Rob Cochran on Shipping a Developer-First Humanoid at Fauna Robotics

Transcript

[00:00:00] Rob Cochran: We're doing something a little unconventional in that we have taken on the humanoid  form, but it is lightweight. It's 45 pounds. It is three and a half feet tall, so it's not a full size humanoid. We went  through many iterations, things that hewed closer to kind of anthropomorphic design. We ended up feeling like  a much reduced, simplified, more abstract form. Allowed for kind of clear expression of ideas without  overloading with the kind of things that you know might cause uncanny valley sort of experiences.

 

 [00:00:37] Brian Heater: Welcome to Automated. My name is Brian Heater. I am the managing editor at the  Association for Advancing Automation. We have a very fun episode for you today. Rob Cochran is a founder  who, uh, I spoke with actually a few years back in the very early days of what would ultimately become Fauna  Robotics. We also caught up with him, uh, about a month or so ago. As the startup officially came out of stealth  and launched its developer robot, Sprout, lots of interesting insight into the company that's doing something a  little bit different in the humanoid space. If you're joining the show, don't forget to like and subscribe and rate  and review us wherever you do the two aforementioned things. And now please enjoy this conversation with  Fauna Robotics. Not all robots work in predictable environments, but every robot depends on predictable  motion. In robotics, motor level precision drives long-term reliability. SANMOTION from SANYO DENKI  delivers deterministic multi-axis control with synchronized EtherCAT communication, stable one-millisecond  cycle times, and motors, drives, and absolute encoders. Engineered to perform as one system. Advanced  robotics needs motion you can trust. SANMOTION from SANYO DENKI — Precision in Motion. Learn more at  sanyodenki.com/america.

 

 [00:02:00] Brian Heater: I dunno if you're aware of the fact that, uh, PitchBook seems to think that you're an Ag  tech firm,

 

 [00:02:07] Rob Cochran: a misnomer. Somehow Flora and fauna frequently get confused, but we are Fauna  Robotics, so, uh.

 

 [00:02:14] Brian Heater: Animal and robotic life?

 

[00:02:16] Brian Heater: Well, first of all, I've never launched a  startup, obviously, you know, I've been on the, um, I was at TechCrunch for a long time and I was on the  journalist side of things, but I'm really curious what that period is like. Um, the stealth period, especially when  people are actually like actively looking at you, watching you, following you, and trying to figure out what it is  you're doing.

 

[00:02:38] Rob Cochran: You may or may not remember, you and I spoke, I want to say in March of 2024  really, when we were just getting out of the gate. Yeah. It was a brief conversation and you said, you know,  even in these early days it might be worth, you know, talking about it, doing some press and

 

 [00:02:51] Brian Heater: Yeah, yeah, yeah.

 

 [00:02:53] Rob Cochran: We weren't really sure exactly the story we wanted to tell yet, but you know, as things  evolved, you know, we built what we felt was a really great team. We had a very clear perspective on the  products we wanted to build, and we knew that it would take time to actually get there. We felt there was  frankly no advantage in starting to share that story ahead of actually doing the work and being ready to go. So  we're now fortunate to be in the position where we, you know, have paying customers, we're shipping robots,  and, and now this exciting, exciting launch coming up and it feels like the right moment to really share, you  know, our vision and, and what we're up to.

 

[00:03:28] Brian Heater: I do remember that, and I remember that on the strength of just, um. You know, like,  oh, these, you know, the control labs, you know, Xed guys are launching, uh, robotics. And I, I, and I think at, at that time it was even clear that it was like a humanoid robotics company. So clearly this is something, you  know, that TechCrunch is going to want to cover. [00:03:47] But yeah, again, it was still too early for you to  really want to talk about at any kind of detail. But at that point, so. Uh, I guess now, but two years ago, um, in  two months, uh, what, what was the vision for the company?

 

[00:04:02] Rob Cochran: Josh and I had crossed paths at Control Labs, which, you know, uh, wearable  Neurotech company that I, I led product at, and he did some early research and I think shared this idea that  some of the technology. Around general purpose robotics was, was converging to a point where you could  build something really interesting and exciting, but it felt like there was no way to really express that on a  platform that you could put into human environments deploy safely among people. And, and you know, we  were really excited about the consumer space and remain quite excited about that, uh, long term, but felt like  there wasn't really a straight line path to get there without a chance to really get. Get to shipping products. So  the, the way we evolved over time was to say, you know, how do we get to the north star of a, you know, mass  market, consumer humanoid product? And it felt like if you look at personal computers, smartphone, a lot of it  comes through developer programs and the ability to kind of unlock the innovation of, you know, thousands of  people building, you know, many, many concepts. And so we thought you start with a platform that's. Safe,  engaging, affordable enough to be deployed at scale. And that could be the right starting point for kind of  igniting an industry and moving towards that ultimate goal

 

 [00:05:14] Brian Heater: as, as we're recording this, um, I, I just spoke to, to Josh, God, I guess it was  yesterday. [00:05:20] Ty. Time is, um, I. I don't know. You know, you say March, 2024 and I, I couldn't even, I  actually had to do the math on how long ago that was. Um, but, you know, he does have that background. I think he was at, you know, deep mind. So, you know, it's, it's clear how you get from deep mind to humanoid  robots, but your path is, I guess, a little less clear, you know, going up and down your cv. Um. How, how, how  do you eventually get from electromagnetic materials and electromagnetic theory and numerical methods to  humanoid, well to neural interfaces, to humanoid, to Goldman Sachs, to humanoid robots?

