Marc
-- Your mouse has moved. Windows must be restarted for the change
***************************************************************************
Marc Collins
to take effect. Reboot now?
***************************************************************************
Don't let the b*st*rds get you down. Or should that be:
Don't let the b*st*rd AI get you. :)
Any tracks that you'd recommend for offline racing, or are the AI still
just plain sh*te?
Actually, my new neighbour has GP3, so I thought I'd bung mine back on
my system so I could whip his ***as a 'Welcome to the neighbourhood!'
kind of thing. Anyway, I know its not the same kind of racing, but I'll
give Geoff Crammond his dues here, his AI programming ain't too bad.
They still get predictable and do some bone-head moves now and again, as
if you're not there, but I love it when they make a complete pants of a
corner or drive into each other whilst dicing, something I've hardly
ever seen a PAPY NASCAR AI driver doing, except when there's a spin
ahead and they drive into each other as if they're blind.
--
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 -21.77
> Don't let the b*st*rds get you down. Or should that be:
> Don't let the b*st*rd AI get you. :)
> Any tracks that you'd recommend for offline racing, or are the AI still
> just plain sh*te?
> Actually, my new neighbour has GP3, so I thought I'd bung mine back on
> my system so I could whip his ***as a 'Welcome to the neighbourhood!'
> kind of thing. Anyway, I know its not the same kind of racing, but I'll
> give Geoff Crammond his dues here, his AI programming ain't too bad.
> They still get predictable and do some bone-head moves now and again, as
> if you're not there, but I love it when they make a complete pants of a
> corner or drive into each other whilst dicing, something I've hardly
> ever seen a PAPY NASCAR AI driver doing, except when there's a spin
> ahead and they drive into each other as if they're blind.
> --
> 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 -21.77
Jason
>>if you're not there, but I love it when they make a complete pants of a
>>corner or drive into each other whilst dicing, something I've hardly
>>ever seen a PAPY NASCAR AI driver doing, except when there's a spin
>>ahead and they drive into each other as if they're blind.
>If Papy's AI started making contact with each other and the player
>cars I could already imagine the horrible feedback. "WTF, Jeff Gordon
>dove inside me at Turn 3 at Bristol and spun me around, what horrible
>AI!!"
Marc
> > >I want to know if I should relegate this to online use only or keep
> hoping
> > >that reasonable AI may be coming soon....
> > Don't let the b*st*rds get you down. Or should that be:
> > Don't let the b*st*rd AI get you. :)
> > Any tracks that you'd recommend for offline racing, or are the AI still
> > just plain sh*te?
> > Actually, my new neighbour has GP3, so I thought I'd bung mine back on
> > my system so I could whip his ***as a 'Welcome to the neighbourhood!'
> > kind of thing. Anyway, I know its not the same kind of racing, but I'll
> > give Geoff Crammond his dues here, his AI programming ain't too bad.
> > They still get predictable and do some bone-head moves now and again, as
> > if you're not there, but I love it when they make a complete pants of a
> > corner or drive into each other whilst dicing, something I've hardly
> > ever seen a PAPY NASCAR AI driver doing, except when there's a spin
> > ahead and they drive into each other as if they're blind.
> > --
> > 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 -21.77
Precisely. It's obviously not too easy to get a full grid to act like
humans.
Set your expectation level at "videogame" (because that's what it is) and
you'll be less disappointed.
Regards,
Mark Davison
V8 Thunder OVAL Deputy Admin
Car #69
http://www.v8thunder.com
Brian Oster
>>if you're not there, but I love it when they make a complete pants of a
>>corner or drive into each other whilst dicing, something I've hardly
>>ever seen a PAPY NASCAR AI driver doing, except when there's a spin
>>ahead and they drive into each other as if they're blind.
>If Papy's AI started making contact with each other and the player
>cars I could already imagine the horrible feedback. "WTF, Jeff Gordon
>dove inside me at Turn 3 at Bristol and spun me around, what horrible
>AI!!"
>Jason
I think imho that the reason the AI don't seem to have problems with
each other is because of 3 things:
1) the AI driving doesn't fluctuate as much as the human driver
resulting in fewer green light calculations for overtaking.
2) When a green light calculation has happened for the AI car behind,
because all AI calculations are happening 40 to 50 times a second, the
AI car in front reacts to this overtaking move almost spontaneously, so
avoiding a collision.
3) Not enough fuzzy logic is in there to make things variable enough
allowing for AI miscalculations.
