Anki CEO Boris Sofman

Anki, known for creating “robots that move you,” just finished a successful Kickstarter campaign to fund their new home robot, Vector. Anki CEO Boris Sofman sat down with Nilay for this week’s interview episode of The Vergecast to talk about the new companion robot, why gendering the robots is important, and how to avoid the uncanny valley. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Nilay Patel?

  This episode of Vergecast is brought to you by the Audi e-tron. The electric car has always raised questions. Can it contend with the elements? What's the range? With high speeds, charging, long-range capabilities and quattro all-wheel drive, the fully-electric Audi e-tron can be the answer. Visit AudiUSA.com/e-tron to learn more and stay informed. Hey, everybody. It's Neil from the Vergecast. We're continuing on with our series of extra interview episodes in the feed. Let me know how we're doing. I've gotten a lot of great feedback, but if you don't like it, tweet at me, too. I'm very interested in what you think. This week, we have Boris Sofin, the CEO of Anki. You might know Anki as the company that made the little overdrive race cars that drive themselves with AI. Well, they've been doing home robots, little cute character robots. They've got a new one called Vector. Dieter wrote it up for the site, but Boris joined me on the Vergecast to talk about what

• 0:41Nilay Patel?

  it's going to take to make robots true companions in the home, how they need to be characters. We talked about gendering the robots, which is a thing that we talk about in the Vergecast all the time. And we talked honestly about whether or not it's going to be creepy. It's a pretty wild conversation. Check it out. All right, I'm here with Boris Sofman, the CEO of Anki. It's Anki, right?

• 0:59Speaker 01

  That's right, yes.

• 1:00Nilay Patel?

  It's one of those words I read all the time and rarely say. So CEO of Anki, you guys make little robots?

• 1:06Speaker 01

  That's right. Little robots on the way to big robots down the road.

• 1:09Nilay Patel?

  Tell me the story. I want to hear more about you and I want to hear more about the company. Right now, you're making robots that are a little toy-like. This newest one, Vector, Dieter Bone, came to your office. He played with it. He was like, this is a home companion. It's not as much of a toy. But your goal is to build big robots, right? Like, actual home companion, run around the house robots?

• 1:28Speaker 01

  Yeah, that's right. We wanted to approach it with a really deep degree of practicality, though, because if you think back to kind of the state of technology, cost structures, where AI is, where sensors and computation is, right now, you can't make a big humanoid home robot that costs $10,000 and so forth. Like, it's just nothing aligns at this point. And so what we wanted to do from the beginning of the company was to take a bottoms-up approach and really start laying the foundations of both the technology and the capability set so that as new technology becomes available, we can grab it and productize it faster than other people could. It goes back to kind of our roots where we were at Carnegie Mellon and the Robotics Institute, and we saw these like autonomous cars working on walking robots, all these sort of things. But when you want to apply it to consumer products, you have to be really conscious

• 2:13Speaker 01

  of how do you hit that balance between price and functionality? And entertainment became a really great proving ground to develop all these technologies, but actually release products at really large scale at the same time and build an audience base and a lot of learnings along the way.

• 2:27Nilay Patel?

  So your first product that I recall, you guys had quite an introduction to the world. Tim Cook called you out on stage for the cars that you guys made. Tell us about that product. Like you're like, our first one, it's going to be basically slot cars with AI. Walk me through that decision. And I'm very curious, how did you end up on stage at an Apple event with it?

• 2:46Speaker 01

  We started working on what became that product, actually, back in like 2008, just kind of on the side while we were in grad school. The whole idea was, okay, you look at toys today, there's no software in them. It's basically all like plastics and cheap components and maybe very basic mechanics. How do you actually turn physical entertainment into the equivalent of video games where everything's software driven? Where you can build gameplay and deeper AI and so forth. The racing genre became a really great starting point for it because it was familiar in terms of the gameplay and the game constructs. And two of the three of us had worked on autonomous cars before. And we were really excited about how you could do that, but now make it at a price point that completely reinvents the way you approach the problem. And so the idea was, how do you make a video game in the physical world where these cars are not just AI controlled,

