Information for Customizing AI drivers in AMS2

Is a downforce scalar planned to come in the future as well?

 

Only drag, power and weight also affect player's car.

 

That's what I suspect, given it's never been documented otherwise, but my main point is that <vehicle_reliability> should also pertain to the vehicle, which might now be implementable, give the new vehicle/livery parameters do.

 

I think a fuel capacity limit would also be necessary to generate effective Balance of Performance (BOP). For example, in GT3 Gen2, the Aston Martin can complete more than 43 laps in a stint with a full tank, while a BMW can only manage fewer than 28. This difference is too significant if you want realistic racing, as the Aston Martin would start with too much of an advantage over the BMW, directly proportional to the defined race distance.

 

I think a slider for fuel consumption itself necessary in a case like that to reduce the weight advantage gained from running that much less fuel, maybe even in addition to fuel tank capacity.

 

The advantage isn't really in the weight difference; the advantage lies in the fact that, depending on the race distance, Aston Martin might not need to make a pit stop, while BMW might have to because it runs out of fuel. That's the big difference. In iRacing races against the AI, you can define the percentage of fuel capacity each car has before a race, giving the user complete freedom to create races with entirely variable but realistic pit stop strategies.

 

A fuel consumption slider does literally everything you want there, while limiting the fuel tank size in this case results in the Aston additionally being able to start with about 25kg less fuel, that is a pretty big advantage.

 

What I'm trying to say is that it's easier to apply and calculate it based on fuel tank capacity than on fuel consumption, especially from a computational standpoint. That's why iRacing, for example, chose this method. Because in real-world conditions, this is reliable; just look at the system the WEC uses to implement the Balance of Performance (BOP) in pit stop strategies: Virtual Energy Tank (VET).

 

Isn't that realistic though that one car has a bigger tank than the other?
Speaking of which - I wanted to ask, and can't really find a good place, in the GT Classics, the 5-litre Corvette is faster than the 3-litre Porsche and can do the same distance on a 90L fuel tank as the Porsche can do on a 100L fuel tank. Is the fuel consumption for the GT Corvette actually correct (not based on the road car), because in real life the Porsches did better than Corvettes in endurance races, though I've not been able to directly find sources that say it was due to more pit stops (so I don't really feel I can report this as a bug).

 

That depends on the class, many classes have a strict regulation for fuel tank size. And since in some classes the fuel consumption of certain cars is still not very accurate unfortunately, a fuel consumption slider would help a lot for these.

 

Question
To make a custom AI driver grid I SHOULD use the new format with the new entries to work (those with the scalar a mean ) or the old one can used as well?

 

You can still use the old format.

The three new scalar entries are optional, so you don’t have to add them for a custom AI file to work. If they are missing, AMS2 will simply use the default behavior for those values.

So the old custom AI format is still valid. The new entries just give you more control if you want to fine-tune things like weight, power and drag per driver.

 

Hey Everyone, I've been building a python to use grok to make the custom ai files, turns out I can just just tell grok my rules and ask him to lock them in its just as effective so here is my logic when making custom ai, it works for me in a way that i like, if you want to use it feel free too and you're welcome to change any parameters you like to get the ai racing in a matter you like. All you need to do is give him the driver livery names and the driver tied to each livery (Like Johns Files) and Grok will do the rest
BROCK'S UPDATED COMPLETE AMS2 AI RULESET
Version: 2026-06-25 –
Update applied: (PURE DRIVER PACE ORDERING) to keep the logic clean and avoid muddying. Sections after it have been renumbered.

MATH LOGIC

On upload of driver information: Provide count of all drivers detected.
If duplicate drivers detected (by name): Remove duplicates and recount.
Split drivers based on real-world Race Skill assumption.
Assign tiers using Golden Rule (pure real-world pace ordering for the selected year or season, fastest to slowest):
Tier 1 – Top 15%
Tier 2 – Next 20%
Tier 3 – Next 30%
Tier 4 – Next 20%
Tier 5 – Last 15%


1. OVERARCHING FIELD SPREAD RULE
Exaggerate gaps when required to create a realistic field spread.
DO NOT cluster everyone within a few points.
Meaningful differences between drivers are preferred over artificial parity.
Backmarkers are allowed to be significantly slower than midfield.
Front-runners are allowed to be clearly separated from the pack.
Goal: Field naturally spreads out over race distance.

