Tips for racing against the AI

[Lorencini]

Hello everybody!

After spends countless hours playing Automobilista, racing against the AI, I start to understand how they behave in longer races (30 minutes or more).
In this post, I summarize some main AI behaviours, and how to configure your race options to get the best possible experience, including practice and qualifiyng sessions, enforcing pitstops, multiclass racing, and so on.

My goal is to make this post a central information point about single-racing in Automobilista, so feel free to post your own tips.


General Tips:

1 - AI Behavior

AI-controlled cars have a simplified physics, and behave differently comparing to real humans. I summarize some behaviors based on my experience racing against them. You should consider the tips below when racing single-player against the AI on your own:

• In Practice, the AI always run with full tank. In qualifying, they always run with enough fuel to run 3 laps (pit exit lap, flying lap, pit return lap) That explains why qualy times drop up to a second in qualy comparing to practice.

• In Race, the AI will use enough fuel to complete the race. However, if a full tank is not enough to complete the race (resulting in at least 1 pit stop), they will ALWAYS start the race with full tank.

• The AI will only make a pit stop if they run out of gas. By the time the AI run with low fuel, the simulator distributes the AI pit entering in 3 laps, in order to avoid all AI cars pit at the same lap, which would result in several cars have to wait their teammate finish his pitstop. Tip: when you a message "Your teammate is in pit", prepare yourself for pitting next lap at most.

• When in pit stall, AI will ALWAYS change its four tyres, no matter how much they wear it; and will ALWAYS fill up the tank, no matter how much time is left for the race to end. Pit stop times can be 32 to 38 seconds, depending of the vehicle.
  
  
• Turn off Tyre Wear: The AI has very little tyre wear effect. As the race goes by, AI time laps will decrease each lap, according to fuel usage: each lap the amount of fuel decreases, the car gets lighter and the AI run faster. On the other hand, AI tyre wear is so little that seems not to be relevant in lap times. So, by the time AI cars are almost out of gas (hence, close to their pit stop), their lap times will be very close to their qualifying times, while the player lap times will increase due to his tyre wear, much more noticeable when running several laps. You can check a detailed example in this post.
  
  Example:
  
  Series: Endurance Brasil (MR18 and MCR2000) (download here)
  Track: Tarumã
  Qualifying lenght: 30 minutes
  Race length: 30 minutes
  Fuel Usage - 2x (with 2x fuel , MCR will race for 25 minutes before run out of gas, so a pit stop will be surely necessary) (MR18 cars can race up to 35 minutes with 2x, so they won't pit in this scenario)
  Tyre Wear - OFF (otherwise, bots will get faster each lap while you get slower each lap due to your wear tyres)
  
  In qualifying, put enough gas in your tank to run 3 laps and to go track. When you complete your 3 laps, get back to the garage (hit ESC) and repeat the process until you feel ok with your times. Then, press Control + T to quickly finish the session.
  
  In race, start with full tank. When you're gonna make a pitstop, fill up your tank and change all 4 tyres.
  
  By doing this, I can guarantee you will have a close race throughout the race lenght (30 minutes), with a mandatory pit stop, in a standart pit-window of three laps.
  
  
  
• Agression: always use Max Aggression when using Reiza's cars (they have a very good car position awareness, specially the Stock Car Brasil -Stock V8). If you set aggression for, let's say, Medium, and run through a corner side-by-side with the AI, the opponent car will try so badly not to touch you that he will deliberaty open a little bit and end up off-track, even if you let enough space for him. On the other hand, using Max Aggression, the AI will still play fair in the aforementioned scenario, but in the edge of the track, he will stick with it and will not run offtrack.
  
  
• Session time: never skip a qualifying session end never click 'Finish Session' to go to race. It seems like the engine need the qualifying to complete all its stipulated time to proper carry on AI performance to race session. So, when you finish your qualifying, press Control + T to fast foward qualifying time, and when the clock reaches zero, then click on 'Next Session'.

