Re: Interesting project about mouse/ gamepad latency
Posted: 09 Jul 2019, 21:54
You can count Linus's clock, it goes to 59 and then 00.
My vague recollection of my proofs from long ago is that under some conditions, pure left/right dodging is optimal. Various physics models here are AD instant constant velocity (Overwatch uses this) and AD instant constant acceleration with exponential damping (I think Quake uses this). So "standing still" would be simulated by rapidly alternating left and right, if the acceleration/position constants in the game make that a valid option under the optimal strategy.
This neglects further physics models such as if jumping evacuates the player's position in less than the aimer's reaction time, or Overwatch's crouching.
The dodger does have some memory, of his own position up to 200 ms ago (if the aimer has 200 ms reaction time). This is hidden state that he can use to make decisions with. So his dodging is not necessarily exponential until after he has moved for 200 ms in the same direction.
I understood your reasoning behind not using position waveforms, and it's true that a direct cross correlation would not work with position, but you can finagle it to make it work with some stupid cutoff tricks. With acceleration, it's much harder to make it work, and your method doesn't suffice. On the other hand, it'll be very hard to work with in either case; you will only be able to get a very loose upper bound on reaction time, since kukkii's aiming is not optimal and won't fit to a model. His mouse tracking is imperfect, like everyone else's in the world. And when the model doesn't fit properly, the results become more conservative.
For those who find the statistics hard, here's an easy visual way to see why:
When Marwan moves in a straight line for a while, kukkii wobbles his mouse like crazy but no correlation is being done.
When Marwan starts alternating his position, kukkii keeps wobbling his mouse, just like before, and now we get some really spurious correlations with Marwan's alternating position.
So when analyzed by the method in the article, kukkii's crazy wobbling gives him a big reaction time advantage when Marwan alternates, and doesn't hurt him when Marwan keeps moving straight. You can think of it like constantly guessing on a reaction time test, where the test doesn't penalize for wrong guesses.
My vague recollection of my proofs from long ago is that under some conditions, pure left/right dodging is optimal. Various physics models here are AD instant constant velocity (Overwatch uses this) and AD instant constant acceleration with exponential damping (I think Quake uses this). So "standing still" would be simulated by rapidly alternating left and right, if the acceleration/position constants in the game make that a valid option under the optimal strategy.
This neglects further physics models such as if jumping evacuates the player's position in less than the aimer's reaction time, or Overwatch's crouching.
The dodger does have some memory, of his own position up to 200 ms ago (if the aimer has 200 ms reaction time). This is hidden state that he can use to make decisions with. So his dodging is not necessarily exponential until after he has moved for 200 ms in the same direction.
I understood your reasoning behind not using position waveforms, and it's true that a direct cross correlation would not work with position, but you can finagle it to make it work with some stupid cutoff tricks. With acceleration, it's much harder to make it work, and your method doesn't suffice. On the other hand, it'll be very hard to work with in either case; you will only be able to get a very loose upper bound on reaction time, since kukkii's aiming is not optimal and won't fit to a model. His mouse tracking is imperfect, like everyone else's in the world. And when the model doesn't fit properly, the results become more conservative.
For those who find the statistics hard, here's an easy visual way to see why:
When Marwan moves in a straight line for a while, kukkii wobbles his mouse like crazy but no correlation is being done.
When Marwan starts alternating his position, kukkii keeps wobbling his mouse, just like before, and now we get some really spurious correlations with Marwan's alternating position.
So when analyzed by the method in the article, kukkii's crazy wobbling gives him a big reaction time advantage when Marwan alternates, and doesn't hurt him when Marwan keeps moving straight. You can think of it like constantly guessing on a reaction time test, where the test doesn't penalize for wrong guesses.