2. MAIN PACE RULE (GOLDEN RULE)
Race Skill: Full range based on points 3 AND 3.1 depending on dicipline
Backmarkers sit in the slower end of this range.
Front-runners in t1
No smoothing, no parity. Pure driver pace ordering only.

3. TIERED RACE SKILL SYSTEM – OPEN WHEEL

Tier 1 – Top 15%: Race Skill 0.86 – 0.99
Tier 2 – Next 20%: Race Skill 0.68 – 0.85
Tier 3 – Next 30%: Race Skill 0.52 – 0.67
Tier 4 – Next 20%: Race Skill 0.44 – 0.51
Tier 5 – Last 15%: Race Skill 0.35 – 0.43

3.1 TIERED RACE SKILL SYSTEM – SPORTS CARS

Tier 1 – Top 15%: Race Skill 0.87 – 0.99 Elite Pro
Tier 2 – Next 20%: Race Skill 0.77 – 0.86 Pro
Tier 3 – Next 30%: Race Skill 0.62 – 0.76 Silver
Tier 4 – Next 20%: Race Skill 0.50 – 0.62 Bronze
Tier 5 – Last 15%: Race Skill 0.40 – 0.50 AM

4. QUALIFYING SKILL RULE
Qualifying Skill must stay tightly aligned:
Qualifying Skill ∈ [Race Skill − 0.03, Race Skill + 0.03]
(No bigger than ±0.03 delta)

5. SILVER RULE (WET SKILL)
Wet Skill is completely independent of Race Skill.

Normal range: 0.40 – 0.65 (Most drivers will sit here)
Top 3% all-time wet masters: 0.70 – 0.90
Bottom 5% all-time wet duds: 0.20 – 0.40
No pace linking. No correlation.

6. DRIVER ATTRIBUTE PHILOSOPHY
Elite / Average / Poor scale (correlate to real-world habits):

0.70 – 0.95 = Elite
0.35 – 0.70 = Average (most drivers land here ±)
0.10 – 0.35 = Poor

Attributes using this scale: (docoupled from race skill and quali skill to mimic real world habits)

Stamina
Consistency
Start Reactions
Mistake Avoidance (avoidance_of_mistakes + avoidance_of_forced_mistakes)

Fully decoupled attributes (0.10 – 0.99 range):

Tyre Management
Fuel Management
Vehicle Reliability
Weather Tyre Changes

7. RACECRAFT RULES

Aggression: 0.20 – 0.73 (hard cap)
Defending: 0.20 – 0.55 (hard cap)
Blue Flags: 0.70 for EVERY driver (locked)

8. DECOUPLING & PERSONALITY RULE
All attributes must feel unique to the driver’s real-world style.
Personalities remain distinct even inside the same tier.


9. XML OUTPUT TEMPLATE (Standard for all drivers)

10. ----<weight_scalar>1.000</weight_scalar>---
----<power_scalar>1.000</power_scalar>--
---- <drag_scalar>1.000</drag_scalar>---
----<setup_downforce>0.50</setup_downforce> <!-- Stock Setup preference: 0.50 = neutral; higher = prefers more downforce -->
----<setup_downforce_randomness>0.05</setup_downforce_randomness> <!-- 0.000 = fixed bias; 1.000 = high variation weekend-to-weekend -->

these can be utilised to assist in real world pace by season

XML example

<driver livery_name="example">
<name>Example Driver</name>
<country>XXX</country>
<race_skill>0.xx</race_skill>
<qualifying_skill>0.xx</qualifying_skill>
<aggression>0.xx</aggression>
<defending>0.xx</defending>
<stamina>0.xx</stamina>
<consistency>0.xx</consistency>
<start_reactions>0.xx</start_reactions>
<wet_skill>0.xx</wet_skill>
<tyre_management>0.xx</tyre_management>
<fuel_management>0.xx</fuel_management>
<avoidance_of_mistakes>0.xx</avoidance_of_mistakes>
<avoidance_of_forced_mistakes>0.xx</avoidance_of_forced_mistakes>
<vehicle_reliability>0.xx</vehicle_reliability>
<weather_tyre_changes>0.xx</weather_tyre_changes>
<blue_flag_conceding>0.70</blue_flag_conceding>
<weight_scalar>1.000</weight_scalar>
<power_scalar>1.000</power_scalar>
<drag_scalar>1.000</drag_scalar>
<setup_downforce>0.50</setup_downforce> <!--- Stock Set up preference is 0.50 + = Dirver Prefers More downforce - driver prefers less-->
<setup_downforce_randomness>0.05</setup_downforce_randomness> <!------ A value of 0.000 means there is no randomness and the driver will always use their exact preferred bias.
A value of 1.000 means the resulting setup value can vary widely from one race weekend to another.
Lower values make the resulting setup more likely to remain close to the driver's preferred bias.
Higher values increase variation between race weekends.---->


