My thoughts regarding AI, FFB, weather and what we can learn from other sims

(Disclaimer: I started writing this post two days ago, which was right before the release of AMS2 0.9.3.1, so some information may be based on the previous version - please keep that in mind. What I wrote also comes from a place of love, and despite pointing out some flaws still currently present in AMS2 I am truly in love with what Reiza has achieved so far!)

As some of you may have noticed, I don't post here very often. My motto is "contribute something new or unique or keep your mouth shut". Most of the time other people have already elaborated on things at least as well or even far better than I ever could, so I stick to giving those posts some upvotes (like, agree, helpful, informative, etc.).

Now what has changed? Why am I posting this? Why should anyone care? Well, I can't tell you the answer to the last one, but as for the others I recently tried Assetto Corsa Competizione for the first time, and this has got me thinking and reflecting.

After trying ACC for a couple of hours I am, for the most part, impressed. Other parts... not so much. Now before you think this is going to turn into a SIM X vs SIM Y kind of thread, please don't reach for your pitchforks yet. I am merely trying to explain why I came to certain thoughts and conclusions.

What impressed me about ACC, and actually even more so after getting used to it, is the FFB. After dialing the correct settings for my G25 wheel in I was surprised how similar in terms of general feel it is to AMS2. IMHO this is a compliment for both sims - especially for AMS2 since it is still in Early Access and the FFB has not been fully finalized yet (though the current beta is pretty close from what I can gather). ACC, on the other hand, has seen about a year's worth of updates post-release. So ranking close to a released and refined sim - or even surpassing it - is a great achievement for AMS2 considering it's not even at 1.0 yet!

Going more into detail: Though the forces are generally the same between both sims' FFB, ACC seems more muted and 'refined' for an experience that's not as exhausting, whereas AMS2 delivers more raw and visceral feedback. As a consequence I'd say that AMS2 FFB is a little more tiring in the long run while at the same time delivering slightly more information. In AMS2 I can quickly feel when the rear is stepping out, whereas in ACC that's not the case before the rear is already coming around. Considering that the FFB had always been the bane of PCARS 1 and 2 players it's safe to say that what Reiza have done with the engine so far is impressive.

(Update post 0.9.3.1 - The FFB seems to be even closer to ACC now. Unfortunately it feels as though some of the detailed information has been lost along the way.)


The second thing that really surprised me was how aware of and reactive to the player the ACC AI is. Since I had no idea how quick or slow the AI was I started at the lowest setting (80 out of 100, I believe) and am now slowly working my way up to a setting that is closer to my own pace. In any case, when doing hotlaps during practice or qualifying, I notice the AI actively moving from the racing line to let me pass. In eight hours of playing* I've made my share of mistakes, particularly when learning individual car/track combos, but I've only had a single contact with an AI vehicle when I bumped into its rear end (and it was mostly my fault for not braking a moment sooner!). Other than that the AI gave me a super clean racing experience. I think in all the years I've been playing racing games - not just pure simulations - which amounts to almost thirty years of experience, I can not remember a single game that provided such a clean and fair AI, at least when it comes to awareness of the player.

*I'm now at 11 hours and have had a practice session where the AI seemed hell-bent on ruining my laps. Whiskey Tango Foxtrot?

That being said, from what I have seen - or rather, haven't - the AI in ACC doesn't do much fighting. I don't recall seeing any passing moves aside from the first few turns of the first lap maybe. So why would having a clean, defensive opponent AI be a good idea?

The idea is simple: Which players do you think are more likely to adopt clean, fair behavior themselves (with or without visible safety rating)? Those who do offline racing against psychotic AI that plough through you every other braking zone, or those who practice racing against an aware AI that teaches them how to race with others while avoiding unnecessary accidents?

I know there could be a lot more said about the behavior of AI - and I personally would love to see more human behavior, like more inconsistency for low-strength AI, different AI personalities, different strengths and weaknesses per AI driver, etc... but that would easily take over the entire conversation and not come to a conclusive result.


