Newsgroups: comp.ai.games
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!fas-news.harvard.edu!newspump.wustl.edu!news.ecn.bgu.edu!siemens!princeton!cnn.Princeton.EDU!gw1.att.com!nntpa!not-for-mail
From: nak@gwe486.cb.att.com ()
Subject: Re: "Seeing" you opponent - And some AI stuff.
Message-ID: <DJ9yyt.CLB@nntpa.cb.att.com>
Sender: news@nntpa.cb.att.com (Netnews Administration)
Nntp-Posting-Host: gwe486.cb.att.com
Organization: AT&T GBCS/Bell Labs
References: <49j0oo$p1u@life.ai.mit.edu> <4a1shb$p4v@ulke.himolde.no> <DJ6AAA.38p@nntpa.cb.att.com> <4a6spb$ogn@ixnews4.ix.netcom.com>
Date: Fri, 8 Dec 1995 15:56:52 GMT
Lines: 93

In article <4a6spb$ogn@ixnews4.ix.netcom.com>,
Eric Dybsand <edybs@ix.netcom.com> wrote:
>In <DJ6AAA.38p@nntpa.cb.att.com> nak@gwe486.cb.att.com () writes: 
>
>[nice points on LOS snipped]
>
>>AI stuff:
>>
>>Regions are very useful for your AI.  Consider regions as nodes on a
>graph.

[snip]

>You have some good points on LOS, however I'd like to jump to the
>mention of use of regions in pathfinding.
>
>The use of regions, or hex/cell groups, is a pathing approach that 
>I've considered in the past, and tried and thrown away and
>re-considered, and re-tried, and thrown away again, all because
>of the problems of dealing with them in a dynamic map generating
>game.  In other words, I've not been able to get good performance
>(as good as my old modified A* approach) when the map changes from
>game to game, and the AI must spend the cycles to generate the 
>regions for use in later pathfinding.

Hi Eric! 

I suspected that this would be the case.  So far, all I've done is to 
set up the regions BASED ON FIXED TERRAIN.  Terrain, in the game that I
have good AI in, is fixed.  You get to pick which map you fight on, but it
is fixed.  

Playtest gave us the most useful regions.  The Rhine river bridge example
was clearly a big win.  The analysis was first done by hand and then I
started working on analysis tools.  I hit the same wall you did, that the
analysis requires a human to be effective.  Thus I was glad that I didn't
have to trust the machine to analyze an unknown map.  I haven't had good
luck with machine generated terrain, so I was not heartbroken.  part of the
"artwork" of making a map includes playtest and analysis for choke points
and regions.

This stuff keeps reminding me of field theory (and heat tranfer), back in
the dark ages of my EE undergrad.  If I had a better handle on the math,
I might be able to make a conceptual breakthrough.  Surface normals and
that type of thing on a 3 dimensional surface...  You can never have too
much mathematics.

I have heard of "algorithmic terrain" which is generated so as to have
along with it an algorithmic way of doing shortest path.  

>
>Also, I've found regions to be somewhat less accurate, when the terrain
>of the map is widely dispate (ie. varies extensively within regions).

Yes, the Manhattan map defied analysis.  Skyscrapers next to Central park
next to two story and one story buildings with water on three sides of the
island.  

>Have you had enough success in using regions during pathfinding to
>perhaps address some of these issues?

Only to add to what you note.  "Gentle" terrain that looks like it came
from a real map, tends to work well.  The rivers and roads stand out
clearly as the obstacles and connectors.  "Rough" terrain, like our
Manahattan map, just gave us fits.  

THere's a note of realism here.  Armies do get lost in cities.  And in
rolling hils with rivers, roads, and bridges, it's pretty "realistic" to
come up against a river and then go find the bridge.  And it's also pretty
realistic to tend to follow the road, since it is supposed to go somewhere.

I tend to view the situation as providing "help" to a normally reasonable
ai setup that can usually get from place to place adequately.  I tend to
stay at a low level tactical sim where each unit has pretty decent, if
never inspired, capability.

I view the "screwups" as the generator of "bad luck" and "mistakes" that
somewhat resemble fog of war.  This contrasts with the behavior code which
while never inspired has had all of the newbie mistakes improved out of
it. Most players seem to enjoy fighting a competent oponent suffering C3I
troubles.  That way the combat was a hard enough fight to be worthy of the
effort, but you knew that if you had a plan and could execute it you would
prevail.


I just yesterday got notified that I'll be a speaker at the 1996 Computer
game Developers Conference, so I'll be there.
---

Neil Kirby	DoD# 0783	nak@gwe486.cb.att.com
AT&T Bell Labs  Columbus OH     USA (614) 860-5304
President Internet BMW Riders
The BMW R1100RSL - Because the Britten V 1000 is not street legal.