2 - Multiclass racing

Tip for multi-class racing: ALWAYS do a qualifying. This way, you will split the different class in the grid, assuring that all the faster cars will be in front of the slower ones at race start. If you skip qualifying, simulator will scramble all cars, putting faster cars mixed with slower ones all over the grid. A 30 minute qualifying is recommended to assure all AI cars will do a flying lap according to its class and won't be cought in traffic. When you finish your qualifying, press Control + T to fast foward qualifying time, so you won't need to wait halp na hour to race.


3 - Race Scale and Day/Night Transition

With a race scale of 20x, each 10 minutes in a race will make the time move forward 3 hours. So, if you want to simulate day-night transition in a, for example, 40 minute race, set the race scale to 20x, and the race start time by noon (12PM).

• 10 minutes racing will be at 3PM (you will notice the shadows start to move);
• 20 minutes, it will be 6PM (and you will notice the golden color of the sunset upon the track and cars);
• 30 minutes will be 9PM (headlights will be required, the light will dim gradually, to a point where the sky is almost black and you will have some hard time to find your marks);
• 40 minutes will be midnight (the sky will be pitch black, and if the track does not have native light posts, all you can see is what your headlights can lit, and the other cars lights).

4 - Night Races

• Racing at night can bring performance issues in some rigs, mainly due to the headlights of your opponents. But I found some parameters in the PLR file that solves the issue:
  
  Max Headlights="2" // Max headlights visible relative to your car.
  
  The default value is 12 cars with visible lit headlights. In my tests, I change the value to only 2 cars, so I can see headlights from on car in front and 1 can behind. Thats enough to keep the track visible enough, and I don't suffer any more performance issues.

• Another issue when racing at night is cockpit ilumination: other cars headlights also iluminate your cockpit, making it too bright, sometimes white. To solve this, use this parameter in PLR
  
  Headlights On Cars="0" // Headlights illuminate other cars.
  
  The default value is 1. Set this value to 0, and headlights will iluminate only the track (tarmac, grass, signs, etc). Your opponents will still be visible but darkened, with their taill lights in highlight. And, most important, your cockpit will be dark, and you'll only see the led lights in your panel and wheel.

First Event Issue

I've noticed that in the first ever session I enter after starting AMS, the AI runs slower lap times. Then, if I quit session, go back to menu screen, and re-enter the same race (same AI%, same parameters), AI gets back to its original performance, and immediately runs 1-2 seconds faster, depending on track/car combo. After that, while AMS is still open, AI retains its best performance. But if you close AMS and open it again, the first event will experience slower AI.


[Edit - Aug 9th 2017 - the "Blue Flag Issue" has been resolved at release v1.4.5, with the new revamped AI. Now faster cars overtake slower ones easily.]

 

[schlitty]

Good point about the different types of tires. I think it would be somewhat safe to assume (but still assume) the AI would more ummm 'robotic-like' handle the tire wear easier though. And in turn, get less affected by it because of their predictable behvior. Like say the way they formerly handled spinouts and crashes in AMS was in video game terms; "godlike" before Reiza tweaked that fairly recently. It surpassed human ability. There's not really much to sugar coat, the AI is indeed very predictable and also lacking in some areas. And that's coming from a guy who only plays singleplayer for well over 100hrs now. There's enough there for me personally and I can take it as is. But, to put it nicely, its also at a level it can only get better. It HAS only gotten better though! I'm confident that trend will continue.

 

[Alex Sawczuk]

Alex, you are implying that the AI slow down to account for (some degree of) tire wear. The OP is implying that the AI speed-up due to lighter fuel load due to fuel consumption. Which is it, guys?

In reality, both should be happening at the same time, partially cancelling each other out. The effect of one versus the other will vary based on how powerful the car is and how big its fuel tank is versus how good the tires are and their degradation rate (with wear and/or abuse). Given this, I would expect quite a range of outcomes depending on which cars we're talking about. Remember some tires are designed to last multiple races or even a whole season for the poor-man's version of the sport. Not everything is F1-grade tires that last 100 km!

You seem to have misunderstood me, towards the end of a race the fuel levels should be similar to qualifying, but the tyres more worn.