 

[00:06:02] Rob Cochran: Yeah. It's, uh, you know, I've really followed, followed my interest, which has not  always been in a straight line, but the, the two through lines through my career have been. Uh, interest in  Frontier Technologies. So whether that was working on meta materials research, uh, when I was, you know, in,  in, in undergrad, frankly, uh, doing some work in, in self-driving cars early on as a researcher, and then really more moving to the business side. But, uh, Amazon Web Services worked in Internet of Things, technology  and enabling software for companies that themselves built connected devices. In that capacity, I met a guy  named Thomas Rearden who was starting Yeah. Control Labs and got fascinated with, with neurotechnology  and, and the sort of interactions that that might allow for. And so, and then even at Goldman Sachs, you know,  the chance to work on what we were calling the financial cloud, but actually externalizing much of the, much of  the infrastructure and, and moving to this kind of modern platform for AD API development. Felt like at each  stage, getting to touch on something that was transformational to transformational to how businesses and  industries could work. And then the other piece, so that, that's kind of the frontier technology side. And the  other piece is developer platforms. We built control kit at Control Labs, so it was actually a developer. Our, our  very first product before becoming a part of Meta was, was a developer platform that allowed you to integrate  neural control into, into any computer application. And that's was kind of my key responsibility when I first  onboarded there. Um, at Amazon Web Services, we built the internet of things platform for, for developers,  Goldman Sachs. We built the, the financial cloud, which was again, a, uh, externalized, API interface to much  of the, the financial institution. And so coming here it felt very natural to then have, having seen many of these  cases where. You could unlock the creativity of, of many people by opening and extending the platform felt like  a very natural fit. [00:08:00] And then tied into, you know, my interest in Frontier Technologies, humanoids is a  really exciting space.

 

[00:08:05] Brian Heater: I'm sure I'm not the first person. I'm, I'm definitely not gonna be the last person to bring  this up, but, you know, meta has, um, and. You know, somewhat publicly has been doing work in, in robotics.  You know, we've seen a lot of the research come up. Um, and, and you were there for a while through the  acquisition. At the time though, did that, did it feel like that wasn't the place for you to pursue an interest in  robotics? Well,

 

 [00:08:31] Rob Cochran: the thing that united Josh and I was shipping a product. And it felt like, uh, given how  the space is evolving, the best way to do that quickly and in a highly opinionated way was in a startup, right? It  gave us a chance to really put on paper, uh, together an idea we had for a product and in, you know, two  years, get it out the door. Big companies can put a lot of resources behind something once something's kind of  working, but I think the, the flexibility and, and kind of opinionation that a startup allows for was the right forum  for us.

 

 [00:09:05] Brian Heater: Yeah. What, what is, what does a highly opinionated mean in that context?

 

 [00:09:10] Rob Cochran: Well, we're doing something a little unconventional in that we have taken on the  humanoid form, but it is lightweight. It's 45 pounds. It is three and a half feet tall, so it's not a full size  humanoid. It's physically soft to the touch, so it's covered in foam like materials. It's focused on a lot of really  interactive features, so it has articulated eyebrows and an LED array around the face. So we've taken a real  eye for design and in how we express this idea and. What that means is it's limited in some capacities. It's not  appropriate to factory work. It's not gonna lift a 50 pound tote, but it does allow for something that you could put  in an office space and have work right by your desk and be able to deploy a novel motor control policy onto it  and not worry too much about putting a cage around it for safe operation. And so that felt like something that  would really unlock a lot of, uh. Really exciting work from us and others, but it was, and is quite different from  what is available today.

 

 [00:10:07] Brian Heater: It's unconventional, sort of like in the broader industrial sense. So it's unconventional,  like certainly like compared to like any like, you know, like a, like a big Fanuc arm or something that we, that we  are used to seeing in, in, in factories. But it's also, it's interesting to be at a point where we feel comfortable  suggesting that there is a convention for humanoids that this runs counter to.

 

 [00:10:30] Rob Cochran: It is the earliest innings of, uh, of an industry that has yet to, I would say, broadly find  product market fit. Right there. There are a number of use cases in, in industry and in commercial applications.  You know, I think many have ambitions for the home, but none of this is yet realized. Right? Yeah. And so what  we said is what is the market that actually exists today? And I think there are some proof points for this, but.  There are many developers who are excited about robotics and looking to extend them into human spaces and  are excited about that. And so that felt like something where we could actually get the experience of building  and shipping something at some scale. And, and that's the chance to then start iterating and building from  there.

 

 [00:11:12] Brian Heater: Yeah. I, wanna talk about the market is really interesting and I know that the part of  the answer to this is that, you know, that. That this is the first of what I expect will be many different robots. You  know, even like looking at, there's different iterations that I can see behind you right now, but the robot that  we're specifically here to talk about is. I, I would say like a research robot, right? Is a, is a developer platform  robot. A few analogies come to mind, both on market now and from the past. Pollen's got the robot, you know,  works with Hugging Face, like specifically for, for developers. In the past we've seen robots specifically for  research. Also other ones not for research that have become research robots, you know, very famously. Yeah,  by us. If you go to NYU or Columbia, you'll see, you know, probably one of the PR2s from Willow Garage. Still  over there you'll see some Fetch robots. You'll see some NAO robots. But there's a reason specifically why, why there aren't a lot of research robots, and it's because like there just hasn't been a market fit for them in the  past.

