PDA

View Full Version : AI Level per track?



lokid
12-02-2018, 13:45
With the AI knowingly at different speeds at any given track relative to actual %, is there a chance to have a track specific ai difficulty option. Or basically it remembers your level each track?

Zaskarspants
12-02-2018, 13:58
I am not sure I understand you correctly but the difficulty and aggression level is set globally.

I do find I have cars / tracks where I am running higher AI, up to 120 at Oulton park gt5, simply because I love that circuit and have raced on it most often.

A AI per track slider would just add more confusion in my view for no real gain, also it begs the question, what about AI per car settings and it becomes never ending.

You versus the AI will always be a complex interplay of your skills and theirs and the variations are a natural and unavoidable result of that.

Perhaps in another five years each AI car will be thread running in some complex spiking neural network that can learn adapt and react as your skills grow, until then some player input will likely be needed to get the perfect balance.

Mahjik
12-02-2018, 14:11
SMS is still going through all the tracks and variations. There were changes to some in the recent patch, but there are still more to come. It will take some time as you saw from Doug's response on the Formula C that the tuning is every car and track combination.

lokid
12-02-2018, 14:17
i suppose, dougs interaction and help makes me want to continue to play and support PC2 regardless of my rants and moans.. At the moment I’ve taken to writing down my ai skill level.. tbd most sit around the 90-95 apart from silverstone anyone find that a lot harder than the rest, but like how brands was?

Mahjik
12-02-2018, 14:21
Also, it doesn't help that other changes to the AI forces SMS to revisit some tracks to re-tune all over again. ;)

lokid
12-02-2018, 14:24
as much as i like sims and driving them and would love to know how to make them,, thats one job i think i rather pass on lol.. AI ha

falm
12-02-2018, 14:33
Also, it doesn't help that other changes to the AI forces SMS to revisit some tracks to re-tune all over again. ;)

IMHO this problem cant be fully solved until the next generation Consoles. Alone that fact that AI is running on a different physics engine leads to tweaking, the closer the AI physics will come to the player physics the better the overall balancing (just think "puddles").
And as @Zaskarspants mentioned also we players run not all track / car combo's with the same skill, leading to the observation that the AI "seems to be off".

Overall I'm happy with what we have in PC2 and I do not expect any game changing improvements. :)

Maybe we see in the next title some sort of adaptive AI so that we can skip the fiddling with the AI speed slider completely, only aggression could stay.

lokid
12-02-2018, 14:38
i do mean Ai level per track which is obviously editable with the player, not stuck? like rf2. Just so we dont have to take to making notes.

Mahjik
12-02-2018, 14:57
I don't think it's an unreasonable ask. However, I don't believe that is the direction SMS is going. They've worked hard to try to eliminate users needing to configure anything on a per track basis other than setups. I don't foresee them changing this direction, at least with PC2.

gelfie
12-02-2018, 23:37
Be cool if you based AI lap speed on a players qualifying speed. Perhaps make the AI range +5% and -5% (or user adjustable %) from the players qualifying time on each track to sort of "Auto balance" the AI.

If no qualification is done, just use the current AI strength setting.

lokid
13-02-2018, 07:03
Thats basically what racerooms adaptive ai does i think. It works but you need to train the AI first.

OddTimer
13-02-2018, 09:44
I'd love to have a save button to allow keeping ai and agression locked for a track/car combination. I started taking notes yesterday so there is no time wasted on testing the ai strength.

GTsimms
13-02-2018, 15:35
Other titles have the AI level issue of being slow on some tracks and extremely fast on others.

Schnizz58
13-02-2018, 15:43
JMO, but I would hate to have to adjust the AI for every track. Not only that but would it have to be for every car/track combo? E.g., AI might be fast in Ginetta Jr. at Oulton Park but slow in the Formula Rookie. I'd rather that SMS go through the process to tweak all the AI as good as they can and just live with the results.

ETA: Just to add to the above in case it's helpful to anybody...
If SMS can get the AI tweaked to the point where they're fairly consistent, that's good enough for me. Then I can refine the difficulty to suit my abilities. There are going to be places where they're better than me and others where I'm better then them. That actually improves the immersion because it's like real life. As I go through the career progression, I adjust the difficulty upwards if I podium and downwards if I finish out of the top 10. Over a series of races, it settles on a value that feels right to me. I want it to be challenging but not impossible.

mr_belowski
13-02-2018, 15:43
yeah, it's not as simple as tuning the overall laptimes (which is what Raceroom's adaptive AI does). This tends to exacerbate existing issues with AI speed relative to the player, so they can end up being way too fast in some corners, so even with adaptive AI you still need to spend *ages* making sure their corner speeds, braking distances, exit speeds, apex speeds, acceleration, fuel use, tyre use etc etc are similar to the player. Factor in all the possible tyre compounds and weather conditions and it's a huge task

GenBrien
13-02-2018, 16:16
Would a dynamic system be possible?