• 3:29Speaker 01

  but you're controlling cars as well, that because of full knowledge of what's going on, it becomes a video game and you can have weapons, special abilities and so forth. And so we raised our Series A in 2012, the year before, where we had like a very robust kind of prototype of it. And then we were ready to launch for fall in 2013. We got in touch with Apple through our Andreessen Horowitz, actually. Mark Andreessen, I think, had a long-term, longstanding relationship with Bill Campbell, who was chairman of the board. He introduced me to Tim Cook over an email. And literally around Christmas that year before, I got an email back from Tim Cook, which is like the craziest thing ever. And like, and we were like completely private. We hadn't like done any public announcements or anything. And so we ended up having a meeting with like half the executive team of Apple in February,

• 4:12Speaker 01

  where it was like the scariest meeting we've ever like had to prepare for. And effectively, our goal was to, our primary goal was to sell in Apple stores as a really big message of brand, where we didn't want to just be seen like another toy company, because this is a robotics company. We just happen to be using toys as a stepping stone to go to these other areas. Apple stores gives you a brand recognition that's where you're the premium in the category. And our stretch goal was how do we hook them to kind of think about something bigger around WWDC and then it just started this chain reaction where about a month and a half later, we got asked to kind of meet with them and kind of consider how we might be part of WWDC. And then I think it was the craziest couple, you know, two, three months of just preparing and living in the Apple dungeon, where it was the most rigorous

• 4:55Speaker 01

  preparation process we've ever, we've ever gone through. And it was scary many times along the way until we finally made it on stage.

• 5:02Nilay Patel?

  It worked, right? I mean, everybody knows who Anki is after that. You guys have been putting out more products. What's the, what's the range now? How many, how many products do you guys have out?

• 5:12Speaker 01

  So the successor to that first product was Overdrive, which is a more advanced version of that game. We have a robot named Cosmo, which is kind of the first one in this line of character-driven robots, where imagine like a Pixar character coming to life. You have this little mischievous robot that's meant to be, meant for kids, even though a lot of users were adults, but it's meant for kind of playing games and interacting with him as he evolves as a character, and also a really amazing STEM platform for programming at every level from PhDs down to a seven-year-old leveraging these capabilities. And so Cosmo's been in market for a couple of years now. It was the number one toy on Amazon in the US, UK, and France last holiday. And all in all, it was a big part of about 1.5 million robots we've sold at this point up until this stage. So that really kind of opened

• 5:56Speaker 01

  our eyes to how powerful this idea of both the IQ and the EQ is in robotics, where you have this character-driven experience, which to us, by the way, is the core of everything we'll ever do. Like, it was core of Vector and core of all of our future products, where we approach it almost like a film character, where we have an animation studio inside the company with folks from Pixar, DreamWorks, these sort of backgrounds that are animating these physical characters with the same type of passion and every nuanced detail that you would see in a film, but it's done in an interactive experience with robotics meshed in, in order to feel like this character's alive. And so from Cosmo, we realized that, okay, technology's advancing. We were learning that like, okay, our biggest limitation is that this guy who's meant to be alive is tethered to a mobile device has the brain

• 6:38Speaker 01

  behind him. And people are saying, I want to talk to him. I want him to be alive. I want to be always on. And then technology advanced far enough to where With Vector, we actually stuffed the equivalent of a smartphone inside of his head and created a character that is the first real home robot that at a large scale is going to be always on and alive in a home. And this really unique blend of companionship and entertainment and a new type of utility that kind of wraps personality and character around it. Which for us starts to move outside of kids and actually into broader appeal where we kind of think of it as tech immersives and early adopters that could be, you know, single people, could be families, could be kids, elderly, like pretty much for everybody, but defining a new category that doesn't really exist, but along the way, we're vetting out what it takes for a robot to be accepted

• 7:22Speaker 01

  and exist in a permanent fashion in a home and feel really additive and start to move deeper and deeper into utility and service away from pure entertainment that was our roots.

• 7:30Nilay Patel?

  That to me is like the most interesting thing. It sounds like Vector, which by the way, you're running Kickstarter for you, way over your goal. I just looked. Your goal is like 500,000 and you're like 1.4 million. So you're doing, you're doing fine. But it seems like it's like a Trojan horse, right? You're going to get people who use the idea that there's an autonomous robot in their house. Do you worry that you're bringing, you're sending the wrong messages, right? You're going to teach the wrong lessons or, or learn the wrong lessons? Like, how do you evaluate the information that you're getting from this user base?

