Runboard.com
You're welcome.
Community logo






runboard.com       Sign up (learn about it) | Sign in (lost password?)

 
mojofilter1499 Profile
Live feed
Blog
Friends
Miscellaneous info

Registered user

Registered: 10-2012
Posts: 36
Reply | Quote
Speeding up AI


Wondering how to speed up the AI without having to run AI strenth at 110%. I have two tracks, both around 1/3 mile, one i'm 1 sec. the other i'm almost 2 secs. a lap faster. Also is there a way to make the cars string out a little bit? The one track is like a short track plate race, the car set I use works fine on all other tracks.
11/6/2015, 8:24 pm Link to this post PM mojofilter1499 Blog
 
fortine00 Profile
Live feed
Blog
Friends
Miscellaneous info

Registered user

Registered: 01-2012
Posts: 68
Reply | Quote
Re: Speeding up AI


Re: ai's laptime/speed
1) Increase the [ ai_track ] ai_grip_modifier value.
 This would be the usual way, the easiest way.
2) If the ai are "sluggish" as well as slow, you could increase the [ ai_track ] ai_line_modifier value. This increases the ai corner speed. The AI line modifier changes the AI's perception of their grip limit, i.e. at 1.00 they see 100% of the actual grip they have available to them ( as set by their grip value in the track.ini ). Raising the value above 1.00 will make them run like they have more grip than they actually do, and they will carry more speed into and through the corner, which will also yield more speed on the straights.
3) Creating a new race.lp may increase the ai's speed if the new lp contains physical information that allows the exe to impart more speed.
Also: ai...WHAT'S THE TROUBLE, WHAT'S POSSIBLE, HOW TO DO IT/Part 2/ai speed... too fast, too slow

Re: breaking up pack racing at non-plate tracks
Yes.
However, to avoid unintended consequences, a particular edit(s) for this situation would require determining the reason for this behavior. It could be track.ini values, the lp's, or despite your supposition, even the ai ratings. The specific track and track.ini would have to be examined.



Last edited by fortine00, 11/7/2015, 1:31 am
11/7/2015, 1:24 am Link to this post PM fortine00 Blog
 
mojofilter1499 Profile
Live feed
Blog
Friends
Miscellaneous info

Registered user

Registered: 10-2012
Posts: 36
Reply | Quote
Re: Speeding up AI


Thank You very much! Started tweaking and already see an improvement. I've played with the track.ini a lot before, but have never had this much of a speed differential. As far as the car ratings, I rate my cars into 5 groups, all starting with a base rating, then make tweaks for how the drivers are in real life for aggression, consistancy, finishing, etc. My fastest cars are rated at a base of 70-85, and my slowest 0-15, with 50-85, 30-60, and 20-50 inbetween, and its like a plate race. the 0-15 rated cars keep up with the 70-85 cars where on other tracks they're being lapped between 10-20 laps depending on track length up to 1/2 mile. But i'll keep playing. I appreciate the help.
11/7/2015, 2:37 am Link to this post PM mojofilter1499 Blog
 
fortine00 Profile
Live feed
Blog
Friends
Miscellaneous info

Registered user

Registered: 01-2012
Posts: 68
Reply | Quote
Re: Speeding up AI


If I understand your ai ratings correctly, I don't recall ever seeing those particular ratings values or groups. I don't know what you're trying to effect, but the numbers seem a little extreme to me.
Nevertheless, I don't necessarily think your "pack racing" is being caused by the ai ratings, I was just noting I never rule anything out as a possible cause for ai behavior.

I would first suspect track.ini values and/or lp's, but I really can't suggest what track.ini value(s) to adjust, let alone a specific value, without observing the ai in action. I'd be willing to give it a look, but to do that, I would need to know the track* at which you are experiencing these results.
* The specific name/version, and possibly where you got it if it's available from a lot of locations. The track.ini values (and sometimes the lp's) vary from site to site; it often depends on why a particular site has it available for download, e.g. league use.

11/7/2015, 3:37 am Link to this post PM fortine00 Blog
 
mojofilter1499 Profile
Live feed
Blog
Friends
Miscellaneous info

Registered user

Registered: 10-2012
Posts: 36
Reply | Quote
Re: Speeding up AI


Yes, the numbers are extreme. I run 99% short tracks, with real cars and real series. Through months of tweaking, I have the AI cars of each series running the lap times of the real cars. There are some guys that get lapped every 10 laps or so. I've got ratings for many different series, some more extreme then others.

The tracks are Dells Raceway Park, and Marshfield from Wild Kustoms and cars.
11/8/2015, 9:23 am Link to this post PM mojofilter1499 Blog
 
fortine00 Profile
Live feed
Blog
Friends
Miscellaneous info

Registered user

Registered: 01-2012
Posts: 68
Reply | Quote
Re: Speeding up AI


The tests.

