Docstoc

Principles of Planning in Artificial Intelligence

Document Sample
Principles of Planning in Artificial Intelligence Powered By Docstoc
					    Principles of Planning in Artificial Intelligence
      Lectures: Dr. Jussi Rintanen
                 Mondays 1pm-3pm, Wednesdays 1pm-2pm
                 (Oct 14 - Dec 18, Jan 6 - Feb 10)
     Exercises: Mr. Malte Helmert, Wednesdays 2pm-3pm
   Lecture hall: building 101, room SR-01-018
  Participation: lectures, exercises
         Exam: compulsory for ACS students (Feb 12 ?)
      Textbook: No! Lecture notes & articles available on the web page
    Web page: www.informatik.uni-freiburg.de/˜ki
                 /lehre/ws0203/aip/

Jussi Rintanen                                 October 14, AI Planning   1/19
                      What is Planning?


    In order to cope, an organism must either armor itself (like
    a tree or a clam) and “hope for the best,” or else develop
    methods for getting out of harm’s way and into the better
    neighborhoods of the vicinity. If you follow this later course,
    you are confronted with the primordial problem that every
    agent must continually solve: Now what do I do?

(Daniel C. Dennett, ”Consciousness Explained”)


Jussi Rintanen                                      October 14, AI Planning   2/19
• Choose your actions (next action) based on your prediction on
  what their consequences will be.

• Objective: achieve some goal, for example,
    – fix the broken kitchen window, or
    – keep your garden beautiful.

• Emphasis is on general-purpose problem representation and
  general-purpose solution techniques.
  (Alternative: tailored algorithms for every problem separately.)

Jussi Rintanen                                   October 14, AI Planning   3/19
                 Why is this very difficult?

Complexity of the world, requirements on plans:

• Not all facts concerning the world can be observed. (What is
  the current state of the world?)

• Unpredictability: nondeterministic events, other agents

• Restrictions on consumption of time and other resources.

• Plan optimality.

Jussi Rintanen                                 October 14, AI Planning   4/19
            Where AI planning is or could be used?


1. intelligent robots

2. general problem-solving (solve Rubik’s cube, play chess)

3. autonomous spacecraft (NASA Deep Space One)

4. transportation planning for large-scale military operations

5. make long-distance phone calls for free (∼ CAV)

Jussi Rintanen                                   October 14, AI Planning   5/19
            Jussi−in−FR      Jussi−in−ST      Jussi−in−BA
         suitcase−in−FR   suitcase−in−FR   suitcase−in−FR



            Jussi−in−FR      Jussi−in−ST      Jussi−in−BA
         suitcase−in−ST   suitcase−in−ST   suitcase−in−ST




            Jussi−in−FR      Jussi−in−ST      Jussi−in−BA
         suitcase−in−BA   suitcase−in−BA   suitcase−in−BA



Jussi Rintanen                              October 14, AI Planning   6/19
                     G




                 I

Jussi Rintanen       October 14, AI Planning   7/19
Question: If planning is just finding a path in a graph, do we
really have to have a whole course on planning?
Answer 1: The problem is far from trivial, because these graphs
are not represented explicitly as graphs (adjacency lists or
matrices), and they can be so big that they could not be stored
in any existing computer simply as adjacency lists or similar.
Answer 2: More general types of planning are not simply finding
a path in a graph.




Jussi Rintanen                                 October 14, AI Planning   8/19
(:action drive
  :parameters (?car - car ?from ?to - location)
  :precondition (and (at ?car ?from)
                     (road-between ?from ?to))
  :effect (and (at ?car ?to)
               (not (at ?car ?from))
               (forall (?x - obj)
                   (when (in ?x ?car)
                         (and (not (at ?x ?from))
                              (at ?x ?to))))))



Jussi Rintanen                     October 14, AI Planning   9/19
                     G




                 I

Jussi Rintanen       October 14, AI Planning   10/19
                     G




                 I

Jussi Rintanen       October 14, AI Planning   11/19
                 Have dinner




Jussi Rintanen                 October 14, AI Planning   12/19
                 Have dinner



                 Dine out




Jussi Rintanen                 October 14, AI Planning   13/19
                 Have dinner



                 Dine at home




Jussi Rintanen                  October 14, AI Planning   14/19
                                Have dinner



                                Dine at home


                 Prepare meal   Eat meal             Clean up




Jussi Rintanen                                 October 14, AI Planning   15/19
                                Have dinner



                                Dine at home


                 Prepare meal   Eat meal             Clean up


                  Order pizza


Jussi Rintanen                                 October 14, AI Planning   16/19
                                  Have dinner



                                  Dine at home


                   Prepare meal   Eat meal             Clean up


Get ingredients      Cook         Set up table


  Jussi Rintanen                                 October 14, AI Planning   17/19
                 Our initial plan for the next weeks...


1. Representations of transition systems

2. Planning with deterministic actions, 1 initial state
 (a) Search in the transition graph
 (b) Planning as reasoning in the classical propositional logic

3. planning with nondeterministic actions, hierarchical planning,
   ...


Jussi Rintanen                                     October 14, AI Planning   18/19
                     Questions?



           Next lecture on Wednesday at 1pm.

Jussi Rintanen                       October 14, AI Planning   19/19

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:27
posted:5/25/2010
language:English
pages:19