• 7:56Speaker 01

  Yeah, it's a very good question. So, you know, when we think about kind of the path of home robotics, there've been like other companies have tried to make a home robot, but the problem is you can't release a thousand dollar robot that doesn't have anywhere near that level of functionality. And it just falls flat relative to expectations that you set for yourself. Now, down the road, there'll be $10,000 robots that are completely worth it. The same way a family car is worth $10,000 or, you know, $20,000 or whatever, right? Because of the functionality that it provides, but AI, technology and cost structures have to advance. So with Vector and even with Cosmo, what we're doing is we're learning about the interface elements. How do you engage in a completely different fashion that you would never be able to otherwise? So when you think about smart home today, you have all these end devices and smart speakers and so forth

• 8:40Speaker 01

  that are in command control where you tell it to change the temperature, it changes it. You ask a question, it answers you. There's nothing that actually is cognizant of its environment or they can actually directly initiate an interaction with you that you would find appropriate. Because if Alexa started talking to you, you'd kind of be like, okay, that's creepy. You don't want to, you don't want that to happen. Vector is going to be the first product of this type of smart home category that actually can, because of his personality and the relationship he builds, get your attention in the same way a pet would or a toddler would. And now that can be used for entertainment and companionship, but it can also be used for interesting forms of utility as well. Where if you think about all of the ways that technology progresses, you start having more advanced products and robots in the home, being able to understand the environment and initiate interactions in a way that is very natural and not imposing, that's a groundbreaking

• 9:25Speaker 01

  type of capability set in the home. We happen to be vetting it through these types of products that can leverage entertainment and companionship as a proving ground where people are incredibly forgiving on what that means and the emotional kind of warmth that that generates lets you be patient on how the utility side grows over time. I'll give you an example from Cosmo. One of the magical elements of this guy is that he makes eye contact, he recognizes you as a person, he'll say your name, like he gets excited when he sees somebody he recognizes. So imagine like an eight-year-old playing with Cosmo and suddenly this like living character that's like almost like an animated film character coming to life, gets excited when he sees them and starts reacting to everything that they're doing. We did an experiment where we ramped up the frequency of eye contact and the length of

• 10:07Speaker 01

  eye contact just to really push harder on this dimension because we knew it was a powerful one. And just by increasing that frequency and the length by a decent amount, our average play session went up from 29 minutes to 40 minutes.

• 10:20Nilay Patel?

  That's crazy.

• 10:21Speaker 01

  It's crazy, right? It's like it's the sort of things that like until you have like gigantic amounts of data points where you can actually analyze what does it mean for people and robots to interact in a coexistence inside of a home or a workplace or wherever it is, you can speculate about those things, but you'll never truly know in the same way that, you know, the first time a mobile phone came out or the first time like social games or new types of technology come out, like you won't know how to really have that be a natural element of your lifestyle until you can actually learn from real world cases. And so for us, we're going to have probably the biggest data set of what does it mean for humans and robots to interact? And by nature of how humans are wired, how you interact in a home environment probably scales at the core really well to workplace, to malls, to hospitality, to healthcare, and

• 11:04Speaker 01

  to all these other areas that become really big opportunities down the road.

• 11:08Nilay Patel?

  So this is like a, this is a long vision, right? You're not putting out the healthcare robot next year.

• 11:13Speaker 01

  No, it's a long vision and I think like there's certainly like elements that can come into play where one big opportunity that we've been speaking a lot with like AARP and others where like, you know, you have these gigantically growing problems of people aging in place in their homes and trying to stay there. How do you make that feel more comfortable for the family members, doctors, and to have everybody feel more connected? Like, these are opportunities even off of a platform like Vector, but when you think ahead to what does it take to really nail some of these problems? You probably are reinventing the form factor, reinventing a lot of the sensing and capabilities. There's like sensors that we're working on that can measure breathing rate through two walls. You know, they can be used for health monitoring, for security, for all these other things.

