AMS2 AI Developer Q&A - Physics & AI

A while back it was mentioned that we might be able to have the players AI co driver start the race. Is that still on the cards? Would be a nice immersion increaser for endurance racers.

To follow up on that, will it ever be possible to edit our co drivers attributes in any way?

 

Hi Renato, firstly thanks for this Q+A session, it is greatly appreciated and hopefully will prove useful to all involved and the development of the game.

I hope this counts as one question and if so a couple more will follow, but I might very well be being cheeky as it's a big question with many sub questions! Feel free to say enough.

I would like to know more about the damage model and physics, specifically relating to player to AI contact, and both player and AI contact with barriers and the ground/curbs. Which setting(s) give a level playing field between player and AI if any? Off and visual is hopefully a no brainer, then there is performance impacting and full, each with low, med and high. Experience suggests both the latter are at low. Can there ever be equality between player and AI at the highest setting?

I currently use full/low, (offline) however think that medium is a more representative level with respect to reality. I feel to go any higher (than low) gives the AI an unfair advantage, especially as they can be quite reckless in their choice of corner to make overtakes, (Bathurst for example) and the relentless attitude of the car behind. Slightly of topic, but it does play in to the subject of damage, the AI have little patience with respect to longer more endurance/grand prix style races. They more realistically represent sprint style racing. I hope in time a balance of aggression/patience can be found for longer races. Part of a winning strategy is knowing when to push and when to bide your time and be patient.

Currently (at damage setting full/low) half a dozen or so gentle taps to the rear of the players car from an AI car will completely detach the rear wing instantly with no warning of if the next hit will result in that happening. Could a warning be given on the hud with the option to replace the rear wing at a pit stop with applicable repair time?

When the AI suffer rear wing loss they aren't penalised as much as the player is. Even in low speed corners the player suffers, I would argue, too much oversteer whilst the AI, albeit slower, are not slowed enough, particularly in high speed corners. There is also no representation of oversteer in how they appear to drive, and sometimes don't even pit for repairs. Can/will this can be addressed?

Is or can floor/diffuser damage be modelled?
If I spin and hit a barrier backwards the 100% rear wing damage and detachment is believable, and depending on severity perhaps floor damage should also be incurred. Moreover, such impacts for example in formula or prototype cars, should result in floor damage and a loss of downforce. The floor/diffuser generates a large amount of the total downforce. Current generation F1 cars develop most of their downforce from the floor by ground effect, and side impacts can have a considerable effect on the downforce generated. Likewise strikes to the ground/curbs can also cause damage and resulting downforce loss.

Front wings have a more believable scale reported in the HUD, gentle impacts result in say 2-5 damage, harder 10/20 or more, with severe detaching the wing. The damage however seems to penalise top speed more than downforce generation. Can and is this possible to revise/implement?

Finally, and it's a pet peeve, can the AI be made to pit for repairs at night when an incident breaks their headlights? Currently they drive around with perfect night vision until a fuel/tyre stop, only then are the lights repaired.

A big thank you for both you and your team's dedication to the sim racing genre. Together with the community this game goes from strength to strength.

Edited for clarity and typo's.

Should you ever get to replying, I've added some telemetry showing rear wing loss. As shown, rear downforce is approximately reduced by 1/3. The car is the P3 Rocco. I have no idea of the real car's areo split between floor and wing, only you (or the manufacturer) can comment on the accuracy. However by losing the rear wing top speed was reduced by 32km/h (13%). Again I can't comment on accuracy, however I'd argue that losing the rear wing should result in a reduction of drag and a subsequently higher top speed, rather than lower.

 

Does the game simulate the effects of wind, on vehicle performance?

 

Will there be improvements in the rain physics and its force feedback? In the loss of grip when driving over the rubber line on the wheel or on the curbs?

 

Does following another car closely or not so closely for many laps have effect on engine overheating and brakes heating?

If I see engine damage at 1 or 2 - how big an effect does that have on car performance if any or it is simply signifying impending engine failure?

Is floor damage (running over kerbs) modeled in AMS2? If not, is it in the plans?

Thank you.

 

Although it's not as bad as it once was, will you be revisiting the 'magnetic' side by side (when 80%-100% car side to car side) contact that effectively 'holds' cars together after side on side contact for just a little too long (clearly just saloon cars not open wheelers)? This happens in single player and MP.

 

When it comes to programming and refining the physics of the cars in ams2 what do you use for reference? For example, how do you determine what the differences are between the LMDH cars?

Another question for you, which sim would you say has the best implementation of real life physics? Not including the AMS titles lol

 

Hi, can we expect more detailed aero simulation/modelling for the AI? Currently they race as if dirty air is non-existent. This is very apparent with combos such early 90s F1 cars on tracks like Barcelona where the AI are sticking to each other through the middle sector and are hounding me while I'm struggling massively behind other cars.

