Newsgroups: alt.comp.hardware.homebuilt,alt.computer.consultants,comp.ai,comp.periphs,comp.programming,comp.robotics
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!news.sprintlink.net!redstone.interpath.net!hilbert.dnai.com!nic.scruz.net!cruzio!bandit
From: bandit@cruzio.com
Subject: Re: Help ! maze solving
Reply-To: bandit@cruzio.com
Organization: Cruzio Community Networking System, Santa Cruz, CA
Date: Tue, 29 Nov 1994 08:21:16 GMT
Message-ID: <D00sJL.J8B@cruzio.com>
References: <3avo57$kut@news.kth.se> <3b05f8$mk0@handler.Eng.Sun.COM> <CzzBtz.Gx@watserv2.uwaterloo.ca>
Sender: bandit@cruzio.com (bandit)
Lines: 46
Xref: glinda.oz.cs.cmu.edu comp.ai:25586 comp.periphs:10010 comp.programming:13111 comp.robotics:15777


There are some good Cellular Automata algorithms.
They assume (from memory) a grid, ability to remember/place a CA on a cell,
ans communication between the cells.
Of course, as long as there really is a grid, and the mouse can remember
exactly where it has been, the rest can be simulated.
Obviously, too, you need a well-defined start && end points.

Basically, start at the start.
The CA looks for all open spaces immediately adjacent.
In those, it places a copy of itself.
This is a good time to check for the end case.
More accurately, the cases are:

1. empty space - clone itself into it.

2. at end: propigate a message to it's "parent" that it's at the end,
and change it's state from "looking" to "found".

3. hits a wall: change state from "looking" to "wall" && inform parent.

4. Any parent which has only one "looking" child, and that child changes
state, adopt that state && propigate the message back to it's parent.

Note that an ally can be treated as one cell, irregardless of length.
Junctions are the important things....

Blind-ends have "wall" propigated back to a parent with at least
one other "looking" child.

The whole thing ends when the original CA gets either a "wall" message
(no solution) or "found", in which case just follow the path of "found"
CA cells.

In any case, I would recommend getting previous mazes,
looking for topographical similarities and differences,
and building a simulation first.

... bandit


-- 
bandit Gangwere                     systems and embedded engineering
topdown design associates           "Waltzing where mere mortals fear to look"
bandit@cruzio.com
(408) 458-9228                      Box 7830, Santa Cruz, CA 95061