 

[David Wright]

I think it would be somewhat safe to assume (but still assume) the AI would more ummm 'robotic-like' handle the tire wear easier though. And in turn, get less affected by it because of their predictable behvior.

The AI tyre wear is independent of the player tyre wear. You can make them more affected by tyre wear than the player if you choose. Of course some players will wear their tyres quicker than others through their driving style. Its not a simple job to match player and AI tyre wear.

 

[PieterN]

You can make them more affected by tyre wear than the player if you choose.

Hi David, how can we do that for the V8 Supercars for instance ?

 

[Marc Collins]

You seem to have misunderstood me, towards the end of a race the fuel levels should be similar to qualifying, but the tyres more worn.

I didn't misunderstand you--just wanted 100% confirmation that you were saying the opposite of the OP. That the AI do experience, and slow down, due to tire wear.

 

[David Wright]

Hi David, how can we do that for the V8 Supercars for instance ?

OOPS - I'm sorry I wasn't very clear. I work with isiMotor2 physics files which are not encypted, and forget that AMS files are encrypted. I'm afraid the only people who can adjust them is Reiza.

If you are using a mod, then look for the line AIWear= in the tbc file and increase it.

 

[Alex Sawczuk]

I didn't misunderstand you--just wanted 100% confirmation that you were saying the opposite of the OP. That the AI do experience, and slow down, due to tire wear.

Yes, the AI certainly wears it's tyres and slows down as a result.

 

[Lorencini]

@Alex Sawczuk, first of all, thanks for joining the conversation. It's always nice to have a team member talking with us about AMS.

Everything I've posted so far is based on my experience with the sim. I'm playing AMS non-stop almost all 2016, did some league racing, but almost all my time spent on it is offline.

Thanks for your info about the AI tyre wear. I'm gonna slightly change the OP to reflect your input. I'm gonna make even more tests about this issue (problably put tyre wear X3, and turn off fuel consumption and change AIMistakes parameter to zero to eliminate those variables, and check AI lap times). Until then, I keep my opinion about it, since it's working pretty well to me so far.

Second, would you give your opinion about the other matters in OP? I really would like to know if my conclusions are somewhat in the right path. This could be a good thing for the community.

Last but not least, please do not take my observations the hard way. I'm not here to criticize AMS; I really like the sim, I play it almost every day. I just want to share some info I've gathered so far, so people that play single-player like me can overcome some issues about AMS and take the most fun out of if. I'm sure I'm having a great time playing it, no question about it.

Regards!

 

[Alex Sawczuk]

AI is not really my department, I don't extensively test them so I can't answer your questions. I just corrected what I had confirmed for sure. There is definitely AI tyre wear.

 

[Renato Simioni]

Firstly thanks to the OP for taking the time to test it so thoroughly and posting such detailed feedback Some of the issues we´re aware of and should be able to address as we do some further AI code work in the upcoming months.

With regards to tire wear, as Alex says the AI tire wear functions do indeed feature, and they´re generally slightly higher than the player´s - easy to tell running a F-Reiza race with 7x tire wear for instance, their lap times will drop off very quickly as will the player. When testing for AI tire wear is also important to keep track progression set to none, otherwise the increased track grip as the rubber builds up might offset losses from tire wear in cars where tire wear is not as extreme.

As you note though AI physics have some custom parameters so the AI may be affected by the tire wear differently to the player. Players themselves might also vary as to how their performance is affected by tire wear - one might lose more time for instance simply because he´s not as used running on worn rubber as he is on new ones.

If you do find specific cars that seem to have a very clear discrepancy in AI performance with or without tire wear, please let us know about it, preferably in our official forum so we can review it

 

[Lorencini]

Thanks @Renato Simioni for your feedback and your tips for testing the AI. Didn't really thought about track progression, definelly gonna turn it off in my tests. Depending on the results, I will surely rewrite some tips in the OP, which mainly focus on how to set up all parameters (both in-game and in PLR) to have the best experience when running long races, in order to have a very competitive, close race with pitstops and time scale.