 

Hi all! Small questions, I skimmed through the thread hopefully I'm not putting in any duplicates, but before, small consideration:

While of course I love racing, I am a casual player (And I love how the game plays on controller if you tweak it right). I am also the special kind of aut- ehm target audience who loves witnessing cpu bots interact with each other, or in general a game that makes me feel I'm more in a sandbox rather than me having special advantages or disadvantages.

Also, as a game designer in training and code in my day job, I am extremely curious in how these systems are implemented in videogames and so happy this thread exists!

With this in mind:

1. What's the weight of individual driver stats on behavior?
2. Is AI subject to under or oversteer in the same way a player is?
3. Can/Will the AI be able to tweak car setups "individually" according to driver stats?
Maybe exploiting the practice session to do some tweaks? (A core loop could be -> AI drives -> monitors car behavior -> tweaks attributes if it is over acceptable parameters -> tries again, rinse and repeat while there's time on the clock)
4. How is the AI integrating race strategy? Are they/Will they be able to adapt to sudden changes? (Full course yellow, sudden weather change). Are red flags going to be a thing perhaps?

Thanks!

 

All my questions have already been asked. For my part, I can only confirm that improving the AI damage model promises the greatest increase in fun. I'm happy to accept any improvements to the game. But the AI damage model lags significantly behind all other developments. There's still a lot of fun hidden here that can greatly enhance other bugs.

 

Physics/Tyres/Grip...
Right, I'm NOT imagining this, serveral members of the community have confirmed it too. Since the update on the 22 March 2025 – 10:17:51 UTC - the F309 no longer has random momentary "shopping trolley wheels" on the rear with added pendulem effect - it is now incredibly good around Mosport+Donington, the JCW mini also no longer has momentary "shopping trolley wheels" on the rear even on cold tyres (although you can still envoke lift off oversteer).
AND having spent 5+ hours in the Formula Classic Gen4 Model1 although intense (as it should be)... it's VERY, VERY good!

All I can say to the Reiza dev team is what ever you "fixed" it has made EVERY car I've tried in the last few days BETTER on the rear axle - thank you!

However, on my setup I feel there is an inherent small increase in understeer across almost all cars (including the Formula cars mentioned), I would guess the "fix" to the rear axle/tyre behaviour has highlighted a slight imbalance in favour of understeer - maybe 5% more grip to the front axle would nuetralize this - yes I know some cars have understeer built in to them for safety but at the moment it's a little too much in my humble opinion.
F.Y.I.
I have not reset/deleted car setups. I have not changed FFB settings.
P.S. If you have done nothing (not possible) to improve the rear axle/grip then some weird magic has been cast across my rig/AMS2/my brain

 

Hello Renato, Reiza staff and everyone else.

Oval racing in AMS2 with tintops has been improved a lot in the last patch. Finally I think it is definitely worth to race with them. I don't know if I want any special feedback (I write some questions in the last paragraphs, though). This worked more as motivation to record a race so now that I set up OBS and stuff, perhaps I can do this more often in the future to do some feedback of this kind. BTW, the recording is not the best, I realized after finishing I had the borders shrink. I corrected this for the future... Also I must say it worked so well overall that I have much less to say than I would like. That is a good thing, though

The race is a 40 lap race at Fontana with the superV8. I think it is one of the best cars to do this at this point (spool diff, no P2P, etc). I hope we can have something even more suitable at some point . The race starts at 13.25

I use x2 fuel consumption and tire deg. Scheduled caution at 25%. A full tank lasts about 23 laps so this allows different strategies. Some did one stop and some did 2, which is reasonable. I did not find a big difference on pace wrt AI due to tire deg/fuel load. This has happened with other cars (specially old F1 cars) .Tires do not degrade a lot and changing them is faster than put some fuel so they should not be an issue in this combination, which is also reasonable (this happens with indycars in real life for instance), so this is not "tested" a lot in the combination.

There are some persisting issues that I want to state here.

When pitting in a pace lap, at some point each AI car decides to pit, and the UI tells you to follow the next car that will not pit. There are at least three issues with this: you have a lot of time to decide what to do with that information (irl, you know at the very end which one will pit),
and much worse, you can overtake the cars in front of you before the pit lane as long as you do not overtake the next car not pitting, gaining advantage. What I do in the video is to just follow the next car pitting to avoid cheating, and this introduces a third potential issue: the gap with the "FOLLOW ME" car in front could be too big and get penalized (I am pretty sure this happened to me in Gateway in the past, were the apron before the pit lane is very long, since they go down in the backstretch).

