Board Thread:Game Discussion/@comment-35180316-20181011013659/@comment-31433854-20181016162246

HeadFool wrote: Mugwumps think about it from the simulation developer's perspective. You've got to manage (in some cases) 40 bots, simulating thier dynamics in real-time, on an increasingly aged and slow variety of devices. If you can simplify the required calculations, you should, you would. Even small efficiencies would yield big gains when multiplied 40x. I'd bet (though cannot show) off-screen bots are simplified even further. It's what I would do, and is not atypical of simulations of all types.

Also, think the car's PR is, by itself... and even considering specific upgrades, insufficient to explain the different user-simulated bots. That's how FM claims to simulate other drivers with thier "time shifted multiplayer". It surely isn't with their different lines. All the bots follow the exact same line (with a couple of caveats), so the bot's lap is scaled to an individual player's laptime. Where you say the bots are based on PR of an event, I'd agree but replace PR with laptimes. How FM chooses bot parameters is a whole 'nother discussion.

I'll agree that people are reading a lot more into bot behavior than is actually there. They're generally stupid.

I'll also agree that they don't go beyond what is available by FU... but I wonder if a laptime generated by a lap with considerable cuts will result in a bot that runs faster than any vehicle could on it's own path could. I figure FM has some sort of cap on how fast they can go, because even in much-abused perpetual endurance races, the bots may be quick, but they're not insanely/cut all the corners quick. It is possible they use a simplified physics for the bots, but what I don't think is going on is that it changes over the course of the race depending on where the player is relative to the bot.

As for TSMs/bot selection, it appears to work based on the distribution of recorded lap times. I assume (and the monkeys have to some degree confirmed it), they run simulations before the update to benchmark where certain lap times (or, really, elapsed times for the entire race), are expected to fall and which bots should be called in those cases. This distribution then has TSM data added to it over time as players race and the cut points for bot selection are shifted. I would guess it is mostly accurate, but for some races it clearly gets skewed when there are lots of TSM data. The 10-lapper is a good example. Try to get Auralpod without running cuts. It doesn't happen, even though you may be lapping the field. Once you start cutting, you will get harder bots. Why? Because so many people have run this event so many times with cuts/trying to set as fast a time as possible, that it has extended the tail of the distribution. Do this enough (and it's helpful if you have fast teammates/friends), and you will hit the limit of what the bot simulation can do (~2:49/lap for the '14s). It seems pretty clear that cutting, etc. is included in the TSM data (obviously it is for non-simulated events, like speed records or enduracnes).