General AI Performance Discrepancy Reports (No Discussion)

@Renato Simioni One quick question, should we determine that baseline pace in quali or race trim or does it not matter? I guess qualifying, as that is what it says to test in the post, but there are sometimes differences on heavy fuel that aren't there on low fuel. Also, any specific track conditions we should adhere to?

(Feel free to delete this, but I think it's important for this thread to know for sure)

 

What about possible discrepancies between player and fatest AI sector times (one way or the other) and they somewhat "cancel out" each other for an entire lap, resulting in total discrepancy being < 1s for the lap ?
During races this would be mostly observed as best AI taking some curbs slower than player and accelerating out of them faster. Or if using qualifying, the sector times in the results would show these discrepancies as well (for example: S1: -1s, S2: 1s, S3: identical).

 

Cool, thanks for doing this. Should we use Barcelona with or without chicane?

 

Note.
This above link is dead.**
delete this post when appropriate! link fixed.

**Maybe this thread?
https://forum.reizastudios.com/threads/general-ai-discussion-questions-feedback.11329/

 

Baseline Strength: 114/High Aggression
Car Used: F-V10G1/Identical opponents
Track: Spa-Francorchamps 1993
Session: Qualifying

Can confirm the report made about this particular combination elsewhere. I'm doing low 49's here, the AI do mid 51's in qualifying (medium grip, I was not cutting Bus Stop despite the fact you very much can). The AI were consistently off my pace; 6 tenths in S1, 9 tenths in S2 and 3.

 

Seeing that Stock Car Pro Series 2023 is used as the baseline I set a time at Barcelona and set the difficulty accordingly. The AI more or less matched the time differences compared to me on 3 other tracks I've tried. (knowing that Barcelona is one of my worst tracks, I turned difficulty up a few clicks)

So the AI seems to be very well tuned for the Stock Car Pro Series 2023 class all around! It was really fun racing them and they provided a nice challenge

From an initial look it seems that the AI struggles the most with high downforce cars that can be pushed a lot in corners, but I'll post some proper reports when I have some time.

 

Baseline Strength: 114/High Aggression
Car Used: F-RetroG1 McLaren
Track: Interlagos 1976
Session: Qualifying and Race

AI is much slower out of T2 (around 5-6 tenths time loss) and out of T4 (you can drift that corner, which is faster, AI don't do that and lose 6-7 tenths on the exit of it).

Baseline Strength: 114/High Aggression
Car Used: F-RetroG2 Lotus
Track: Interlagos 1976
Session: Qualifying and Race

AI is much slower out of the last corner. From there on, it's full throttle for the player all towards T3. The AI either don't do that or are slower on the straight, because they lose almost a second there.

 

Baseline Strength: 108/Max Aggression
Car Used: F-RetroG2 Brabham BT46
Track: Brands Hatch GP
Session: Qualifying
They are slow pretty much everywhere here, but more so on the first and third sectors, I think the class itself is a little slow as in Barcelona I was 2 seconds faster than the other Brabham.

 

Baseline Strength: 105/Medium Aggression
Car Used: 2023 Stock Car Corolla
Track: Barcelona and Interlagos
Session: Qualifying
Conditions: Light rain/wet track, real time. Track started soaked and moved to damp as dry line formed

AI was off the pace in the wet compared to baseline dry. Their times improved over the session as the dry line formed. Their main weakness was braking too early and they were slow early/mid corner. Corner exit was a bit better and straights were ok.

 

Caterham Supersport vs Superlight:

• BA No.6 track, 42 degrees, dry, default date
• Superlight AI are 1.5 secs per lap faster than Supersport RELATIVE to player (me) at same settings
  • Slight set-up tweaks to Superlight that has substantially more understeer in default set-up than Supersport (which has none) reduced the gap a bit

 

Baseline Strength: Used to be 114 but for this class I used 110/High Aggression
Car Used: F-Reiza
Track: Barcelona
Session: Race
Setting: Fuel accelerated

AI goes from bang on my pace to 1.6 seconds quicker per lap while also seemingly suffering no tire wear (which is very high for the player; after 7 flat out push laps the soft tires are done).

They gain the most in T4 and the third to last corner (both very long right handers, might be a clue).

 

Baseline Strength: 108/Max Aggression
Car Used: F-RetroG2 Brabham BT49
Track: Barcelona
Session: Qualifying/Race
Similar to Dicra, I am losing 4 seconds from my pole time to the start of the race, while the AI only loses 2, 66 lap race, started with 180l (70 laps) of fuel. This happens pretty much with every formula class, bar the Retro G1, which are perfect.

 

Baseline Strength: 108/Max Aggression
Car Used: F-Classic G1/Lotus 98
Track: Barcelona
Session: Qualifying/Race

Unfortunately I couldn't upload the photo from the race, too big. But this class is gonna be tricky to balance against the player, a lot of variables here. I lost about 8 seconds from my qualy time to the beginning of the race (1.31.7) managed a 1:24.3 in qualy but my ideal was 1:23:4, so I left a lot on the table, very tricky car to drive, but very nice nonetheless.

The tricky part is that Mr Silva did a super uber lap on qualy with a special 2k bhp car. I know he was pretty good in qualy, but thats too OP. He did a 1:26.5 on the race, while I managed a 1:31.7 on medium tires which are about 1 second more or less slower than the softs, but Mr Marshall did a 1:26.7 so he only lost 3 seconds from qualy to race and only 1.3 seconds from practice to race and practice was run on a green track. Track temperature in the race was 92f, so while it should be faster than the 85f of practice, it's still too much of a difference when compared to the player.

The player is at a huge disadvantage in this class, if you factor in the tire choice and wear, fuel weight and having to run the lean fuel map because otherwise you will run out of fuel, its simply not possible to race the AI at the same difficulty that you qualify in a full length race in this class. Granted I did not continue the race past lap 10 to see if the AI would pit or lose a lot of pace (will do that tomorrow) but it's still not very immersive having to deal with such disadvantages at the beginning of the race.