You can see this at 22.38, and 34:53. Look in the second how the distance with the car in front is a lot at the end (and I stop the one behind me). After each pit stop they go back to the banked track right after the pit lane, avoiding the apron there. I would expect exactly the opposite! That they cut entering, and take it easy going out.

After 30.00 you can see how I fight with 2 other cars. It is not clear at every lap in the video, but in the last corner (4th, entering the main "straight") I can gain a bit of time (perhaps even more clear is at 28.11 with the car in front). Even when I do not gain, they break (or lift) late always similarly, where I can do the corner differently. Also it is noticeable how after this moment they pass each other many times, it would be nice to "work together" in this kind of scenario (though this is much better than permanent nose to tail behaviour). Also, in all the video it is obvious that I could be a bit better with a bit less drag (perhaps one less click in the wing).

What is the relationship between everything I said in the previous paragraph? I think at this point, were AI is working like a charm, a bit of "personality", and "shuffling" would be nice. I would love some of them taking a different line, some of them with more downforce, some of them with less drag (question 1: do they always use the same setup? how AI setups work?, Q2: do you plan to make possible to modify AI power/aero in custom AI files?), and that they do some non-trivial racecraft like pushing with another car (even the player) to close a gap (Nascar Heat's "draft partners" are an interesting concept related to this). Q3: Do you plan to add something like this at some point?


Edit: I forgot one. Look at 5.55, I hit the wall and the car gets stick to it. This is related to a previous question, that happens vs other cars too. I generally think collisions are the area where AMS2 has room to improve physicswise. For reference, Raceroom feels like in the opposite corner, cars are like equally charged particles...

Edit2: I forgot to state that AI difficulty was 100 with high aggression. Medium aggression did not change their behaviour with respect to they attacking each other too much (today I did a 60 lap race with that being the only change). As seen in the video I was at the back at 100, while I fight races at 110+ in vintage formulas on road courses. In Indianapolis I am at the front at 100 though, so apparently it is a Fontana thing. Oval racing is a different discipline and users skill can vary so I do not expect perfectly comparable AI performance among disciplines to me, but I am pretty sure it can be improved in the oval-to-oval comparisson.

 

What are the possibilities of opening up the tuning options a little so that for instance we could build a rally setup with gravel tyres for say the BMW 2002 or the old Porsche 911, or maybe run a Lotus 49 on a modern slick tyre.
There was one time in WMD when they were developing the PC2 tyre models and we had access to a modern slick on the Lotus 49 and it was the most awesome car to drive. They took it away after a couple of weeks but once you've tasted it, well you never forget

 

Hello Renato, three questions:

01. How can I imagine the creation of AI for a track in AMS2? Set outer boundaries and then create a spline and set specific braking and acceleration strengths for certain zones (similar to Assetto Corsa)?

02. Would it be possible to improve the AI actually using AI (we see AI take over quite a few spots in our lifes right now, ChatGPT for example is very helpful in some cases)? Could it support or refine it somehow or is it still a completely manual task?

03. Is there still a feature / capability you'd like to bring to the AI which isn't implemented yet or is very difficult right now?

Thanks for taking your time.

 

Several situations happening in one video for you guys

Group C AI being too fast at spa for the player (Grip strips maybe)
A couple of AI cars getting stuck togeather in a odd way with 1995 carts
Loss of rear downforce after a pit stop/AI Stint/weather change with a GT1 car at Nurb... (Not sure what happened really but am confused)

 

I did not expect I've posted in this forum 25 times already, I thought I wouldn't have been able to participate here

Anyway before getting to my questions

Unfortunately it still happens, and it happens a lot. If you watch my video below, after 1:30 the AIs would start clumping together and get nose-to-tail until either a pit stop cycle or the race finishes. Admittedly this was run without qualifying but it still happened even with qualifying, and probably shouldn't have happened even without qualifying. (Note: this was run without custom AI)

As for my questions :
1. Related to AIs being locked into each other, once the race settled after the first lap everyone will be in a sort of "pack" where they cannot get away from each other. Even when an AI overtakes another it will not be able to get away from the car it overtaken and then could be getting re-overtaken, resulting in perpetual cycle of overtaking with the same car instead of chasing the next car to overtake. Is this something you guys aware of and is it gonna be looked into?
2. Is there any chance that we'll have separate parameters for big and small mistakes? I like to decrease avoidance of mistakes parameter in custom AI to get them to do small mistakes like slightly overshoot the apex or having slower exit which encourage more overtaking, but that also causes them to off track more often than I'd like.
Cheers

 

And then nobody saw him again... (jk obviously, I hope everything's alright)