@speed1: "You're talking of baked drivers not true talents..." "If you need telemetry to be fast you've already lost." I do not think you understand what is meaning of "AI". Suggest you to take some time to learn "AI". AI has been widely applied in many industries today, with a great sucess. @Domi: "Using telemetry like in that video you linked earlier to improve laptimes is not rocket science,..." No, but using telemetry data from AI driver is challenge and pretty new and great potential and benefits. In fact, I would say is of revolution. @Mrslfrsl: "ask the aliens if they would share their telemetrie for a specific car on a specific track you are practicing on? Pretty sure people would help out...." You missed whole point. No, I need no human help. I only buy AI driver for help. Yes, a simple click on my mouse. I careless about real human telemetry data. That shall belong to ISI to use for their modeling. In fact, good AI driver will be many times better and reliable than human. In general, here are steps how AI get trained and learned and perform (not necessary in race Sim SW): 1) ISI make an AI drive model by using same physics, basically list all non-linear variables as inputs; 2) Train the AI with real human data (could be several sets of telemetry data by "best" or "very good" drivers); 3) let AI learn and test, then refine. 4) store the models and implement to be used.
Meh, like with setting up a car, telemetry is something to support what is already in good development. i.e. a reasonable fast/consistent driver can improve quite a bit using it. bad/new drivers or those with little time will gain much less as they are still going for the basics. A good setup does not help much if you don't know how to use lines and zones, studying telemetry is equally redundant in such cases. And don't forget telemetry has a wide range of options, allot of people focus on certain bits (usually where they know they should/could improve) and not the whole thing all the time. Its all about finetuning and not about 'becoming fast' imo. that already has to be there to a certain extend. Edit: hm, maybe we are saying the exact same thing
If AI would drive 100% player physics, which is no even close at moment, then right, there could be a benefit to improve setups from AI telemetry. You could optimize rake etc. and other easily measurable settings. The quickest simacers in world, for example Morand and Huis from FSR, set up their car almost purely based on feel. The same happens in F1 often. For example simulations showed improvements in suspension which Ferrari added mid-season 2008, but it made Räikkönen much slower, which cost him his championship campaign, whereas it benefited Massa. So even a car that is better based on every telemetry measurable could still feel worse in your hands.
Joe don't get me wrong please but i don't need the AI to understand and trying to learn from the AI or telemetry. It is like you are trying to run with the help of a computer without to exercise mate. First you need to reach the limit of what is given than you can look for better shoes homie. I'm with you, i see you are intrested but thats the wrong way to go. Don't take those talent free drivers as example. Gesendet von meinem GT-I9300 mit Tapatalk
Yes me shortened a bit. I would have to add some more details and a lot more to say about but it's late and i'm jumping from one pc to another, to a smartphone and had already some good old brandy's, so you have to excuse me when i'm out for the first.
No way. AI cars (in all games) use all sorts of cheats in their code to drive at competitive speeds, that kind of driving isn't for humans. If you want to improve, you need humans telemetry.
One problem with that is aliens typically have a lot of simracing experience and have fiddled with endless setups. In prior years it meant exploitative setups, these days there could still be setups exploiting some areas of the simulation yes, but above all these setups can be quite tough to handle. I'm basing myself off an extreme here with GPL, being a complete novice and taking a Huttu or other setup and then having to dance on braking or accelerating wasn't the best.
YOur "AI cars" is different things than I mentioned. See my post above. My AI driver is trained and learned by real drivers' data. Hence, it can drive to limit. Human cannot (maybe 1% of your guys can). A good AI driver learned by only a few sets of real driver's data can actually outperform most of real drivers.
Your AI driver doesn't exist, and won't exist in the next 10 years at least, there isn't yet enough knowledge and computing power to expect to have an AI that drives naturally like a human.
You mixed apple with organ... Given a SIM with same physics models applied for both AI driver and a human driver, an AI driver will outperform most human. Say, John1, John2, and....JohnN are best human drivers out there. Their data will be used for trainning and test on an AI driver. In "non-linear programming" models, a set of objective functions will be imposed. In this case, maybe like, least laptime, least slip, largest radius on cornerning, least over/under steering, etc....an AI driver will be forced to meet those objective while learning and under realistic boundary conditions within physics allowed. As a result, the AI driver will be more or less with behaviour like John1, john2... JohnN, but perform far better than most of us....
rF2 AI is quite far from the application you describe where you optimize radius with respect to laptime etc. In rF2 you basically pre-record the AI fast path of the track by driving the car yourself in developer mode and then store the data in the track .AIW file, one for each track. It's an interesting topic of course and I know it's something that is studied, I heard a presentation about this few weeks ago. But they were speaking about optimizing something as simple as a few 180 degree corners, so even that can be a difficult task, depending how far you want to aim. rF2 has hundreds of tracks. Also we have to remember ISI has one guy working on the AI basically, it's one of the lesser focus areas I'd say.
Thanks for the info. Interesting. If so, I could not imagine a logic reason why they pre-record of driving non-rF2 car. Say for latest v1.47 Nissan GT-R GT1 car, why not record this car on rF2 (same physics as players do)? I made a video on this, see below and I post my description too: "Nissan GT-R GT1 v1.47 with following changes to previous v1.43: - Chassis flex implemented and new steering parameters. - CG increased. - Downforce reduction and slightly less yaw stable. - Slightly less downforce sensitive in slip-stream. - Updated to use new tyre mass method and slightly more accurate unsprung masses used. - Some tyre tweaks, mostly, but not limited to wear, load sensitivity as well as grip correction for wet conditions. - Automatic shift aid works a little better (less reluctantly). - You now also have to blip on down-shifts as per the real car. - Minor AI tweaks. - Additional radiator setting to prevent overheating in high ambient temperatures. This video is to show comparison between v1.47 and v1.43 in telemetry data run by AI driver on Silverstone. Two sets of telemetry data were recorded: “car behavior telemetry data”, and “driver behavior telemetry data. I ran three laps on v1.43 and v1.47 car, respectively. The 1st set of data was collected while running 1st 1 ½ laps and 2nd set of data was collected while running 2nd 1 ½ laps."
your monitors don`t have the right angle. Go to C:\Users\XXXX\Documents\rFactor2\UserData\Config.ini and edit this line: "SubViews=2" to "SubViews=3" and all is fine.
