Official General AI Comments & Discussion Topic

Spielberg 1988, F-Retro Gen3, I gain a whole second out of T3 on the AI.

 

The M1 Procars are way too fast through Eau Rouge at Spa, with insane grip while I'm braking and shifting down to avoid running wide, highly frustrating.
This is 95%, 75% aggression. Seem quite well balanced everywhere else

 

I'm in the first race of the GT1 championship with rain, and while I'm trying to not loose traction the AI just blast through like the wet track doesnt affect them.
Track: Cascais
AI skill: 98
Agression: 80
In general I feel like it's impossible to race in the rain because of the AI, anyone feel that?

 

It greatly depends on the class. With some of the lower power tintops and open wheelers I find that I'm 5 sec/lap faster than the AI in the rain, and 1-2 sec/lap slower in the dry. As I understand it the recent physics and tyre updates might have exacerbated the issue, but now that things are settled there's a more stable baseline for the devs to work from in terms of balancing performance.

 

I just found a car track combo that hasn't been optimized for the AI at all apparently. Formula Classic Gen 1 at Jacarepaguá 1988. All the AI get stuck behind one another and drive single file at maybe half the speed they need to be going to make the turn. It's infuriating because you come up behind them going Mach Jesus and they swing over to block you and cause a crash. (and it's the very first race of the FCG1's championship)

It does show the amazing difference tuning the AI for a car type can do though. Jacarepagua worked great in the 1983 F1 cars.

Formula Classic Gen 1
Jacarepaguá 1988
AI skill: 90
Agression: 50

 

Version: V.1.3.8.3
Car: Formula USA Gen 3
Track: Road America
AI Strength: 102
AI Aggression: Any
Weather when the issue was observed: Dry
Type of session when the issue was observed: Qualification & Race

AI doesn't seem to be affected by fuel weight. I.e. best AI qualification time was 01:40.820.
The times of the same driver in race were:
01:40.240
01:40.380
01:40.639
01:40.643
01:40.617
01:40.560
01:40.700
01:40.564
01:40.596
01:40.582
01:40.541

It was a 40minutes race, as they were not pitting even at the 20minutes mark, they had to start with +100liters of fuel, and still they were faster in race from the get-go, than in qually.
Weather was similar, track was natural progressing. So race should start at the same condition qualy ended.

 

I posted about this problem in the official bug thread (page 189) but I would like to raise the issue in this thread too.

Problem: There is a fundamental problem with custom AI driver files (except those in Brazilian Stockcars).

In every series, we can change an AI drivers name, livery and, notionally, his attributes. The primary problem is that AI attributes have no discernable effect. Any attempt to improve an AI drivers attributes does not move them up the grid or to the front. They still continue to qualify and race at the back or in the midpack.

However, there is one exception - Brazilian Stockcars. If I award a Brazilian Stockcar Pro Series driver 0.999 across all of his stats, he will persistently qualify in the top 3 on every track. This is the way it should be so, I ask, why are the AI parameters not functioning uniformally across all of the series..? I sincerely doubt this is intentional.

I suspect that recent improvements to AI have had an unintended detrimental effect. It's almost as if the parameters affecting the spread of AI race/qualifying are not in sync with the spread of AI driver attributes. Please, @Coanda, can someone on the development team, review the calculations/parameters used in Brazilian Stockcar AI and replicate them across all of the other series. Until this is done, single-player racing and custom championships are effectively broken.

 

Have loads of custom AI files, they all work.
Race skills/Qualifying skills/Aggression are the most important values. I change values during seasons and it clearly has an effect.
You must be doing something wrong.

There is one issue, the file can not solve: AI in general has problems with overtaking on many tracks and fast cars crawl behind slow ones for ages.

 

Cheers for the feedback. I have tested Super V8s, Mini Challenge and Ultimate Gen2. They all suffer from the same problem that I have outlined - with AI attributes having no discernable effect on improving a drivers qualifying position or lap times. Brazilian Stockcars does not suffer from this and your feedback suggests that historic F1 cars are exempt from this problem too.

I am aware that, in qualifying especially, fast cars can get stuck behind slow cars but the latest AI updates have reduced this problem, which is great to see.

What is the spread of your field in qualifying..? Are you getting a spread of 2 to 2.5 secs (first to last) with fast AI drivers consistently running at the front and slow AI drivers consistently running at the back..? As stated, I can only achieve that in the Brazilian Stockcar Pro Series.

 

To me, it looks like a problem at your end, RayGun. These are timings from today.

MINI JCW at Silverstone National:
Dan Zelos, Championship winner, race_skill = 0.999, 1:03.3
James Griffith, slowest car, race skill = 0.608, 1:05.5

F-Ultimate G2 at Imola:
Max Fastard & Speedy Gonzalez, race_skill = 0.999 & 0.985, around 1:18.8
Nick Lately & Pierre Goslow, race_skill = 0.643 & 0.658, around 1:21.5

Super_V8 at Adelaide:
Craig & Jamie, race_skill = 0.999 & 0.985, 1:23.8 - 1:24.2
Alexandre & Greg, race_skill = 0.65 & 0.66, 1:25.2 - 1:25.7

This is consistent with all drivers across all series and all circuits that I use. The open wheelers are not as consistent as tin tops but this is in line with AMS1. They were better before the overtaking issue arose and when the draft wasn't so obvious.