You set how you want your race to be(easy-challenging-hard/ where you want to be in the car's pack(end-mid-top)
Then the AIs are adjusting their strenght, compared to your time, on the fly

ex:
you set the AIs to be ''challenging'' and to race in mid-pack
Game would set agression to something like 60% and strenght to 80%

you start your practice and after some laps you're in the top 3(so better than you should)
Game adjust strenght to 90% without your imput
Settings would be lock in race conditions, because it would not be possible to win/lose otherwise

Of course, you would still get the option to have a fixed value for strenght and agression(like it is now)

I guess its too much for Pcars2 now but I think it would be a good system

Schnizz58
13-02-2018, 16:34
Would a dynamic system be possible?

You set how you want your race to be(easy-challenging-hard/ where you want to be in the car's pack(end-mid-top)
Then the AIs are adjusting their strenght, compared to your time, on the fly
That would essentially be rubber banding wouldn't it? I don't think I'd like the AI to adjust themselves to me. I'd rather it be the other way around. I want to be continually getting better so that I can keep up with them. They're like the rabbit at the dog track.

Bealdor
13-02-2018, 16:46
Would a dynamic system be possible?

You set how you want your race to be(easy-challenging-hard/ where you want to be in the car's pack(end-mid-top)
Then the AIs are adjusting their strenght, compared to your time, on the fly

ex:
you set the AIs to be ''challenging'' and to race in mid-pack
Game would set agression to something like 60% and strenght to 80%

you start your practice and after some laps you're in the top 3(so better than you should)
Game adjust strenght to 90% without your imput
Settings would be lock in race conditions, because it would not be possible to win/lose otherwise

Of course, you would still get the option to have a fixed value for strenght and agression(like it is now)

I guess its too much for Pcars2 now but I think it would be a good system

Dunno. IMO a dynamic AI would take away any incentive to improve yourself.

GenBrien
13-02-2018, 16:47
I'd rather it be the other way around. I want to be continually getting better so that I can keep up with them. They're like the rabbit at the dog track.

so you're setting it at 120% and hope to catch them one day?

My idea is to eliminate the tedious and long process of fine tuning the AI to have a race like you want

In carreer for example, one race I had the AI at 105% and won by a little margin
the next race after that with the same Ai % I was last by a good 5 seconds...
set a lower %
load the race
make some laps
still not where I want to be
quit the race
adjust %
re-load
and so on....
and so on....

Like I said:
1- it would be optional (so people like you could set the % exactly how they like)
2- it would be only in practice/qual, and not in race, as to not have rubberbanding

GenBrien
13-02-2018, 16:47
Dunno. IMO a dynamic AI would take away any incentive to improve yourself.

because settings the exact % as I want does that?

Bealdor
13-02-2018, 16:58
so you're setting it at 120% and hope to catch them one day?

My idea is to eliminate the tedious and long process of fine tuning the AI to have a race like you want

In carreer for example, one race I had the AI at 105% and won by a little margin
the next race after that with the same Ai % I was last by a good 5 seconds...
set a lower %
load the race
make some laps
still not where I want to be
quit the race
adjust %
re-load
and so on....
and so on....

Like I said:
1- it would be optional (so people like you could set the % exactly how they like)
2- it would be only in practice/qual, and not in race, as to not have rubberbanding

I'm reading this kind of posts quite often but I have yet to experience this myself.
I've raced quite a lot career series already and not once have I seen such huge variance between different tracks in one championship.
Yes, I've raced on unfamiliar tracks and was quite shocked that I was suddenly 4-5 seconds behind the AI although I've won the race before.
But after ~2 hours of practice and good setup work my performance was good enough to be competitive again.

Schnizz58
13-02-2018, 16:58
so you're setting it at 120% and hope to catch them one day?
No, I set it a little bit better than I am and when I can beat them, I set it a little higher.


My idea is to eliminate the tedious and long process of fine tuning the AI to have a race like you want
Yeah I get that but it really isn't that tedious. With my method I can converge to a number in a handful of races.


In carreer for example, one race I had the AI at 105% and won by a little margin
the next race after that with the same Ai % I was last by a good 5 seconds...
Well that's a problem of consistency. So SMS need to get the AI to a fairly consistent level before we can do anything else, which Doug and team are working on.


Like I said:
1- it would be optional (so people like you could set the % exactly how they like)
2- it would be only in practice/qual, and not in race, as to not have rubberbanding
If it's optional I wouldn't be horribly opposed to it but if it doesn't apply to races then I don't quite see the point.

