sure, the music credits are listed in the text description (you'll need to click on the "show more" tab, and they are at the end of the video as well. To save you work, here they are: Music provided by NoCopyrightSounds Alan Walker - Fade: www.youtube.com/watch?v=bM7SZ... Alan Walker: ruclips.net/user/DjWalkzz cheers, Michiel
How likely is it to adapt this kind of work to physical robots, using world sensing technology similar to what they use on autonomous cars? I imagine the mechanical capabilities of current gen robots is the most limiting factor? All of this ignoring the bulky power systems of today, obviously.
I disagree, the most limiting factor of robots today is the software not the mechanics. Deeploco is cool because providing it is fed a terrain map then this will totally work in the real world on real hardware robots. As we have learned in recent times, agility and locomotion is not a hardware problem, it's a software one. We've had the hardware to make drones for decades but it's only recently we've been able to build software to fly them. Drones would not fly without machine learning, it would be impossible for a human programmer to come up with all the variables required for flight.
You only have to look at your own body, it's mechanics are very simple however the amount of "calculations" that go on in your brain to keep you balanced is overwhelming when trying to recreate that using computer programming. The big breakthrough happened when we learnt to build neural networks (or AI as it's called in the media) that could learn to walk in the same way a human would. Before this happened, it was essentially impossible. Deeploco is just one example of a software neural network that's able to learn on its own how to walk, and run across terrain (as opposed to a dead flat surface). This is something that we take for granted because we don't have to consciously think about it, our brains do it for us.
A good example of this is watching a young baby kitten trying to walk, it looks like a complete failure, however fast forward 8 weeks and it has turned into one of the most agile creatures on the planet. It has the same hardware as when it was a kitten however it's "software" learnt.
lmao the music choice is golden. never thought NCS would make its way to academic stuff
I think I'm in love with just watching it move.
Where to download? I could be using this for videos! What would happen if it tried parkour? What if it free-runned?
indeed DL is going to make a huge improvement on animation techniques
Wow! Great work guys. Mesmerizing
What kind of simulation framework do you use.
And really impressive work, I like it!
Should really put the music credits too ;), mostly just because I want to know the soundtrack.
sure, the music credits are listed in the text description (you'll need to click on the "show more" tab, and they are at the end of the video as well. To save you work, here they are:
Music provided by NoCopyrightSounds
Alan Walker - Fade: www.youtube.com/watch?v=bM7SZ...
Alan Walker: ruclips.net/user/DjWalkzz
cheers, Michiel
Looks like me when I try to walk on ice.
How likely is it to adapt this kind of work to physical robots, using world sensing technology similar to what they use on autonomous cars? I imagine the mechanical capabilities of current gen robots is the most limiting factor? All of this ignoring the bulky power systems of today, obviously.
I disagree, the most limiting factor of robots today is the software not the mechanics. Deeploco is cool because providing it is fed a terrain map then this will totally work in the real world on real hardware robots. As we have learned in recent times, agility and locomotion is not a hardware problem, it's a software one. We've had the hardware to make drones for decades but it's only recently we've been able to build software to fly them. Drones would not fly without machine learning, it would be impossible for a human programmer to come up with all the variables required for flight.
I see. I didn't know robotics were advanced enough. I'm totally looking forward to seeing some of this applied to physical machines.
You only have to look at your own body, it's mechanics are very simple however the amount of "calculations" that go on in your brain to keep you balanced is overwhelming when trying to recreate that using computer programming. The big breakthrough happened when we learnt to build neural networks (or AI as it's called in the media) that could learn to walk in the same way a human would. Before this happened, it was essentially impossible. Deeploco is just one example of a software neural network that's able to learn on its own how to walk, and run across terrain (as opposed to a dead flat surface). This is something that we take for granted because we don't have to consciously think about it, our brains do it for us.
A good example of this is watching a young baby kitten trying to walk, it looks like a complete failure, however fast forward 8 weeks and it has turned into one of the most agile creatures on the planet. It has the same hardware as when it was a kitten however it's "software" learnt.
That does make sense. Good example.
Why is it always so funny when AI falls??? XD
Great!
WHY!!! at 3:16
abdullah kepceoglu ahahahahahahah
I immediately scrolled down to see if anybody made that comment)
Because reasons.