One other point I would like with regards to AI is how important it is to be able to adjust their pace over a wide range to accommodate an equally large number of different player levels. When reading through this forum one could come to the conclusion that everyone is an alien since so many have said that even at 120% the AMS2 AI doesn't provide an adequate challenge for them. However, recently I've seen a number of others point out - rightfully so - that the AI currently possesses inhuman ability and, in some instances, is still too fast even with their skill set to 0. I know I've seen this myself.

What fills me with hope as someone who exclusively races offline, is what Renato has said about the topic:

So it looks like this issue will mostly be solved once the proper AI logic is done.

One thing I am still a little worried about is AI performance in dry vs. wet conditions. IMHO one pitfall of many sim developers seems to be that the AI severely outpaces the player in the rain. This can have a wide range of reasons - simplified AI physics are certainly one factor, but I assume that the psychological factor is another big one. And last but not least visibility. Under wet conditions, particularly when it is raining heavily and there's a lot of standing water on the track, a human race driver is more likely to stick to 99% or 98% of their possible pace in the wet, just to be sure. Unexpected occurrences aren't unheard of in motorsports, and while those are bad enough under optimal conditions they are certainly something you'd want to avoid when it's pouring like crazy.

From what little I have been able to gather in ACC, so far, is that I found the pace of the AI in both conditions to be roughly the same with regards to my own pace.

Just for some comparison I also did a quick 10-minute race in the sprint car in AMS2. I chose a track I raced a lot recently and know fairly well, which is Brands Hatch GP. Conditions were Medium Rain, and the AI was at 70% by default. As expected based on my previous experience I didn't stand a chance. AI lap times were varying wildly (as is expected in the rain), mostly somewhere between a little and several seconds faster per lap than my best lap time. It didn't help that the player starting position was still set to "last place" by default.

Next try: AI down to 0% - same conditions, but starting in the middle of the pack. This time I could somewhat keep the pace with the midfield, but once the initial first lap chaos had subsided, overtaking was out of the question in these conditions. The AI in front of me was slowly pulling away despite my best efforts, whereas the ones behind me kept pressuring me.

So afterwards I wondered what it would look like in dry conditions? Same AI strength and same settings as before:
This time I could more easily retain the pace of the mid field and slowly gain a few places. Probably would've made my way half-way to the podium if I hadn't made a mistake.


This was all before 0.9.3.1 was released, so I did a few additional test runs today.

This time: Ultima GTR Race Car, Jerez (no chicane), AI at 100% / 50. 15 minutes, three weather slots scaled to session: Light Cloud, Random, Random. Player starting in the middle.
This one was quite interesting. The AI once again slowly pulled away post-start, but things got really crazy when the weather suddenly changed to Rain and Heavy Fog around minute 4 or 5. Within half a lap the track was flooded and I could barely keep the car on track (still on slicks). I pitted first chance I got but had still lost considerable time on the AI. Around minute 9 or so it stopped raining, but since the track was still rather wet I decided to stay on wets.

I decided to take a closer look at what the AI opponents were doing in replay post-race, and that was quite enlightening...:
Right from the start the leading AIs were braking far too late before several turns, and also carrying too much speed through some of the corners as well. One that stood out in particular was the approach to turn 8, (Curva Aspar I think), where AI cars not only started braking far to late but also managed to brake far into the corner. Try that type of trailbraking in the Ultima GTR yourself, I dare you... It won't end well.

Additionally I also noticed that some AIs where apparently pitting twice and still managed to stay ahead of me (?). Others only came into the pits right after it had already stopped raining - basically when changing to wets was pretty much a waste of time.