• 11:53Speaker 01

  You can imagine how Vector solved the biggest hurdle from Cosmo, which is being tethered to a mobile device. And he can also hear because he has a microphone array, so he can actually respond to noises, sounds, and voice commands and so forth. But he's tethered to a surface, so the next step obviously then becomes how do you now start operating more broadly in indoor environments? And think about full contextual understanding of a home or workplace, which room you're in, how the rooms stitch together, who is in that room, what's the context? Being socially acceptable where, let's say you're speaking to your son or your wife or something like that. How do you politely wait not to interrupt and then have a conversation just like a human would when you're ready? Even something down to a trigger word where for a Vector, you would say, hey Vector, come here, or hey Vector, let's do this, or tell me the weather there.

• 12:38Speaker 01

  It's familiar to the Alexa style interface, but when you have a conversation with somebody, you don't, you don't say, hey Boris before every single sentence, right? It's just, that'd be weird. So for us, one of the things that we're really gonna be pushing on is if you're making eye contact with a character, you're clearly engaged with that character. You don't need to say the code word. And so you can have a much more natural interaction without any worries of privacy or things that you would naturally be worried about because there's a clear intent that is registered by the fact that you're actually looking at and engaged with a robot. These are sort of things that are completely novel, with the same way interface completely changed with mobile devices. This is an even bigger jump where being comfortable with technology in your life in this way,

• 13:21Speaker 01

  it comes down as much to on the EQ side as it does on the IQ side.

• 13:25Nilay Patel?

  You've described a lot of really hard problems. I'm very curious how you're going to solve some of those problems. But it strikes me, you've been describing Vector as a character. A joke we say on the Vergecast all the time, it's a joke we kind of mean, is like, don't gender the robots, right? Alexa is not a she, Siri is not, these are just voice assistants. You're very directly gendering Vector. It's a he, it's a character. How did you make that decision? Are you, are you going to make she robots as well?

• 13:51Speaker 01

  So we will. And I think like, you know, down the road, I think the he was just a little bit easier because we could reuse a lot of more elements from previous, from Cosmo.

• 13:58Nilay Patel?

  Yeah.

• 13:59Speaker 01

  But down the road, we'll have female robots as well.

• 14:01Nilay Patel?

  What, what does it mean for our robots to be male or female?

• 14:04Speaker 01

  You know what? It's the same thing that when you think about, think back to like all the animated films that you've seen, or any movies in general where you have robots. If you have a character, having a gender makes it easier to make that character feel human-like in its traits, even if it's not human. And so Wally was a he, Eve was a she in, in Wally, right? You think of any sort of film that you would make, almost everything that comes to mind has a gender as a way to, to really amplify some of the character traits that fit in the character in the end. And so a lot of this is driven by our animation team where we have character directors that work in film that are thinking not just of gender, but and everything from the backstory, the personality traits. How would he react in every circumstance? What are his goals in life?

• 14:48Speaker 01

  And so Cosmo is this like rambunctious little toddler of a robot, whereas Vector is meant to be more subtle and almost like an older brother that is calmer and you would be okay having him always exist in the home. So you can imagine the storyboards that happen behind the scenes that model that character. But to really bring that to life, it's almost like saying Wally doesn't have a gender. That would be weird. He has a gender. It's not like a gigantic focus point of the story, but at the end of the day, by having a gender, it changes some of the mannerisms, some of the industrial design, and makes it a little bit easier to connect as a, with human-like characters, even if it's not a human.

• 15:24Nilay Patel?

  I mean, this goes right to your, your point about EQ. Do you think you can put robots in the home that don't mimic sort of like human social structures in this way?

• 15:33Speaker 01

  Well, and this is where the key is that you want to figure out what you want to mimic. So we've always been puzzled on why there's such an obsession with humanoid robots trying to replicate like skin and smiling and emotions. And you see some of the like projects from Japan where you have like some robotic secretary that like looks realistic, but doesn't actually like, you know, do anything AI wise. You you run into a trap there where you're stuck with all the strengths and weaknesses of a human and you're measured against a human where anywhere you come up a little bit short, you hit uncanny valley where you become almost creepy. By making a character that is a robot that clearly isn't trying to be human in look and form, we actually kind of put the advantage in our court because you can play to your strengths and avoid your weaknesses where we have a lot fewer degrees of freedom than