 

 [00:12:16] Rob Cochran: I think this time is different. The space has evolved meaningfully over the years, and  so, you know, if you take some of the work Josh did at DeepMind, you know, they, they were using the OP3  platform, which is very much a, a small research device, and there wasn't anything really hardened for  continuous operation in a research setting that you could then train to do useful things and actually. Test those  deployments in, in real spaces. They

 

 [00:12:42] Brian Heater: actually built their own in-house too, right? I mean, they had Everyday Robots, so they  actually just went ahead and made their own.

 

 [00:12:47] Rob Cochran: Yeah. And I, I think that was maybe a separate, separate division. Yeah. And you  know, there are different teams doing different work. Uh, but, you know, I think the, the point is, you know,  there, there are a bunch of different questions you ask from the outset, right? There are tabletop options. There  are mobile options. Uh, we've chosen a form that we think is quite flexible to deployment in human  environments, so it can operate across dynamic and unstructured settings. It has capable locomotion, it has  grippers out of the box so it can open a door, grab something out of the fridge. It's physically capable of doing  many useful things. And it's also, you know, small enough that in a lot of settings you can actually safely. Try  those things out and deploy them at scale. And so what we started with was the right form factor allows for  many, many ideas to be expressed on the platform. But you pair that with an SDK, a set of AI capabilities out of  the box that give developers of all types, whether that's a deep research group doing motor control or someone  interested in character design and expression, a range of tools that let them get started right out of the box. So  if you just want to change the voice or change the, the, the kind of character that sits behind that voice, you  can do that and still rely on mapping and navigation capable locomotion that lets the, the robot operate. But if  you want to dig into how, you know, your mapping and navigation system operates, you can, you can deploy  that on our robot as well. So we wanted to give that flexibility and I think that's a, a key difference to, to what  has existed previously.

 

 [00:14:18] Brian Heater: So I have all these, um, I have these robots behind me. You can kind of see them like  blurry, um, maybe a few of that, but one of them, um, in the front right there, he, he. He stopped working a long  time ago, but if it, uh, the Anki, Cosmo, do you remember Anki? Oh,

 

 [00:14:31] Rob Cochran: yeah, yeah, of course.

 

 [00:14:33] Brian Heater: I remember meeting with them when they launched and one of the really sort of, I, I  would say innovative, but prohibitively expensive ideas that they had at the time was, Hey, let's hire a bunch of  ex Pixar, Industrial Light & Magic animators to create these facial expressions. And they were able to. Create  this like really expressive, interesting robot, and that was probably the highlight of it. And then looking behind  you, looking at all these iterations and designs, you know, it strikes me that you probably, maybe you've taken  a similar route. I mean, you brought creatives in to help you make this like expressive system.

 

 [00:15:09] Rob Cochran: Yeah. Well we're, we're lucky to have a very talented industrial designer and a, and a  team of folks working on this, but I would say from the outset, we, we set out to build the kind of characters  from science fiction that we liked and grew up with. You know, WALL·E, Baymax, Rosie the Robot. These were  kind of inspirational designs that. Felt comfortable to be around, felt engaging, endearing, something that you  could connect with, uh, on an emotional level. And I think, yeah, Anki you mentioned, did a, did a wonderful job  of that. And so we took cues from, from much of robot history as, as well as, you know, science fiction and  other kind of storytelling capacities. But yeah, you can see on the wall behind me, we went through many  iterations, you know. Things that hewed closer to, to kind of anthropomorphic design. We ended up feeling like a much reduced, simplified, more abstract form allowed for kind of clear expression of ideas without  overloading with the kind of things that, you know, might cause uncanny valley sort of experiences.

 

 [00:16:07] Brian Heater: And the head is a router, right? Am I, am I off on that?

 

 [00:16:12] Rob Cochran: Uh, router wasn't the, wasn't the motivation. Okay. But, but we pulled from a lot of,  yeah. Other, other design ideas.

 

 [00:16:21] Brian Heater: The marketing material is interesting specifically for this product in that it is a  developer product, that it is a research product, but obviously you are showing that you're showing it with kids  as this sort of like family friendly system. I know that you have, again, you were at Meta, you were at Goldman  Sachs, you know, you've got Kleiner Perkins and Lux and these big investors involved. So I, I'm really just, I'm  curious like. From the outset, like what is the marketing strategy and, and like, who, who are you going after?  With this first system,

 

 [00:16:53] Rob Cochran: we have a, a really broad range of customers who are developers, so they are  businesses. And individuals with technical skills who have ideas that they want to express on a robotics  platform. So we have, you know, Disney and parks and entertainment looking at accelerating and doing a lot of  iteration on character based experiences in parks. Uh, we've got, uh, Boston Dynamics, uh, that is interested.  You know, obviously they have deep history in industrial robotics, but is also interested in more interactive  human spaces type of work. And so they're a customer. We've got a number of NYU, UCSD, we've got  academic research labs building and extending on the platform. We've got AI world model labs building on the  platform. So a really broad range of use cases. And again, I think, you know. Part of the reason for a humanoid  form is to allow for a highly generalized set of use cases, and I think the power of what we're doing is that  though it is a research prototype device, it allows for the kind of user testing and experience work that then  gives you confidence you can actually deploy it in the real world. And so, you know, my hope is very much that  we see, yes, videos from labs, but also a lot of experiences of the robot and people very broadly getting to  interact with it in the world.