Race starts are another issue: I can get off the starting line about as well as the AI in dry conditions, but when the track is wet the AI exhibits the same levels of grip while a human player struggles to get moving and control any excessive wheel spin. Not to mention some kind of erratic maneuvers the AI is doing on a wet track - if you try to pull that kind of crazy swerving off you'll meet the gravel trap faster than you can say "OH SH-". Having a closer look at what the AI was doing when it started to rain also exposed the fact they are merely slowing down on straights and throughout corners, but still exhibit roughly the same grip as before.

Oh, one more thing: After trying this one more time with static weather (Heavy Fog with Rain) I noticed that the air temperature was something like 7 degrees Celsius, with the track temperature not being much more than that at 11 degrees.
I thought Jerez was known for its mild climate...?


Finally I ran a 15-minute race on Donington GP yesterday. Sprint Cars, static weather (light clouds), same AI settings. This time I actually got a pretty decent and (somewhat) fair race out of the AI. I had to fight with tooth and nail for every position but made it from P16 (I think?) to P5 in the end. So I guess not all hope is lost yet.


So what was all of this about? I'll try to come to a conclusion - sorry for this being so long, I started writing this post the day before yesterday and it got a little out of hand.


Since improvements to the AI are planned next and Renato has already confirmed that, until then, the strength/aggressiveness sliders aren't working quite like they should, I am genuinely hoping for huge improvements to the AI behavior and scalability (in both directions, so that both average people and aliens can enjoy offline racing against AI opponents!).

But, in conjunction with the dynamic weather we have in AMS2, I am also hoping or rather proposing two more things:

First off, going back to ACC, all weather conditions in ACC are still very much raceable. Heavy Rain is never nice and I guess most race drivers don't particularly enjoy these kinds of conditions, but at least in ACC there are no conditions where a race would be unfeasible or have to be aborted in real life (at least as far as I have seen).
AMS2 recently had its race date option removed, and while I was kind of surprised by it I actually understand and agree with the decision. No more racing in freezing conditions. As far as I remember there was a snow option (and ice track) in PCARS2, and how many people did actually use that?
...
Right, so what I would like to see is a removal (or adjustment) of all weather conditions that are beyond what would be considered safe and raceable in real life. This means that maybe Thunderstorms need a bit of adjustment, and Heavy Fog with Rain should be considered for removal. In real racing there are safety cars and races are regularly stopped or even aborted in extreme weather conditions. We don't have that kind of thing in AMS2 (yet? ), so I don't think these conditions serve much of a purpose except the novelty factor and some extra challenge for masochists.
Dynamic weather should make things interesting - as it does in real life - but not automatically punishing. From my limited experience with ACC I actually enjoy a bit of unexpected rain in that sim during a race, because it means I have to make a choice when (or if) to pit for wet tires and, most importantly, the AI does not automatically receive a boost because it has unrealistic levels of grip in the wet. Yes, I need to pay more attention to track conditions, yes I need to be careful, but it also offers a chance for an even more interesting racing experience. That's the kind of dynamic weather (and AI balance) I want to see for AMS2.

The other option I would like to be added is another slider for AI adjustment: A rain performance multiplier. Now I do realize that there is most likely a lot more involved to determine their wet performance than simply changing a single value - I am not that naive. But I believe Reiza could make it work by exposing one player-adjustable value that impacts several other important parameters in one go.

And before you get out your pitchforks and scream "git gud, noob" and "this isn't realistic" keep in mind that adjusting AI difficulty isn't realistic in the first place. If you are too slow in RL you can't just tell your opponents to take it easy for your sake. Worse yet, if you are one of the aliens you can't just expect your measly mortal opponents to magically match your pace if beating them with one hand tied behind your back gets boring!

tl;dr please add a slider to enable slight modification of AI performance in the rain (as a multiplier based on the selected AI strength value). The huge advantage of the AI in changing conditions has ruined several other racing games and sims for me before. Unbalanced AI rain performance is one thing I am more afraid of than their psychotic, unaware behavior and that's quite something.


