General AI Performance Discrepancy Reports (No Discussion)

Baseline Strength: 100/Medium Aggression
Car Used: Formula Trainer (both)
Track: Silverstone national
Session: Quali/Race

AI slow in general with both cars. I can easily overtake them in turns and straights - being more than 1 sec faster than them

 

Bathurst 83
F1 1991
100% medium aggression
AI brake too early on first corner, after the long straight in the beginning of the lap and on the last corner too. Generally AI is a bit too fast over twisty section.

 

Stock car 86
Brasilia
100% medium aggression
On wet conditions they are too slow, I took pole and 2nd place was more than 6sec (!) slower.

 

I'm not particularly sure the supposed "benchmark for calibration" - F-UltimateG2 around Barcelona - is actually calibrated too well. Especially in the 3rd sector the AI just behaves very off compared to the player.

I've raced them at 112 strength, which may be a little below my baseline, but it's hard to tell because they're quite far off to what I'm doing.

They tend to be too slow compared to me at the apex of T1, then be a tiny bit slower on the exit of T4, take a slightly different line through T7/8 (braking deeper and accelerating more slowly - this is actually a common theme for these corners across all classes; it's not egregious though), and then we arrive at Sector 3 where the biggest discrepancies happen.

The AI miss the apex of T10 slightly, which allows the player to consistently get up their inside. They are also significantly slower than the player on apex. However, then they get an insane launch out of this corner which gains them 3-4 tenths on exit so the discrepancies more or less even out. They proceed to lose most of that time around T12, where they again carry significantly less apex speed. Then, they gain a lot of time again because of a very good launch out of T12 suspiciously similar to the one out of T10.

It should also be noted that the F-Ultimate G2 in particular is also too slow in these slow-ish long corners on other tracks, notably Luffield at Silverstone and Dunlop at the Nürburgring GP.

It is strangely one of the not so brilliant experiences for AI calibration, which is why I'm not at all sure it's a good benchmark for anything.

See this video at 6:10 for further impressions on their driving. I will note I had to change the setup for this one because my car kept scraping the floor in the last corner, which led to me raising ride height by 2 for the front and by 3 for the rear. I also put clutches to "4" to have less trouble with the rear.

 

I was about to report this now, the only thing I experienced (in 112 difficulty too) that you didn't mention was that they were significantly quicker than me through the final corner, especially on the exit. Also, I had massive issues with my left-front tyre overheating and consequently wearing very quickly, that tyre specifically (I was using softs, matching the AI's) barely lasted 10 laps, while some of the AI did over 20 lap stints without any issue in the test races I did.
I also did a few races with GT3 Gen2s there, and, in those, they not only do basically the same things mentioned earlier, but also take a very weird line through T9, carrying way too much speed in the entry and massively losing in the exit, and break too early into T10.
This was all on 1.5.5.5 .

 

Their tire wear compared to player seems completely off for this class, yes. 12-13 laps are the maximum that can be done without significant compromises on cornering speed, whereas AI can push full throttle and seem to have no issues going double the distance. Same thing happens at Silverstone, 20 laps for me, easy 29 for the AI.

That final corner thing is caused by your car scraping the bottom of the track. It's not an AI calibration issue, it's a default setup issue for this one and can be fixed by raising rear ride height, as described in my post above.

 

The current frustrations i have when it comes to AI reports is when i see people comment "The AI just crashes me out" without even putting up a video showing the incident, and that just seems to give Reiza the idea of AI having to be dialed down and give as much space to the player as possible and beyond, when we have the aggression sliders which should be a simple way of adjusting that parameter.

It also confuses me as to what the expectation of some people even is when racing AI - to never EVER be hit by them ever ? Is that not unrealistic to never be in a collision with other drivers ? I don't envy Reiza for having to dig through all these conflicting reports and then trying to make sense of it. Like my two cents are the AI is not aggressive enough.

 

The AI is much faster on the Nordschleife than on the Nürburgring, both modern and classic.

 

I posted this on the Beta thread, but this one seems to be the official one for this reports so will repost here.

59 lap race at Watkins Glens Long with the F-RetroG2, 110%AI, high aggression, Lotus 79

The tire wear in the F-RetroG2 is pretty unfair for the player compared to the AI, my fronts where completely gone (as in completely black in the hud and with horrible understeer) by lap 30 on a very cool 76°F track temperature, the rears had about 1/4 left. Some of the AI managed to do the whole 59 laps without pitting and managed to do very competitive laps towards the end. Some pitted, though I cant really confirm if it was for tires or they had some sort of mechanical issue, but at least the winner (Spector) did one stop and managed to win, so I assume he only changed tires. If so, that was pretty nice to see some AI being able to finish the race without changing tires and others having to do so. Still the wear rate for the player is too high for this class, only some super tire saving would have made it possible to maybe finish the race, but I doubt it.

On a side note from another test, on a wet/damp track with slick tires in the HitechG1 and RetroG3,the AI are simply too fast on acceleration and fast corners, on medium and low speed corners its fine, until the exit. But due to their OP grip out of corners, its impossible to keep up without crashing or losing control. I dont mind the puddles much, most appear on the inside of corners so I just avoid them and the AI do slow down quite a bit, but their acceleration is too OP.

