Question AI mid race speed up

Firstly, I think the AI in RaceRoom is the best of any game I've played and the quality of the racing itself is the number one reason I would recommend R3E over any other title. (Physics comes a close second, followed by sound ).

This is more of an observation than a question but I had to pick a prefix...

Using a fixed AI level, I have noticed that often, but not always, the lead AI car suddenly takes off mid race as if they've just been toying with the field to that point:


Roberto and Mueller above are both FR3 AI cars who finished first and second respectively at BHGP (AI=96). Roberto led from lap 3 but on lap 5 he turned on the afterburners going SIX SECONDS faster all of a sudden. Mueller was in a close battle for second and continued at Roberto's early pace.

Here's another example (WTCR 19 at Ningbo / AI level = 107):



Again, Priaulx (AI race leader) went from consistent 1:57s to low 1:54s from lap 6 onwards.

This tendency to "take off" occurred both before and after the current update. I have found that if the lead car "senses" a challenge from behind (either AI or human) they do not launch into hyperdrive but if they get a bit of a gap then off they go.

It's not really a problem as you would expect the leader to be faster with clear air in front and behind. This is more an observation that the speed difference when the AI breaks free is perhaps too much (especially the 6-second difference at BHGP).

Another observation - my manually set AI level (set to match my own pace) at Ningbo was 107 and at BHGP was 96 albeit in a different car. It would be great if the AI pace were more consistently set across tracks. I realise its complex with multiple car/track combos and very likely my own personal differences from one track to another.

 

I can confirm this "Take Off". I did a 12 min championship race with wtcr19 fixed AI at 101% at Falkenberg. Andy priaulx finished 12 Seconds ahead. Rest of the AI were closely batteling. I came from 10th and finished 3rd. At the other 3 Championship rounds there wasn't such a big gap for the winner.

 

about the spread across different tracks with different cars:
I currently note down for which car class at which level at which track I run about.
I think the attachment is quite self explaining (not too much data yet). You can see that my personal spread is not as big as yours, but not too far away. At DTM92 it seems I'm pretty bad (just one track yet) but I do not run with ABS (I never use TCS nor ABS) and AI, I think, does as by factory setting).
I have one clue: At car classes, where you run against "real" drivers (so all WTCC/WTCR, PC germany, ADAC GT Masters, DTM etc.) the AI seems to be stronger. But I don't have enough data yet.
Or maybe it is just because these cars (in my case DTM92) are simply more difficult to drive.
For slower cars, like the new scirocco even 102% was a little bit too easy for me whereas 97% at DTM was too hard.
Cause of that spread I tried AAI in the past but it does not work. AI does not adapt propperly enough to level out these gaps.
Anyway: AI is the best we have in any sim and as long as you are not way too fast or slow then it is fun to battle with them (and seeing them battle with each other), my only point is to find a level where I can run at what I want. I prefer to run at least inside the points, so top 10, cause batteling for 19th in a championship race is not as much fun as bttle for positions where you get championship points for. But that is my personaly preference.

 

I also noticed that during a single race the ai (number 2 car) was about 3 seconds faster per lap from the middle of the race, this was a GT3 race at Spielberg...

 

The problem is that the AI currently loses too much time when fighting each other.

 

True, but I assume that P1 driver was much faster than in qualifying. But I die not remember the lap times.
And that AI loses time whilst fighting is OK, If it ist too much or not ist hard to was.

 

Another example of AI launching themselves mid-race. Note the 18s and 19s to lap 4 then 13s to 14s thereafter. Note also that Muller was leading from lap 3 so to some extent was therefore the pace-setter. Not really a whinge just an observation to add to the mix. WTCR 19, Bathurst, AI=100.

 

In my observation, they don´t speed up, they only reach their normal speed when they have free road ahead of them.
Another thing is that AI speed is always about an estimated 2 notches higher in race 2. Did anyone notice this, too?

 

But you can't explain 6 seconds increase in pace this way...

In these examples I noticed the "driver X finally took P1 and then sped up" theme. As if there's a piece of code in their programming that's supposed to increase and decrease their aggressiveness/focus depending on a situation, but it gets stuck in one state and only makes them faster, over and over again.

 

Yes, although I think the point is the relative lap times - whether that's "faster" when clear or "slower" when being challenged from behind it boils down to much the same thing.

