it's bimonthly "felthry is amazed at the fact that inverse kinematics is computationally easy" day again
-F
I mean, if you build a little neural network, yeah. Good luck otherwise though, what with the non-closed-form solution thing. :P
@zetasyanthis i'm talking about its use in 3d graphics for positioning limbs with reference points only for the feet/hands, specifically! it probably gets a lot harder with an unbounded number of articulation points
-F
Oh gosh, yes, that would definitely make it harder. You'd need a bigger and bigger neural net. (Last time I played with them, a 3-5-3 feed-forward was enough for a limb with three articulation points. NASA has a paper on it somewhere an used the same thing for the space shuttle arm.)
@zetasyanthis they've been using IK stuff since the early days of 3d graphics though, before neural networks were a thing people researched. There's IK in the original n64 release of Ocarina of Time!
-F
Oh, those have been researched for a long time. They just haven't been the omg hotness in software engineering until recently.
Not surprised! We ran 4 copies of our neural net (for a four-legged robot with three articulation points on each leg) on a little PIC micro-controller after training stuff on a regular computer.
@zetasyanthis now i want to look at the code for ocarina of time or mario 64 and see how they did it
i thiiiink mario 64 had some IK?
-F
@zetasyanthis i won't, though, because i have better things to do this weekend than slog through MIPS assembly for hours.
-F
Yeaaah. As much as I find computers fascinating, chunking through assembly for hours is not my idea of fun. XD
@zetasyanthis assembly is neat but we are 100% entirely unfamiliar with anything to do with MIPS so it would be a particularly arduous process
-F
@zetasyanthis Well, either that or neural networks have been a thing used in video game programming for way longer than we thought
-F