Wow, this presentation is actually brilliantly entertaining. These creatures are too adorable. I think the way they trot on their pointy feet makes my brain happy.
@@divinephanes ah. ohhhhh ive misinterpreted this comment then my bad. in that case yeah i do agree this is like something that would get posted there. whimsical science. whimscience.
Your "chicken" tet reminds me of Diogenes' "man" but inverted. You explain the essential attributes of a chicken, then present something that has all those attributes but obviously isn't a chicken.
Having a penalty to erratic movement/output is something I've wanted to see people do in alife/AI for a while now. When you watch people doing AI toy examples or getting an AI to play some game, said AI is constantly jittering back and forth, probably from some tiny perturbations in the input causing radical changes in the output that just happen to still get it where it is supposed to go.
What you mentioned about walking motion being less about direct neural control and more about the natural physical dynamics of the motion was fascinating. It has so many implications regarding hand-eye coordination that I hadn't thought about. Cool stuff!
God this video has so much meme potential 9:36 my body, on 7 cups of coffee, giving up after powering through entire nights to finally finish a month long assignment thats due in 3 days.
Very cool! I always thought embedded cognition and balance were neat topics when I was studying cognitive science, even if I ended up taking a different stream of the program. It's nice to find work in machine learning that's not so cynical. I'll have to check out Vehicles too.
Yes! It's not just that the brain is somewhat hands off, it continuously engages the PNS and muscles as part of thinking. Embodied cognition really means what it implies.
All fun and games until one falls over in a way that keeps its non-foot points off the ground and vibrates toward the point - algorithmically siring an entire lineage of vibrating worms that evolutionarily capitalize on the simulation's physics weaknesses.
An interesting modification would be giving each spring multiple sine waves and adding them up: it's a known mathematical result that you can approximate continuous functions with a sum of sine waves of differing frequencies and amplitudes. Since walking is such a periodic motion, probably 10 sine waves per spring would already give very rich motion control, potentially unlock far more efficient motion. (Especially if using 2 or more tetrahedra per leg to give more range of motion to the legs)
I’m absolutely obsessed with the way these little guys move, the whole design of them is cool af, watched the whole thing with a big dumb grin on my face
A missed bit of available movement I noticed: Harmonics of the main frequency can still be in phase with the overall motion, and open the way for complex gaits. For instance, if I want to make a figure eight movement, I move with one frequency in one direction, and double of it in another.
I wonder - what if you could let the creature's body evolve? Let it add tets, assign springs as balance checkers. What would be the most stable creature?
The algorithm is always listening. 😅 A penny here, sone location data there, a dev document or two thrown in the hopper and then it say oh yeah, I was supposed to show you this give days ago but you were busy and you sat your project down for a while but hey, now that you're back at it, check out this dude. Oh, remenber how you used to buildnhets and helicopters? Want to see an add on being a pilot? No? Okay, maybe again later. 😂 Yes? Can you remind me why that worked out this time? No? Its clasified? Got it. 🎉 LOL thanks, Skynet.
Id be interested in knowing if theres a way to incorporate backpropagation as another way of converging on valid solutions. Im kind of a noob in regards to what situations it can be applied, but this intuitively seems like an interesting one to try it with
Have you read the Fabian, et al. nature communications article? The one about dorsal light response in insects, "Why flying insects gather at artificial light" it feels relevant to the zenith navigation method, or I guess how real world examples have inherent flaws.
I do a little coding sporadically . I remember animating a ball under gravity in BASIC then I learned some Javascript & OOP and suddenly it was so easy to have as many bouncing balls as I wanted and even the code looked great ! Fascinating video . Shame about the knuckle-dragger comments from some.