• 16:18Speaker 01

  a humanoid robot, which is why we can sell Vector for $249 and $199 in pre orders, right? So it's like, so the cost structures are a direct result of simplifying the form factor. But as a robot, it's okay to have four degrees of freedom and a huge amount of expression in the eyes and voice. But what we do try to mimic is the personality. So we've basically parameterized the emotions of a human where you think about how do you think, how do you map to happy, sad, surprised, bored, frustrated, scared, all of these like emotional traits and create this AI system where you map from the context of the real world to the right output. And when you beat Cosmo in a game, he gets, he gets grumpy and goes and sulks in a corner. That's kind of a human like emotion. It just becomes even funnier when it's in this like grumpy little robot who's now upset

• 17:03Speaker 01

  because you've like, you know, you've you've wronged him by beating him in a game, right? So you still connect to it on a human level, but we're not trying to be human. And I think that's like a really key and subtle variation where I think a lot of people go wrong by trying to recreate something that's exactly human because you distract yourself and you actually run the risk of hitting the uncanny valley.

• 17:25Speaker 02

  Hello, my name is Carrie Byron. I'm a former Mythbuster and seeker of truth. So I'm always very curious about new technology. Today, I'm going to talk to Sven Thiesen, who claims his life is changed for the better after purchasing an electric vehicle. Sven, you're a self-proclaimed EV evangelist. What is that exactly?

• 17:44Speaker 03

  I feel like I've got this wonderful gift that I want to give the world, and that's driving electric. So why am I so keen on driving electric? One, it's way more convenient. I don't have to go to gas stations anymore. Two, I have zero tailpipes, no tailpipes.

• 18:00Speaker 02

  All right, Sven. So what's the maintenance like on one of those electric vehicles?

• 18:04Speaker 03

  It's pretty much windshield wipers and windshield fluid. If you look at the number of parts, there's on the order of a couple thousand moving parts on a gasoline car. And on an EV, there's on the order of less than 100. So which one is going to break down first?

• 18:21Speaker 02

  How many EV converts do you think you're responsible for?

• 18:24Speaker 03

  Well, I'm famous for grabbing the guy with a muscle car parked beside our EV and made him drive our EV around the block. And he said, "An EV is so in my future," because he loved the instantaneous torque and the way it handled.

• 18:41Speaker 02

  Well, there it is. I've seen the light. To learn more about going electric with Audi, check out audiusa.com/etron. That's audiusa.com/etron.

• 19:00Nilay Patel?

  As you think about extending the capabilities of these robots, right, over the long period of time that you're describing, are you worried that giving them emotions is actually dangerous? You know, 50 years from now, there's the large robot from Aki in my home, and I beat it in a game, and it like flips over a table. Like, are you, do you have like ethicists on your staff that are, are thinking about these kinds of problems?

• 19:21Speaker 01

  Yeah, Cosmo can be a sore loser, the gigantic robot cannot. It's a right question. What we found is that everything we've done with emotion is hugely disarming for people, where you would typically be scared of something that has camera, a microphone, sensors, lasers, all these other things. You give it an emotion, it actually becomes disarming because you can more clearly read its intent. And we've seen this where you have this like, imagine like a large humanoid robot that has a huge arm that like loads and unloads a dishwasher. It's not only kind of imposing, but if it makes a mistake, you're also thinking, okay, why did that robot make a mistake? Something's wrong with it. Whereas if you have a character, not only do you immediately connect at a human level on what its intent is and what it's doing,

• 20:05Speaker 01

  but it makes you more forgiving of any limitations it has, because if he makes a mistake, as long as he's aware of it, you are way more forgiving because you personify it in the same way like if your five-year-old made a mistake or something like that. It's okay, because you know exactly what he was attempting. When the robots get more and more advanced and you have more sensing and more capability, emotion becomes even more critical, and it'll be more subtle. But people will naturally become alarmed if you start having like very advanced forms of technology, particularly when motion is involved. You're okay if a computer is very advanced, but if something's actually moving, just kind of from a primal standpoint, you're kind of more cautious of it. The right character makes all the difference, because if you move in a very confident manner, but a smooth manner, and you're not robotic, if you make eye contact and you acknowledge

• 20:50Speaker 01

  a command and you show warmth, you immediately, like we've seen this in our tests, it immediately disarms people. And people forget about the fact that it's a machine. They don't worry about the fact that there's these sensors that are required for the things that are happening. They think of it as a little robot who's super cute, and they start interacting with it like they would with a puppy versus a machine. So I think it'll become a key down the road. And, you know, it goes towards kind of the AI, bigger AI question where it's on us to program the right types of boundaries and use cases of this, of these emotions. But it becomes as much of an interface as like the PC only taking off when graphical interfaces became popular.

