If you read my lengthy explanation a few posts above, itâs due to a number of factors, but the underlying reason is an unrestricted (âwrongâ; âbrokenâ) PI is being generally applied to all the race courses designed by PG that include all surface types. So, certain AI cars are performing way out of expected bounds from where they should be performing if it were just one course that was sampled, and/or adjusted more accurately by humans to reflect realistic racing constraints.
Here is a more simple explanation:
In the Course Creator, you can create a race type called âAnything Goes.â This means all PIs (car classes) are allowed to compete together on a certain course, regardless of real world, or in-game PI restrictions. This allows you to race C class cars one time, and then S2 cars the next on the same course without having to create a separate C-Class Only race and a S2-Class only race every time you want to race different classes of car. More importantly, this system allows the game (AI) to know how each class âshouldâ perform within each individual race type without having to resample the entire course every time you change car classes.
The majority of problems occur when you want to race a S2 class car in an âAnything Goesâ race and the game populates it with some D class cars to fill out the opponent count. The PI for the race is S2 (player car), so the AI will âboostâ the D class cars to perform on a similar level.
This is why we get ridiculous things like a BMW Isetta 300 leading the pack, going faster than an S2 class car at top speed, most often doing a wheelie that spits sparks from the rear bumper scraping against the asphalt. Other visual glitches due to the AI overcompensating are various car models sunk into the ground - literally - That just skate along in front of the playerâs car because the game is using the wrong course PI for those vehicles that were never meant to go that fast.
On top of this, the AI is designed to upgrade itself one level above the players PI to be âcompetitiveâ according to difficulty level. They are given more grip, downforce and horse power (proved in the leaked developer build of FH3). Now, imagine if you are racing a âXâ class car? The AI boosts itself to a class that doesnât exist for players, but the AI obviously has parameters for.
Add the fact AI donât physically exist in the game world if they arenât within a certain radius in front of, or behind the playerâs car, and you can see why all of this adds up to âRunaway Drivatarsâ that get so far out ahead of the pack there is nothing (physically) stopping them from slowing down.
By the way, these same issues apply to races that are restricted to one particular class, too. Itâs because the AI is basing PI off the selected course, and not the cars. The PI system is âbrokenâ, and why fixing the AI might not be possible unless they do a full redesign of how the game works?
I have a theory: Because of the dwindling player count, the game just grabs Xbox Live accounts (even on PC since you have to have a Xbox account) and applies a general AI profile (Level 1) that combined with the wrong PI on courses is whatâs causing them to behave more aggressively?
The base profile was always this way (aggressive), but more players hid this fact because of how real player level might be factored into an algorithm that adjusts what kind of advanced profile is used for in-game races?
Itâs just a theory, but I donât believe for one minute the AI was âlearningâ from players the way the PR material likes to tout. I believe itâs just generic AI that has a complex algorithmic system that when faced with a low(er) player count⌠Leve 1⌠It defaults to a more aggressive, or raw profile, perhaps?
They appear at all levels of difficulty, unfortunately.
Iâve done extensive testing and Runaway Drivatars occur at every single level of difficulty from Tourist to Unbeatable. Itâs because of the flawed PI system and not because of individual cars, themselves.