I like to run with Stock Car Brasil and MCR2000 (with Endurance Brasil skinpack), and also Boxer Cup and SuperV8 as well. I've took my conclusions based on those cars. I will take this week (and problably the next too) to refine my tests using these classes in official tracks.

In conclusion, I did post this same OP in Reiza forums. I'm gonna update it there when I have news.

Regars from Brasil

 

[Lorencini]

New info to OP:

First Event Issue

I've noticed that in the first ever session I enter after starting AMS, the AI runs slower lap times. Then, if I quit session, go back to menu screen, and re-enter the same race (same AI%, same parameters), AI gets back to its original performance, and immediately runs 1-2 seconds faster, depending on track/car combo. After that, while AMS is still open, AI retains its best performance. But if you close AMS and open it again, the first event will experience slower AI.

 

[Lorencini]

@Alex Sawczuk and @Renato Simioni , you may want to take a look at it.

Yesterday I finally could spare some time to do my tests. So, I did the same race I had 3 days ago (Endurance Brasil @ Autopolis), since I was already trained.

Race was set for 30 minutes, with a 30 minutes qualifying session.
I change some setting according to Renato's suggestion and completelly turn off Track Usage and Track Progression, to avoid the track rubbering's extra grip to mask AI times somehow.
Also set Fuel Usage to X2 to force a pit window with MCR in the specified time, and set Tyre Wear to normal, to check it's relevance in AI lap times.

A note for clarification: the Endurance Brasil skinpack consists of multiclass racing with MR18 and MCR2000. I've raced with MCR2000, and ran a qualifying session first to ensure the two grids will be fully separated in race start, thus preventing one class mess with other's lap times. (I explain this tip in the OP).

The picture below shows my findings:

In the left column, we have qualifying times. You can skip the first 10 positions (they are MR18 class, don't apply to our test).
Then, after 11th position, I isolated 4 AI Cars (subjects marked A, B, C, D), which I think could best describe the issue.
The complete qualifying and race results, laps, gaps, positions for this test race can be found here: http://simresults.net/170119-Xo1

In the RACE LAPS columns, I isolate race lap times for ours subjects.
The first two laps can be discarded, since in race start all cars are very close, some AI breaks too much to avoid touch, there's fight for position, etc.

So, after this phase in race, I've marked in green a reference lap time for each AI. You can notice that, for this car/track combo, lap times are 1s to 1,5 higher than qualy times.

As the race goes by, you can see that AI lap times start to decrease. Of course, there are some very slower times, due to AI errors, off-tracks, fights for position. But this behaviour is desirable, that's ok.

Marked in orange are the lap AI performed a pit stop, and the lap after. You can notice that lap times before pit are much faster, closer to qualifying times. Those laps are marked in blue, the same color I've marked qualifying times as well.
Interesting to notice that subject A could even be 1 tenth of a second FASTER than his qualifying lap. In those laps, AI have very low fuel, they are ligher and get faster times.

At this point, we can observe another tip of mine: the 'Pit Window'. We can see that, when MCR2000 AI cars reached a 20-30% fuel left window, they initiated their pit stops, and all cars were distributed in a 3-lap Pit Window, problably to split teammates and avoid them to pit at the same time.

After the pit stops, the AI cars fill up their tanks, which make them heavier. Their lap times as now compatible with race start times (marked as green for better comparison).

We can clearly see that Tyre Wear have INSIGNIFICANT EFFECT in AI race lap times, close to none.

This has been happening from a long time ago, in every car and track I tried (being default content or mods). So long that I could notice, test it, and create this system I talk in OP. As a matter of fact, I've been observing this behaviour since GSCE days. But is was worse then, AMS have fixed other important issues (like fuel usage influencing too much lap times, but this is past).

On the other hand, I've struggle a lot with my tyre wear. Got off-track once. And in my last lap before pitting, my rear-left tyre wear bar marker in HUD was already yellow. I could cleary feel the wearing effects, It was difficult to make long corners, and I almost spin in the harpin in third sector.
The negative tyre effect superceeds the positive fuel consumption for me.

