Road to V1.6 Hype Train (cont.)

I do think at least that, with the latest revisions, they've been able to apply some tools to tackle car physics that they previously didn't have as good of an understanding of. That allows them to improve some cars that have exhibited problematic behaviours for a very long time.

There are several cars in the current public release that don't feel quite as good as others (to me for example the Caterham 620R and to a lesser extent the other Caterhams, or the BMW 2002 Turbo), and they are attempting to bring every single one of them up to standard. However, it appears to be quite a time-consuming process, and it's a process that I'm pretty sure was not planned to happen at all initially, back when they couldn't be sure how effective the hysteresis revisions were going to be.

It's for sure a grueling wait by now, however I think it's not unreasonable to hope that it's the last time we see as much groundwork on physics being done.

 

Compared to project cars 2 the physic/ FFB is in a other League !

 

After doing a number of races against AI my impressions of AMS2 1.6 are: less wtf moments, more fun races.

 

One annoying situation with AI is the difference in skill and/or speed between one track and another. I use Brazilian stock cars more. Lastly, I did the 2024 championship, which was good, but on some tracks, the AI at 110% was good, but on others the AI was much slower. I noticed that on some tracks I have to put it at 112%, others at 109%. This is something that gets in the way of a championship. In your opinion, with the 1.6, do you think this type of situation has been solved, or at least alleviated?

 

It's heavily WIP. Because the physics are still being adjusted, AI calibration is a moving target and not dialed in fully. I expect it to not be perfect for release because for some classes, they will have seen a couple days of AI calibration (if even that) and nothing more when 1.6 is released.

Additionally, AI and physics developers overlap somewhat (e.g. Renato does both), so that's another thing that makes it difficult, especially because AI calibration is very time consuming with the myriad of combinations in the game.

I expect that there will be some progress on this for release - it will definitely not be worse than it is currently in public release - but I also don't think it will be spot on for all classes and tracks; otherwise we'd be talking a 2026 release

With the physics hopefully not requiring as much work after the 1.6 release, I would think that there will be swift progress with AI calibration after it's released.

 

There is indeed a lot of effort going into better balancing AI strength from track to track, and also from car to car, and has been since at least the v1.5 cycle actually. With over 40000 car/track combinations though it's no small task and I expect further tuning will be needed beyond v1.6.

 

I'd actually been considering trying to build some kind of cheeky ML tool that collects observations of users with regard to desired difficulty and "true" difficulty (i.e. the difficulty at which they are comfortable. Then when you want to try a track and car combo, you can plug in your desired difficulty and combo and it'll output the real AI setting you should choose. It'd be a simple model if you could get enough people to respond but... when is that ever going to happen? Plus it'd be useful for the team for finding the worst offenders.

 

I think we have to take into account that we ourselves are not the same percentage strength on each track so even with properly calibrated AI they may appear faster or slower on different tracks relative to us.

I find it more annoying with certain classes that the AI are faster in a race than they are in qualifying which makes setting the AI strength a real headache.

 

I find it annoying the fact that they can brake later in some corners than the player in the same car (chicane at Daytono RC, 2nd chicane at Lemans Etc)

 

And at the same time they are super slow on other corners. Velopark S turn for example.

 

Once again a combo I can essentially replicated in AMS2!

 

This doesn’t sound too optimistic but I appreciate your honest and realistic take on the current state. My expectations for 1.6 have lowered a lot by that, especially since as a single-player user, I was mainly hoping for a cohesive AI experience including working AI in pit stop strategy, realistic AI behaviour in wet races, working AI damage simulation... I really hope that at least new physics will finally be solid as promised. With 1.5 and older updates that were presented as gamechanging, only for more physics changes to be announced shortly after I am sceptical.

 

This is a great example, at Velopark I was one second ahead of the second placed driver, while at Buenos Aires No. 8 I was one second behind the last placed driver, so there really is a lot of disparity in the AI's skill. And I can't help but compare it to Ams 1, where I was at 102% with the 2017 Brazilian stock car. All the races were competitive for me, including on these same tracks, and the times were very close to real life! (only the Cascavel track is not close to real time)

 

Don't get me wrong, AI are a focus of 1.6, but I reckon there's more groundwork being done than actual finetuning - as of yet anyways. Some more annoying discrepancies have been removed, but there's no way they can tackle them all. Of course, if they want to, they can choose to spend another two months calibrating to perfection before releasing 1.6, but let's hope for all our sanity they don't

As for physics:

 

ok I feel better

 

So as I understand it, the AI must be calibrated for every car-track combination, and that cannot be properly completed until the physics are set in stone?

In that case, it explains so much on why the update is delayed to this point, but my question is, is this sustainable for the developers, as more and more content is added to the game?

 

...or wouldn't it be better to develop and implement something similar to Raceroom's adaptive AI? It works very well for me, and they also have a wide variety of cars and tracks, much like AMS2.

 

For sure it would be better. I doubt it's realistic though, otherwise we would probably have hints of it from Reiza.

 

I would think that with 1.6 the physics will be 99% complete and no need for a complete overhaul of the AI calibration in future. Although that has been said with past updates...

 

I guess it makes more sense that the update that will add the career or progression mode will be the one where the AI will be fully refined. To me the inconsistency from track to track is no problem. With a few minutes of practice (if I know the circuit) I can already tell if I need to adjust it, but honestly rarely do it, I find it ok that in some circuits I am for example competing for P5 and in others for p10... It doesn't need to be always exactly the same pace. My problem comes when the AI is considerably inconsistent in different parts of the same track