➡ The company is developing robots, with plans to start production next year. They’re currently testing and improving the product, aiming to make it reliable and safe for use in homes and workplaces. The goal is to have robots in homes within three years, starting with pilot tests in controlled environments. The company is also focusing on making the robots affordable and useful, with the potential to perform tasks like errands and chores.
➡ The robot’s lower body parts become stiffer as they widen, which is beneficial for handling loads and maintaining balance. The robot’s hands, now in their fourth version, have seen many improvements, including better sensors, packaging, strength, and finger speed. This has enhanced the robot’s ability to perform delicate tasks and handle objects like a human, which is crucial for the robot’s overall functionality.

We have, you know, several of them here now, uh, on the floor. Um, they already started doing some tests on it. We’re starting a production line next year. So, um, we’ll start with a few as we, um, you know, have like engineering really close with those customers and making sure that the product really works well. There’ll be like a bunch of bugs and process improvements we need to make to make those like a well-oiled machine for the, uh, for the market. We want to do those with those customers now. And, you know, as we get to some level of product maturity, we’ll, we’ll branch out to more and more customers.

And something that we’re going to spend more and more time on is also like robots in the home. Every six months that goes by, I go to a couple folks here and I say the timeline for us getting in home is accelerating. And one of the things that’ll be really helpful for us getting into the workforce, it’ll really help us get system reliability up, safety up, and the cost down and volumes up for manufacturing. It’ll really help us in the home because you really want, uh, like you’ll probably have like an order of magnitude, uh, pricing collapse going into the home from the workforce.

And you really need economies of scale to get there and you really need a safety certified system, uh, in the home. So like you need a really safe product in the home. So, um, we, we’re using the workforce in a lot of ways to bootstrap that vision for us. And, but I think from a performance perspective, I think I, I think we’ll start doing early work in the home and, you know, and then like, you know, I think the, the home for us is an area where, yeah, I think it’s accelerating in my mind every, every six months.

I would say within the next three years, we’ll definitely have robots piloting in homes. Probably start with some lock homes here in our facilities and start, you know, giving the bugs worked out, understanding how to, the system architectures all work. But, uh, I’m interested in seeing what problems we face that we’re not prepared for that are limiting our ability to get in the home longer term. Yeah. We’re actually starting our production line next year. We’re gonna do it here in California, like close to engineering. So we really work out the kinks, um, on like more of a traditional like kind of pilot manufacturing line or production line.

And, um, and then from there, we’ll, we’ll start announcing our intentions for like, we’re like, like a high rate manufacturing will look, but, um, our production line is being, uh, we’re starting to design it as we speak. Uh, we’ll have robots coming off the production line next year. Uh, we’ll start with hundreds of robots and then thousands like us, uh, we really want to get the product. What’s more important for us than just like having like, you know, here’s 3,000, 4,000 robots out is getting the process really dialed in for how to do that. And then also making sure those robots that we’re producing are working really well.

There’s a situation where, um, anybody easily can get into where we have like too many robots that don’t work well and you’re basically in this like recursive loop, fixing them, uh, always having them down like volume doesn’t really help. So you really need the system reliability to be at a certain point and the production, like the performance of those robots, then into the end of the use cases to be pretty high. Uh, at this point, you know, building, we’re building one a week. So it might seem, you know, and then we’ll start building like one a day and then we’ll start building multiples a day here next full month.

So like, I think, um, but it seems like pretty straightforward path. Um, I mean, we make cell phones almost by hand in the world to make a few billion a year. Like this is more complex than a cell phone, but far less complex in a car. So yeah, I feel like the path to making thousands in the, you know, in the near term is just not super difficult. What’s the difficult part is making those thousands, uh, to a point where they’re really useful and really work. That’s like, that’s the name of the game for me. It’s been all year now looking at how cheap we can make these robots and the answer really lies in like a bottom up bottoms up analysis, call it roughly a thousand parts.

