Date: Wed, 08 Jan 1997 20:22:06 GMT Server: NCSA/1.4.2 Content-type: text/html Last-modified: Sun, 15 Dec 1996 20:55:57 GMT Content-length: 4841 CSE 573 Project Page

CSE 573 PROJECT PAGE

Notes

Please focus on the following problems: PRODIGY-BW.PRODIGY-SUSSMAN, INIT-FLAT-TIRE.FIXIT, D1S1.D1S1-1, D1S1.D1S1-3, D1S1.D1S1-5

Please prioritize as follows:

  1. Most important: Getting some encoding working for the simplest subset of the action language (i.e. the files in
    /projects/ai/planners/ucpop/domains/clean/files/simple-ucpop/
    and getting some data showing how your system works on that one encoding
  2. Next: Testing one of your conjectures about the effect on speed of an encoding variation (i.e. kautz splitting vs overloaded splitting).
  3. Lower priority: Going back and handling a more expressive action language.
Your code should find any solution to the planning problem, not necessarily the shortest. I.e. optimize the speed of your planner, not the length of the plan.

Constants may be specified in the problem goal as well as in the initial state.

UCPOP results

0 3 FIXIT -1 f 0 0 0 0 76320 0 0 0 12 12 23701 14878
0 3 PRODIGY-SUSSMAN -1 T 0 0 0 0 850 0 0 0 6 6 515 272
0 3 D1S1-1 -1 T 0 0 0 0 10 0 0 0 1 1 3 2
0 3 D1S1-3 -1 T 0 0 0 0 10 0 0 0 3 3 11 8
0 3 D1S1-5 -1 T 0 0 0 0 10 0 0 0 5 5 19 14
The key to this gibberish is at the bottom of this page. Find more results in
 /projects/ai/573/project/results.

Team Pages

  1. Druken Masters (Linear)
  2. Medics (Linear)
  3. Wolverines (Parallel)
  4. Kosher Sushi (Parallel? Linear?)
  5. Lawrence of Arabia (Graphplan?)
  6. Causalists (Causal)
  7. pHan (Causal)

Input Language

Your system should handle the following subset of UCPOP/ADL domain descriptions: goals and preconditions with conjunction, disjunction, negation and existential quantification; conditional effects. in addition to standard conjunctive effects, but no universal quantifiaction and no need to reason about side effects vs primary effects. There is no need to handle universal quantification, side-effects/primary effects, procedural attachment (i.e. facts), domain axioms, or safety constraints.

CLick here for the input language BNF.

Look in

/projects/ai/planners/domains/ucpop/clean/files/{simple, partial}
for problems written in this format.

Output Specifications

The outputs of an experiment should be in two files, which should be understandable to a human, PERL, and LISP.

Filenames: <groupnumber>-<experimentnumber>.info
           <groupnumber>-<experimentnumber>.data

The info file has the following items, with some white space after the colons. Blank fields are OK.

Group: <groupnumber>
Experiment: <experimentnumber>
Encoder: <switches for your encoder>
Solve: <any switches you used for the solve program>
Decoder: <switches for your decoder>
Machine: <machine type>
Comment: <whatever>

The data file has one line per run, separated by whitespace. When you fix the maximum number of timesteps, encode, solve, and decode, that's one run. A planning problem may have numerous rows with different timesteps.

Please put a value (or zero) in every column. No parentheses. No quotation marks. The columns are:

Results

Results go in
 /projects/ai/573/project/results.