 

[00:18:14] Brian Heater: Okay, so, so this, this was something, you know, this is, this is news to me and this is  very interesting and obviously, I mean, for different reasons, having Disney and Boston Dynamics as  customers are, are different levels of votes of confidence. You know, Boston Dynamics One, like clearly you've  designed a capable robot if Boston Dynamics is, is interested. Um, but like specifically, let's start with Boston  Dynamics. What does it mean for a robotics company to be a customer of yours.

 

 00:18:44] Rob Cochran: In addition to building robots, they have a whole commercial go-to market arm that  helps Spot get out into the world. At factories, they have Boston Dynamics Consulting, which works with  companies to provide robotics solutions. This isn't at the experimental phase, but. What they want to explore is,  is how do they get robots more into human spaces, not just factories and warehouses, but out into the real  world. And so, you know, they're exploring what that could mean both with, with their customers and, and  internally

 

 [00:19:13] Brian Heater: Disney, you know, I know we've been speaking with them a little bit lately. We've  spoken with Moritz and the work that they're doing over in Zurich as far as, um. They're doing like really  incredible work, you know, around Star Wars and around these roaming robots in the parks. You know, I  assume that that's something that they're probably maybe interested in the consumer space as well. So how  does a system like yours kind of fit into the work that they're doing?

 

[00:19:38] Rob Cochran: Yeah, so the Imagineering team at Disney has done so much amazing work in  robotics, and I think has a real understanding for what it means to take something from concept idea IP and a  movie or on a whiteboard and put it into real spaces with lots of people around, and I think they'll speak to it directly at some point. Sure. There's not that much I can share, but I, yeah, I think the idea is that this gives  them something where they can really freely experiment on something where they can rely on. On mapping  and navigation on an audio system that's reproducible on core locomotion that they can build and extend on.  And so not just the, the form factor, which, you know, I think from a design standpoint is, is appealing. But an  SDK that is well supported, well documented and, and provides a lot of features out of the box that they would  otherwise have to build in house means a lot more character design from their standpoint.

 

 [00:20:30] Brian Heater: So commercial applications and then, you know, NYU, so, so research applications,  um, like, I don't know. That seems, that seems pretty straightforward, right? I mean, that, that's kind of what we  were talking about before as far as like what has been done on the Willow Garage side of systems. Just doing  these sort of like reinforcement learning kinds of work on these robots.

 

[00:20:52] Rob Cochran: From, from an academic standpoint? Yeah, I think, I think it's a, a fairly broad range of  things. We, at UCSD, it's, it's frankly more, more neuroscience research and, uh, yeah. You know, we have  manipulation research, uh, multi-robot collaboration. So, uh, you know, it's, it because it is a quite a flexible  system, I think it will appeal to, to researchers of, of many kinds.

 

 [00:21:15] Brian Heater: I'm curious how your robot crosses over into the neurosciences.

 

[00:21:19] Rob Cochran: Well, I think it's a way to express research in motor neuroscience, right? Mm-hmm.  It's a, a simplified form of how you control bodies that you can simulate and, and reproduce, and so it's a  another means of. Jo, some of Josh's work actually was, you know, published in, in Nature, uh, on, you know,  motor nervous systems in insects. And so, you know, doing work in simulation and then validating some of that  work in, uh, in the real world, robotics provides an interesting platform for that kind of expression.

 

 [00:21:52] Brian Heater: That's really interesting. And, and, and that, I mean, that does seem to kind of connect  back to Control Labs. Is that fair?

 

 [00:21:58] Rob Cochran: Yeah, I think neuroscience, obviously I'm not a neuroscientist, but it's been core to  Josh's, you know, work and career and to, to my passion. So it's fun to get to kind of connect these ideas back.

 

 [00:22:11] Brian Heater: On a very abstract level, like how does that play in here as far as the robot? I mean,  are we talking around like, are we talking teaching the robot to interact with the world? Are we talking about  sort of like teleoperation or just how the robot, I mean like with insects, I'm thinking about how the robot just  functions on kind of a very basic level.

 

 [00:22:29] Rob Cochran: Yeah. So from uh, the beginning we've taken I would say a fairly modular approach to  our kind of AI stack, the way the robot moves in the world. You know, motor control policies are trained at a, a  low level reinforcement learning largely in simulation. Some, some training from video and other sources. Um,  but then we've built a whole mapping and navigation stack, which also leverages models that we've trained but  is not. Trying to connect all of these concepts into a single, uh, single model that kind of rules them all. And in  part that's a, a philosophical choice about the moment in time and how to get something deployable and  working. It's also a choice about the type of developer experience we can offer where. Having modular  systems then allows our customers to separate and work on the pieces that are most interesting and useful to  them while relying on components that are, are kind of hardened and, and usable off the shelf.

 

 [00:23:27] Brian Heater: Yeah. Okay. So deployable and working gets back to something that you were talking  about before because again, two years. In terms of what we're talking about now is not, it's not a lot of time,  right. It's, it is certainly not a lot of time to like build up to scale company to get to where you are, to actually like  announce this thing and, and I, and I guess I, I guess launch it as well. Um, and, and it sounds like part of, uh, both of your and Josh's desire to really, to move as quickly as possible and, and maybe some of that was a.  Backlash from getting acquired by like a massive company where it's a lot more difficult to be nimble.

