rec.autos.simulators

Take care
Nigel
--
Redline Race Controls
http://www.racesimcentral.net/~lakewood/Redline2.htm

I'm not sure I agree with the point of AI is to give nice gameplay
either, unless of course you mean by gameplay the feeling that you are
racing against other 'intelligent' and 'unpredictable' drivers of
course.  :)
--
Peter Ives (AKA Pete Ivington)
Remove ALL_STRESS before replying via email
If you know what's good for you, don't listen to me :)
GPLRank Joystick -50.63 Wheel -25.01

But are the AI cars braking as efficiently as a human driver?  Also, how
do you create a little variety in their braking ability to avoid them
all braking at exactly the same point and at the same rate of
deceleration?

Mind you , I just noticed that you stated later that your AI cars just
understeer at the moment.  I'm sure that goes a long way to preventing
decent laps.  A  human driver probably makes 10 or 20 minor adjustments
from understeer to oversteer whilst taking any one corner.  If all we
could do as drivers was to turn in and hope the car wasn't going to
understeer away from the corner then I'm sure our laps would be pretty
shabby as well.

<snip>
--
Peter Ives (AKA Pete Ivington)
Remove ALL_STRESS before replying via email
If you know what's good for you, don't listen to me :)
GPLRank Joystick -50.63 Wheel -25.01

Sure, it's worthwhile.  Some people don't have a fast enough internet
connection.  Others tire of getting beat everytime they go online.  Having AI
in the game means they can run in a fairly competitive race occasionally.

Eldred
--
http://www.umich.edu/~epickett
Screamers League
IICC League
GPLRank -6.0    MoGPL rank +267.80
Ch.Rank +52.58   MoC +741.71
Hist. +82.34  MoH:na
N2k3 rank:in progress
Slayer Spektera lvl 72 assassin
Slayer Spectral_K lvl 38 Necro
US East

You've been spying on me, haven't you?

Eldred
--
http://www.umich.edu/~epickett
Screamers League
IICC League
GPLRank -6.0    MoGPL rank +267.80
Ch.Rank +52.58   MoC +741.71
Hist. +82.34  MoH:na
N2k3 rank:in progress
Slayer Spektera lvl 72 assassin
Slayer Spectral_K lvl 38 Necro
US East

Todd.

On a serious note, EECH really was only a few major tweaks away from being
the perfect combination of simulation and game. The gameplay side of thing
has been ignored by far too many developers. So thanks for that and good
luck in your current/future projects. As long as you guys are still
working there's always hope you'll get *** rich and return to your
flight sim roots as a hobby to stop you getting bored ;-)

Andrew McP

Are you involved in the open-source EECH development project ? If you are interested I can provide a link/email address. Perhaps
email me personally for more info.

Kind Regards,

Todd.

There's obviously a limit to how far the EECH code can be tweaked by
enthusiasts, but I admire their efforts greatly. What I was looking for in
the imaginary sequel was something that would allow complete (or at
least more advanced) planning of the battlefield... tasking ground forces
and creating missions (including establishing supply lines) to support
them rather than having the campaign engine in full control.

Still, at least it has a campaign engine, that's better than most! And I
still think the virtual***pit of the Hokum is *the* most immersive
flight sim***pit ever. Amazing detail, great sim.

Anyway, I'm wildly off topic for the group. I'll stop leading you astray
down memory lane. ;-)

Andrew McP

I do mean that indeed. :) The point I'm trying to make is that just
because we humans use a sophisticated neural network to drive, your
goal is not to come to a point where you use a similar neural network
in the computer to do the same thing. Your goal is still to fake the
human into thinking the drivers are as human as online players.

The methods too achieve that are irrelevant to the user; they just
want to see it work. From an engineering perspective though, it is
ofcourse relevant. And you can take a lot of shortcuts that make it a
lot simpler to simulate a basic AI driver. Then you start to add
varying behavior, based on the things you miss.
Realise that to do it the Wests' way takes ages and doesn't
necessarily provide a better result. Simple principles are more easily
adapted to include tiny things. Some basic principles even allow for a
whole lot of complex behavior, see the laws of physics for example.

Ruud van Gaal
Free car sim: http://www.racer.nl/
Pencil art  : http://www.marketgraph.nl/gallery/

That's easy, just add a random number to the brake point distance...

Ruud van Gaal
Free car sim: http://www.racer.nl/
Pencil art  : http://www.marketgraph.nl/gallery/

There are two AI racing simulation projects (freely) available:
- TORCS, www.torcs.org
- RARS, rars.sf.net

TORCS is a very sophisticated simulation, it supports also input
devices, so you can drive a car yourself. There is a tutorial about how
to implement a basic AI-driver at www.berniw.org in the TORCS section.

In commercial games the AI-drivers follow more or less a given (by the
developers) trajectory. In some games I have the impression that the AI
   physics is not simulated the same way like the players car.

In TORCS and RARS you can find various implementations:
- Searching a trajectory in a pre-process on- or offline (optimization
problem -> least squares, A*, neural networks, other machine learning
approaches, etc.).
- Reactive behaviour without preprocess.

bye, Bernhard.

--

visit my homepage http://www.berniw.org

I found TORCS and RARS useful when I was researching racing AI, it's a good
place to start. There were some pretty clever methods in there.

There's no reason not to cross the offline and online elements, you get the
best of both worlds - intelligent basic behaviour with individual reactive
elements that players hopefully notice.

What's nice is when you get quasi-emergent behaviour from AI drivers, it's
just that in reality, players rarely notice. One of my main aims (kind of
ongoing research and application) is to look at what players do and try to
emulate that as an extra dimension to the AI. You get much more noticable
reactions from players racing the AI.

We currently use the same physics for AI as the player, even the same car
model, so the AI has the same interface as the player, i.e. via steering,
throttle, brake and gears. In the past we've used simpler models for the AI,
because you can't have 20 cars racing with a full physics and car model. If
we went back to that number of cars I'd look at some kind of LOD'ing scheme.
It's easier with one model though, you don't have to match the performance
across the two, which is a nightmare job.

When (if) we make another sim fitting for this forum, I'll be sure to let
you all know what sort of techniques we use, hell, you'll probably be able
to test the thing!

Jeff.

Jason