/loco/

inverse dynamics

>>1031046
are there any viable alternatives without a "simulation" component

>>1031048
I don't see how. If you're bi pedal you don't have any other legs to stand on while trying to figure out where you can put your foot, you need to be sure of where it's going before it gets there, otherwise you fall over.

You might be able to build a neural net that creates useful heuristic models that can run on less powerful hardware, but that sort of bullshit would be a PhD in software, not a BS in EE.

You could have a 'bipedal' bot that doesn't really have a bipedal gait--just have it stand on one leg that is big enough to balance on easily, then shuffle one foot forward and slowly transfer weight to it until it's 'okay' and then move the last foot.

>>1031052
>If you're bi pedal you don't have any other legs to stand on while trying to figure out where you can put your foot, you need to be sure of where it's going before it gets there, otherwise you fall over.

Lets forget true bipedalism for a moment. My bot is small and light enough that it has propellors and can fly if needed. I'm thinking about getting it to be able to crawl and climb, but remain elegant. I lean towards the neural net more than inverse dynamics, because I've been programming for a long time and it seems to make more sense. But, I have doubts.

Do you know any good books on inverse dynamics?

MIT can barely do this

Boston dynamics spent millions kinda sorta doing it

Make a HEXAPOD. They are INFINITELY more terrifying, lower profile to ground, better payload options, more off-road capable, and you can weaponize it in an end of the world scenario

Look up the tamagotchi or whatever the tanks from Ghost in the Shell were called

>>1031056
Thanks for the hexapod tip anon.