A couple notes: -It looks janky as hell because this is just the first pass proof of concept stage. There's obviously a long way to go before it actually looks humanlike by any means. Besides cleaning up the jitter, I would like to eventually give it multiple forms of inaccuracy as well as at least rudimentary pattern recognition so that if you AD spam, it performs better than overshooting repeatedly. -This was actually really easy to implement because I was able to utilize almost the exact same logic as my lag compensation system. -The end goal here, for me, is just an NPC that aims in a way that looks and behaves pretty similar to a human, with all of the inaccuracies and limitations of a human. The purpose is basically just for funsies.
honestly would love to play a pve game where the bots are human like in their aiming, i used to spent so much time playing against bots as a kid and it was fun but i think it would be more fun if they were human like
yeah probably. it's actually a bit fucked right now tho so i can't really just copy paste it. but the concept is simple enough. record locations for the target, then extrapolate out a predicted position from some of those historic location records (going back into the records an amount of time equal to the reaction speed you want the bot to have, and extrapolating forward that amount)
How do you deal with a player spamming A/D very fast? Surely extrapolating by current velocity wouldn't work Also a thought, human aiming in FPS games is a thing we have gigantic amounts of data for, imagine training a neural network on CS:GO demos to simulate human aiming / reactions.
It extrapolates by using historic velocity. For example, if it has 150ms reaction speed, it extrapolates based on the velocity 150ms ago. If you AD spam, it continuously overshoots like a human would if they attempted to track based purely on reactivity and not pattern recognition. I would like to eventually give it at least rudimentary pattern recognition so that if you do AD spam, it will perform better than pure reactivity. And yeah a machine learning'd aimbot would be cool to see.