</driver>

 

I updated my rule set today

BROCK'S UPDATED COMPLETE AMS2 AI RULESET
Version: 2026-06-25 –
Update applied: (PURE DRIVER PACE ORDERING) to keep the logic clean and avoid muddying. Sections after it have been renumbered.

MATH LOGIC

On upload of driver information: Provide count of all drivers detected.
If duplicate drivers detected (by name): Remove duplicates and recount.
Split drivers based on real-world Race Skill assumption.
Assign tiers using Golden Rule (pure real-world pace ordering for the selected year or season, fastest to slowest):
Tier 1 – Top 15%
Tier 2 – Next 20%
Tier 3 – Next 30%
Tier 4 – Next 20%
Tier 5 – Last 15%


1. OVERARCHING FIELD SPREAD RULE
Exaggerate gaps when required to create a realistic field spread.
DO NOT cluster everyone within a few points.
Meaningful differences between drivers are preferred over artificial parity.
Backmarkers are allowed to be significantly slower than midfield.
Front-runners are allowed to be clearly separated from the pack.
Goal: Field naturally spreads out over race distance.

2. MAIN PACE RULE (GOLDEN RULE)
Race Skill: Full range based on points 3 AND 3.1 depending on dicipline
Backmarkers sit in the slower end of this range.
Front-runners in t1
No smoothing, no parity. Pure driver pace ordering only.

3. TIERED RACE SKILL SYSTEM – OPEN WHEEL

Tier 1 – Top 15%: Race Skill 0.86 – 0.99
Tier 2 – Next 20%: Race Skill 0.68 – 0.85
Tier 3 – Next 30%: Race Skill 0.52 – 0.67
Tier 4 – Next 20%: Race Skill 0.44 – 0.51
Tier 5 – Last 15%: Race Skill 0.35 – 0.43

3.1 TIERED RACE SKILL SYSTEM – SPORTS CARS

Tier 1 – Top 15%: Race Skill 0.87 – 0.99 Elite Pro
Tier 2 – Next 20%: Race Skill 0.77 – 0.86 Pro
Tier 3 – Next 30%: Race Skill 0.62 – 0.76 Silver
Tier 4 – Next 20%: Race Skill 0.50 – 0.62 Bronze
Tier 5 – Last 15%: Race Skill 0.40 – 0.50 AM

4. QUALIFYING SKILL RULE
Qualifying Skill must stay tightly aligned:
Qualifying Skill ∈ [Race Skill − 0.03, Race Skill + 0.03]
(No bigger than ±0.03 delta)

5. SILVER RULE (WET SKILL)
Wet Skill is completely independent of Race Skill.

Normal range: 0.40 – 0.65 (Most drivers will sit here)
Top 3% all-time wet masters: 0.70 – 0.90
Bottom 5% all-time wet duds: 0.20 – 0.40
No pace linking. No correlation.

6. DRIVER ATTRIBUTE PHILOSOPHY
Elite / Average / Poor scale (correlate to real-world habits):

0.70 – 0.95 = Elite
0.35 – 0.70 = Average (most drivers land here ±)
0.10 – 0.35 = Poor

Attributes using this scale: (docoupled from race skill and quali skill to mimic real world habits)

Stamina
Consistency
Start Reactions
Mistake Avoidance (avoidance_of_mistakes + avoidance_of_forced_mistakes)

Fully decoupled attributes (0.10 – 0.99 range):

Tyre Management
Fuel Management
Vehicle Reliability
Weather Tyre Changes

7. RACECRAFT RULES Sports cars touring cars Nascar

Aggression: 0.20 – 0.66 (hard cap)
Defending: 0.20 – 0.50 (hard cap)
Blue Flags: 0.73 for EVERY driver (locked)

7.1

RACECRAFT RULES Open Wheel

Aggression: 0.20 – 0.60 (hard cap)
Defending: 0.20 – 0.40 (hard cap)
Blue Flags: 0.77 for EVERY driver (locked)


8. DECOUPLING & PERSONALITY RULE
All attributes must feel unique to the driver’s real-world style.
Personalities remain distinct even inside the same tier.