 

 [00:24:08] Rob Cochran: I don't know if it's so much that, but I think the, the freedom we had definitely let us,  let us move quickly, uh, align. You know, we've got a really great team here that is spread across, you know,  hardware development. AI model, you know, research, software development, human robot interaction. So we  have a bunch of different disciplines and getting that group to kind of all row in the same direction and align  behind a, a mission and, and do so at, at, you know, very rapid pace, I think is something that's definitely been  special about the, you know, world building is part of what makes it fun to be here and operate how we are.

 

 [00:24:46] Brian Heater: Timelines are really weird around humanoids. Um, this is something that I've been  trying to kind of wrap my brain around because at the same time, companies are pushing to be as like, quick  as possible, but also they're incredibly long and abstract from, from the quick side of things. You know, I won't  name any names, but there there's a handful of companies that come to mind that have been like, how quickly  can we get. Develop a robot that can walk, right? Like, and then that's gonna be, our press release is going to  be, it took us six months to build a robot that can walk. Right. Okay. You know, like, that's impressive, but like,  when are you shipping? When's it gonna be out there? And, and that's like much more abstract down the road.  And nobody really has an answer to that question. Right. I mean, we just, again, I, I just, um, interviewed the  head of the Atlas team and he was like very upfront about. Their, their ML team. And when it comes to like  dexterous manipulation, there's a very broad range in their own room about how long it's gonna take to really  kind of, kind of crack some of these things. So this, this robot that you're launching right now, this, this  developer robot, this, this research robot, um, this is a way for you to, to go to market.

 

 [00:26:07] Rob Cochran: What I'm hearing from you is two things. One is. There are companies that have  specific research ideas that they are eager to express because they are a core innovation in the field of  robotics and getting something out there that shows they have made a meaningful contribution in the space.  Sometimes requires building a whole robot to do that. That's an unfortunate state of the world. And part of what  we've tried to achieve is, you know, we know that robotics is gonna require thousands of, of new innovations to  deliver, you know, humanoid robots into the home and everywhere people are. How can we create a platform  that allows a research team with a really brilliant insight into training world models or VLAs or whatever it is.  Express it on a platform that already exists that's stable as a is, is safe to experiment on a piece of what we are  trying to accomplish with this. Uh, the second element is, you know, timing of kind of announcement relative to,  to progress. And, you know, the project for us is very far from done. I would not say, sure, this is extremely  early, early days, but it was our, our goal to get robots out to customers. You know, before making the  announcement to be actively shipping robots before, you know, announcing and, and really sharing what we're  up to with the world. So wanted to be in a place where we could actually start to scale and expand out from a  commercial perspective before, you know, kind of, uh, making too much noise about it.

 

 [00:27:38] Brian Heater: Yeah, I, I guess the, the, the distinction I would make like with, with, with timelines, like  what, what does, what does done mean? Right? I mean, hopefully it's gonna be a long time before, before,  before you're done. But there's done, and then there's like shipping product, like who knows? Like when any  single company is going to have a reliable. Industrial robot that that's, that's operating out there. Right? So  quick, quick to build these systems. You know, we see them in promotional videos, but actually like having  them in the real world, like that's gonna take a lot, a lot of time to scale them, right? To make them at, at a point  where you can like reliably manufacture them and have them out there. And this is obviously the first stop in  what are going to be, as we said before, many robots. And this is a way to, um. Establish yourself as a company and to to ship, to ship a robot, to get a robot in people's hands.

 

 [00:28:34] Rob Cochran: Yeah, and I think that that experience is, is what we're looking for. I think it's what  matters to, to robotics, maturing. You know, part of that is supply chains, manufacturing, reliability, service models. All of those things, you know, are nascent or non-existent in, in this industry. Part of the kind of early,  early phases of iteration are certainly on how do we build out that muscle, um, alongside better understanding,  uh, of use cases, you know, and development of more advanced capabilities that, you know, earn our, our spot  in new markets.

 

 [00:29:11] Brian Heater: Yeah, I mean, I should say, like I'm saying that, I'm saying this is a positive thing, that  you're getting robots in people's hands. And I, I'm saying this because like. Uh, you know, again, I'm not gonna  name any names and, and this is partially because I don't know who is the right answer to this question, but  there are almost certainly companies, uh, humanoid companies that we're seeing right now that have  announced products that we have seen that won't come out because, you know, it is such a hard problem to  solve, right? And you've identified a space in the market. You've gotten a robot in people's hands years ago.  When Melanie Weiss was still at Fetch, she was talking about their research robot and you know, she was  telling me at the time, like essentially like this thing's a loss leader, right? I mean, we are not making money off  of this, but it makes. Well you do this one 'cause Willow Garage doesn't exist anymore, so somebody has to do  this damn thing. And then two, we're doing this because like, you know, you get people working on our  platform, they get used to our platform and they're gonna be really good at working on Fetch robots in the  future when like they actually enter the job market. So you get your foot in the door with this system. You build  more robots down the line and people are used to programming, you know, with your, your AI on your platform  and you're really well positioned for the future.

 

 [00:30:29] Rob Cochran: Yeah, I mean, I think about how, uh, so many of us grew up on Apple computers in  schools. Yeah. And the familiarity that, you know, our generation now has with that as a, you know, workplace  platform and personal computing platform. And I, I think the same will eventually be true in robotics. And so our  hope is to make this at first, you know, university and PhD level research, then undergrad, then high school,  and eventually, you know, this will be a key part of how anyone learns, you know, computer science and, and.  You know, how to work in the technology world.

