Automobilista 2 V1.4.3.3 & Racin´ USA Pt3 Officially Released!

Tremendous simulation level, which pilots prefer to play it on another SIM, congratulations on your Feed!

 

So you come in here, complain about something, someone gives you a solution, then you complain about having to do that solution.

It's fine to criticize the game. FCY's are clearly unfinished and IMHO shouldn't have been released in its current state, the max number of cars per track is incredibly small and the ai is inconsistent and doesn't feel as natural as it could be.

It's about how you come across. If it sounds like you're posting just to complain about something you don't like instead of genuine wanting it to be an Improvement. It really doesn't seem like that, when ever I post problems and errors with the game I post my specs. And if ever asked I'll post my FFB settings.

Also spoiler alert here but if you're critism is that AMS 2 should have better default FFB settings then a lot of folks here agree with you. It's all about how you come across, if the language you use indicates you don't want to enjoy AMS 2 (regardless of what you may actually think) then that's what people are going to think.

We can't read minds, and we are also not perfect but you've had a lot of opportunities to share your FFB settings and prove that you're actually willing to see if it improves your experience. But you haven't, and all that says to us is that you don't want to try to improve your experience.

 

Worth noting that the more physics you lend the A.i., the more CPU not only to run the physics but the more CPU to have the A.i. make appropriate decisions. I think ACC is a good example, I find the A.i. somewhat 'boring'. Because if they were to dance about like in AMS2 you'd need a CPU thread per car (at least). AMS2 A.i. definitely needs work (A.i. cars braking on straights is annoying instead of making a pass) but indeed that would be a lot of re-work if you intend to update the physics of the engine. I personally think they are fit for purpose for now. If Reiza were allowed to make it easier for modding (my understanding is their Madness Engine license restricts this), the community could attempt to solve it. Just my subjective opinion.

 

Yes right,but all cryed "give as ovals"- no ovalracing possible without FCY - Now we got Ovals and a not so perfect W.I.P FCY.It's a process like adjusting the AI.Takes time and can only get better.
And you can switch it off.Isn't that great?
Btw: Reiza implemented this by themself.

 

I absolutely agree.

But then we get to the question: can this simplified physics make the AI reach a level where it can at least look more natural and closer to what the player is subjected to? If not, is the trade off worth it?

Doesn't mean that if AMS2 had simpler physics the AI would be any better though. As stated in this conversation, ACC AI has the same physics as the player and it is even worse than an already inconsistent AMS2 AI.

Does iRacing's AI follow the same physics of the player? Or is it also simplified? Since it is being dubbed as "the best AI" out there, it could show the way where proper AI should follow. What is iRacing doing that makes their AI great?

 

They release it very slowly, a few car&track combinations at a time, which gives them lots of time to calibrate it. If Reiza worked for 13 weeks on the AI of the Porsche Cup at Donington they would have it just as perfect. All that subscription & item sale money probably doesn't hurt either, maybe they even have somebody dedicated to developing the AI?

 

I mean there are obviously problems with it and that isn't to say I don't respect the devs for the work they've put into it. But I think there are enough problems where I would've rather them waited on pushing it out. I would've been fine with ovals not having them as long as it's a planned addition

 

Agreed, slow release cycle + oodles of money to invest due to their <controversial> disgusting </controversial> pricing model. It's almost certainly a simplified physics model, just with a lot more dev resources poured into refining it.

 

Did a back to back test with all the Stock cars today, they all feel much the same aside from the subtle differences in performance, ( mostly aero and mechanical grip differences)
The 2019 (2017*) has a lot more grip due to its Soft tyres , and feels more planted because of it , it’s configuration is the faster of all the stock cars , the newer ones are a bit more edgy . But that’s how the are in comparison to older regulations by what we are told.

Stick the hard tyres on the 2019 and it’s not a lot different really.
They all feel a little snappy on default setup’s, I tend to reduce the amount of clutches in diff settings to make them more predictable.
The driving position and fov takes a bit of get used to , you definitely wouldn’t want to much camera movement and some aspects of scale can fool the mind, the rear view mirror seems extremely large in relation to the rest of the cockpit for example.

But overall they all drive similar.

 

iRacing's AI is based on "big data" or machine learning. It's a unique approach that takes advantage of the vehicle tracking in their online system to "learn" the racing lines and approaches of their thousands of customers...and then "train" the AI to mimic the same. It is by definition the most naturalistic AI because it is based on probabilistically copying what the humans do when, e.g., they are at the same corner of the same track in the same car.