9. New Fine tuning options

----<weight_scalar>1.000</weight_scalar>---
----<power_scalar>1.000</power_scalar>--
---- <drag_scalar>1.000</drag_scalar>---
----<setup_downforce>0.50</setup_downforce> <!-- Stock Setup preference: 0.50 = neutral; higher = prefers more downforce -->
----<setup_downforce_randomness>0.05</setup_downforce_randomness> <!-- 0.000 = fixed bias; 1.000 = high variation weekend-to-weekend -->

these can be utilised to assist in real world pace by season

10. XML example

<driver livery_name="example">
<name>Example Driver</name>
<country>XXX</country>
<race_skill>0.xx</race_skill>
<qualifying_skill>0.xx</qualifying_skill>
<aggression>0.xx</aggression>
<defending>0.xx</defending>
<stamina>0.xx</stamina>
<consistency>0.xx</consistency>
<start_reactions>0.xx</start_reactions>
<wet_skill>0.xx</wet_skill>
<tyre_management>0.xx</tyre_management>
<fuel_management>0.xx</fuel_management>
<avoidance_of_mistakes>0.xx</avoidance_of_mistakes>
<avoidance_of_forced_mistakes>0.xx</avoidance_of_forced_mistakes>
<vehicle_reliability>0.xx</vehicle_reliability>
<weather_tyre_changes>0.xx</weather_tyre_changes>
<blue_flag_conceding>0.70</blue_flag_conceding>
<weight_scalar>1.000</weight_scalar>
<power_scalar>1.000</power_scalar>
<drag_scalar>1.000</drag_scalar>
<setup_downforce>0.50</setup_downforce> <!--- Stock Set up preference is 0.50 + = Dirver Prefers More downforce - driver prefers less-->
<setup_downforce_randomness>0.05</setup_downforce_randomness> <!------ A value of 0.000 means there is no randomness and the driver will always use their exact preferred bias.
A value of 1.000 means the resulting setup value can vary widely from one race weekend to another.
Lower values make the resulting setup more likely to remain close to the driver's preferred bias.
Higher values increase variation between race weekends.---->

</driver>

End rules


Grok: Must always count driver names and remove duplicates.

Grok Must always when a % remainder applies round down to the next tier to fit in the full grid.

Grok must always give me a tiering summary before creating the XML file

Grok Must always Wait for my command GO GROK to create each NEW XML

Grok MUST MUST MUST ALWAYS provide a full untruncated XML Output regardless of size

 

So if I understood well at “power_scalar: Multiplies the vehicle engine power output”
if I put a value as 1020 that’s means that I have multiple the power by 20% ?

 

"The effect of power_scalar may vary depending on the vehicle's powertrain configuration. For naturally aspirated internal combustion engine vehicles, the resulting performance change will generally be close to the specified scalar value. However, vehicles using turbochargers, hybrid systems, electric power assistance or certain driveline and clutch models may not exhibit a proportional change in performance."

20% feels like a lot.... for a 0.20 attribute change... reiza hasn't provided the exact formula....

 

Now you put more pressure on me , my mind just blended . I just want too increase a little power to Milano g36 car so by changing its class to gt3 gen 2 make a custom grid of modern GT3 cars but by default lack almost 70 hp for the class . So with this way I tried to increase it . But am l done it? Perhaps @gian can help us on that ?

 

yeah sorry dude i'm working on a suck it and see method with it atm hopefully we get a more scientific formula

 

Regarding that , Reiza gives us a ratio from 1000 which is default value up to 1100 which is the maximum for increasing that value , so automatically we can come to this conclusion that 1015 is 15% 1067 is 67% etc etc . Is a fair thinking I believe.

 

If you have a value of 500 and multiply it by 1.1 it is a 10% increase. So this gives you the option to reduce or increase the power by up to 10%.