P.S.: I realize this post does somewhat sound like I want to praise ACC, but I am not. There are so many things AMS2 already does better, but it never hurts to look elsewhere for inspiration and things to learn from! Learning from others isn't the same as copying them, and in the end it's all meant to benefit AMS2 so it can become the best overall racing sim in the market.

 

Very good post! Thank you!
Only thing I don’t understand is why you want a separate slider for AI performance in the rain.
If AI is unrealistically strong in rain they should just fix it. So no need for a slider IMHO...

 

Yep I agree with pretty much everything the op said , btw it would have taken me at least three days to write that post too . AI must be so hard to balance , especially in changing weather conditions .
Using pc2 as an example I feel that the Devs never fully got the best out of it , can't help when the AI are playing by a different physics set to the player either .
I for one would be happy to lose the extreme conditions if it meant that the Devs could spend more time balancing the AI pace in normal weather conditions .

 

Longest post I have ever seen on a Sim racing forum.

 

No offense but why do you have AMS2 when you seem way too pleased with ACC and want AMS2 be another ACC?

Not to mention all this bringing up of other sims all the time.
these folks made Automobilista, and many other sims themselves, i think they know what they are doing and have their own vision, not follow the ACC, RF2 or whatever other Sims vision.

 

First off, thanks a lot for the positive feedback everyone! I was starting to think I was crazy but it is reassuring I'm not the only one with these thoughts.

The reasoning is the same as for having a general AI performance or aggressiveness slider in the first place - 'one size fits all' just doesn't work for everyone. The rain performance span modifier doesn't have to have a huge range, but at least allow for some fine adjustments, IMHO.

Yup, but I remain hopeful that Reiza can once again work their magic and get AI & weather interaction to a level that hasn't been seen before. For PCARS1 and 2 it was probably a case of "oh well it's good enough", especially with the amount of time and effort needed to get it even remotely close to perfect. To be honest I would've been happy enough if the rain wasn't such a mess, both in terms of physics (puddles and aquaplaning behavior!) and AI. Single-handedly ruined both PCARS games for me.

For real?

I may have gone a little overboard because I've been having trouble with my voice for three months now. Haven't spoken a word for 4 days when I posted this, so I probably needed to compensate for that...

None taken!

Though I'm surprised that after reading through that wall of text that was your key take-away.

Of course I could have talked about all the things I don't like about ACC, but what good would that have been? I mainly meant to talk about a few key points I think we could learn from (although, to be fair, there's always something to be learned from the bad as well).

Besides, if I didn't believe in Reiza, if I didn't care about AMS2, does it seem likely I would have taken the time to write such a long text? I've been a fan of Reiza's work since Game Stock Car 2013 and have played Game Stock Car Extreme, Formula Truck 2013, Copa Petrobras de Marcas and Automobilista. In fact I backed during the Simracing Bonanza! crowdfunding campaign on indiegogo (almost five years ago! ) and couldn't be happier with it since it got me AMS & all DLC as well as AMS2.

So apologies if my post came across as wanting to praise ACC or wishing for AMS2 to become exactly the same. That is not at all the case. AMS2, in my opinion, has by far the biggest potential out of all current racing simulations on the market. This is why I would love to see this potential realized, and though praise is well-deserved at this point, we as a community can offer much more by keeping both mind and eyes open for flaws and opportunities. If you've ever worked in software development or as a modder you may have experienced first hand how easy it is to become so narrowly focused on certain aspects of your work (and polishing them to perfection) that other aspects are overlooked or forgotten about. This is what we should try to avoid by offering Reiza constructive criticism in addition to praise and love.

 

Where did he say he wants another ACC exactly?
Personally im a reiza fanboy since day one but even i can admit to booting up ACC more for AI racing.
Does that mean i want AMS2 to be ACC?
No it means i enjoy ACC AI a lot more,the AI right now in AMS2 need a huge amount of work thats pretty obvious and it will come in time.

Great post OP and thanks for taking time to write and post.