00:00You can say, Brett, definitively, over the 24 hours or more, there was no teleoperation.
00:06A lot of people in the comments, as you know, pointed to the idea, and I think we have video
00:10of it,
00:11that the three on shift kept gesturing to the head, which is a telltale sign in robotics of teleoperation.
00:19You pledged that there was none?
00:22There's absolutely no teleoperation into this.
00:24The robots are all operating fully autonomously using an onboard neural network redesign called Helix-2.
00:31Sometimes when the robot takes a turn to left to grab packages, it moves its left hand out of the
00:35way upwards.
00:36You'll see this behavior happen every single time the robot turns for packages.
00:40But we've been running autonomously now for close to 50 hours.
00:44The robot's operating shifts.
00:46There's been basically almost no downtime on the belt.
00:50We've pushed over close to 60,000 packages.
00:54And we're just going to keep going now and see how far this can go.
00:57So this was live streamed, right?
00:59And that was one reason I really wanted you to come on the program, because there's the bit people don't
01:03see.
01:04What's happening behind the scenes, like in shift changes?
01:07Where does the robot go?
01:10Like, does it need maintenance?
01:13For the most part, the robot's operated on a four-hour battery life.
01:17After the battery's low, the robot messages another robot to come out to take its place.
01:23The robots didn't do a swap.
01:24The robot that's just left the conveyor system is going to go charge wirelessly on the stand, while the other
01:30robot continues to do work.
01:32If there are issues, say we have hardware or software issues, the robots can basically walk off into maintenance and
01:38call another robot to take its place.
01:39The goal is to be able to enlist it 24-7 operations with basically no failures on the use case
01:47itself, which we haven't had to date.
01:49So basically, the conveyor system has been running 24-7 since the middle of this week.
01:55I think we're approaching 50 hours of just full, like every single hour since we've launched.
02:01The robots have been basically doing work now on this line, which I think is crazy.
02:07Figure wants to build like, you know, we want to build like iRobot.
02:11We want robots everywhere in the world, in the commercial market.
02:15This is like the first large step to doing that.
02:18Okay, so what's harder now?
02:21Robots getting faster or making them even more reliable?
02:25Because at the moment, you seem to be doing both.
02:27The robot that you're seeing here is roughly operating around human speed, just about three seconds a package.
02:32That's the requirement to operate on this logistics line.
02:35So we're at human parity and speed.
02:38The goal is also to have like 90% success rate on the package flips for barcode scanning.
02:45We're in that as well, in that requirement.
02:49The robots are also like getting extremely reliable.
02:52Like part of this whole process of running this 24-7 with no downtime is to show how reliable humanoid
02:58robots are.
02:59And four years ago when I started the company, humanoid robots were falling.
03:03They were extremely unreliable systems.
03:06We've designed the systems and engineered this now to a point where the robots are, I think, extremely reliable.
03:11I think we're showing that now on the live stream to the entire world.
03:14The big focus for us is like how do we solve for a truly general purpose machine?
03:19And then how do we manufacture at unprecedented volumes similar to cell phones today?
03:25So this week, BotQ, our manufacturing facility, will manufacture anywhere between like 60 and 70 humanoid robots just this week.
03:32And we do it right next door on the figure campus.
03:36So, Brett, I want to get into the idea of this is full stack.
03:39I've been hearing a lot growing speculation that open AI could get back into robotics.
03:44And you have a history right where you had a partnership.
03:47You decided on the software side you could do better yourself.
03:50You know, what do you make of that idea that ultimately a big party is going to want to own
03:57the entire stack that powers the humanoid robot?
04:02To really do this right, like if we want to really build like iRobot, like the movie.
04:07Yes.
04:08You know, basically you have to design the entire hardware system almost yourself.
04:13Like motors, like stator rotors, all the electromagnetics work.
04:17Like you have to do all the battery systems work.
04:18You have to do all the actuator design, sensor design, kinematics, like structures, like which we do all in-house
04:25now here at Figure.
04:26We also manufacture the robots, and then we also test them.
04:30And we do all the AI data collection and all the AI neural net training ourselves here in-house.
04:36So, basically this is a full end-to-end vertically integrated system that we now have out doing real use
04:42case work like humans do.
04:44And we can do this at human speeds.
04:47And we're doing this now for like, you know, this example, most of these shifts run like just eight hours
04:52a day.
04:53But we're doing this like 24-7 just to show how reliable the systems are and how like mission ready
05:00these things are to get out of scale.
05:02So, anyway, to solve this, you have to have like a truly vertically integrated approach from top to bottom.
05:09What about bottlenecks for you?
05:11Is it now money?
05:12Do you want to go public, Brett?
05:13What are you needing to get this out more broadly?
05:17Our largest two bottlenecks are like data for pre-training our Helix neural net and in manufacturing.
05:25We're spinning up manufacturing here.
05:27Our manufacturing facility is called BotQ.
05:31We're now at several, like thousands of run rate annually production that is continuing to scale up.
05:40You know, I think on the data side, we're collecting and training kind of unprecedented models for like our AI
05:51stack here internally that we've ever done.
05:54So, I think we're like we're and we have, you know, we have well over a billion dollars of cash
05:58in the balance sheet today.
05:59So, I think from a financial perspective, we're in a good spot.
06:03We're manufacturing at pretty much unprecedented volumes for ourselves.
06:05And we're building like next generation AI models that I think are, to be honest, are just completely mind-blowing.
06:12And so, the goal is like the goal is to solve the data problem and the manufacturing problem to get
06:16humanoid robots out of scale.
Comments