I'm curious how many people actually use Adaptive AI once they've been playing R3E for a significant period of time? I've been using the sim for about 9 months and had always used fixed AI. Had pretty well settled into mostly revolving around 100, depending on the car/track of course. Decided a couple of weeks ago to try the adaptive AI in a GT4 championship. Did 2x 20 minute races. Started 15th out of a 30 car field each race. I was quite surprised that over the course of time, I basically settled into finishing in the bottom 5 every race (sometimes would be last--10 or 15 seconds behind the second-to-to-last place car). Decided that just wasn't fun and decided to drop the adaptive AI experiment. But was curious is most people use fixed or adaptive?
There are several useful threads on training the adaptive AI to get the most out of it and get it 'up to speed' efficiently. For example: https://forum.sector3studios.com/index.php?threads/adaptive-ai-the-tool-how-to.12888/ My personal experience however is that once you have the AI fully trained you will find yourself running in the mid-field. This suits many people as they get very close racing and lots of competition. On the other hand, I like to set the AI such that if I drive at my best (and perhaps do a PB lap time during the race) I have a chance of winning. I find with the adaptive AI I will have great racing but little chance of winning (and it's possible to have great racing nearer the front too!).
I use fixed AI basically cause of the same reasons as Skidmark. But there is one (small) issue if you do a championship. At few tracks the AI is too fast and at few too slow. At most of the tracks, the AI is fine. As an example. You run a 10 rounds championship and at 7-8 rounds you finish in the same position range. At the other rounds you are far off that position range. If you use adaptive AI, than you have to train them, which takes a lot of time, which I (and many others) do not have and you have boring races. If the AI is trained well, than you run about mid pack. But, like Skidmark, I want to have a chance to win if I do a perfect race, which is nearly Impossible with adaptive AI. Solution would be to have an option for adaptive AI where you can choose if you want to race top 5, midpack or bottom; I assume that last option is not really relevant. Question off topic to the offline racers, who do championships without qualifying: Is the starting grid of the AI random or do the faster ones start at the front? Of course I could test that myself, but if some already knows it, I would be thankfull.
Thanks for the comments guys. Yeah, I decided to experiment with adaptive AI to see if it would automatically resolve the gaps in performance between me and AI at certain tracks. Increasingly I just going with the R3E Open Championship tool and configuring my championships as single events, and just using the tool to keep track of points. Would be nice of R3E could enhance the tool to allow one to edit parameters like AI strength, race length etc. for each individual race along the way (kind of like ACC/AMS2 do).
Indeed - I suggested exactly this in a post several years ago which would solve the problem that at one track I have to run at AI=92 and another AI=110... (although usually at 102 +/-2). I'm still hoping it will happen.
Being fairly new to this sim, I like and can see the merit in the adaptive AI but I think the speed multiplier (if there is one?) is set too high! It starts off ok as you learn a new track and the AI are well within your pace but as you get to your limit the AI seem to find another gear and end up being like 1 sec quicker than your best possible time! If they ran matching your best time, I think it would be great as everyone has different speed but to be out paced by 1 - 1.5 seconds is no contest, like everyone says, forced to run mid to back of the field! Can't they be pegged back a bit?
I did now 7 races with adaptive AI without any qualy and I started 11th out of a 21 field in FRJ. At all races (15 minutes races) I ended up between 3rd and 8th, so I could gain positions. At Moscow (somehow a good track for me) I even did the fastest lap. At the other 5 tracks (Monza, Hockenheim, Hungaro, Aragon, Imola, Mid Ohio) my fastest lap times were slower than the fastest laptimes of any AI driver in the field. At aragon even 1.5 seconds and fatsest AI driver fastest lap was 3 seconds faster. At the other tracks not that much, but I always was the one with the slowest fastest lap. How these races look like: I start better than AI (standing start) and gain about 2 positions. Whilst AI is not so good in traffic I can gain another 3-6 spots and when AI comes to normal speed (which is faster than mine) I mainly need to defend and lose touch to the car(s) in front of me, except they battle as well, then it depends how much they batle and I can catch up or do, at least, not lose too much ground. Or AI drivers make mistakes. At FRJ they "often" have bad corner exists. I think such kind of races are more fun as the way I did it before with fixed AI and qualy. The only disadvantage is, that the starting grid is quite random. So the championship is not as in real life where the same guys are running at the top. The championship, I'm runnign contains 20 rounds. 10 tracks 2 times. I'm a little bit curious how the 2nd part works, when AAI should have adapted a little bit.
