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:
- 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
- Next: Testing one of your conjectures about the effect on speed of an
encoding
variation (i.e. kautz splitting vs overloaded splitting).
- 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
- Druken Masters
(Linear)
- Medics
(Linear)
- Wolverines
(Parallel)
- Kosher
Sushi (Parallel? Linear?)
- Lawrence of
Arabia (Graphplan?)
- Causalists
(Causal)
- 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:
- Group number
- Experiment number
- . (all caps)
- Max of time steps (fixed in the encoding)
- Success? (t or f)
- Number of clauses
- Number of variables
- Number of literals
- Elapsed time encoding (msec)
- Elapsed time solving (msec)
- Elapsed time decoding (msec) (zero if solve fails)
- Mean flips until assign (part of solve output)
- Mean restarts until assign (part of solve output)
- Number of time steps in plan produced (ignores noops)
- Number of steps in plan produced (may be > time steps if parallel)
- Number of plans created (UCPOP only)
- Number of plans visited (UCPOP only)
Results
Results go in /projects/ai/573/project/results.