 

 [00:31:01] Brian Heater: One of the things that's been really incredible about covering the space, you know, I  mean, one, like, it's still a relatively small world, you know, and, and you still, like, like you said, you know, you  and I talked, uh, not too long ago and I ended up just, and I love it, but I ended up interviewing like some of the  same people, like over and over again and, you know, in different companies and they, and they make those  moves. And I used, you know, I always said this about, uh, tech journalism, like. Be nice to people because  you're gonna end up working with them again. You know, your, your competitors are gonna end up working  with them. And, and the, the, the same is definitely true in robotics and I found that like on a whole, everybody  seems pretty supportive. You know, there is some competition, but I think everybody sees there's enough  space in the market. Um, and there are enough places where companies can kind of like work together to build  things up and. You know, I'm looking at a product like this and I'm thinking about, again, Hugging Face's robot  as like a open source repository, like GitHub, you know, there's some NVIDIA hardware on here. So like Jetson  as a developer platform. Yeah. Um, their simulation. I know that you're working with ROS here as well,  obviously, like you need to know. Right. You need to know ROS to work in robotics. I guess this is a way in  which it's not like Apple, like you can't right now operate in a fully closed ecosystem in robotics.

 

 [00:32:27] Rob Cochran: That's right. Yeah. So I, I think there are a few things there. One is from a ecosystem  perspective, you know, a rising tide lifts all ships and we are in a phase of the industry where innovations  anywhere help everyone, right? Seeing what's possible. Uh, you know, that sort of collaboration across  industry I think is really important because we all want to see a world where robots can help us in our, in our  factories and warehouses, but also in our daily lives and, and nobody's solved that yet, so.

 

 [00:33:00] Brian Heater: Mm-hmm.

 

[00:33:00] Rob Cochran: You know, collectively starting to solve that problem and we're making our  contribution to that, I think is a. Uh, a key piece of it.

 

 [00:33:08] Brian Heater: What does it mean to make an accessible robot? Um, you know, there's like, there's  accessible for UCSD and NYU and they, but they, I assume have deeper pockets than, you know, somebody  who's like a recent graduate from those schools who wants to sort of just go out and start their own startup.

 

 [00:33:24] Rob Cochran: Right now the robot costs about $50,000. So, you know, not, not inexpensive for, for  an individual certainly, but for an institution that's, that's focused on this sort of work and relative to. To other  platforms that are available. It's a, you know, a very, very achievable price point. But a another piece of that is,  you know, we layer a lot onto it. So it's not just hardware you're getting out of the box. It's a, a developer  experience that includes, you know, uh, infrastructure that supports simulation work, that supports, you know,  training models that, you know, mapping and navigation stack voice interaction. All these pieces get, get  layered into a kind of a, a single system. And we keep updating that software and you always get the. The  software updates from there. What I would say is we're operating at, at low volumes today and, and our  ambition is to ship many more. And I think, you know, driving cost curves down to where this is well under  $10,000 is, is very achievable in the fairly near term. And so that's very much our goal and, and I think we will  continue to drive the product in a direction that is lower cost and more available, uh, and more accessible, uh,  to a broad audience of developers.

 

 [00:34:35] Brian Heater: So, so, you know, if I, if I'm looking to launch a robotics startup, may or may not be,  you know, um, and, and, and I want to like, you know, uh, say like, deploy an LLM on a system for like voice  commands or, you know, I want to start building out like VLAs or, or VLMs. Like, is is that, that's who you're  going after with a system like this.

 

 [00:34:59] Rob Cochran: There are a range of customers that is, uh, among the set of use cases that would be  a, a great fit for this platform. You know, we have AI research labs that have a commercial product in world  models or, or other, other things, and are looking to start experimenting in robotics. And this is a very, again,  accessible platform, not just from a cost perspective, but from an ease of use perspective. It's something that  you could have in an office. It sits on a little chair, it's, you know, can sit right by your desk. And it's fairly  approachable to, to get started with, even if you are new to robotics. But then we have people at, uh, more at  the application layer who are thinking about, you know, uh, a retail experience or like you said, changing,  changing the voice models and doing more character based experiences. And so some are doing that sort of  deep VLA type research and others are, are more experimenting with the experiential aspect of robotics.

 

 [00:35:54] Brian Heater: So when you talk about retail, are we talking about like Aldebaran was attempting to  go after with Pepper, like that kind of thing? As far as like greeting people?

 

 [00:36:03] Rob Cochran: Yeah, I think the, the kind of, uh, I guess industry term would be greeter and guide:  someone who welcomes you in, shows you around, has awareness of space so it can actually move around  with you, can connect to a CRM system so it kind of understands the customers and their needs, can  understand the space and what's available and, and connect the dots on a lot of those things.

 

 [00:36:24] Brian Heater: I start to develop these models, I start to develop AI, start to collect data, and then I,  then I build my hardware based on those models. Is that kind of the rough idea?

 

 [00:36:36] Rob Cochran: Yeah, so I, I think it actually would be, you know, even more straightforward than that.  Out of the box, you get, you get the hardware, you get, uh, you know, voice interaction, uh, reasoning,  capabilities, mapping and navigation. So if you were deploying it in a, in a store, you could, you know, build a  map of the space by driving the robot around, uh, you know, give it various locations that it can show  customers. Uh. Objects, build that into the context window of the, the AI model you're working with, and very quickly be off to the races in terms of, you know, building out a, a proof of concept experience at least. And  then

 

 [00:37:14] Brian Heater: okay.

