I've been spending a lot more time concentrating on the AI lately since the radically increased raciness and assertion has really highlighted the way the AI are unfairly advantaged relative to the player. Obviously they run on a simplified set of parameters relative to the player, and tweaking these parameters to give the illusion of human like competence must be an enormous task. So, according to my observations, are AI affected by: Driving off the ideal racing line: apparently not Rain/wet track: more than player Puddles: less than player Kerbs: less than player 2 wheels on grass: not at all 4 wheels on grass: much much less than player Penalties: apparently not Collisions: much less than player Mechanical damage: much less than player Tyre wear: apparently not Track/tyre temperature: apparently not Fuel load: apparently not Reduced visibility (heavy fog/darkness): apparently not Anyone else got observations to contribute? I'm happy to be proven wrong on any of the above! I'm heavily impressed by how well AMS2 is developing, especially regarding the physics and driving feel - which is just sublime. But the good feelings evaporate when rage quitting against AI who just don't seem to be constrained by the same things as the player.
Driving off the ideal racing line: apparently not; Agreed Rain/wet track: more than player Scaling levels and Track dependant Mostly agreed Puddles: less than player: Agreed Kerbs: less than player: Agreed 2 wheels on grass: not at all. :Agreed 4 wheels on grass: much much less than player: Agreed Penalties: apparently not: Disagree I had them disqualified. Dont Know what for Collisions: much less than player:Agreed Their Hardy/Sturdy in this respect Mechanical damage: much less than player: Mostly run visual (for sake of on online early quiters only so same as me. Tyre wear: apparently not: Agreed Track/tyre temperature: apparently not: Agreed Fuel load: apparently not: Agreed lol Also they have accurate gauges I ran out of fuel 100 mtr from finish last night in league race 21laps and 22in tank Reduced visibility (heavy fog/darkness): apparently not. Agreed probably in mid 20's age group Agreed Short comment: picture/vid/ speaks a thousand words, like indisputable "4 wheels on grass" (other thread) Tricky parameters for devs to get right. Constant adjustments are working, swings and roundabouts some times as expected. I would add Effected/React to Flags to your survey My answer would be not affected
Broadly agree with all of that. I have to say I'm not feeling much affection for the AI in this game at the moment. The driving side of the game has blossomed beautifully over the past few months and the work done on the physics and FFB has delivered a strong range of different types of car that are now very distinct and enjoyable to drive. The AI, imo, is now the weakest part of the game. Whilst there's a lot of fine tuning of cars performance at each track to be done (captured in the official AI thread), for me, I think it's mainly the lack of individuality and errors that makes them unenjoyable to race against. Most races I have against them, they just stick together as a pack for the entire race and other than a few overtakes amongst themselves, nothing much ever happens, no spins, no crashes, no overshooting breaking points, no mechanical failures - just the entire pack finishing as a bunch, separated by a few seconds from 1st to last. This possibly varies depending upon which class you drive as I know some folk have reported AI carnage (eg. Formula Retro). And they can race 2 or 3 abreast for lap after lap with none of them making an error - yet they'll happily collide with your car as if it's not there and knock you off. Which brings me onto stability under contact, where levels of contact that result in the user's car being sent into a spin and/or punted off track don't seem to affect the AI cars. The only positive thing I remember about the AI in PCars2 (I'm assuming the base AI in AMS2 comes as part of the Madness engine, and so started as the same AI PCars2 was built using) was that they seemed to be equally as vulnerable to losing control under contact as the user's car was. I don't see that vulnerability in most of the AI cars in AMS2, other than perhaps the Opala 86, which is quite easy to accidentally turn round with rear contact. It's like they've been given an artificial boost in stability and grip, which just adds to the feeling of them as being bots with an unfair advantage. I know Reiza mentioned breakdown of previously global parameters to allow car-specific tuning of various aspects which comprise their overall performance, so hopefully there's plenty more that can be done to create more individuality and vulnerability. The AI in AMS1 set quite a high benchmark against which to be measured but I still have hope that it can be achieved again (albeit with a different engine and those talent files were a big factor).
I did a race last night at spa in the 2021 stockcar and another with the sauber C9 with fuel use at x4 and I had to pit after 4 laps with the stock car or I would have runout of fuel and the AI were amazingly able to do an extra lap before Pitting, no idea how they did it.
Do you think it depends upon the car? The 2017 Stock car series seems to be very realistic. I don't pay attention to the lines that the AI drive, but it does seem that they slow down a bit (tire wear?) over a long race.. also, one AI pitted after the pits were closed and I "think" he was DQ'd but I don't remember for sure. Overall they seemed realistic. I really haven't played any other of the cars other than a little GT3 which also seemed realistic to me. I had to turn their aggression down to 50% in order to have them not wreck me or other AI lol.
Yesterday I did the first Stock Car 2017 Championship race at Goiana, 100% strenght and 50% aggression. While I managed to qualify 3rd, the race was a different story. AI cars trying to overtake everywhere, going by weird racing lines, grass, etc. without losing speed and/or car control, groups of AI cars racing side by side the entire race, AI punting you from behind, etc. Generally, I experienced the same problems as discussed in this thread. I didn't test tyre wear and fuel useage because I rage quitted 20 minutes in the race because the described problems, but at race start my lap times were 1 to 2 secs slower than qualy with the tank full of fuel, while the AI lapped all the race at or fastest than their Q lap time, so not penalized by fuel weight it seems. May be it would be better at 90/95 strenght, but there is no point to do that because I will always qualify 1st. and be faster on Race, avoiding close racing the AI, but it would be boring as hell and not a challenge at all. @tpw @Maser V6 I 100% agree with your comments.
Yep, something is definitely amiss when it comes to stuff affecting the AI. I just did a 45 minutes race in the Gen 1 Formula Classic, and while I was competive in practice and qualy (top 5), I got overtaken left and right during the race. After looking at lap times, the AI was basically doing the same times as in qualy... with 150Kg of fuel more... On such light car this just not realistic. Ever more damning is the look on the AI telemetry. The green lap is the AI best qualy lap. The red lap is the third lap in the race, of the same driver. So, in Donington the AI is able to take the Old hairpin at the same speed, in race, with 150Kg extra, as in qualy. This next chart is the speed difference between those two lap plotted over a lap. During qualy, the AI was actually slower in braking zones, than in race. So yeah, running AI, atleast with fuel consumption, looks to be pointless. And I'm definitely turning it off.
