The sphere feet are somehow easier to balance on, as explained in a previous video. I do not understand why or how this works, but it does make it look like they’re skating which is funny.
@@bugzilla318Without knowing the exact intricate math, I would guess that the reason is that the feet would effectively require two additional joints and the AI would need to be trained to use like actual feet (otherwise it would just be tippy toeing around and be pretty much the same as the ball feet AIs). The ball feet are likely a case of "good enough" approximation of what humanoid locomotion would be, without needing to spend an inordinate amount of time replicating the functions of ligaments and tendons that allow us to use our feet the way we do.
@@NopeNaw Also, reducing the number of inputs and outputs speeds up the learning and thinking process by an order of magnitude, especially on non-specialized graphics cards or cpus.
yea i think actual feet would be better(imagine walking with wheels as foot) but i remember he said something about it in the bowling video being too hard for the AI
Fellow Aussie here, I'm proud that you've gone from one successful video on BTD6 AI to one banger after another. My favourite so far has to be the cannons that never miss video because of the personality in the scenarios. Keep it up man, your animation skills and video production are excellent.
I started chuckling when the continuous stream of rainbow candy started going into your guy's mouth and nearly stopped breathing when the taser animation came in. All I can say is so true.
I know you have hundreds of thousands of subscribers, but i still dont think you have nearly what you deserve. Your content is so expertly made, being funny, entertaining, and informative all in one. Keep making such great content!
Goddamn hilarious to watch that AI struggle with the spherical feet you gave it, wow Does perfectly well standing in a small area but the second it needed to move any kind of distance without control over the arms (which did practically ALL of the balancing) it flopped over
<a href="#" class="seekto" data-time="107">1:47</a> If four burgers equals 1 pound then 773 divided by 4 would be 193. But 85 kg equals 187 lbs. so there’s a 6 pound difference. So technically 6 x 4 would be 24 then 773 - 24 = 749. There should only be 749 burgers as 749 divided by four is 187.
I think the jerky small movement is because you punish distance travelled by limb, not acceleration which is what actually wastes energy - a single big movement of a leg is less wasteful than constantly jerking it back-and-forth and having to apply a lot of force to achieve that acceleration. Nice video anyway, subbed.
b2, yours is the first youtube channel I think I've ever encountered (certainly the first AI channel) to consistently make pause the video because I'm *creased* up laughing. The witch's spell got me in the bowling vid, and now.... Will Chamberlain, eat your heart out! 🐟❤
I feared the fast limb movement punishment would make the ragdoll walk by taking rapid, short steps. It did end up doing that, but looking at how the face is moving, maybe it's not that big of a deal
Half of the impressions for this video must have been used on me. I see this video everywhere I go and all the time. I didn't look like a bad video but just a video I wasn't in the mood for. I finally caved and clicked on it. I enjoyed it
If the AI is sneaking data into the neck movements, why not just give it a stored value it can edit? Wouldn't that achieve the same result without the jittering?
I think this raises a bigger question about what constitutes programming? Because you aren’t actually programming anything, you’re just defining the parameters an algorithm used to program something.
<a href="#" class="seekto" data-time="70">1:10</a> Gotta love how its going to be the most dexterous creature alive once its finished training .....But only in its right arm lol
you should use stencil shadows for the lighting in your videos instead of shadow mapping, shadow mapping is noticeably blurry and low res and sometimes leads to the "peter panning" shadow bias effect at <a href="#" class="seekto" data-time="29">0:29</a> where when the ragdolls fall you can see the shadows disappear into the ground. Stencil shadows, often called "shadow volumes," are based on the actual geometry of the meshes instead of a camera image taken from the light itself and as a result have none of these problems. There's probably a plugin for these in whatever game engine you're using, it'll look so much better.
<a href="#" class="seekto" data-time="317">5:17</a> Hey, it's a SUVAT equation! Didn't expect that, I would've expected a Euler integration with Newton's laws of motion, which is the usual way people do this.
I'm curious to see how an AI would handle an "energy consumption" problem. Seeing them walk with as many taps and movements as possible in the arms and head is funny, but also only possible in a state of infinite potential energy. It could force the AI to learn how to use assumed positioning with bigger (more natural looking) movements, in other words, kinda like rollback netcode in online games
I see these walking AI videos often, and I notice they are never given feet, they always either have spheres or the end of a square leg. I wonder how it would go if they were given feet, even just simple rectangles that they could move.
<a href="#" class="seekto" data-time="29">0:29</a> "Table Tennis is a lot harder to solve than... being Spiderman" Not a quote I ever thought I'd hear, but here we are. Honestly this channel is hysterical, really enjoy the AI content!