>>On Wed, 14 May 2003 02:01:04 +0100, Peter Ives
>>>if you're not there, but I love it when they make a complete pants of a
>>>corner or drive into each other whilst dicing, something I've hardly
>>>ever seen a PAPY NASCAR AI driver doing, except when there's a spin
>>>ahead and they drive into each other as if they're blind.
>>If Papy's AI started making contact with each other and the player
>>cars I could already imagine the horrible feedback. "WTF, Jeff Gordon
>>dove inside me at Turn 3 at Bristol and spun me around, what horrible
>>AI!!"
>>Jason
ai_drafting_distance = 1.6
ai_dlongpad_scale = 1000
ai_dlat_pad = 1.4
These values are for Watkins Glen and may still need tweaking for other
tracks. Basically the idea is to force them to keep a little bit more
distance from you (ai_drafting_distance and ai_dlat_pad) and to force them
to decide earlier in the braking zone whether or not they want to try a
passing move (ai_dlongpad_scale). This is all trial-and-error of course, so
your mileage may vary. I would really like to hear from other people
whether these values work for them or not.
> Which means that the AI routines aren't very forgiving. It seems as
> though the algorithms, once they've given the green light for an
> overtaking move, don't applying any recalculations to temper that move.
> I think imho that the reason the AI don't seem to have problems with
> each other is because of 3 things:
> 1) the AI driving doesn't fluctuate as much as the human driver
> resulting in fewer green light calculations for overtaking.
> 2) When a green light calculation has happened for the AI car behind,
> because all AI calculations are happening 40 to 50 times a second, the
> AI car in front reacts to this overtaking move almost spontaneously, so
> avoiding a collision.
> 3) Not enough fuzzy logic is in there to make things variable enough
> allowing for AI miscalculations.
> >>On Wed, 14 May 2003 02:01:04 +0100, Peter Ives
> >>>if you're not there, but I love it when they make a complete pants of a
> >>>corner or drive into each other whilst dicing, something I've hardly
> >>>ever seen a PAPY NASCAR AI driver doing, except when there's a spin
> >>>ahead and they drive into each other as if they're blind.
> >>If Papy's AI started making contact with each other and the player
> >>cars I could already imagine the horrible feedback. "WTF, Jeff Gordon
> >>dove inside me at Turn 3 at Bristol and spun me around, what horrible
> >>AI!!"
> >>Jason
> --
> 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 -21.77
If that was the case then I think you'd also find the AI cars doing the
same to each other because they are unable to calculate how big the car
is. This just doesn't seem to happen.
But I believe the dive bomb pass comes from the initial green light
calculation for the AI based upon the speed of the human driver. I
believe this miscalculation of the distance required is because human
drivers don't brake evenly, unlike the AI drivers.
When a green light calculation is made to pass a slower braking AI
driver it works fine because AI braking is generally very even and
therefore the calculation will be a 'good choice' on most occasions.
Because human drivers don't brake so evenly, what appears as an initial
green light passing opportunity soon becomes a non-passing opportunity,
but because the green light for an overtaking manoeuvre has been given
(based upon even braking) it appears to go ahead anyway. No
recalculation is made to determine whether to change the AI drivers mind
which, like I said, against the other AI drivers is fine because in most
cases no further calculation would be necessary. If the AI driver is
slow at the start of the braking zone then they're going to be slow all
the way through. No problem. Human drivers just aren't like that.
Maybe I'm wrong, but if they are further back then all that is going to
do is reduce the number of green light overtaking opportunities, because
the actual overtaking algorithm won't allow this now that you are just
that little bit further ahead. Unless you start braking a little bit
harder or earlier than you were before, that is.
If that is the case then I would expect that you're going to find fewer
overtaking manoeuvres in general, both against human and AI drivers.
>message
>> >Umm the problem is that the AI are a bunch of wreckers... to the
>> >human. They will stick their nose under you and then drive right
>> >through you and spin you around at just about every track. They don't
>> >seem to do it to each other though...
>> Which means that the AI routines aren't very forgiving. It seems as
>> though the algorithms, once they've given the green light for an
>> overtaking move, don't applying any recalculations to temper that move.
>> I think imho that the reason the AI don't seem to have problems with
>> each other is because of 3 things:
>> 1) the AI driving doesn't fluctuate as much as the human driver
>> resulting in fewer green light calculations for overtaking.
>> 2) When a green light calculation has happened for the AI car behind,
>> because all AI calculations are happening 40 to 50 times a second, the
>> AI car in front reacts to this overtaking move almost spontaneously, so
>> avoiding a collision.
