Adaptive AI: Should be able to adapt in test/qualifying session

[Le_Poilu]

Hi,

First post here but long time reader. I'm really enjoying R3E and DTM 2014.
As I'm in a middle of the DTM Campaign in DTM solo mode, I have little grief about Adaptive AI (witch is not only a DTM E concern I think).

I noticed that in test and qualifying session (even in training mode) AI never improve their fastest times. Once it's set (in one or two laps) it's set till the end of the session.
Sometimes I manage to take a full second gap faster than AI from the beginning of the session, but it doesn't matter... If I do this I can end session and save 10min of my time, I already know AI will never improve and make a better lap than me.
It fell a little bit wrong as we all know that lap times improve over time, and the most often you'll have the fastest lap at the end of the session.

I think this is because, with Adaptive AI ON, AI doesn't adapt during session, it's a fixed settings from the launch or the session (or even all the race "week-end").

Can it be possible to have a "in session" adaptive AI ? So AI could adapt when you run +1" faster from the start, and levelup a little, being more challenging.

 

[Sean Kenney]

Good idea

 

[Andi Goodwin]

this is probably not what you want to hear but think about this ;(in my opinion on how it works best)

adaptive ai is loaded at the beginning of race weekend/session and its written to after the race ,at present that would mean reading and writting to it constantly and im not sure if this is possible ; as with all championships they work better with adaptive ai , if you do a race on that track before entering the championship mode ... i tend to practice a little on each track then run a short race (doesnt take long) at that point i know my setup and i know how im going to drive etc and it gives the ai a solid base to work from; also remember as the season goes on you will get some weight penalties if you do well ....
i think you get out of the adaptive ai what you put in , the more you feed them info the better they are at adapting
i know its a long winded one but once you have adaptive ai tuned in they are pretty good and make for interesting racing

 

[Le_Poilu]

We're agree on the way Adaptive AI work and that it should be better the more you race against it.

But there's still the issue about test/qualifying time.
The fact is that I have pretty close races against IA. In race I never win easy, I have to fight. Sometimes, I have to fight hard. (I have a medium driving skill)
But when it come to test/qualifying session, it's far too easy in most case. Only once in the DTM campaign I lost the pole position for 0.02" (it was on a track I never raced, witch I had hard time to learn). Other time I'm more about 0.5" ahead, even 1" on track I know well.
I could live with that if I felt the IA try to catch me, but they don't. IA stick with it's initial lap time, never improving: frustrating.

To be fair this is how it's for me when I start a new race:
If it's a track I don't know I go to training mode, for one or two session. I try to catch IA lap limes. When I'm confident enough about my driving on the track I start the "week-end race"
15" of test session => IA set it's time in one or two laps.
I drive until my lap time is better if I know I can do it (almost every case)
When I do so I try to keep the pace, and try to be the most regular around this lap time.
Once I'm here: I know i'll be in pole position. No more suspense. IA will not improve or do a much better time in qualifying, sometimes it can be 0.2" better, never more.
The 3 qualifying sessions in DTM became boring, I don't have to push my skill more than needed.
I just have to make one of the 10 best time in 1st session, skip the session, and one of the 4 best in 2nd session and skip the session again. No need to wait until the end, having the fear being pushed outside the qualify zone.
In third session, IA still not improve, only need to make a clear lap to be good for pole position. Again: no stress at all. because I already knew I'll be able to do it.

As I said: you know in test session if you'll be in pole or not.
But during the race it's not the same. I'm still a little bit quicker, but as can't make flying lap for all the race (tires wear, etc), IA is more difficult to fight against. Even if you're 0.5" quicker in a flying lap you can loose a race

At least Adaptive IA should adapt itself between session.
Is other sim like GSCe, IA level is a bit higher for qualify session than race session. It's way more challenging, you never know if'll be able to be on first line in the grid. Here it seems that IA is set for all the week-end session, once for all.


In fact, seems that adaptive AI adapt only from races, not from training, test or qualify session. It would be great if all running session are used for adaptive AI. No need to have a constant read/write during a race, only need to adapt after every session, regardless of the type of session.

 

[Andi Goodwin]

realistically its not important for me if ai adapts for practice sessions , im not racing them ; its not important when im doing a track test ; im not racing them ...;
Race yes
Quali yes (but ill be honest i prefer to race most of the time from mid pack so i avoid doing quali and start about 12th) saves time

but when i do do quali our times are very close / i make it into the top 10 on most tracks

Andi

 

[Le_Poilu]

You're way of enjoying it

Myself I like to have some challenging qualify session, because it's a full part of the thing

BTW: I also like race against IA in Training mode. I already had some great moment, driving in groups with 5-6 IA, overtaking them, them overtaking me, etc. It's a good ... training for futur races

 

[Shovas]

I agree it should adapt on after every session (every lap if that were possible), practice, qualify x3, race. I wrote this exact idea in the ideas thread.