 

Formula Reiza
Hockenheim (full modern layout)
112 difficulty / medium aggression
AI drivers still cut turn 1 and at the moment they don't get any penalties for doing so.

 

So, after the changes to Silverstone in the last update, I did a few tests with F-Ultimate Gen2, GT3 Gen2 and LMDh cars there.

Surprisingly to me, I had only one issue common across all classes: They miss the apex at Aintree, which, in most classes (GT3 Gen2 and LMDh in this case), forces the AI to lift to avoid going off track, making them quite slow on the Wellington Straight.

With the F-Ultimate Gen2, they:
- Brake a little bit early into Village and The Loop;
- Have insane traction coming out of The Loop;
- Are really slow mid-corner through Luffield;
- Carry insane speed on Copse, (they gain 0.5-1s there) but weirdly only on the first 2 laps of the race, I'm not sure if it's a me problem;

With the GT3 Gen2 (the best class I've tested on this track):
- Unnecessarily lift on Farm, where you can go flat out;
- As mentioned before, they lift through Aintree due to missing the apex;
- Carry a little bit too much speed through Beckets and Chapel, gaining roughly 0.2-0.4s there;
- Are slow through the final corner, after the Club Chicane, they probably lift through there, where it's not necessary;

And with the LMDh, that definitely need a lot of calibration:
- They brake too much through Abbey, and lift on Farm;
- Get even more insane traction out of The Loop than the F-Ultimate Gen2, but lose it all immediately by lifting on Aintree;
- Carry way too much speed through Copse, Maggots, Beckets and Chapel (gain 0.5-1s);
- Are very slow through the final corner, like the GT3 Gen2, but much worse.

 

Le Mans de la Sarthe AI very slow on entry to pitlane after chicane. GT3 Gen 2, not driven anything else there yet...

 

Formula V10 Gen 2 is too slow overall

Testing the Formula V10 Gen 2 at Interlagos 1993 and Silverstone 2001 with AI at 120, i can tell you the AI are too slow from 2 seconds per lap to 4 seconds per lap against the human i am. (2 seconds being the fastest AI ; 4 being the slowest).

That happened in both tracks. I asked my good friend @Dicra and he confirmed to me that the FV10 Gen 2 is too slow everywhere, even at 120 strength. It's just more noticeable at some tracks.

So this is what i call a performance discrepancy.

 

Road America
LMDh way to slow
100% low aggression 1:53s-1:54s
120% low aggression 1:49:9 best

fast human sprint race 1:49:0, my sprint race pace 1:50:x,. hotlap best 1:47:8
mix some GTE in and watch them snuffle in the trunk. The GTEs do not keep distance to each other, looks very weird

 

RX cars AI is dead slow in non-RX tracks. My baseline calibrated on Barcelona/Stock car 2023 is 105. I used 107 at Hockenheim Short A and Ibarra (screenshots below after 3 or so laps) and I am multiple seconds faster. It's very obvious they are way too slow. They take curbs at snail pace.
These cars are very fun to throw around on non-RX tracks so I'd really like to see it improved.
I understand these cars must be a nightmare to calibrate for good performance, as they do not handle like regular racing cars and a human player can push these crazy fast in curbs.

 

Baseline Strength: 110/High Aggression
Car Used: F-VintageG1 (probably G2 as well, but need to test it properly)
Track: Spielberg 74
Session: Quali/Race

So I know I have been persistent in this issue, but the Vintages are one of my favorite cars and this issue makes it impossible to enjoy any race longer than 30 minutes with them. Its actually 2 problems here, first they are overall just too slow for the respective difficulty, I am far from an alien driver but still can get pole comfortably on most tracks after 5 laps or so at maximum 120% difficulty.

The second is what I have reported before, the fuel weight not affecting the player as much as the AI, I lose a second , while the AI lose 3 seconds, so it makes long races a cakewalk and impossible to lose, no matter the difficulty. Below the qualy results and the pace difference after 2 laps, at 120% difficulty.

 

Track: Hockenheimring 1988
F1 Classic Gen 3
AI Difficulty: 117
Aggression: Medium
Laps: 20
Weather: sunny

The AI here are pretty woeful & extremely slow. They’re painfully slow on the long straights and at the start, offering no resistance & are overtaken with ridiculous ease. Starting in Pos. 4, I managed to overtake all those in front, after the first straight, & was already 2.5 seconds ahead after the first chicane; it was as if the AI were lacking their turbos, or had engine problems, they were that slow, at diff: 117!!! The AI also cannot handle the first chicane, & wipeout regularly, causing mass collisions. I’ve had some excellent experiences with the AI in AMS2; but the combi’ classic F1 (Gen 3) at the Hockenheimring 88, is dire. I’ve also tried the other classic F1 gens at this track, & they’re all very slow at the start/straights. Typically the MP4/5B reached approx 330 km/h at hockenheim back in 1990 on the straights, with the others around 320-25 km/h, qualifying in 1:40.198 (Senna). The AI struggles to reach 300 km/h. I hope this can be looked into, as the classic F1 cars are just awesome, as are the tracks. But in this state, this combi’ Hockenheim 88/classic F1 is not raceable.