Planning Methods In Artificial Intelligence
(Notes from the ASU Planning Seminar)
Online version of ASU CSE TR 96-004
Page accessed
times
since 2/17/96.
Abstract (of the hardcopy version of the Notes)
This technical report contains the lecture notes and some handouts
from the course on AI Planning Methods that was conducted at Arizona
State University in Fall 1995. For most of the topics and classes,
lecture notes were taken by designated students and were
revised/extended by me. These notes are also available via WWW at
http://rakaposhi.eas.asu.edu/planning-class.html. For the sake
of completeness, I am also including the hand-written transparencies
for the those topics that did not have designated note-takers.
It is hoped that these notes can serve as an opinionated
reference guide for advanced graduate students or other researchers
interested in AI Planning methods. I do not make any claims of
completeness of coverage. In many cases, I went for newer unifying
research directions and accounts, over more standard treatments (even
though the former have not yet been tested by time).
Here is a list of papers from the Yochan Planning Group.
Contents
- Dynamical Systems as abstractions for understanding issues in Planning (8/28)
- State Variable based representation of World and Actions in Classical Planning
- Refinement
Planning Overview (1/28/97) (Eric Parker
- Forward, Backward and MEA
planning in the space of states
- Plan Space Planning:
Basics
- Plan Space Planning:
Extensions -- using operator schemas, using least commitment to
objects, negated goals, conditional effects, quantification etc.
- Refinement planning as a unifying framework; combining
state-space and plan-space approaches using UCP. (See class notes).
- Slides from invited talk on Status and Prospectus of refinement planning given at AAAI-96
- Graph Plan Algorithm and SAT based approaches...
- Graph plan Summary (by Xiuping Yang; Revised by Rao)
- Reconstructing Graphplan from forward projection (Rao)
- Slides of a talk given at KR-96 which discusses the relations between traditional refinement planners, Graphplan and planning as satisfaction algorithms.
- Approximate Planning (see the class notes and discussion).
- Scheduling
-
Introduction to Scheduling, and discussion of iterative re-scheduling (by Jim Neudorf)
- Essentials of Constraint Satisfaction and dynamic backtracking (notes distributed in class)
- Efficiency and Complexity Issues.
-
Controlling non-backtrackable decision in UCPOP (Role of Constraint Propagation) (By Rao Kambhampati)
-
Viewing Planning as Constraint Satisfaction -- summary of Descartes system (Robert Guttman).
-
Classifying problem and domain complexity (Jay Noh)
- Learning to improve planning performance (see class notes).
- Explanation based learning of search control rules. (Notes from Naccache--to be added)
- Issues and Explanation based and
Inductive learning of search control rules (Notes by Billal;
Revised and Extended by Rao)
- Plan Reuse and Replay
-
Execution and and Handling Incomplete Information.
- Execution issues (Srivastava; Revised extensively by Rao)
- Sensing and handling of incomplete info (Holly Coast; revised by rao).
-
Planning with metric time and continuous change.
- Zeno summary (Brian Hartwig; Unrevised)
-
Planning with stochastic dynamics (Probabilistic Planning)
- Buridan Summary
- Markov Decision Processes Summary (Leanne Vander Meer Dunn; Revised)
-
Applications
- Software agents summary (Eric Lambrecht)
Here are the notes from Spring 1994 Planning Seminar. They are a bit dated, though..
Subbarao Kambhampati
Assistant Professor
rao@asu.edu