So, with all that, I'm standing my position about my tip: to have a fair race against the AI, turn off tyre wear, since with that option on, as race goes by, you're gonna get slower, and AI will be faster, almost to the point of qualifying times.

 

[PieterN]

Thanks for testing and writing about this @Lorencini !

 

[Andy_J]

@Lorencini Much appreciated. So for racing against the AI turn of tyre wear. That is clear enough. Hopefully Rieza will fix this in future updates.
Cheers.

 

[P*Funk]

I've always felt this way about the AI and tire wear as well. I knew they had tire wear it just was obvious to me that the effect of fuel burn was always much more pronounced and that my pace would always slow as theirs would either remain steady or even increase. This is why I always end up setting AI to a level in longer races that sees me relatively faster than them at the start because I know by the time tire wears sets in my pace will fall as theirs climbs leading to a sort of characteristic early race charge followed by late race defend pattern for me.

 

[Lorencini]

@P*Funk I use to set AI harder in the beginning just like you, before I begun to play with parameters. I never really like this way tough, because I fell obliged to pass as much cars as I can in race start, in order to create a gap comfortable enough to hold position to the rest of the race.

 

[P*Funk]

because I fell obliged to pass as much cars as I can in race start, in order to create a gap comfortable enough to hold position to the rest of the race.

That is definitely how it plays out for me. Some cars are better than others. I've found with the SuperV8 I was much better at managing the tire wear as I got better driving that particular car and I think in many ways I think the AI issue with tire wear is exaggerated by our own amateurish handling of worn tires. That said it would be nice if the level of ai performance could also be matched to a sliding scale of relative competence. Really it would add a lot to the AI if there were some AI cars that represented really professional drivers who were the aces and others who were more inconsistent and who suffered worse under tire wear like the lesser drivers that they are.

In many respect the overall homogeneity of the AI is part of the issue. Its not just that the AI gets too much faster as their tires wear but that they all behave identically so that everyone's pace increases or decreases in proportion. Would be nice if there was more individuality to the AI so that you felt like you were racing individual opponents with different qualities (meaning AI settings). Would be neat if the guy in front were slowly falling back into your clutches but then the guy behind you starts making a push on you so you have to defend rather than attack.

 

[Lorencini]

I think the AI issue with tire wear is exaggerated by our own amateurish handling of worn tires

I think I got what you say. As a matter of fact, I'm no alien, but I'm no "Forza player" either. I've raced simracing championships for 2 years in Brazil (www.f1bc.com), and when training with my team (real-life racers included), normally my tyre wear was on team's average. So I think there could be some margin for adjustment in that area.

In respesct to the rest of your post, I agree. It's true we have talent files to try to configure different 'personalities' of drivers (I indeed edit them in my Endurance Brasil skinpack), but even that is not enough in today's AMS to configure what you suggest. That would be a nice feature in game, for sure.

 

[wowbagger]

This is a great thread. For a long while now I have been thinking a sticky about improving the single player experience would be a good thing.

One thing I've done is copy unabashedly Keith Windsor's amazing work from his fantastic 1988 season and use the same idea, albeit in a cheap and dirty ghetto manner, to give fairly believable performance differences between AI. I find talent files alone don't really make a noticeable difference.
Just make an _Upgrades.ini file for a few teams at the top of the grid, and a few teams at the bottom of the grid and then play around with these settings:

Spoiler: upgrades.ini

UpgradeType="Engine Power Changes"
{
UpgradeLevel="Engine Power Change"
{
HDV=[ENGINE]
HDV=GeneralTorqueMult=1.1
HDV=GeneralPowerMult=1.1
}
}

UpgradeType="Drag Tests"
{
UpgradeLevel="Drag"
{
Description="Drag"
HDV=[BODYAERO]
HDV=BodyDragBase=(0.250)
}
}

You don't even need to mess about with the midfield. I've used this for a '94 season of a certain championship (which shall remain nameless, Bernie) and it works very well. I wish I had the time, patience, and talent that Keith possesses to do it properly, but given how quick it is to bang up these few files, it works surprisingly well.