CSE 574 Planning and Learning Methods. 

Schedule 

  Lecture:  M,W, 3:15-4:30 (JWS 330A)

Instructor: Subbarao Kambhampati 

  Office:       GWC 374
  Telephone:    965-0113
  E-mail:       rao@asu.edu
  Office Hours: ;;to be determined. Will be on the class days..
  Home page:     http://rakaposhi.eas.asu.edu/planning-class.html
  Mailing list:  planning-class@asu.edu


One of the central problems in Artifcial Intelligence is that of
planning--deciding a sequence of actions that can take an agent from a
given initial state of the world to a desired goal state. This course
will provide an introduction to the theoretical and practical aspects
of AI planning and scheduling methods. It will cover various issues in
representing and reasoning about actions and plans, existing
architectures for synthesizing plans, and interactions between
planning, learning and execution. It will also emphasize techniques
for improving planning performance through a variety of
speedup-learning methods.


There is no required text book for this course. Readings will be drawn
from a set of tutorial papers as well as conference/journal papers
(see syllabus). Parts of the book "Artificial Intelligence: A modern
approach" would be relevant to this course, and the book is thus
recommended (not required). 

Online synopses of the class discussions will be put on the class
home-page throughout the semester. 

Prerequisites: CSE471 -- Introduction to AI (or consent of instructor)

Grading: Grades will be based on two exams, class participation, a
set of homeworks and a project. The relative weights are:

   1.Class participation (keeping up with readings, revising online notes) [~15% of final grade]

   3.Midterm, Final [~40% of final grade] 

   4.Computer Assignments + Course project [~45% of final grade]


Class participation involves attendance, active involvement in
discussions as well as being part of weekly note-taking teams. 


Approximate Syllabus (see homepage for a more detailed syllabus):

Course Syllabus

   1.INTRODUCTION/MOTIVATION
              Dynamical systems
              Planning vs. Scheduling 
              Action and Plan Representation 
              overview of approaches

   2. Classical PLANNING APPROACHES 

              Conjunctive Refinement Planning
                    State-space planning 
                    Causal-Link Planning, 
                    Task reduction planning 

               Disjunctive refinement planning
		    Graphplan, SATPLAN, Blackbox

               Transformational planning

               Unified Frameworks
                    Planning as Refinement Search

   3. CONSTRAINT SATISFACTION & SCHEDULING
	      CSP problems
	      Enforcing local consistency
	      Dependency directed backtracking
	      Explanation-based Learning

     4.CONTROLLING SEARCH IN CLASSICAL PLANNERS
              Static Search Control, Loop pruning, Abstraction
              Operator graphs, Greedy regression graphs, relevance analysis
	      Domain Analysis;  Constraint propagation

   5.LEARNING TO IMPROVE PLANNING PERFORMANCE 
              Learning Control Rules 
              Case-based planning:    Reuse/Replay/Retrieval

   6.INTERACTING WITH THE ENVIRONMENT 
              Sensing and Incomplete Information, 
              Conditional Planning 
              Interleaving planning & Execution 

   7. EXTENDING THE REPRESENTATION
	 Handling time
	 Handling metric 

   8. Stochastic and Decision Theoretic Planning
	            Probabilistic planning
                    Sequential decision problems and Markov decision processes
                    Reinforcement Learning 

   9.PLANNING APPLICATIONS 

              Planning in the Physical World, 
                    Robotics and Path Planning
                    Assembly Planning 
                    Process Planning 

              Planning for Software Agents
                    Softbots 
                    Information Gathering