And then we start understanding at real scale, um, how are we going to procure those parts, uh, whether we build or buy and what, um, contractual, uh, you know, volume estimates we can get with price prices there. So yeah, I think like, uh, over a long enough period with enough high volumes, I think you’re getting these costs down like sub $20,000 a unit, like really cheap. If these robots can do everything a human can, I have to think that we’d be able to put, um, uh, three to five billion in the workforce. And I don’t see any reason why every human wouldn’t want to have a humanoid like you do a car or phone, perhaps even more important than a car phone where you can just do all the work that you just don’t want to do all day.

Whether it’s, you know, walking the dog, getting coffee, uh, doing errands, um, doing laundry chores. Like I think the rule of thumb here is that like roughly you need probably minimum three hardware versions to get to a point where the hardware is, um, uh, relatively commercial and reliable, like bug free. And our goal is to in the limit, make this basically a software limiting issue for us. Um, which means we need a really capable hardware that’s really reliable. It’s safe, it’s low mass, um, and it’s like low cost and we can manufacture really well. Um, that’s like a light, a lot to bite off in like the first version hardware to get that right.

It’s just too difficult. I mean, um, so we basically want to be on this continuum of like rapid, uh, like hardware iterations where we’re, um, basically looking at different heuristics of things that we need to mature over those hardware continuum and making all like necessary improvements, uh, so that the hardware is at some point, um, very mature. And I think our first generation hardware, uh, figure one was mostly trying to get the roughly the architecture trades, right? So all the details of like the engineering system for, you know, um, as a battery as an example, um, what is the energy? Is it going to be hydraulic? It’s battery powered.

What type of battery cell chemistry, um, from there, what type of path type of cell is it cylindrical? Is it, um, prismatic, uh, pouch, um, you know, how are we going to, how are we going to pack those, um, thermal propagation, like all those different trades you have, that’s just the battery alone. And then you have like the rest of the whole system. So that, you know, there becomes like an order of like a hundred to 200 decisions you have to make to go out and build a robot. And I think we got most of those right on figure one.

We did even better job on some of those decisions on figure two. Figure two was really about getting to a feature complete robot has all the systems on it, whether we’re going to build or buy it on the robot that are working. We built most of it. So that’s software with those firmware, but it’s systems control software, um, uh, all the hardware systems on board, uh, you know, the actuators, electronics, wiring, battery systems, cameras, sensors. Um, so, and then, you know, um, so we think we got roughly to, um, uh, like feature complete hardware complete on figure two.

And so we’re really excited there. And then, you know, future generation for us will be, how do we get the costs, um, down by well over order of magnitude from around now? And how do we get the ability to manufacturer like unprecedented scale? And, um, we have, um, right now we’re manufacturing about one a week. In our facility here in California. And, um, and we have, you know, several of them here now, uh, on the floor. Um, so here’s, you know, a quick look at, uh, figure two robot that we’re, um, already started doing some tests on.

First is we tripled the amount of CPU and GPU on board, uh, just for more overall compute and inference. Uh, the second is we almost doubled the battery to about 2.3 kilowatt hours. Uh, it’s all on board the system in the middle of the torso, uh, gear next to basically the compute and, uh, and GPU. Uh, we had all the wires all internal, uh, that’s really for reliability and for overall packaging. Um, we also have exoskeleton structure. So all the outer shells of the robot, uh, actually take loads. We also have, uh, six onboard cameras. So we have more perception.

We have them in the head, in the back and in the lower torso. Yeah, it’s basically like, uh, like overall, like, um, the parts will get a little bit stiffer as they get a little bit wider. Um, we found that, um, having one structure for both crass loads and stiffness, uh, is the right ideal mass trade. These are our, um, uh, fourth generation hands now. And, uh, we’ve made like quite a lot of improvements, uh, over the previous generations. Um, better sensors, better packaging, better for mass, better strength. Better speeds of the fingers. Um, overall just better dexterity and control of like fine grain manipulation that we’re doing, uh, on board the robot.

Uh, we need to do like human-like applications. Uh, so the more here that we can, um, do human-like tasks and grab human-like objects, the better for, uh, for generalization of the robot. [tr:trw].