My 43 car roster consists of 3 very strong cars, a smallish group of strong cars, a rather large group of good cars, a smallish group of field fillers, and a few 60/40's (min50 for Track-type, Consistency, and Chassis).
There are a maximum of 27 ai opponents allowed at both tracks, the poorest rated cars are not being selected. My ai roster uses car numbers assigned by points ranking, the lowest rated car selected was the 31st car.
I ran the races @ 100% Opponent Strength, 4x Pit Frequency, Cautions disabled.

@ Dells Raceway Park
The leaders started getting single file about lap 3, and the spread worked it's way back through the field. By lap 15 the field was more or less single file, although cars randomly dropped to the low line for no apparent reason; I'll assume it is partially due to ai_squeeze_pcnt = 0.12. That's not a particularly high value, but this is a sub 1/2 mile short track and these tracks create their own set of anomalies.
The leader caught up to the back of the field around lap 18, but it took to lap 23 before the leader started the passing process. The other fast cars followed. In the 35+ laps I observed, there were no accidents.

@ Marshfield
The leaders started getting single file about lap 5, and the spread worked it's way back through the field. By lap 12 the field was more or less single file, with a few in the back half of the field running side by side. By lap 20, there was mostly single file, but there were cars attempting passes behind the 1st 1/3 of the field. The leader has not caught the back of the field by the time he pitted on lap 27. After the pit cycle the leader was behind the 21st (of 27) place car. There were no accidents.

2x Pit Frequency
The leaders started getting single file about lap 11, and the spread worked it's way back through the field. By lap 16 the field was more or less single file, but there were cars attempting passes behind the 1st 1/3 of the field.
The leader caught the back of the field on lap 27 and immediately started passing. The other fast cars followed. In the 34 laps I observed, there were no accidents.

4x Pit Frequency
GurtonBuster_AI ratings - His ratings consist of 4 groups, 80-90, 75-85, 60-75, 4 @ other. Of the 27 ai selected, 17 were top tier, 10 from the second tier. His ratings are designed to yield a different (race to race) winner (and top 15-20) from a large group. In limited testing, I have observed it produces the desired results on ovals. The results are somewhat less random on RC's.
The leaders were single file by lap 12, and the spread worked it's way back through the field. By lap 20 the field was single file, but from position 13 and back, cars are continuously dropping to the lower line (as allowed by the maxrace.lp) and passing the cars running the higher line, a combination of the race.lp (on the straights) and the minrace.lp (through the center of the corners).
The leader didn't catch the field before pit stops.

--------- --------- --------- --------- ---------

Due to the "NR2003 low line is always faster for the ai" bias, the ai will drop to the low line every time the exe determines the ai is being held up, thus causing a fair amount of side by side racing. However, there is really no close tail to nose proximity and little to no 3 wide. I wouldn't describe what I observed as "pack style" racing.
The results seem to indicate your "short track plate race" experience is not track induced.
Observing the exe's roster selection, almost exclusively the best 27 ai, and the two rosters I used, I don't think it's ai ratings. The 27 ai would have to have exactly the same ratings with a very minimal (<5) min/max differential, and still overcome the exe's internal vesting of more prowess to ai of equal ratings, i.e. the exe tries to impart a pecking order to the field in addition to the ratings.

That leaves little else,
1) some modification to the track.ini, lp's, the fast setup, a modified papy_ai.ini.
2) it really is caused by your ai ratings (but that still seems highly unlikely to me).
3) your description of Plate racing may be slightly exaggerated ? emoticon emoticon


ATM, I have nothing else.
11/8/2015, 6:44 pm Link to this post PM fortine00 Blog
 
nr2k3tracks Profile
Live feed
Blog
Friends
Miscellaneous info

Head Administrator

Registered: 01-2012
Posts: 442
Reply | Quote
Re: Speeding up AI


Car ratings drive the bus. For example, the default 70s mod car set sucks. They are 2-3 seconds slower than BGN.

I've never heard of anyone making car ratings lower than the default 40/60. Good god, man! at 0/15 you've got rolling roadblocks on the track.

To increase driver agressiveness, you have to mess with some very tricky numbers, especially at short tracks. ai_line_modifier adjustments will make cars hunt for the best line. Most short tracks I made had cars moving around every lap, looking for an opening, looking for a best line.

I can't over-emphasize how much LP lines play a role either. You make slow LPs..... you get slow sluggish, non-hunting cars. you make them too tight, nobody moves. The wider you make the min and max race, and the more aggressiveness you give the AI will make them move and hunt more. You have to give the AI space to move left and right off the race.lp. But that min and maz lp better be damned fast or they just fall back in line.

dlong_pad and dlat_pad will also affect how cars behave. The default N2k2 and N2k3 AI would run right thru each other because apparently the papy geeks didn't understand their own control values.

And as pointed out, squeeze_pct also plays a role.

Now...............

Any time you change any ONE of those variables I just mentioned, please realize that it directly connected to every other variable I mentioned. you make ONE change, you gotta change them all.

Have fun...........


Last edited by nr2k3tracks, 11/19/2015, 7:04 pm


---
steveb73
11/19/2015, 6:59 pm Link to this post PM nr2k3tracks Blog
 


Add a reply





You are not logged in (login)
Back To Top