GenBrien
13-02-2018, 17:04
If it's optional I wouldn't be horribly opposed to it but if it doesn't apply to races then I don't quite see the point.

because, by the time you've done the practice + qual, then the AI would be set to the correct %



I've raced quite a lot career series already and not once have I seen such huge variance between different tracks in one championship.
mostly the differences are from dry to wet/ vice-versa

gcx
13-02-2018, 17:05
This is a never ending topic.

I think that AI level per track is an almost lazy approach to the problem, but it's one that could work, for very cheap (implementing this should be easy). But that doesn't solve the issues with track conditions (wet, cold, hot).

Having AI balanced across the board with global settings is the more courageous approach (it's what SMS is doing), as it requires tweaking the AI for all possible conditions, which must be thousands, if not hundreds of thousands, of possibilities.
We can see how this can fail unless the AIs are using the exact same physics as the players.
I don't buy the argument of simplified physics for AI being the problem (although it must be a challenge).
Teaching the AIs to drive like humans using real conditions seems like an incredible technical feat, I'm sure that AI research on the subject isn't that advanced yet (Doug can, and probably will, correct me).

Having a system that tries to challenge the player based on scenarios and adapts in real time seems smart, but it poses many game design issues. How do you reward a player who wishes to win no matter what?

What I would like to see is a big data approach. Store as much data (splines with velocity key frames...? Probably many other variables) as possible from real players (online races, single player races, leaderboard times), for all track/car/condition combinations and extrapolate a difficulty from that. It's the Google approach, it's working quite nicely for them. Sadly, they don't make racing sims.
If there is no data, then your race will be weird. The nice thing is, the more you try, the better the AI will get.

As for aggression, I think that it's pretty great with patch 4. Doug (and his team...?) delivered on that subjet. It's the best racing AI behaviour I've tried so far. It's just speed balancing that needs work (and I'm afraid it will never be good enough, but no other sim is). All the aggression problems I've encountered at 75 aggressivity were caused by the AI being too fast compared to me (damn you first corner on Daytona road course!!!).

Lastly, I find the argument of AIs being rabbits at the dog track to be nonsense, as they can go quicker than is physically possible. The real rabbits are on the leaderboards.

Bealdor
13-02-2018, 17:07
because, by the time you've done the practice + qual, then the AI would be set to the correct %



mostly the differences are from dry to wet/ vice-versa

That's basically the same IMO. How long are you typically practising a particular track in the rain compared to dry conditions?

GenBrien
13-02-2018, 17:17
That's basically the same IMO. How long are you typically practising a particular track in the rain compared to dry conditions?

what I meant is I can race for example 4 different career races @70% dry and finish top 3, but on wet I need to set it to 90% to get the same result
the problem perse is not me/the player but the differences in time from the AIs

Mind you, I dont have patch4, so I dont really know if its ok now

My point was that I can spend about 1 hour/ race to fine tune the strenght % to get them like I want, and its long

Schnizz58
13-02-2018, 17:23
because, by the time you've done the practice + qual, then the AI would be set to the correct %
The way I do it, the AI would be set to the correct % before the practice.

GenBrien
13-02-2018, 17:34
The way I do it, the AI would be set to the correct % before the practice.

-GroupC @ Lemans /2hours / clear skies

- GTE @Watkins Glenn/ 30min / light rain

- FR @ sneterton 200 / 20min / overcast than rain

how would you set the %, wihout doing any practice, and not be the 1st nor the last ,and having to fight for the whole duration of those races
Just for fun

I would like to have your trick because I would not know

Schnizz58
13-02-2018, 17:45
I already explained that.

Tar Heel
14-02-2018, 05:31
Be cool if you based AI lap speed on a players qualifying speed. Perhaps make the AI range +5% and -5% (or user adjustable %) from the players qualifying time on each track to sort of "Auto balance" the AI.

If no qualification is done, just use the current AI strength setting.

If they went for that though it would always put the player at a huge disadvantage if they sold out during qualifying. Either that or the player could take advantage and sand bag in qualifying for easy wins.

Honestly adaptive AI always sounds great on paper, but look at RaceRooms adaptive AI. You literally have to train the AI for each and every car/track combo. It becomes a huge chore and if you don't do it properly the game will just shoot the AI difficulty to the max and you'll have zero chance and competing unless you're an alien.

I'm not saying SMS have a perfect system, I actually think the skill gap from track to track is still too great even after patch 4, but I don't think an adaptive AI is the answer unless they find new ways to have the AI adjust to you.

I've taken the approach that some tracks I'll dominate and some I won't. Hopefully future updates will make the AI behave more consistently from track to track in the future.

Having said all that I think the AI is actually pretty good compared to some of my other favorite titles to play. I had some tight racing with them yesterday at COTA and Road America in the Formula C. Fingers crossed the AI continue with this upward trend.

yannara
14-02-2018, 05:33
With the AI knowingly at different speeds at any given track relative to actual %, is there a chance to have a track specific ai difficulty option. Or basically it remembers your level each track?