I used to race with adaptive AI a lot, then switched to fixed AI for a while, but lately I returned to Adaptive. Both has its advantages and disadvantages. One problem with Adaptive AI may be that you may loose a feeling of how good/competitive you really are, especially when you don't run time trials, competitions or online races. It could be helpful that the equivalent AI strength chosen by the program is displayed somewhere, before and/or after a session. After a race there could also be some kind of change indication, e.g. 94-->96. This could offer some motivation to race, some pseudo career mode. Another idea is that instead of having an Adaptive AI the fixed AI values could be coloured depending on the json files already there, e.g. I select a track/car combo and then some AI strength values would be green, others yellow and the rest red (too slow/too fast). In this scenario you had to pick a value each race, ok, but you would have more influence on the fine tuning (do I want a quite easy race or a real challenge?), and you would see your actual level. Admittedly that woul work for single races, not for championships. On the other hand a championship with Adaptive AI is a rather strange thing, as - if the system is adapted - you have a bunch of similar races, regardless if you are better at some tracks.
Good ideas. For your last point. A "slider" for AAI would be fine where you can set if you want to run at the top of a championship or more mid pack. That races are similar is like in the real world except you do some crazy stuff like weight penalty and reversed grid like BTCC. That makes BTCC fun to watch, cause much action but if you have bad luck in race 1.... So maybe not always the best driver wins the championship in the end. But that's off topic
I started R3E with the adaptive AI as it seemed an attractive choice. But I could never find the balance besides many hours of training the AI. Same car/track combo, one race I am fast, next one I finish way way back last. This ping pong effect would go on and on. I gave up and settled with the fixed option and never looked back. So much fun. I know going into every race what I am getting from the AI. It feels far more natural on long stints. Yes with the fixed AI you may come across a track they fast. I compensate with an aggressive setup to make up time.
I see some weired things in general. Few tracks where AI is too fast or too slow. For few car classes it is the same. E.g. I'm averagely fast at portimao and gt3 at about 102%. With WTCC I could do 109%, whereas with similar car WTCR I also do about same as gt3. So when you switch a car on same track and you do not have trained AI, then learned values are taken for AAI from similar combinations, from what I understand. And another factor is time. Then you have a factor on which track you are fast as the track fits to you better than others. Setup is a factor. So it is getting difficult. I have found a fixed AI strenght for FRJ for my average fast tracks and racing makes fun. But then there are 30+ others classes I like..... How should I train the AI for each class in each track???? So what would help. Make the AI speed in each track more equal and Car class AI more equal. And then make a good algorithm, which can predict AI strenght for next Car/track combo well. Sorry for long and maybe confusing post.
But isn't that the whole point of the adaptive AI? Regardless of car or track, the AI should match (or be in the region of) your fastest time for that car/track combo, no matter how fast or slow you are? I just don't understand why they are up to 3 sec quicker than that? Doesn't make sense to me
I've been running championships with aai since discovering the sim, and I love it. That's one thing that keeps me sticking with RR. I never know what kind of race I'm going to get... I win some, I battle mid-pack on some, but I'm always in the hunt. I have yet to win a championship but it's come down to the last race several times, with me either having a narrow lead or within 3-10 points. I do win some, and I never have the field leave me, so I don't know what to say about that. I love aai and just racing without worrying about tuning ai strength... Makes it much more immersive to me
@Muzarati That is what makes me worry. My race at Oschersleben. At start I gained 2 spots, so I was 9th. The rest of the Race I just was overtaken. In the end I was 17th. My fastest lap was 3 Seconds slower than the slowest AI. The fastest AI was more than 4 Seconds faster.... I think that is a lack of data for that track. I never drove there with FRJ but I have dozens of races with FRJ on other tracks. But the AAI strenght was set by the game for that Car/track combo. But way too strong. But why? When I did another Championship with other class and fixed AI and Oschersleben was Part of it, I cannot remember that Oschersleben was one of my worst track. Anyway. My next race in my AAI Championship with FRJ was at Ningbo. Never drove there with FRJ before. I finished 2nd. I was about 7th (started 11 as Always) after few corners. And within the next 2 laps I could take over AIs and was 2nd. I could even Attack 1st and move away from 3rd. After 4 laps the 1st placed Driver escaped and won by 8 seconds, he Had free Road and did those Magic laptimes. In the end my fastest laptimes was about the 5th or 6th fastest. But why was I in the Mix at Ningbo (never drove there before with FRJ) and way Off the pace at Oschersleben? Just for your understanding. My Last fixed Championship with FRJ, which did right before my AAI Championship, was at 109% and I could mainly run top 5. So I'm not the slowest and, I think, I do quite consistent laptimes.
