Dear ISI, If this is already implementet into RFactor2, then feel free to give me a kind reminder and a verbal slapping. But I have noticed something with the AI that I find rather problematic. When the AI closes a corner they have a way of handling and braking, that gives them a huge advantage. I have noticed for example that they can break the car with so small consecutive tap's on the pedal, that it almost functions as ABS. The same goes with their way of handling the steering wheel, although it is not as apparent. It would be nice of the reaction timing of the AI could be slowed down, so that they would have to brake the car in longer strokes and use the steering wheel more fluidly, like any human driver. Cheers
That's typical for many games that don't have AI properly "layered" to split decisions from reactions. So something like this: "I want to get there", "...and since I'm currently here", "...I need to go that way now", "...but I'm turned the other way", "...so I need to turn right now", "...this means I need to have the steering wheel turned to the right", "...so I need to turn the steering wheel to the right". Sounds a bit weird maybe, but If you think of it, that's the decision chin you need, because this lets you separate decisions from reactions. First you start from decision "I want to go there" and it eventually leads to producing steering input, but there are other things considered in between. Of course any average algorithm will do the same, but the trick is to make all these layers run separately and not synchronously, like an algorithm. Just set up different reaction and decision times / delays / frequencies for each layer (varying by some small random number each time) and you got something believable. Many games have single "AI frequency" and it just "does the math" in one pass. I think rF2 has some separation (well, you have decisions like "I'm overtaking", "I'm pitting", etc.) but perhaps it needs a few more layers in between.
Mh, maybe it needs a few more layers as you said,...... in my eyes all math and algorythm seem good, they only have to smoothen out.
AI performance will always be relative. Even if you slow them down, they still use none of the new tire model stuff, they use the old tire model from rF1, so it's not really comparable laptimes to human because they drive an entirely different car. It's just about finding a good AI strength and aggression setting for each track and use that for benchmark.
Its not about slowing their speed down, but slowing their reactiontime down. It seems reasonable that I would never be able to compete with opponents that effectively can drive and break like they have perfect steering and ABS breaks.