Track layouts to receive AI line redos

Velopark and Sprint cars. Back straight exit speed from the preceding esses is slow for AI. (They take the esses simply too slow.)

Aggression 100, Skill 105.

 

Long Beach AI is generally a bit inconsistent and a bit slow even when set to 120.

Most of their problems are around T6-T8. They seem to have more traction than the player coming out of T6 through T7 (Pine Avenue) but they slow down a lot on the exit of T8 making it very easy to overtake them on Seaside Way. They also seem to be bit slow going around the final hairpin of T11 and never defend there so you can just keep divebombing.

 

Silverstone 2020 AI with F-Ultimate Gen2:
- The exit of Becketts the AI are consistently slow and the player can easily overtake
- When the AI are battling into Stowe they regularly (but not all the time) go on to the grass and bring dirt on to the exit of the corner.

Unlisted YouTube video:

 

Stock Car 2023 (I drove the Toyota) at Cascavel

Default date, default progressing, 5 minute Q, 5 lap race, AI @ 100/High, default set up.

I qualify about 0.8 seconds clear (S1 +0.05, S2 +0.35, S3 +0.42) and win the race easily without using boost. In the race the AI are just a little slow in T1 then so slow through T2 that you can line up a pass every time by grabbing the inside line for T3. They are slow out of T5 - again you can line up almost every time for a pass on the inside of T6 - and they are slow through T7 and thus all the way down the pit straight. Other than their outright pace, they drove well and I only noticed a believable level of going onto the grass.

 

Opala 1979 at Cascavel, default everything, 5 minute, 5 lap race, AI @ 100/High.

Fastest AI beat me by 0.7 seconds in qualifying, but in race I got fastest lap by 0.5 seconds.

I repeated but with the race in the rain where I was 0.3 seconds faster. In the wet they are very slow at T5. And they have rear lights on but no front lights.

Same again with Opala 1986:

I can get pole by almost a quarter of a second and in the dry race the fastest AI lap was equal to mine. They are slow at T5 and T7 though.

In the wet race I was 2 seconds a lap faster - they are slower at T2, T5 and T7 in particular. I guess they found some budget to replace blown headlight bulbs at some point in the intervening seven years. ;-)

EDIT: Forgot to add that in the rain race the rev counter needle glitched and spent the whole race jumping around all over the dial.

 

Formula Dirt at Ascurra Dirt, default everything, 5 minute, 5 lap race, AI @100/High.

I got P2 in qualifying, 0.2 seconds behind, but won the race very easily, with fastest lap by 0.3 seconds.....

Incidentally, I then tried again using the "shortcuts". Whilst none of them will count your lap as valid, you only get a cut warning for taking one at the Dirt/RX split, no warning at all for turning left at the start, and a penality fo trying that one the other way round - ie going from sector 2 straight to the start area.

Formula Dirt at Foz, default everything, 5 minute, 5 lap race, AI @100/High.

I qualified P1 by 2.56 seconds (+2/-0.3/+0.8 by sector) and won the race by 8.3 seconds even after being pushed wide at the start. I had fastest lap by 3.3 seconds!

I guess it's possible that I was 'cheating' by straight lining most of sector 1, but even when I followed the AI line I was still much faster.

 

Are ovals left out of this list for some reason? Or can we suggest them?

In Fontana AI slows down a lot in T1-2 then are too fast in T3-4. I'm sure similar things happen in other ovals. Tomorrow night (GMT -3) I'll be at home again to do some tests.

Edit: Oulton park AI behaved really bad in chicanes last time I tried. As I said before editing, I will do a proper test session tonight.

 

Goiânia - Miolo (3 loop) - Stock Cars 2023 - Aggression/Skill 100/105

Mostly the opponents choose the outside lane into the loop. It warrants better positioning into the next two turns, but it bears no fruit as they drive too slowly in this turn, compared to the rest of the track.

 

Ok, I did some tests with the 2019 Stock car. I've had similar experiences with other oval cars. 105 AI.

AI cars slow down before the tri-oval corner at Daytona:

AI cars slow down at 1-2 and are too fast at 3-4 at Fontana:

AI cars tend to be slow when exiting 2 and 4 at Indy (much more subtle than the other two cases)

Apparently Daytona and Indy misbehavior is due to the AI braking sometimes (like an accordion effect that occurs for some reason, to avoid collisions). The Fontana one seems to be somewhat more "fundamental".

Is it enough to report this here or is it worth reporting this on the AI bug thread?

 

I just did a race in the Superlight at Oulton Park International (100% & Med Aggr). Two of the AI got stuck at Britten's (separate incidents; maybe the second was later penned in by the first) and just parked it there throughout the rest of the race. All the rest of the AI (and me) managed to avoid the two retirements across several laps.

 

Something I mentioned in another thread about 10 days ago:

• I raced the Indy 2023 cars at Gateway oval last night (AI @ 100%) and was effortlessly cruising past multiple cars at every corner, particularly entry-mid corner at T1 and exiting T4.

 

Daytona RD bus stop. Tried quiet a few different lines, breaking points, aero configs but cannot match the gte and gt3 AI there.

Donnington both courses, the AI always drive on the grass at the pit exit area when driving close to another ai.

Having fun with the Caterham super light at Brands, but the AI drives on the grass several times a lap.

 

German Group A at Nurburgring Sprint S 2020, default everything, AI 100/high. 5 minute Q, 5 lap race.

AI often ride (or rather bounce over) the big kerb on the right at T1 and are slow through the tight left hander at T2 (but very fast exiting it). They are very slow at the right hander before you head down the hill on the fast section, then they often drive on the sausage kerbs at the chicane. Also saw one clip the tyres at the final turn whilst fighting another AI, during quali.

In the race the safety car came out after a BMW landed on its side at the chicane... Initially the car behind hit it and stopped, but then they all ghosted through it. In fact at one point they were pretty much all in a line overlapping each other. This is in the replay - I don't know if they ghosted like this in the race as I was leading at the time.

 

So does this mean we should expect more AI line updates to Spa 2022? as in the 1.4.9.0 release changelog you stated this:

Apologies if I came off a bit antagonistic. Just curious if this was a mistake or not

 

Modern Montreal, even with the new adjustments the AI is slow at the S after the bridge. Corner entry is fine but exit is slow as if they were trying to avoid gettimg to close to the wall.

Overal comment at this track is that the AI drives on the grass way too often and take some funny lines when attacking the car in front. For example the chicane before the wall of champions where they hit the sausage curbs too hard. Tested with the F309.

 

Daytona Road Course : AI are too fast in the chicane which brakes the oval section

 

Long Beach. The AI are slow out of turn 8 going onto East Seaside Way

 

Track: Bathurst
Class: Hypercars
AI 100, Aggression High

AI is unrealistically fast (rocket ship) at the Mountain Straight and Conrod Straight.

 

Sprint Race at Brasilia, default everything, AI 100/high, 5 minute Q, 5 lap race.

I got pole by 0.42 seconds even with being 0.4 slower in sector 1. S2 I was 0.28 faster and S3 0.53 faster. They were closer to my pace in the race but that might just be because after pulling away from them on lap one I let a few of them pass me. They are slow in turns 4 and 10 (so both of the right hand u-turns).

I noticed too that when fighting they would often climb several of the big kerbs and pick up grass.

This track is on "the list", so maybe not so useful feedback.....TSI Cup at Cascavel, same session settings - all fine except that they are very slow at T5. (The third to last corner). They are good in the final corner though, which has been a problem for several classes up to now.