>> 3) Not enough fuzzy logic is in there to make things variable enough
>> allowing for AI miscalculations.
Yes, but as you point out further down (snipped, sorry), the AI are more
predictable so they don't need as much padding between each other as you do
between an AI and a human.
Didn't seem to work that way. Somewhat fewer perhaps because they were no
longer attempting foolish passes, but they were still making passes, just
doing it more cleanly.
>message
>> Maybe I'm wrong, but if they are further back then all that is going to
>> do is reduce the number of green light overtaking opportunities, because
>> the actual overtaking algorithm won't allow this now that you are just
>> that little bit further ahead. Unless you start braking a little bit
>> harder or earlier than you were before, that is.
>> If that is the case then I would expect that you're going to find fewer
>> overtaking manoeuvres in general, both against human and AI drivers.
>Didn't seem to work that way. Somewhat fewer perhaps because they were
>no
>longer attempting foolish passes, but they were still making passes, just
>doing it more cleanly.
Here are the original WG values from the track.ini
ai_drafting_distance = 1.15 ; car lengths behind while
racing
ai_panic_decel = 4.5 ; G's of extra deceleration
inducing panic
ai_dlongpad_scale = 1.8 ; scales padding in braking
zones (smaller = more scaling)
ai_dlat_pad = 0.8
You have increased the distance between cars while racing from 1.15 to
1.6--that makes sense given the problem we are trying to address.
You have changed the dlongpad_sclae from 1.8 to 1000, which dramatically
lessens the scaling re: padding in braking zones. OK, what the hell does
that mean? What is "scale padding"? I know what a braking zone is. I
think I know what padding would refer to. But what exactly this value does
if you go up or down from the default is a mystery to me at this point. The
fact that you have changed the value by a factor of 500 and it hasn't "blown
up" the AI suggests it doesn't do much.
Lastly, you have increased dlat_pad from .8 to 1.4. Presumably the pad part
refers to padding again, but what is the dlat referring to here? I may be
mistaken, but dlong and dlat are distance latitude and distance longitude.
So would this value be attempting to pad more distance between the sides of
the cars so the AI don't turn into the human so carelessly in corners?
This may be a hopeless venture, but if the AI can finally have some "I," the
game will be worth playing offline. I actually doubt it is possible for two
reasons: i) it hasn't happened in the previous ump*** versions and ii) Papy
and the testers would have figured it out by now--this isn't a first-time
out rush job.
Marc
> ai_drafting_distance = 1.6
> ai_dlongpad_scale = 1000
> ai_dlat_pad = 1.4
> These values are for Watkins Glen and may still need tweaking for other
> tracks. Basically the idea is to force them to keep a little bit more
> distance from you (ai_drafting_distance and ai_dlat_pad) and to force them
> to decide earlier in the braking zone whether or not they want to try a
> passing move (ai_dlongpad_scale). This is all trial-and-error of course,
so
> your mileage may vary. I would really like to hear from other people
> whether these values work for them or not.
> > >Umm the problem is that the AI are a bunch of wreckers... to the
> > >human. They will stick their nose under you and then drive right
> > >through you and spin you around at just about every track. They don't
> > >seem to do it to each other though...
> > Which means that the AI routines aren't very forgiving. It seems as
> > though the algorithms, once they've given the green light for an
> > overtaking move, don't applying any recalculations to temper that move.
> > I think imho that the reason the AI don't seem to have problems with
> > each other is because of 3 things:
> > 1) the AI driving doesn't fluctuate as much as the human driver
> > resulting in fewer green light calculations for overtaking.
> > 2) When a green light calculation has happened for the AI car behind,
> > because all AI calculations are happening 40 to 50 times a second, the
> > AI car in front reacts to this overtaking move almost spontaneously, so
> > avoiding a collision.
> > 3) Not enough fuzzy logic is in there to make things variable enough
> > allowing for AI miscalculations.
> > >>On Wed, 14 May 2003 02:01:04 +0100, Peter Ives
> > >>>if you're not there, but I love it when they make a complete pants of
a
> > >>>corner or drive into each other whilst dicing, something I've hardly
> > >>>ever seen a PAPY NASCAR AI driver doing, except when there's a spin
> > >>>ahead and they drive into each other as if they're blind.
> > >>If Papy's AI started making contact with each other and the player
> > >>cars I could already imagine the horrible feedback. "WTF, Jeff Gordon
> > >>dove inside me at Turn 3 at Bristol and spun me around, what horrible
> > >>AI!!"
> > >>Jason
> > --
> > 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 -21.77
