Re: [TuxKart-devel] Ideas for AI

[Aly M. <mer...@ac...>]

On Tue, Jun 29, 2004 at 11:38:37AM -0600, Ryan Flegel wrote:
> On Tue, 29 Jun 2004 10:21:14 -0400, Aly Merchant <mer...@ac...> wrot=
e:
>=20
> > 3. Knowledge of World
> >=20
> >   Again tied to decision making.  I am thinking that the AI would only
> >   know about things it can "see".  It could keep a list of points along
> >   the boundaries as checkpoints it has to pass through, and it would
> >   have a list of points of interest for stuff like other cars,
> >   obstacles, herrings, etc.  Are there other suggestions, or is there
> >   any compelling reason to know about the entire track?
>=20
> As far as the road goes, AI should know *exactly* where to go. We
> should assume that the AI player has played each track hundreds of
> times and knows it off by heart. Even after I play a track for 5 laps
> (depending on size) I know every single turn by the last lap. If the
> AI players are still having trouble remember turns after I've played
> the game a hundred times, it'll probably be quite easy to beat. Of
> course low-level difficulty can be a little sloppy on this.
>=20
> Not sure if this is exactly what you meant by entire track, but they
> should know about turns in advance (because human players will).

What I mean is how much knowledge it actually requires to do whatever it
needs to do.  The way I see it the AI will probably only care about the
track it will reach in the next 5 seconds, or maybe it will track up to
the next 2 turns in the road.  It's not going to be doing a full search
on the entire course -- so what does it need to know about to be a
decent AI?

In my opinion considering the entire track on each decision would be a
waste of time, it should be able to get away with making a decision
based on its current surroundings.  In the implementation we're probably
just going to keep the track in memory, but here I'm just talking about
the algorithm.

Aly

--=20
Aly Merchant   mer...@ac...   http://cobe.dyndns.org