 

Had a look in my folder, there are 40 custom files right now. From Mini up to modern F1 cars. Some are fantasy stuff, others try to portray a certain season of the series, which can open a gap up to 4-5 seconds between fastest and slowest driver, depending on the car.
In general, I give the fastest driver 1.0 in race/quali skills and the slowest backmarker 0.35. Makes a huge difference.
Also the aggression level is high for fastest drivers. But as "Gian" stated in the thread about customising AI, you should set aggression to 50% in game settings, in order to have them exactly the value you set in the file.

That's what Gian (Reiza stuff member) said about aggression:

"aggression:
Driver aggression. It is scaled by the "Opponent Aggression" slider setting. Examples:

• At 50% "Opponent Aggression" slider setting, the driver aggression will be identical to the overriden value.
• At 0% "Opponent Aggression" slider setting, all drivers have 0.0 aggression.
• At 100% "Opponent Aggression" slider setting, all drivers have 1.0 aggression.
• At 80% "Opponent Aggression" slider setting, the 0-1 aggression value is mapped into a 0.6-1.0 range.
• At 20% "Opponent Aggression" slider setting, the 0-1 agression value is mapped into a 0.0-0.4 range."

And yes, when I alter the values for some drivers in order to see them perform like they did in certain races of some season, it works with no issues for me.

 

Again, thanks for the feedback, guys. I do appreciate reading constructive views.

Everything you've listed though, with regards to key skills, aggression levels, etc, are things that I am well aware of. None of it is new information to me. I took a great many hours to review the explanations and descriptions offered by Reiza before testing over and over again. Only then did I post my comments here.

Earlier today, I deleted my AI custom SuperV8 file and made a new one from scratch (being extremely careful with spellings and references) but nothing has changed. Drivers who should be front runners are still in the midpack while others who should be at the back of the grid are claiming pole. Brazilian Stockcars work perfectly for me but not all other series do.

@Romeo Foxtrot, if you could share your SuperV8s file then I will give it a go because I'd like to get to the bottom of this.

 

SOLVED..!!

My issue was a relatively new one so at some point during recent AI updates, reading of custom AI attributes has been made case sensitive with regards to the file name. Some of my xml files did not begin with a capital letter and after I corrected this, AMS2 is now reading my AI driver attributes correctly.

In short, AMS2 will read a custom file for livery updates and driver name changes as long as the file name is correct. However, it must be spelled correctly and case sensitive in order for attributes to be updated. Just happy that I solved the issue.

As I posted about this issue in the bug error thread is it the correct protocol to delete that post or edit it..?

 

I'm glad you've solved your issue, RayGun, but just to clarify: the operating system controls filenames and they are not case sensitive in Windows. ARC_Camaro.xml and arc_camaro. xml both function correctly.

XML tags are case sensitive and there is a difference between race_skill and Race_skill.

There are a number of bug reports in the bug thread that are not bugs but user errors. You may edit the post or leave it as you see fit but I don't know if you can delete it. I could see quickly it wasn't a bug so I'm sure Reiza staff can spot these just as quickly.

 

Since one of the latest updates, F3 cars have much less grip compared to older builds.
But AI has some kind of magic physics and is much faster than before. Could easily compete with them at 105% in older builds, but now, they beat me by miles, as I can not achieve the same laptimes then before.

 

This (F3 physics) will be addressed in the offer end-of-month update as Renato explained in the June development update as something is off there.

 

In V1.3.8.3, P1 class is universally exceptionally slower on the straights of Long Beach than I am in my MetalMoro AJR Honda on 100 setting.

 

Caterham Academy AI are exceptionally difficult to compete against in 1.3.8.3. As I understand it, this car received updated tyre physics, but the AI didn't get the message: they can take corners with unmatched grip and will routinely pass they player on the outside with no problems. I need to turn the AI down to 80/50 to have a chance of competing.

 

Group A / 90s DTM
Snetterton 200
22 lap race
Practice 20 Min
Real Weather / current date (20.07.22)
AI 100 / 50

• AI loses much time in the long corner before start/finish ("Coram"), I am able to pass two AIs here
• AI could be a bit faster through "Brundle" and "Nelson"
• some AI swerve left right on the start/finish straight, shortly before the first corner "Riches"

 

Disregard... I cannot read numbers... :-/

Version: V.1.3.8.3
Car: Formula USA Gen 3
Track: Long Beach
AI Strength: 103
AI Aggression: Any
Weather when the issue was observed: Wet (Storm)
Type of session when the issue was observed: Race

Best AI qualification time in dry, sunny day, was 1:05.659

On the second lap, in stormy weather, track soaked wet, not a sign of dry line, it did 1:15.559... the AI is faster in wet, then in dry...