Fix AI Strength missreported
In the 1960s series I have noticed that at 100% AI strength the AI will post lap times much faster than historical lap times. For example they will get around 3:20 to 3:22 at Spa in dry conditions. This is 8 seconds faster than the times posted at the real Spa F1 GP in 1967 and 1968, the era the 60s F1 cars are based on.
When adjusted to 95% AI strength (the default) the AI are getting 3:27 to 3:29 at the same track. I assumed that 100% would equal normal real life times, instead I find that the default 95% is realistic. For ages I was assuming that my 3:26 was 5 seconds of the real pace of F1 1968 times. It turns out I was 2 seconds faster.
Now this might sound like a trivial request, after all 95% is the default setting but I feel that the 95% setting should be renamed to 100% for logical reasons. When I read 100% in this context I take that to mean actual real life lap times and speeds.
I just think it would make things easier to understand when setting up races.
Last edited by ICDP; 07-15-12 at 07:15 AM.
Trouble is the AI is dependant on so many things; the quality of the tracks AIW, the talent files, other individual AI settings. Simply renaming 95 to 100 might cause problems at another track where suddenly the AI are too good at 90 for example.
Personally I just just the Strength as a guide and dont take 100% as Gospel/Perfect, whatever setting suits me is the right setting.
What is needed (and I think has been Wished before) is to have AI Strength changeable on the fly whilst at the track.
I agree with ICDP. It can easily be changed. I assume AIW for ISI spa and monaco are good so, and although there are various talent files, we can always take the fastest one AI; this one fastest should be able to do close to real time (with some groove on track, dry sunny).
Overall AI seems a bit too fast anyway, its the same at monaco and monza, so yeah, make them few seconds slower at 100% setting and it will be more realistic imo.
You can change the AI strength "on the fly" during a race weekend from the garage in options. You need to click "restart weekend" for it to take affect though.
Originally Posted by MarcG
aahh, I did notice that but presumed it did'nt work as I never tried to restart the weekend. That then is excellent, thanks for the heads up
Originally Posted by buddhatree
It seems that the AI strength varies quite a bit from track to track. At Monaco 1966 the AI set at 95% are getting 1:22 - 1:24, a full 6 seconds faster than the real drivers got in 1967/68. So this means the AI need individually adjusted for each track.
Originally Posted by osella
EDIT: I just tried adjusting the AI strength for testing and it no longer has any effect whatsoever on AI lap times. I haven't tested since build 85 as I found the 95% setting to be closest to reality.
Can someone else please test and verify this?
thats exactly my point, changing the strength to suit one track wont suit another so IMO theres no point messing with it. AIW is a big player in the whole process and its something that needs to be extracted from the MAS file so it can be edited easily enough by anyone as unfortunatley some tracks will be released with poor AIW......already wished something long those lines before anyway!
Originally Posted by ICDP
I disagree, the community should not have to extract and edit core files to fix official content. If the AI strength is set at 100% then that 100% should = reasonably accurate times for all tracks and classes. The developers should not leave it for the users to adjust the AI every time they load a new track or series. Let alone extract MAS files and edit them.
Originally Posted by MarcG
IMHO It is not conducive to a good racing experience if each time we load a new track/series we have to test and adjust the AI until the lap times they get are accurate. Personally I believe the AI using official tracks and cars should be getting somewhat accurate times at 100%, not 95% (or any other percentage). If an individual want to adjust the AI up or down to suit their skill level then fair play to them. I still think 100% should be a base reference roughly = to real life for that particular track/car/conditions combination.
For example. At Monaco 1966 the F1 AI are around 6 seconds too fast compared to the real GP drivers of the era. Without researching the actual lap times achieved it would be safe to assume that 100% AI at Monaco were accurate. With lap times at 1:22 they clearly are not realistic.
I guess I am trying to say that for me the real life lap times are a yardstick I aim for when I purchase/drive any racing sim.
When playing rFactor2, to get an idea of what the real life lap times are for a track/car class I have to search the internet. Then I have to adjust the AI strength and adjust until it is just right. Then I can get an idea of what yardstick to aim for. This then has to be repeated for each track/car class.
NOTE: I am not discussing car physics or speeds, just the AI strength setting. I just think some logic, accuracy and consistency should be applied to the AI strength setting for official content.
Last edited by ICDP; 07-15-12 at 03:11 PM.
Yeah sorry I was leaning more towards user created tracks with my post, agreed that official content should be as close as possible but again it'd be hard to do so I think especially if you have a lot of tracks to take into account.
Its another argument for Mod Specific AI that I spoke about here:
that would help get the AI cars closer to reality as the AI would be based on that mod and not a basic form which it currently is.
It does not matter whether we talk about official or user created tracks, currently the AI is just too fast, period. Whether it's accurate on user created track X or Y doesn't matter, since on official tracks the AI racing line is correct, it is supposed to do real laptimes on those tracks.
If they are 6seconds faster at monaco, that's seriously wrong. 6 seconds at monaco is difference like between no downforce and moderate downforce car. Why not make them 6seconds slower at monaco - you might say that then they might be 2 or 3 seconds slower than real at spa, so what? Still much better than now.
In rF1, many user mods are able to achieve incredibly realistic times on most tracks where their reallife counterparts raced so it must be possible in rf2 too.
Not saying it must be fixed in the next build - woudln't be surprised if AI has currently low priority - but it must be in final.
Last edited by osella; 07-15-12 at 06:39 PM.
But you can not compare IRL laptimes to rfactor2 laptimes? I mean 100% strengh should mean the AI have speed and skill that reflects reality. Not laptimes.
If the track is not 100% correct (maybe some incorrect corners make the track slightly longer) and the handling and grip ingame is not 100% like it was when they did that laptimes in 1966, well, then you can not adjust 100% AI strengh from lap times.
Anyway, I'm racing the AI at 105-110% and they do certainly not make 3:20 times. They will do 3:25 - 3:28. If their laptimes where limited by real-life laptimes, then they would not be very competitive.
Am I the only one that see the absurd in this wish?
The AI strength is broken in build 90, I mentioned this earlier in the thread. The AI get the same laptimes no matter what I set strength to. Also the idea that AI should match speed and skill and not actual laptimes is a non sequiter. Ideally if the AI speed and skill are correct for a given track/car/conditions combination, then the laptimes should by definition, be accurate.
Originally Posted by Hedlund_90
If the AI are wrong, it will result in unrealistic lap times.
If the car is wrong, it will result in unrealistic lap times.
If the track is wrong, it will result in unrealistic lap times.
If rFactor2 is released with one or more of these areas being significantly wrong then the results will be an unrealistic simulation of motor racing.
EDIT: Unless I am mistaken, 100% in the context of simulations = accurate based on real life data. Why should the AI strength be any different. As a total newcomer to rFactor2 it is logical to assume that the AI set at 100% should = realistic laptimes, regardless of track/car selected.
So, no it is not an absurd notion to ask that 100% = as accurate as possible based on real data.
Last edited by ICDP; 07-16-12 at 02:03 PM.
On the topic of AI: Does anyone observe what seems to be the AI's ability to brake a little *too* well in some mods? I've been able to observe this most blatantly in the 370Z. Or, is this simply an indication of me being way off on the setup? Can't seem to get a stable enough setup under braking to hold a candle to these guys..