@vvPULSEvv I can somehow confirm that some of my races are like the ones you describe. But few are weired as described. So you use own Setups or default? Do you Drive consistent (no offense)? My Goal is to understand what values are taken by the game to Set AAI strenght. More transperency would help. It is clear, from what others Post, that If you would only run only one Car track combo, that the AAI would have adapted after about 3 Long enough races (10 laps?) and you would end Up running mid pack. But what values does the game Take when you go to a nee track, either with Same class or other class. I think I know Adaption does Not start at 80%. Anyway. At the end it is about fun. And doing a race where you only defend can be fun as well. I have done now 9 rounds of that Championship and rounds 11-20 are the same Tracks from rounds 1-10. So I will See how AAI has adapted. So maybe they not 2-3 Seconds faster than my fastest laptime and that Gap has closed a bit.
1) adaptive ai starts at level 80 2) do 4 to 5 races of 5 laps so that it adapts to your level I exclusively use ai adaptive
Initially it Starts with 80%. But it adapts also to other track or class/track Combi. And that is the point how this is precalculated. And 4-5 races with 5 laps with about 30 Car classes at about 50 tracks (some of them with multiple Layouts)... Much time required. So my points are. Few tracks have wrong AI Level and few classes have wrong AI level. What the reasons are??? What Parameters with which weight are used for AAI? Example. With class A I Drive 5 races with 5 laps at track 1. AI should be adapted quite well, let's say 100%. Then I Switch with class A to track 2, at which I already have good AAI strenght but for class B, let's say 103%. Does the Game use the 100% or 103% or a combination? And if you have driven many many combinations before, do newer ones count more for precalculation of AAI strenght for the next race? For sure it does Not start at 80%. Then it also comes into Play, that some cars are harder to drive but it does Not look as if it is the same for AI.
I always use adaptive AI as once it’s trained to recognize your level you can have some great close racing lap after lap.
I did my last AAI Adaption File reset years ago. And drove plenty of races, so it should be trained quite well. I switched Car classes quite often. Maybe that is a factor. Last race at Nürburgring my fastest lap was still 1 second slower than 2nd slowest. Even tough I finished 6th (started 11th). So a typical race again. Gained 7 spots after 1st corner. Then I could keep Up with top 4 for about 2 laps till they did not battle and then they drove away by 0.5-1 Seconds per lap. Driver behind me catched up. Short battle till he overtook me and drove away. I checked Interval times and at least the gap to 9th and 10th was not changing that much. So I seldom have races where I'm the attacker. Most work Like described above.
My understanding is the AAI will look at players times in the selected car/class at the selected track first. If there is no entries for that combo, then it will search for times in player and selected car/class in other tracks. To use AAI at a selected track, you need to have done at least 2 x reasonable length races at that track. This seems to work but like I mentioned earlier, for me the only issue I have is the AAI are just set too fast. From fiddling around with AI in other sims, I would imagine there is a speed multiplier in the AI files that says something like "AAIspeed=1.1" or something like that? If this was set closer to 1.0 and AAI pace was closer to the players top time, I think racing would be closer and better...IMHO