 

 [00:37:14] Rob Cochran: You know, kind of maturing it from there.

 

 [00:37:16] Brian Heater: Yeah, I guess that's kind of what I'm getting at is like, you know, are you eventually  white labeling this product?

 

 [00:37:22] Rob Cochran: I think we're certainly open to it and we've spoken to a number of kind of, uh, what  you would call channel partners who are interested in applications in various industries and would look at  commercializing them directly and running the operations of those deployments mm-hmm. In various spaces,  like retail could be an example of that or the kind of greeter and guide use case. You know, I think eventually  we are particularly interested in the home and doing that, uh, directly ourselves, but the, you know, in, in many  industries, and again, starting with a general purpose platform should allow for, for broader deployments. And I  think we'd be excited to work with partners, uh, across a range of those industries.

 

 [00:37:59] Brian Heater: I'm, I'm glad you use home and eventually in close, close proximity to one another.  Like, that's a very honest way of, honest way of putting it. Um, because you know, it is, we, we talk about like  scaling, like that's what's gonna bring the price down and, and you know, and this is why people are going after  industrial, before the home, right? Like, these are relatively easier problems to solve. When, when you talk  about going after the home, are you talking about. Like a, like a companion robot? Are you talking about, you  know, like a, a useful robot doing, you know, sweeping up around the house?

 

 [00:38:43] Rob Cochran: Yeah, so I mean, I guess a few points there. I, I would challenge that industry is an  easy, easier set of use cases. I think the level of precision operational resilience required in those settings is, is  extremely challenging. Um. You know, you don't have to deal with some of the unstructured environment and,  and human interactive elements, but you do have to deal with high precision, high repeatability, and a bunch of  other challenges. So I wouldn't minimize that. But I would say that the opportunity is 10% of labor is in  manufacturing. 80% exists in service sector jobs. Jobs that interact with people in human environments. And  so, you know, there's a tremendous opportunity, you know, with aging workforces and, and kind of specific  sector, sector level shortages in, in ability to hire. So I think there's a, there is a massive opportunity, um, in the  home. I, I do think that there's a great opportunity for a companion and light utility robot. Something that can  grab you a drink from the fridge, tidy up, help out around the house, uh, and, and be a, you know, cheerful and  enthusiastic companion. Mm-hmm. Uh. I think it'll be some time before we get to the thing we all want, which  is, you know, yeah, cleans, does the dishes, cleans out the gutter. These are hard, hard challenges, but part of  what we're trying to build is a platform that allows, you know, others to start solving that and, and we'll solve  some of that ourselves as well.

 

 [00:40:10] Brian Heater: Yeah, I, I, I like it. You're kinda, um, it's kind of like a Tom Sawyer, like getting the  neighborhood kids to, to paint the fence, sort of like, almost like crowdsourcing to like, let, let's solve some of  these bigger, these bigger problems together. You've created this developer platform, you're getting your  system out there. You're having people like collect data and solve problems, like specifically for this robot. And  in a way, you know, as they're working to. As they're working on their own things, they're also kind of like  working to maybe help you out down the road.

 

 [00:40:41] Rob Cochran: It's very hard to say what this system looks like long term, but obviously the idea of a  skills or an app store where you can actually engage a developer community in the way a product gets  expressed out to the world would be, would be really exciting, and I think is definitely possible. You know, it shifts the, the kind of responsibility of what works and what doesn't work, uh, around in a, in a way that needs  to be figured out. But. You know, I think that would be, uh, very much an exciting idea for, for us to pursue.

 

 [00:41:09] Brian Heater: What is your relationship with. Data. Right? Because that, that's like the big, that's the  big question mark in a lot of this, right? That's, that's one of the biggest problems with dexterous manipulation,  right? Right now. Right. And that's the big problem with these robots actually operating the real world is there  just isn't that much data out there. If you get, you know, the more robots you have out in the world, the more  data you're collecting. Are you actively, or will you be actively collecting that data from these robots?

 

 [00:41:36] Rob Cochran: We don't proactively collect data from our customers' robots. Uh, you know, our  customers' data is their own. Uh, and, and that's a key piece of the value prop. There's no kind of faustian  bargain for returning data back to us. Mm-hmm. We have a number of partners we work with also for, for data  collection that's kind of a, uh, dedicated line. And we have a lot of data on our platform that will be useful to us.  Um. Data comes in a lot of forms, and I think a key piece of data we'll get over the next year from everybody is  the, is the kind of experience and product data of how this robot gets used, what, what works, what doesn't, et  cetera, what use cases catch on, which don’t. And I think that can then guide more focused application of large  scale data collection in a way that could be quite useful.

 

 [00:42:14] Brian Heater: how's, how's your team doing on the dexterous manipulation front?

 

 [00:42:25] Rob Cochran: I would say very early days. It, it hasn't been our core focus. Um, again, we, in getting  to a platform that was, was broadly useful, you know, making sure it could effectively get around core  locomotion capabilities. It can sit in a chair, it can get back up, it can get down to the ground, get back up. It  has full body teleoperation out of the box. So it comes with a, an app for a Quest headset, and you can go in  and pilot the robot directly. It has a full mapping and navigation stack. It supports voice interaction, uh, long  horizon reasoning and tool calling. So we've built all of this in as kind of a developer framework and. We'll do  some of the manipulation work in-house. Uh, but we also are excited about our customers who are, who are  building that sort of work. I think we'll continue to focus on things like, you know, opening and closing doors,  interacting with elevators, you know, basic object manipulation, which we can provide again at a platform level  such that our customers can then integrate it into broader workflows.

