Realistic AI drivers skills in offline races

Just a topic about the AI drivers performances in offline races.

It seems to me that there's a script to make some AI driver slower than other, James Thompson or Monteiro in WTCR 2018 are awful... Kozlovskiy, Homola, Demoustier in WTCC 2015 are nearly 3 seconds slower than every team mate

My question is why Raceroom don't fix this to make the offline races/championships results realistic?
If Tom Coronel ends WTCC 2015 in 13th position with 7th as best results, why in offline gaming he is nearly the fastest one?

Also if in some championships the results are nearly random, than winning would be easy for the player... not having consistent rivals.


Please fix this,
A true fan of yours and buyer of every content.

 

There is no "script". AI drivers have a spread of skill levels throughout the field based on the AI level you choose. That ensures realistic racing rather than a bunch of robots all at the same skill, running the exact same lap times. Whether their skill level is tied to their name or not I don't know, but I doubt it. It's probably randomly assigned.

Tying skill to individual named drivers would mean separate AI profiles for every driver in the sim, as well as every individual class and season they competed in since skills and job roles change over time. With the research and testing that would require to get "realistic" race and season results, I don't really think it's feasible.

 

I heard that on top of AI skill you choose before a race, AI drivers have their own skill levels and then they are also assigned a random modifier for each race to imitate a natural change in pace.

 

That is what I read somewhere as well. And covers my experience.
So AI drivers have a base level and a "daily form", meaning a good base level AI driver finishes within e.g. top ten and If has good daily form he is capable of winning. I find it very good to have some variation. Of course it would be awesome If these skills would cover real world for those classes where we have real driver names.

Your 2nd part about too easy wins:
Simply use AAI and it is fixed. But you need about 2-3 4-5 lap races per class/track combo and you need constant laptimes.

I have gone through my AAI file and e.g. for Porsche Carrera Cup my average AI Level is about 100 and for FRJ about 110 and then I have some bigger differences at same class but different tracks e.g. F4 at Red Bull Ring at 104 and at Paul Richard 110

 

Shouldn't be that hard to rank the drivers AI skills in the same order as the season-ending points. But if not, give the users the ability to modify AI parameters in the way that AMS2 and iRacing do.

 

I think that when you have some licensed championship you have to be realistic in the "AI drivers" speed.

Also Raceroom is maybe the only game with so many licensed drivers and names... they need to exploit this.

When you buy an F1 game you don't see Mazepin first. When you buy an NBA game Michael Jordan is the best player. In FIFA Ultimate Team Mbappè is stronger than Federico Bernardeschi.
You have a licensed game, you try to make it a bit realistic.

 

This discussion already had place on Discord, so I'll just rephrase what I said there.

First of all, this isn't only about this real life driver being faster in the game than that real life driver. That would be nice as well but the issue is deeper.

Right now the AI is shuffling randomly between each race. This leads to drivers who were clearly on pace being backmarkers next race - even on the same track. Now that doesn't matter in a single race, since you don't really care who's where as long as you're racing a competitive field. But when you're dealing with a championship, drivers shuffling randomly means you will most likely be the most consistent driver and you don't even need to win races to win the championship. I've ran a few championships and AI very seldomly surpasses half of available points, which is generally rarely the case IRL on high level. So you're not really competing in a championship, because you don't compete with a set of consistent drivers over a span of several races, but with a different set of drivers gaining points for the same cars instead. That's not a championship.

The AI behaves very well over a span of a single race, but it clearly lacks consistency over a span of several races. Since Raceroom has a championship mode, it should aim to simulate driver's performance over a span of entire championship, not just a single race. Right now it's the equivalent of a driver switching from being the fastest on the track to the slowest just because another lap began.

And I doubt it's hard to implement. Multiclass races work fine and that's what they basically are - one AI driving faster than the other, consistently, race after race. Allowing to slice the field to multiple skill "classes" shouldn't be a problem as far as I can tell.

 

I posted a similair concern back when DTM 2020 came out. R.Rast was dead last in my races and custom championship. It was R.Kubica dominating all races. It made no sense whatsoever and had to stop playing with that class.

 

Is there any class that already manifests the intended behaviour? Or will the DTM21 be the first one?

 

WTCR is not so bad... Ehrlacher in Lynk is fast, Muller and Tarquini are top in 2018... but for example Ma Qing Hua and Filippi dominate and they never did this in real, also Hondas are really slow, Michelisz too... so they can improve

WTCC 2014-17 instead are awful... you see Chevrolet and Lada drivers humiliate the Citroens that won nearly every race for 3 years...

 

3 dtm 2021 qualyfying simulated... Gotz is the fastest Mercedes, Lawson is ok too... You say great, everything fine
NO.
cause in every qualyfying Mercedes (and Ferrari) are above all the others... so we get in the first eleven 9 Mercedes and 2 ferraris everytime

Why this beautiful game let down who wants offline simulation in this way? It would be easy to fix I think

 

I cannot confirm that at least for f4 and frj. Within a Championship there are "Always" the same AI drivers in Front/mid/rear Pack. Sometimes a top driver is mid Pack due to bad qualifying etc. At f4 there are Always the 3 same drivers much off the pace. So I deleted them for the Starring grid cause they also would affect AAI level.

As far as I can remeber, I have "never" seen an AI driver in a Championship, who was dead last in one race (except he had a Spin) and in next race was top 3 or vice versa

 

I can honestly confirm this @anno900
It seems ok with fake championships
I did an F-US championship and the bests drivers were nearly everyrace on top

The real problem is with licensed Classes... like wtcc, dtm or wtcr
You expect realistic results or something similar
The game is no way near it

 

Interesting. I definitely had such cases in FRJ, but maybe there was some exceptional circumstance at play that I didn't notice. I'll take another look at that, since I'm starting another FRJ championship soon anyway.

An actual clarificarion from devs regarding this would be nice as well.

 

Just my experience and I have done dozens of Championships and I could not see such a randomness. But no offenes just to make that clear

 

AI driver named gunnasron at f4 maybe such a single Case. He is sometimes at the end of the field and sometimes wins. But if devs can make a Statement it would be fine. My understanding is that each Driver has different base skill and a varying "daily" skill

 

I'm sure they have a base skill... so please just put the mighty Tom Coronel behind legends of WTCC like Lopez, Muller or Tarquini

 

The small number of championships I completed, I can recall the top 5 drivers were about the same in each race with a few drivers getting randomly in the top 5. Since I am also not familiar with all their names I never took notice if the rest of the field was consistent or if it matched their IRL skill levels. I'd imagine this would be pretty difficult to maintain and updating player rankings periodically like Madden and Fifa does. Quite frankly, RRE has plenty of other tasks on their list that need fixing or implementation. There are more variables and unknowns in motorsport compared to say a soccer player skill attributes to then come up with a guestimate for a number and then figure out how to translate that into tracks that these drivers may of never turned a lap on.

What should be done is all AI driver has a base level variable for any category which is then adjusted to match the AI level chosen in the settings. Then when either a single race or a championship is chosen, the field of drivers is then split into 4-5 groups where some drivers will get a boost in skill points, some a bigger boost, some with just the base level attributes, and then the remaining groups have varying negative boosts. These modifications then stay for the remainder of the event or series.

 

I don't think it's that difficult to solve this.
As you say, some drivers in non licensed championships are faster than the others.
So with a bit of attention this should be always transferred to realistic ones... not only in terms of drivers but also cars results.

It you buy, investing some good money, DTM 2021 you will find anytime 2 Ferrari and 9 Mercedes in the top 11.
Audi and BMW totally humiliated.
If you want to play offline that's awful.