AMS2 Q&A General Discussions & Feedback

My apologies, misunderstood then. How could I misread so bad?

 

My major concerns have already been posted and answered by Renato, so I just put here some comments.
AMS2 v1.6 took 7 months to get out but also took AMS2 to another level. Now the AI is the weaker aspect of AMS2.

I don't like the ghosting on pits either, but that's more of a 'nice to have' kind of feature (very high in my hierarchy, but still a 'nice to have').
But getting out of the pits and see a car pass-thru us (Imola, Nurburgring,...) is a "must" to correct, in my oppinion. That's pure arcade.

Renato (kindly) expose the number of variables required to set AI but still not aware that's exactly the kind of work Machine Learning is good at. Those are weights ready to be processed by a ML algorithm so every time physics change a processor can optimize AI overnight (without allocating human resources to do it again and again...).
The investment in ML knowledge is not lost, it is scalable, versatile, adaptable to other (future) sims.

I think current AI is not plagued by bad decision making process as much as it is with bad physics, particularly the effects of collisions (but also unrealistic spin recovery). I've been expressing this concern for some time, so I won't elaborate it further.
In this aspect, AMS2 lags behind others in a non-negligible way.

 

I made a new recording of the current AI after testing a 45 minute race session with the LMDh, GT3 Gen2 and GT4. I was driving the GT3 Gen2 class. I know some of these issues have already been described in the start of this discussion thread but i am still putting them in this video because it is the up-to-date current version of the game and just informing which problems are still present.

But this video also shows some good things that i noticed, so this video is not all criticism.

 

Have to say I think the Ai makes the right call at 1:05 and 1:09, there is no significant overlap, at 1:20 the cars 3 places ahead do have an overlap and make the move stick.

6:43 maybe the Ai is concussed Though those rejoins and so bad.

 

At 1:05 the AI can totally make the overlap happen but decides to almost never do so. It also becomes a problem if the AI does the exact same type of look and yield again several laps in a row in the exact same way. Like sure two or three times the move did not work out but i would expect that by the fourt time the AI would go "Okay i should go for it now because i am really close so no point in lifting so early".

 

The problem with your video is the cars are all different, with different performances. Try the same in a single make class, like the Brazilian Stock Cars. You will then see the AI struggle to use the slipstream effectively, pulling out of the tow far too early and losing what little momentum they have gained.

 

That is a different issue entirely. And that issue is with all car classes how they regularly pull out too early and do not make best use of the draft. But also never follow the player's draft.

 

One interesting follow-up question would be, are they as cautious near the end of the race, as I'm sure Renato mentioned that the AI get more aggressive closer to the end of the race.

 

I did a test with just a 15 minute race and indeed the AI does get more aggressive and some of them take it two-wide into that corner . Thing is i would hope that in a 1 hour race the AI would already start doing this on lap 1 because out of 48 drivers only about 5 went like this in the short race test. In the last 15 minutes i would be expecting a lot more drivers doing this.

 

Hi, I noticed the "Relative times" table, the one indicated below:


This reflects the instant gaps, so the numbers are jumping around in many cases seconds. I understand this is correct if they reflect the real time differences as I assume, but in my point of view this make this gaps not very useful. I think should be implemented the same that you can see in real F1 or other sims like iRacing, that they apply some kind of "average" because the number are much more stable. Yes I know then it is not exactly the gap in that precise millisecond, but honestly having a number that in one moment indicate 2.5sec for example, in the next couple of seconds indicate 3,7, and after some more seconds 2,3 is not very useful in my point of view, at the end to make some use of the numbers you need to make a "mental" average while racing.... Assetto Corsa 1 had the same issue and honestly I think if the developers apply a average like many other sims this gaps would be much more useful.

 

Agree with your point at 6:05.

AI rejoins are a big issue I've noticed. Here's me slamming into the back of another car because instead of going around on the completely clear side of the track they brake and wait for the rejoining car to get going:

 

Replying here as per the thread rules, but I think you need to lower the difficulty for the group C. Below a quick 5 laps I decided to record from the replay.

Pretty good racing against the AI, they might be a little bit too quick through Eau Rouge but not like in your video. Sometimes the AI "issue" is simply trying a too high difficulty.

Default setup
111 AI difficulty
High aggression
Only auto clutch aid used
No custom AI values, just the real names

 

Yeah you might have a point. Turned it down to 85 and had a much more reasonable race. (Although they still gap me at Eau Rouge and overslow for the turn at the end of the straight.)

 

I think it's also class based as well, just had a race in the LMDh where the AI and I were relatively the same speed up Eau Rogue and the straight.
But in GT4 they were doing like you said and pulling a massive gap up there and the straight.

I don't feel like lowering the difficulty is really the solution because that lowers their laptimes which are relatively similar to what I'm doing.

 

The rears of the opponent cars are not always rendered right at night, when they should be lighted by my headlights. Sometimes they are and sometimes they are not. Not sure if it has to do with the distance. The ongoing change is very noticeable, so it distracts quite a lot.

 

Consoles are no longer PS2 era performance so why is that argument used as if it meant anything ?? Just say if the game is more or less realistic because realistic sim titles are on consoles since at least the PS4 era. Is ACC arcade just because it is on console too ? AC EVO will become arcade when it will release at some point on PS5 ?

 

I see where you're coming from, but let's say if car physics (tires, aero, grip etc) are using a mere 5% of an average CPU power (complete guess work, I've no idea how much it uses) that'll mean you can only have 19 opponents before your PC runs out of power, and that's before any power is utilised for any other use.
So to achieve that the physics model would need to be simple. Now think of a 30-40 car grid, AI decisions, graphics, antivirus etc running in the background. It all adds up.

 

Well ACC on the normal PS4 had a 20 AI limit with a locked 30fps and acording to Kunos they use the same physics as the player, so I think there's an argument to be made there.

 

Yeah, i second this. It's unreal the gap they pull on me in that sector.

 

Is AMS2 SMT-aware (schedules important threads only to even/'real' cores for physics) ?
If not, does it work correctly with manual affinity assignment (some titles don't like it) ?
Thanks!