This is an innovative approach that requires a massive effort up-front, which is why the AI were rolled-out in dribs and drabs instead of across-the-board as in all other sims. Of course, if the starting point was many years of no AI, it is a bit more marketable to take this approach than for other sims. But other sims don't have the competition data to work from in the first place, so kudos to iRacing for thinking outside the box.

In terms of whether traditional sim AI can be naturalistic, they answer is yes, but it also takes a ton of effort tweaking and massaging both global and track- and car-specific settings. This task is made more complex when dynamic weather and track grip is added into the equation.

When resources are limited, you cannot attack all aspects of the problem at once. The solution will be iterative, with some global settings, some track settings and some car settings each successively contributing to an eventual finished product. For example, there is still a tendency for AI to put two wheels on the grass too frequently. Is that a global issue or specific to those areas of tracks where it occurs or does the type of car also affect it (or does every series and vehicle do the same thing at the affected areas)...or is it a combination of all three?

Oh, and when all of this is figured-out, it has to be scalable in terms of speed and aggression, so we can all have an experience tuned to our personal needs and skill levels

 

Fair point.

Not only the money influx and the size of the company help them with that, but the great MP refrain users like myself from complaining about terrible AI, or about the slow development.

I think AMS2 is short on both ends (SP and MP), hence why all the whining (myself included). Must be truly annoying for Renato and the rest of the team, but it is what it is.

 

An approach that was used in rF2 back when ISI was still running the show... Albeit without all of the online data... There was a point before S397 porked the AI that rF2's AI was top notch and learned from itself and the player to create nice real track grip lines instead of the tram tracks that rF2's AI originally created... Sure it would take a couple of hours of on track time from the player, but it made modded tracks come alive with the AI after those couple of hours... No matter how bad the AI paths were to begin with...

It'd be nice if Reiza could implement something similar for AMS2... There's probably some limitation within the Madness engine that stops this, as Reiza was working with ISI when rF2 had the learning AI so may have already had some insight into how that worked...

 

They may have used machine learning (A.i.) to design and help refine the algorithmic parameters but I'd be very surprised if they're using it during the race. You'd need a GPU (or tensor-based accelerator) per opponent. Unless their car A.i' is cloud based and consequently would be rather expensive to run. I would have expected them to charge extra for such a feature to cover costs.

 

The iRacing machine learning is not related at all to the old AI training from ISI. That was a much simpler process of refining the AI path based on repeated passes by the AI itself (with whatever physics limitations were inherent).

And, I don't think S397 intended to ruin anything. rF2 hit the same issues versus rF that AMS 2 has versus AMS--dynamic weather and track make things much more difficult. Even rF2's more limited level of "dynamic" doesn't prevent the necessity to do the difficult work behind the scenes to calibrate and refine the AI behaviour.

 

The machine learning is done in the background and in advance. Instead of one optimal racing line like most titles (or one line with some associated secondary lines), you have many (how many is probably a trade secret) to choose from...dynamically as the conditions warrant. And those lines are based on "good" quality racing as practised by thousands of people over many thousands of laps, so known to be useful and successful rather than based on a theoretical optimum.

 

It's still machine based learning... Same concept, different approach...

S397 didn't intend on ruining anything, it's just what happened... They thought they could do better than ISI and were wrong... Nothing wrong with trying, just have to live with the results...

 

I've noticed a little unseasonal Christmas grumpiness creeping in again. So, at the risk of trolling AC, I invite you to gaze in awe at sim-physics thrown heedlessly to the wind.
Glorious!

 

RF 2 is my main sim, not sure what you mean by the ai is ruined. I let the ai practice with me, and then qualify and the ensuing race is almost always excellent. I take it a step further by making my own custom ai grids but I do that because I want to, not because I have to.

The best thing about the rF 2 ai is that it seems to know I’m there and races respectfully. It seems like such a simple thing but apparently it’s hard because most other games mess it up.

There WAS a more or less visual only bug with ai brake and throttle application but it was fixed in the last patch.

 

S397 ruined what ISI had and brought in all that jerkiness and on/off brake and accelerator during the corners that bugged through their updates... That even after that November update has remained within it for users who have made complaints about the AI post that update on forums like RD... It was more than just a visual bug as it affected the audio as well...

Before S397 got their hands on it I would of agreed with your statement that the AI knew where you were, in a very similar fashion to where AMS2 is now with certain cars and tracks... But during the S397 era that wasn't the case until this year, and even then it hasn't been fully stamped out...

 

Holy **** the audio on the majority of those cars is horrible.