 

 [00:43:24] Brian Heater: This is an interesting thing that I've seen happen as, um, as robots are getting more  mainstream coverage. Um, you know, I, I saw it happening around the 1X robot a lot and I think that there are,  um, maybe misconceptions or misunderstandings around, around teleoperation, you know, and this idea that  maybe teleoperation is like necessarily like cheating with a robot if it isn't like fully operating, uh, autonomously  all the time, obviously. Teleoperation is, is gonna be a big part of training data. It's gonna be a big part of  collecting data as we're seeing your robot in its current state interact. Um, you know, it, it, it, to what degree is  it, is it operating autonomously at all at this point? You know, what degree is it teleoperated?

 

 [00:44:10] Rob Cochran: It comes out of the box with full teleoperation capability. So that is one thing it  supports. It also supports, uh, autonomous mapping and navigation. So you can drive the robot around, build a  map of its environment. Uh, it can build a semantic layer of, of that environment. So recognize refrigerators and  couches and tables and chairs, and, and you can reference those things and it can navigate among its  environment to do that. It can recognize people and remember things about them, address them by name. And  so we've built in that sort of autonomy out of the box and we'll continue to layer on things again, like, you know,  grab me that drink from a. Or something like that. And, and that's exciting to us.

 

 [00:44:49] Brian Heater: We were talking a little bit about value prop before, you know, and, and 50,000, like,  it's, it's not, I mean, it's, it's not a lot for humanoid, you know, it's, it's a lot, you know, for, maybe for the average  person, but, you know, I, I, but I, I guess it's all contextual and. If your frame of reference is, for example, like a, a low end Unitree robot, it's maybe a more difficult sell. But if those right now are people's primary frame of  reference as far as like a, a humanoid robot that I can currently buy, what is the selling point for this system?

 

 [00:45:22] Rob Cochran: One, I'd say it's important to make an apples to apples comparison 'cause there's a  different price listed on the website than what you actually pay when it arrives at your door. And I think from a  price standpoint, we actually are quite, quite competitive.

 

 [00:45:33] Brian Heater: You're

 

 [00:45:33] Rob Cochran: talking like

 

 [00:45:34] Brian Heater: shipping tariffs, things like that, or you're talking about upgrades?

 

 [00:45:37] Rob Cochran: Yeah. And getting grippers, for example, included in the robot then it's substantially  more expensive. So our robot comes with grippers included. But what I would say is that, you know, we offer a dramatically different developer experience. So one is a robot that is aesthetic, that is much safer to be around,  that is soft to the touch, engaging, has articulated eyebrows, articulated head, LED array around the face. So  it's something where you can prototype a categorically different set of interactions and deploy in different  spaces than what's possible on other systems. And then building in a set of capabilities that are available to  developers out of the box. So, uh, at risk of repeating myself, mapping and navigation. Yeah. VR teleoperation  app, uh, core, you know, locomotion, primitives, voice interaction. These are things that are well documented.  Dockerized, deployable on, on the robot, which comes with a, you know, a Jetson Orin, uh. NVIDIA computer  on board. Um, we leave plenty of headroom for the developer themselves to build and extend applications, but  it comes with a lot of core software that works out of the box. And so that's not the kind of, you know, uh,  developer experience you get, uh, with, with other platforms. And also the, the service model is quite different  because, you know, we sell directly in the US and around the world. And if you have a problem, you speak to  us directly and that's, you know, uh. I think categorically different.

 

 [00:47:03] Brian Heater: Is that scalable?

 

 [00:47:05] Rob Cochran: Yeah. I think the reasons that others work through resellers is, is not necessarily  scalability, but uh, you know, geographical restrictions.

 

 [00:47:13] Brian Heater: Yeah, that's fair.

 

 [00:47:14] Rob Cochran: So, so, so for us, I think long term, having direct relationships with our customers is  very core to what we do. And, and over the next year as we scale, making sure our earliest customers are  extremely successful is. Is the only thing I care about. Um, and so we'll be very focused on making that a great,  great experience as we scale.

 

 [00:47:33] Brian Heater: And you're already talking about international sales.

 

 [00:47:35] Rob Cochran: We have made international sales.

 

 [00:47:37] Brian Heater: So you're going quickly, it sounds like already at this point

 

 [00:47:40] Rob Cochran: we're moving as quickly as we can. It's early, it's early days, but, but we are, are  moving as quick as we can.

 

 [00:47:45] Brian Heater: Great. Well thank you so much for taking the time.

 

[00:47:49] Rob Cochran: Absolutely. Yeah. Very excited to, to share a little bit about Sprout and Fauna  Robotics with you and, and for developers to get our, their hands on this.

 

 [00:47:58] Brian Heater: Thank you so much to Rob and Fauna Robotics. The company is doing some very  cool things. Highly recommend you check them out. Uh, thanks to you so much for tuning in. If you're enjoying  the show, don't forget to like and subscribe and check out our sister newsletter over at Automated.fm. We have  a bunch of very cool interviews lined up. Uh, we've got, uh, Gary Cohen, the CEO of iRobot. We've got, uh, Ali  Kani from Serve Robotics. We've got Boston Dynamics. Physical Intelligence is coming up, uh, ESI plus one  Mad. Uh, so stick around. And with all of that, uh, we will see you just about this time next week for another  episode of Automated.