It’s not exactly predicting player input, the idea is to correct input lag (say, because the processor is under heavy load) by making microadjustments in the timing so that an action happens when the player intended it to happen, instead of missing (ex:) a jump to another platform because you had been correcting for lag and, when the lag stopped, you speculatively had hit the button few frames early. This system would theoretically correct for that discrepancy, so the user input would happen when the user intended for it to happen.
It’s… not the worst application for AI I’ve ever seen, but that’s really not saying much. It seems like you could implement this with ~150 lines of assembly instead of a full ML model, but what do I know…
you can predict, but you can’t be right all the time. the more constrained the inputs are and the more you observe the behavior, the better you can get at the predicting. and this thing just has to predict the extremely near future.
How can you even predict player input?
Even when trying to do the exact same thing, you cant
It’s not exactly predicting player input, the idea is to correct input lag (say, because the processor is under heavy load) by making microadjustments in the timing so that an action happens when the player intended it to happen, instead of missing (ex:) a jump to another platform because you had been correcting for lag and, when the lag stopped, you speculatively had hit the button few frames early. This system would theoretically correct for that discrepancy, so the user input would happen when the user intended for it to happen.
It’s… not the worst application for AI I’ve ever seen, but that’s really not saying much. It seems like you could implement this with ~150 lines of assembly instead of a full ML model, but what do I know…
you can predict, but you can’t be right all the time. the more constrained the inputs are and the more you observe the behavior, the better you can get at the predicting. and this thing just has to predict the extremely near future.