You have to remember in real life too, tracks aren’t just one temperature. Track temp changes throughout a race from karts running on it. Everything is dynamic in the real world, and humans will naturally be better at feeling those changes than a computer. Every lap, even every corner, the conditions to factor in an “ideal lap” changes. Temperature of the “common line” compared to between the tires is even going to be different by a few degrees. Tire temperature and wear changes, amount of rubber laid down increases with every kart, wind, etc.
Sure, we might have an ideal lap factoring all those variables plus everything else, but nobody is ever going to actually come close to making it there. As soon as you have any variability in being able to identify and adapt to those changes, the margin for how to drive through the corner to make your fastest lap is going to widen, allowing different ways to attack the corner and use your driving style and preferences to your advantage.
But the track is made up of an impossibly complex mix of tiny aggregate, with bumps, cracks, varying levels of grip from inch to inch, different surfaces, pieces of dirt, dust, etc. The track is not constant in any measure. What about how a driver situates their body through the corner and the constant motion, weight transfer and floppiness of the sack of meat in the seat? What about tire deflection?
Maybe you could get an approximate simulation of a “good” line through a corner, but it would be assuming all conditions were “perfect” or “constant”. A glass smooth track. A statue for a driver. A tire that reacts the same way every time. There are so many variables to account for, that even if you could simulate the “perfect line”, it would matter much because that millimeter perfect line doesn’t exist in real-life and the conditions it was derived from have already changed by the time you input the data. It’s all sort of a moot exercise in futility.
Even F1 drivers talk about how their simulators aren’t close enough to real-life to be much use sometimes. Hamilton barely used the sim until this year. Max just noted in the Red Bull reveal that he hasn’t driven the new car in the sim that much. Even racing teams with the biggest budgets known to man can’t produce a sim that is close to enough to derive perfect data from. They might be useful for other things, but they aren’t good enough to simulate the intricacies of driving fast down to the nano level.
I agree with all of this. That being said there MUST be a testing methodology/process that is deeply detailed and idealized. The budgets of these teams demand it. Too bad we don’t have any f1 quant people here to tell us what’s up with how they approach this. We probably don’t have the ability to do more, currently.
Don’t forget unknown unknowns, things we don’t even know we don’t know about.
Take a read through of TJ’s response, it covers this part pretty well. Basically, even if you get perfect data from a lap of an F1 car, measuring everything down to the nanometer, thousandth of a degree, etc. The conditions you measure are imperfect the moment after you get them, as the conditions will always be off from there.
I think teams have simulators primarily to get over the initial “acclimatization” hurdle that drivers will have when going to a new track, or a track they haven’t driven in a year, or a new regulation set, or an upgrade that makes major changes, or just to keep the driver sharp during a week or two of downtime.
I guess why I push back a bit here is everyone says “no”. Yet the stakes are so high and the funding so big that it makes me think that they must be seeking to understand and as much as possible given limitations of human ingenuity.
Yes I agree that real-world is complex and cannot be modeled for all situations. Imperfect modeling can have disastrous results due to things that aren’t supposed to happen, happening, (the fat tails of a distribution curve).
Simulations are used to verify if the direction of the change seems correct. There is a ton of correlation work that is done to adjust the simulators and historical data used to judge to if a change will provide better lap times. However, even if the visual system and the vehicle/road modeling were perfect with very minimal transport delay, the driver still would not be able to drive the simulator at 10/10 because of limited motion cues.
So I just texted my IndyCar friend, he told me the simulators they have are actually very accurate compared to real life, like better than anything the public would be able to access. I would guess they haven’t tried programming a perfect lap in though, since they have so many other projects going on at the same time. I’ll try to get a better conversation and come back to this next time I see him in person.
Depends on how much I’m allowed to know. There’s many topics that he isn’t allowed to talk about, and many more that we operate with a “these discussions stay within the walls of this tent/text messages/etc.” where it’s not necessarily classified information but would be better if the entire racing world didn’t find out about it from a 22-year-old go-kart racer. I try to keep most of our discussions on that level of courtesy for him.
Makes sense. Teams are secretive and espionage is likely a thing. Having an edge, matters. (Like when I was the first assassin in DAOC to implement multiple damage type weapons to exploit limitations in armor resistances on my server. It was a huge advantage before the rest of the gang got wise).
“the driver still would not be able to drive the simulator at 10/10 because of limited motion cues.”
That’s a good point. I would think that modeling a human being is probably a couple of orders of magnitude more difficult than a car/track but maybe a simplified or task optimized model of a human could be used to research a concept or to verify performance. The F1 guys might be doing this already, so that they can simulate w/o a driver.
“I am probably waaay out in left field here and having fun thinking about this but I suspect it’s annoying. So I shall stop.”
Yes, I’m not trying to advocate that simulator use is pointless. I was meaning specifically within the context of using a simulator to reach a “perfect lap” as that’s where the discussion had been going to.
They do use the simulator to get baselines for setups and trying adjustments before they get to the track to find a direction to go in. They’re also able to run simulations without the driver to check how changes work as well, I’ve heard of engineers maintaining “this change will give us 2 tenths because the simulation said so” and then the driver goes slower because the change, while maybe being faster, made the car less drivable overall.
The sims are absolutely useful to teams, or McLaren/Ferrari/Red Bull/whoever wouldn’t put as much money into them for development. However, their usefulness is not to find the perfect lap and translate that to real world application.
@Bimodal_Rocket I’d also like to add, how are we considering programming an AI? Is this programming a machine learning system to discover the absolute limits of the driving system on its own? Or are we talking about creating a system that can do exactly what we want exactly when we want it to.
In that first scenario, that’s similar to what any autonomous driving programs are doing right now, whether it’s the autonomous IndyCar stuff or Tesla’s Autopilot.
The second system, theoretically, could be used to brute-force the fastest laptime reachable in a sim. I don’t know if you follow TrackMania speedrunning at all, but they have a Tool-Assisted-Speedrun category where the competitors program inputs to a program that runs the track. Since the game starts in the same conditions every time, and every input will have exactly the same result, they can fine tune to frame perfect inputs to get wild shortcuts, boosts from weird programming, etc.
I’m not sure how iRacing’s physics work, if for example 80% throttle with a perfectly straight steering wheel from a standing start will have the same exact wheelspin and result every time, but if that were the case, you could program and find the ideal lap. If it does not behave like that, and instead factors in for other variables that could cause the car to spin in either direction or for any amount of time after doing 80% throttle and exactly 0 steering input, you would have to go for a machine learning system that adapts to the situation. A much more difficult and time consuming method.