I see your pain, but in real life, some tracks are better for some drivers and some other tracks suits better for other drivers. Also, I noticed, that if you really practise a new track for entire 60min, it will give you results.

GenBrien
14-02-2018, 10:44
Either that or the player could take advantage and sand bag in qualifying for easy wins.



I'm always laughing really loud each time I read about difficulty in this game and how you could cheat your way :D

You can already do the whole career with 0% if you want, so ''cheating'' by slowing down or cheating by lowering the slider is really no different

Tar Heel
14-02-2018, 17:33
I'm always laughing really loud each time I read about difficulty in this game and how you could cheat your way :D

You can already do the whole career with 0% if you want, so ''cheating'' by slowing down or cheating by lowering the slider is really no different

Yeah you make a good point. I guess my issue with adaptive AI is more so on the premise that some AI will always be faster than your time. In my experience it just makes it almost impossible to win because if there is always an AI that will be faster than your best, chances are you would never win a race unless something goes wrong. I want a challenge from the AI and I feel adaptive AI in R3E for example is worse than just finding the right difficulty.

MaximusN
14-02-2018, 17:51
Yeah you make a good point. I guess my issue with adaptive AI is more so on the premise that some AI will always be faster than your time. In my experience it just makes it almost impossible to win because if there is always an AI that will be faster than your best, chances are you would never win a race unless something goes wrong. I want a challenge from the AI and I feel adaptive AI in R3E for example is worse than just finding the right difficulty.

Outside of career mode I don't need to win at all. There is only one winner and me winning every time is far from realistic. Fighting for the top 5 is just fine.

And in career mode it should not be a bad thing if I end up in 5th place. A lot of very good real-life racers would be happy with 4th or 5th in their car(because it's the best it can do, period). Let's not try to say everything lower than podium is meaningless, because that's a high expectancy approach to racing IMHO.

So yeah, I'd want an adaptive AI(not to be confused with horrible rubberband), that will give me a tough(but achievable) challenge every time. So I don't end up in front of the pack or chasing the back.

Tar Heel
14-02-2018, 19:17
Outside of career mode

I'd be in full agreement with you that winning wasn't everything, but in career mode with the way you have to place to move up, it kind of is.

I don't disagree with your ideology of winning not being everything. I spent 18 years of my life swimming. Yes winning is the goal but there were times I didn't win that I was still thrilled with my performance. I'd love to have that same approach to career mode and try to as much as possible, but I also have to have the AI set at at level that gives me a chance to win. An adaptive AI that was programmed to always have a certain amount of cars to be faster than me would mean I would probably be stuck and never advance in career mode.

Again all I'm saying is I personally haven't experienced an adaptive AI that works well so IMO unless SMS figured how to implement it properly an adaptive AI is not the answer.

MaximusN
14-02-2018, 20:17
I'd be in full agreement with you that winning wasn't everything, but in career mode with the way you have to place to move up, it kind of is.

I don't disagree with your ideology of winning not being everything. I spent 18 years of my life swimming. Yes winning is the goal but there were times I didn't win that I was still thrilled with my performance. I'd love to have that same approach to career mode and try to as much as possible, but I also have to have the AI set at at level that gives me a chance to win. An adaptive AI that was programmed to always have a certain amount of cars to be faster than me would mean I would probably be stuck and never advance in career mode.

Again all I'm saying is I personally haven't experienced an adaptive AI that works well so IMO unless SMS figured how to implement it properly an adaptive AI is not the answer.

Totally agree with that, but I kind of implicitly meant that career mode should let that 'need top 3 to continue' go a bit. You're only fooling yourself anyway, because you are in control of how fast the AI is. So people will set it so top 3 is easily achievable(which will translate to people winning most separate championships, I'm sure). And I admit I do this myself too. It's part of the reason I enjoy non-career races more, because a good fight in the midfield is actually fun, and not winning or being in the top 3 has no consequences. The only consequence is lots of fun, which is the most important reason why enjoy sims.

Tar Heel
14-02-2018, 23:37
Totally agree with that, but I kind of implicitly meant that career mode should let that 'need top 3 to continue' go a bit. You're only fooling yourself anyway, because you are in control of how fast the AI is. So people will set it so top 3 is easily achievable(which will translate to people winning most separate championships, I'm sure). And I admit I do this myself too. It's part of the reason I enjoy non-career races more, because a good fight in the midfield is actually fun, and not winning or being in the top 3 has no consequences. The only consequence is lots of fun, which is the most important reason why enjoy sims.

Totally agree! I wish it would be let go as well for the reasons you stated. It would be cool if the career mode was more immersive and opportunities to move up came in more ways than just placing top 3.