I guess my point is that the difference of 6 seconds a lap seems (unrealistically?) high and makes it difficult to choose an AI level that will give me a good race. (For various reasons I prefer setting a fixed level than adaptive.) I know battling for position slows the field down - my question is, is it really this much? I'm happy to be told its realistic, I honestly don't know.

 

Last week I had a 15 minute race with wtcc 15 at 109% with 10 oppenents. I qualified 2nd. First 2 laps we had a battle for 1st with Lopez, Thiago and me. In laps 3 I battled with Thiago for 2nd and Lopez could escape and was from there on 2-3 seconds faster per lap, whilst Thiago and me did not battle anymore, Thiago was slightly faster than me. In qualy Lopez was about 0.3 seconds faster than me.

Anyway. There are races when the fastest lap of an AI Driver ist suddenly 2-4 seconds faster than any times Driven before. So also much faster than qualy times. I never reviewed If these unnormal laps are just done by one AI driver or If others come close. These magic times happen every 2-4th race, so Not always.

From what I can remeber it is either the Leader or a driver than spun before and is chatching the pack again.

It is somehow good that others See this Strange behaviour as Well, but it is not a deal breaker.

And that AI is faster If Road ahead is clean is also ok but these laptimes are too magical sometimes

 

Just a further observation on this. I think the difference in time (as others have noted) has something to do with AI having clear air in front AND behind. When this happens off they go...

Perhaps the big difference has to do with over-defensiveness. In the example below (see pic) the AI (Catsburg) moves to the left and takes up a defensive position in front of me going into Murray's Corner despite the fact that I would have to be Danny Riciardo and Max Verstappen's love child on steroids to make a move on him from that far back (30-40 metres?).

The result is I took the normal line into Murrays' and passed him into Hell Corner (T1) quite easily. He lost loads of time defending unnecessarily.



Anyway - its just an observation as to why the AI are so much faster in clear air and slower in traffic.

I offer it in the spirit of making the best AI in sim racing (IMHO) even better and definitively not as a whinge or complaint.

 

What's the easiest way to get per-lap results like yours? I guess I need to install an overlay or a telemetry app, but there are so many of them I can't be bothered to compare, setup and get used to them.

 

I use Second Monitor: https://forum.sector3studios.com/index.php?threads/secondmonitor-timing-status-app.9587/

At the end of a race it creates an Excel spreadsheet with lots of info including laptimes.

 

Just experienced this for the first time as well running p2 (and gt3 to get some traffic) at 115 around road america. Around lap 6 or 7 (of 14) the leader suddenly went from what seemed like low 1:54s to low 1:51s and at the end had a best lap of 1:51.0. Meant that i ended up driving around alone for almost half the race.
(On an unrelated side note the p2 ai lost quite a bit of time lapping the gt3s. I made up around 5 seconds, overtook and gained another couple seconds on the then second place car in the space of like 3 laps. Also ai generally seem to be faster than could ever be achieved by players (p2 2.5s faster than georg ortners leaderboard time, gt3 2s faster than davide toccacellis leaderboard time, both of those guys not exactly slow) and there wouldve still been 5 more ai levels possible. Not a huge issue as you can just choose a lower level but a little strange to have like a quarter of the scale faster than humanly possible)

 

Just another observation on this little bug-bear of mine . Here's the top ten finishers (see below) from a WTCR race at Nords 24-hour.



Note that the lead car, Bjork, was on average almost 13 seconds per lap faster than everyone else and his fastest lap was 15 seconds quicker. It's just another example of the tendency for the lead car, if they get a bit of a break on the field, they then turn on the after-burners and disappear.

The racing for me back in the mid-pack was bloody brilliant though! Love these cars.

I have noticed some improvements to T1 performance of AI as well (thanks @Alex Hodgkinson) although turn 1 lap 1 is still very messy (AI seem to think track limits are optional). About to try today's update!

UPDATE

The AI are a lot better through turn 1 (as long as it's not lap 1) and there seem to be improvements in AI consistency across the lap. I say seems to be because I am quite open to the possibility that's its me having trouble with certain corners rather than the AI being too quick - thinking here of exits of Bergwerk and Galgenkopf which seem much better now (AI used to get super-human exits I thought).

Interestingly in my latest race with all settings identical, the fastest AI lap was 16 seconds slower than the fastest AI lap in the previous race. After burners not on this time.

I did a PB myself.

 

Thank you Sir. Your responsiveness is just brilliant and really appreciated.

 

Any news? This still happens..