That way you don't feel like you're wasting your time in qualifying.

It would be so great to see them keeping up and putting in a good race each and every step through the championship without having to fake it with small

And why not? Is it technically inefficient to record AI data every session or lap? Certainly not per session...

 

[Msportdan]

impossible , ai will forever be beating your time...

 

[Le_Poilu]

The thing about being Adaptative is to be adaptative
Adaptative doesn't mean "always do better", but "adjust your level in regards of the player skill".

If you only beat the AI by 0.01" on a Flying lap, it means your pretty close, AI doesn't have to adjust. But If you have more than half a second at the beginning of qualify sessions, there's some room for adjustment.

 

[JayJayRaceinho]

thats what i though adaptiveAI is all about, i do some laps in practice or qualifiaction or whatever ..... and the AI takes my best lap as AI-Level strength with +/- ~2-3 sec. difference on the grid. Not sure what the sense of the actual AAI is all about if i have to complete races and then race again maybe over and over

 

[Andi Goodwin]

i suppose you have a point but then maybe ... if you run a short race , isnt that as quick as a practice /quali .... its not your grid position that counts etc its your lap times, i know when im practising in general , i do a lap or half a lap tweak something go back out , try a different line the same line , get cut track warnings etc , in quali i rag the **** out of it , smash curbs etc etc , its not representative of my times , but once all that is sorted my race pace is consistent with me in traffic , racing , ahead , behind .... but maybe they already thought of all that and thats why the emphasis is on the ai learning from your race pace ...

call me old fashioned but its probably best that way

Andi

 

[JayJayRaceinho]

i see your point, ofc, qualification laptimes are always or atleast in most cases the fastest laptimes you hit and would probably never reach it in a race, however, it's just a hint where the AI Level could belong to, gridlevel 90%-95% AI Strength.... 100%-105% AI Strength ? ....

... oldfashioned, yes, qualiratio/raceratio in race07 ?! racepace of AI a bit slower than qualipace.

 

[Andrew Macpherson]

also remember as the season goes on you will get some weight penalties if you do well ....

This seems a bit silly - you get weight penalties to slow you down, but then the adaptive AI will slow down to match you.

I have trouble with the concept of adaptive AI because it makes wins and losses feel meaningless. At least if the AI is tuned (say to the developer's pace) you have a fixed target. I can try a season at 95%, then if I win try 97% and so on. But this is no fun if the AI isn't consistent from track to track at a given level.

 

[Andi Goodwin]

thing is is a season in real life is long, people get better so by the time you come to the end of the season you have improved , your competitors have improved .
i hated adaptive ai , it didnt work blah blah , i couldnt see the point in it , but as im beta testing it was part of the program so i did , i tested it now i have ai that are within tenths of me (better and worse) so for me it works ; and as for the weight penalties well only DTM rule makers to blame , but it works aswell as its not always you who wins (except when its a team mate off course )

Andi

 

[Shovas]

So, I finally took the time to see how long it would take for Adaptive AI to adapt somewhat close to me. I remember @Georg Ortner recommended doing 2 or 3 short races to let the AI adapt to you.

Well, it actually took 11 6lap races for the AI to adapt at Oschersleben. The first race I was terrible because I was new to the track. The second race the AI decided to let me win by 30sec because they thought I was terrible (rightly so, but maybe not by 30sec). The third race the AI thought I was Schumacher so I lost by 30sec to 11th in last place. Then began a series of slightly closer matches which were bouncing up and down, now 10sec win, next 10sec loss, etc. One race I'd win the next I couldn't keep up but they were closer matches. Finally, by the 11th race I was middle of the pack.

I just wish I knew what the AI was thinking and how it averages your AI level over laps. I guess the question is, what is the goal of adaptive ai? Is it to put you in a position of a win? Or top 3? Or middle of the pack? At some point I would assume it should stop raising the AI level so that you have a chance of winning. But I have no idea what's behind adaptive ai besides having to run fake races to get it to work.

I don't want to race 10 races every single new circuit in a DTM championship. I wish it would just adapt during practice and qualifying - it seems like so many laps gone to waste not to use it to adapt the ai...

On the plus side running 60 laps against AI teaches you a little bit about driving with these maniacs. I still think they're more ridiculous than the worst iRacing human rookies in terms of aggression (in a negative way) or 'lack of respect' for drivers around them.

I don't want to use a static AI level because that seems fake. For a little added realism, it should feel competitive based on my actual driving not a static AI level.

