00:00So here we're making robots that are made of robots, which is why I call them metamachines.
00:06So these robots are made out of modular parts that are themselves robots.
00:13And this means that if one part of the body is damaged or lost to injury, the rest of the
00:20body is fine.
00:21It survives. It continues to function.
00:53So this is all autonomous. There's no remote control.
00:56You can see it changed how it was moving once it got onto the concrete.
01:03So as a very vivid example, you could imagine chopping the robot completely in half.
01:11And what you would get is simply two robots.
01:14Chop any other technology in half, and what do you get? Two pieces of trash.
01:19So this is one advantage of a machine made out of machines, in which each part of the machine is
01:27capable of surviving on its own.
01:30So we really wanted to create robots that were more resilient and that could evolve.
01:36And to do so, we turned to nature.
01:39Which means it could be redesigned by someone else or evolved itself.
01:44Movement might seem like an overly simplistic task, but as far as we know, brains evolved to control movement.
01:54And so nature tells us, if you want to create an intelligent agent, start with movement.
02:01We evolved these robots to move themselves through the world with a little bit of athleticism.
02:09So more athletic than any other modular robot has been on land, more athletic than any other evolved robot has
02:18been.
02:19Non-modular and non-evolved robots historically have been much better at movement.
02:24Of course, they couldn't rearrange themselves or survive being chopped in half.
02:28So we really wanted to combine these two fields into one at the same time in coordination.
02:36The problem is that the design space, the space of possible robots you can build with these modules, with just
02:43a few modules, is absolutely enormous.
02:46So just with two modules, you can make almost 500 different two-part designs.
02:53With up to five modules, there are already hundreds of billions of possible valid body plans you can build.
03:02You don't know which design is good or bad until you give it the opportunity to learn, which takes a
03:10little bit of time.
03:20It is usually very difficult, if not impossible, to predict exactly what a robot will need to do before building
03:28it and deploying it into the world.
03:30So it would be incredibly useful if those robots could be redesigned and rebuilt in the field on demand.
03:48Yeah, it's very happy when it gets on the phone.
04:01See ya!
04:02You
04:03You
04:06You
04:21You
Comments