) © 2004 S. J. Russell From: Reused with permission.

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Local search for CSPs Hill-climbing, simulated annealing typically work with "complete" states, i.e., all variables assigned To apply to CSPs:  allow states with unsatisfied constraints  operators reassign variable values Variable selection: randomly select any conflicted variable Value selection by min-conflicts heuristic:  choose value that violates the fewest constraints  i.e., hill-climb with h(n) = total number of violated constraints © 2004 S. J. Russell From: Reused with permission.

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Alpha-Beta (  -  ) Pruning: Modified Minimax Algorithm Adapted from slides by S. Russell UC Berkeley

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Expectiminimax [1]

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Expectiminimax [2]

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Expectiminimax [3]

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Expectiminimax [4]

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Lecture Outline Today’s Reading: Sections 11.4 – 11.7, 12.1 – 12.4, R&N 2e Today and Wednesday: Practical Planning  Conditional Planning  Replanning  Monitoring and Execution  Continual Planning Wednesday: Hierarchical Planning Revisited  Examples: Korf  Real-World Example Friday: Robust Planning, Uncertainty, Planning-Like Problems  Planning-like problems: design; scheduling; tutoring, critiquing  Why probability?  Planning and reaction  Planning under Uncertainty

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Planning and Learning Roadmap Bounded Indeterminacy (12.3) Four Techniques for Dealing with Nondeterministic Domains 1. Sensorless / Conformant Planning: “Be Prepared” (12.3)  Idea: be able to respond to any situation (universal planning)  Coercion 2. Conditional / Contingency Planning: “Plan B” (12.4)  Idea: be able to respond to many typical alternative situations  Actions for sensing (“reviewing the situation”) 3. Execution Monitoring / Replanning: “Show Must Go On” (12.5)  Idea: be able to resume momentarily failed plans  Plan revision 4. Continuous Planning: “Always in Motion, The Future Is” (12.6)  Lifetime planning (and learning!)  Formulate new goals

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Review: Clobbering and Promotion / Demotion Adapted from slides by S. Russell, UC Berkeley

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Review: POP Example – Sussman Anomaly Adapted from slides by S. Russell, UC Berkeley

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Hierarchical Abstraction Planning Adapted from Russell and Norvig Need for Abstraction  Question: What is wrong with uniform granularity?  Answers (among many)  Representational problems  Inferential problems: inefficient plan synthesis Family of Solutions: Abstract Planning  But what to abstract in “problem environment”, “representation”?  Objects, obstacles (quantification: later)  Assumptions (closed world)  Other entities  Operators  Situations  Hierarchical abstraction  See: Sections 12.2 – 12.3 R&N, pp. 371 – 380  Figure 12.1, 12.6 (examples), 12.2 (algorithm), (properties)

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Universal Quantifiers in Planning Quantification within Operators  p. 383 R&N  Examples  Shakey’s World  Blocks World  Grocery shopping  Others (from projects?) Exercise for Next Tuesday: Blocks World

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Practical Planning Adapted from Russell and Norvig The Real World  What can go wrong with classical planning?  What are possible solution approaches? Conditional Planning Monitoring and Replanning (Next Time)

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Review: How Things Go Wrong in Planning

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Review: Practical Planning Solutions Adapted from slides by S. Russell, UC Berkeley

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Conditional Planning Adapted from slides by S. Russell, UC Berkeley

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Monitoring and Replanning Adapted from slides by S. Russell, UC Berkeley

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Preconditions for Remaining Plan Adapted from slides by S. Russell, UC Berkeley

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Replanning Adapted from slides by S. Russell, UC Berkeley

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Summary Points Previously: Logical Representations and Theorem Proving  Propositional, predicate, and first-order logical languages  Proof procedures: forward and backward chaining, resolution refutation Today: Introduction to Classical Planning  Search vs. planning  STRIPS axioms  Operator representation  Components: preconditions, postconditions (ADD, DELETE lists) Friday: Robust Planning, Uncertainty, Planning-Like Problems  Planning-like problems: design; scheduling; tutoring, critiquing  Why probability?  Planning and reaction  Planning under Uncertainty

Computing & Information Sciences Kansas State University Wednesday, 15 Oct 2008CIS 530 / 730: Artificial Intelligence Terminology Classical Planning  Planning versus search  Problematic approaches to planning  Forward chaining  Situation calculus  Representation  Initial state  Goal state / test  Operators Efficient Representations  STRIPS axioms  Components: preconditions, postconditions (ADD, DELETE lists)  Clobbering / threatening  Reactive plans and policies  Markov decision processes