I keep thinking adaptive AI shouldn't be this bad. Why not just make it set a plus/minus around your average lap time in previous races, practices, and qualifying?

 

[Shovas]

Three more races: (1) 11th/12, (2) 1st/12 by 12sec, (3), 12th/12

So frustrating.

 

[Shovas]

I gave up. I put in 12 or so more races, somewhere close to 100 laps in total including the races I mentioned before, and I can not get consistency. I think this is due in mpart to ai but also to a few other things: I think it also includes difficult racing with overly disrepectful ai, and the poor ffb center deadzone I experience with my G27 (iRacing fixes this with its Min Force setting).

I gave up and went with the static AI level. I did a few races with that but found I was just doing a binary search to get it to a point where I had a chance to win - but the racing was still too inconsistent.

One problem with the adaptive ai is it's too eager to raise the level when you win. For example, I'll have a bad race and it'll adjust so I can win easily the next time. That's not so bad but the very next race the ai is so hard I'll end up very low again. While it's good for the adaptive ai to change more when I lose a race by a lot it doesn't make a lot of sense when it moves dramatically when I win a race so that I don't even have a chance on the next race. There should be some kind of averaging going on so that over times it gets to know what I'm consistent with. I'm sure there is but it should be tweaked.

Maybe put the mathematical function(s) in the ini file so we can find something that suits us better?

 

[Pär Öqvist]

Interesting. Just started a championship with adaptive AI. Now Will the AI take the results from Oschersleben into the equation and be more competitive at Zandvoorth?

Running the Oschers race I was 1,5 second quicker first race. Second race was closer due to the weight penalty only perhaps

 

[Pär Öqvist]

Hittade denna post ett år gammal så vet inte hur aktuellt det är.

[Sector3] Sonat Ozturk [utvecklare] 17 mar, 2014 @ 3:53pm
We are working on improving adaptive AI at the moment but in the meantime I can clarify how things work currently.

- The AI only adapts to your performance on race sessions. This is mainly because player performance is quite different between practice, qualify and race.

- The adaptive AI works on a per track basis. It's not an overall rating that applies to all tracks. E.g. Your performance in Hockenheim does not affect the AI in Brands Hatch.


"How can you realize if the adaptive ai is working at all?"
Imagine starting a game in easy difficulty and then playing through the game again in normal and then hard. The AI at the beginning will have a fixed performance based on the difficulty you have selected. Once you have completed a race session in the track, next time you are on that track, they will adapt to the average of your performance from previous race sessions on that track. The more race sessions you have, the more adaptive they'll become to your average performance.

"What time, compared to you, is the ai supposed to drive?"
There is no fixed system as such. They will on average race around the same level as you do. Sometimes close to you and sometimes faster than you.

"Does this only affect the training? Are there changes in race or during championships?"
The adaptation is not realtime, meaning it doesn't happen during the race. It happens after a race session has ended.
E.g. If you do a single race (no practice or qualify) in Brands Hatch, exit and do another single race in Brands Hatch again, the AI will adapt because it now has data of your performance on that particular track.

In championships, with the exception of Hockenheim, you only race once per track, so the AI doesn't have data of your performance until you have done some races on those tracks. Unless you've done some Single Race sessions beforehand, your very first Championship will be against AI with fixed values.

"Why are they so slow in race? Can you change this?"
Probably due to the reasons I mentioned above. We are improving them as we speak though so hopefully they will be much more challenging from the beginning.

"On some tracks the adaptive ai seems to work, at least as I suppose it should work, on some tracks it does not seem to work. "
Unless it's a bug and not due to player having no previous race sessions registered on that track, it should work. If it is indeed a bug we would love to know.

"Setups? Useless anyway, or why can´t you save them? "
Not sure what you mean. Do you mean car setups? You can save them but I agree we can improve it a lot and maybe enable multiple save slots.

 

[Pär Öqvist]

Yes that is how it seem to work. Tried another race at Oschers and the AI was 3 second faster. I don´t recall ever doing a 26 . Race was tough as heck qualified 14th ended up 13th after some misshap and taking advantage of some misshaps. That was fair.

They seem to be more competitive at Zandvoort though. 39.0 don´t seem super slow looking at the leaderboards. I managed to get a pole on the second qualification and I could challenge in race though not finding a way past the pole sitter. They don´t brake as good from high speed here I noticed maybe to at least give a fraction of a chance to overtake?

I got engine damage for some reason it seems. Not sure if you can check temperatures lost almost a gear but AI did have the same issue as I despite doing stellar attempts and succeeding here and there. Replays look so real

Anyway I hope it will be as tight as Zandvoort I will try sticking with adaptive though I guess it´s just the same as running the default setting on classic whatever that is except for tracks I done laps in.

Is it enough with the same class or must it be same car?