00:00Best place to start, with all of the abstract discussion on humanoids this week, frankly,
00:05what is it that's changed, that's allowed you now to say we have a production version of a humanoid robot?
00:11Well, we've been building robots a long time, and so we've figured a lot out about how to, A, build a reliable machine that you could mass produce.
00:19And that's what we've introduced at CES this week.
00:21We're showing our research version.
00:23It's been running, that's the two-hour wait to get into our booth.
00:26We've been running demos all week long.
00:28But we announced our mass production version that we're beginning to build out now.
00:34And we're going to scale production with Hyundai factories so that we'll be building these in the tens of thousands by 2028.
00:43There is always going to be the question, how real is this, right?
00:46So you and Hyundai are partners.
00:48They're a customer, but you're also working together to continually test.
00:53What is it that Alice is going to be doing in that facility?
00:56Is it revenue generating?
00:57In other words, is this a real commercial deployment?
01:00Absolutely.
01:01So look, our humanoid is our third commercial deployment.
01:05We've already got two robots out generating revenue today.
01:08And so the commercial opportunity is real.
01:11What's special about Hyundai is that they're both a producer.
01:15They can help us mass produce.
01:16They can also be a consumer to use robots profitably in their factories.
01:21You know, they've defined their own software-defined factory initiative.
01:26And robots are going to be a part of that initiative to make their factories more flexible and productive.
01:33The brain of Atlas, at least in the inference context, is NVIDIA.
01:40But a lot of this work you've done yourself on the software side in particular, where would you say Boston Dynamics' core competency is and how Atlas reflects that core competency?
01:51We've always been the whole body coordinated motion first, right?
01:56We basically introduced robots like humanoids that could really, in an agile and coordinated way, move around in the world.
02:04And so our core competency is in sort of the motion control and the low-level stuff that's in close proximity to the hardware.
02:12We're going to be teaming with Google DeepMind for the high-level cognitive functionality, so that tasks that go over minutes or even longer will be orchestrated by Gemini Robotics.
02:24So we're going to kind of split the low-level motor control and the high-level cognitive control.
02:29How you train Atlas in the first instance, particularly in the industrial use case, is fascinating.
02:35There is a great emphasis on virtual worlds, digital twins.
02:38I think the reality is teleoperation plays a role here, but how complex is it to get the humanoid robot to interact in a factory facility, taking into account the actual physics that's being applied to it in that environment?
02:54Well, this is where AI has really provided a breakthrough.
02:57We can use simulation tools to coordinate the low-level motion of the robot, so reinforcement learning.
03:04We use teleoperation.
03:05If you can teleoperate a robot to do a task in a factory, you can pretty much get an autonomous policy doing that task.
03:12So we've already demonstrated that multiple times.
03:14What we need to do now is generalize that so that in a day or two you can train a whole new task.
03:21Because ultimately, humanoids are going to need to do hundreds of tasks in a factory.
03:25So therein lies the question, like, right now, what is the problem we're solving for?
03:28Literally labor shortages, which Jensen Wong's been talking about all week, or productivity, efficiency gains?
03:35You know, what is it that you're selling with your products?
03:37Well, the high-level goal is to onshore manufacturing that requires enhancements in labor productivity.
03:43And I think robotics have a strong role to play in that context.
03:48You know, demographics are uniform the world across.
03:52We're seeing declining populations.
03:54It's difficult for manufacturers to get the talent that they want.
03:59So the robotics are going to take the dull, dirty, dangerous parts of the jobs.
04:03The humans are still going to do the high-level cognitive high-value tasks.
04:07The Google DeepMind relationship, like, it's a research relationship.
04:10But Hyundai has the potential to grow to something commercial at scale, right?
04:15I think that's kind of your vision.
04:18We're talking a few hundred units at first, but would you just kind of outline a roadmap, a timeline of how you see scaling?
04:24So the relationship with Google DeepMind is both a research relationship, but also a commercial relationship.
04:31So our plan is we're going to build 200, 300 of these robots.
04:36We're going to use them to collect data in factories at Hyundai, also at Google DeepMind facilities, also at our facilities at Boston Dynamics.
04:43Together, we'll develop the technology, and then we'll commercialize them together.
04:48And the first application, again, starting at Hyundai, will be to deploy robots in factories, doing part sequencing, and eventually final assembly.
04:57We will also work with some other customers along the way in logistics or other manufacturing applications.
05:02I guess my question is, we're talking a couple of hundred this year, maybe a few more hundred next year, thousands the following year.
05:07Like, what's the pathway?
05:08So by 2028, Hyundai's going to build us a purpose-built factory designed for mass manufacture of our humanoids.
05:15And that's going to have a capacity of between 10,000 and 30,000 units per year.
05:19And so that's the goal.
05:20By 2030, we'll be building tens of thousands of robots per year.
05:24I've always wanted to ask you, and I'm going to ask you, your analysis and impressions of Tesla's Optimus program, and your assessment of what they're doing and what they've done.
05:35Well, I think they're a very serious competitor.
05:39They've done a lot of very impressive work.
05:41They also have the advantages we have, that they have a consumer in Tesla to use those robots.
05:49Internally.
05:50Yeah, and I think it really requires both aspects.
05:53You need the technology from robotics and AI.
05:57You also need that customer and that consumer.
06:01And so automotive is a natural fit.
06:03And so I think Tesla is actually very well placed to succeed here as well.
06:08There's a different vision of the world.
06:09And all those people that are waiting two hours to see Atlas might have that vision, where a humanoid robot's in their home, child care, elderly care, manual tasks.
06:20Is that something that you are working toward as well, Robert?
06:22Absolutely.
06:23We're starting in manufacturing because we think that's the right place for the cost point, safety, and the capabilities of the robot.
06:32But as we mass produce these, we're going to drive the cost out.
06:36We're going to increase capabilities because now we're going to have tons of data to train these robots on.
06:40And they will enter the home.
06:42And I'd say five to ten years before they're in the home.
06:45The next five years, it'll be focusing on manufacturing.
Comments