• 21:31Speaker 01

  Or the, you know, the role of an operating system in a mobile interface where, you know, it doesn't fundamentally change the underlying programs or functionality you're trying to do, but the interface makes it more approachable and eliminates friction and increases adoption. So that's the way we see it is that it allows people who are totally non-technical to feel like this is as natural as interfacing with a human or a pet or a computer today.

• 21:54Nilay Patel?

  I feel like the idea of adding emotions to decrease friction, I feel like a lot of people could take exceptions to that, right? Like, a lot of emotional interactions are actually quite confusing. Like, do you worry that you need to constrain sort of the boundaries of how emotions are communicated or what emotions are possible in order to actually reduce friction instead of increasing friction?

• 22:16Speaker 01

  For sure. And, and part of what we've done is like, we try to simplify it to kind of the normal extremes that you would think about and then modulate it for the circumstance, right? So a lot of it comes down to, you know, the social cues when you see somebody you recognize, right? Something that simple, you know, imagine not having that, like how much that would like kind of throw you off in your day if like there was no like visual or, you know, kind of physical response to somebody that sees you. You know, when you think about the way you move, if a robot is meant for entertainment, it moves in a certain way and has a certain personality. If it's meant for like, let's say, security and monitoring, but it still has to like coexist among people for an extended period of time, you're going to have a much more serious personality where you would think of it almost like a dog in the home that

• 23:00Speaker 01

  can be imposing, but he's super friendly to everybody that's around him. And so I think for us, we acknowledge that humans can have very confusing emotions and a lot of those becoming distracting. We have the ability to structure this and create the kind of probabilistic mappings on what you do and when and why to where if these characters feel alive, but they don't feel confusing or distracting, where it truly becomes additive and then you can really appreciate the, what they were meant to be.

• 23:25Nilay Patel?

  All right. Well, I want to, I want to stay in philosophy for a few more minutes with you. And then I actually want to talk about vector and the product itself, but. You described, you know, you've got this long range plan where the robot knows to come up to you. It knows not to interrupt. It's mobile around your house. That, just at first glance, sounds like three to four huge problems, right? You've got a mobility problem that a lot of people are working on. You've got almost like a general AI problem where people, you have to understand this constant stream of input. It makes sense of it. You've got an interaction problem where the robot is interacting with you different ways. Are you working on all that stuff now? Do you have a team that is looking at, you know, the Honda robots falling down the stairs?

• 24:07Nilay Patel?

  And you're like, we can do better. Are you, are you like in talks with iRobot to build a dog? Like, how are you thinking about all of those problems?

• 24:13Speaker 01

  Yeah. And so we are actually already investing in R and D on all of these sub problems. And a lot of it actually is going to come directly from the work that we're investing in for Vector. I mean, it's funny, you mentioned like, you know, Osmo, it's a perfect example of where you have billions of dollars of investment going towards product line. But it's on such a long timeframe where it becomes obsolete every three, four years as technology advances. So I'll give you an example, like just in the last year, there have been such amazing advances in sensors where the sort of sensing technology that you would need to robustly understand an indoor environment costs thousands of dollars. And it was just prohibitively expensive in a, in a product that you would sell at a reasonable price. Now, all of a sudden, in the last couple of years, you now have depth cameras that are effectively like, imagine a camera

• 24:55Speaker 01

  that instead of just giving you like RGB pixels, like color pixels, it gives you a depth for every single pixel as well. That's like the holy grail of perception because now it's the equivalent of what autonomous cars use on the street to understand the environment around them. And now you can build 3D models for localization, for mapping, for navigation, for obstacle avoidance and so forth. And these sensors now cost from tens of dollars for some of the high-end ones to as little as a few dollars for the low-end ones because they're being put into cell phones for face detection. You look at something like that, suddenly, just like deep learning, just like cloud functionality, voice interface, that becomes a sort of transformative technology where you can start thinking about, how do I put this into a platform that now for a very reasonable amount of cost can do things that you could never do before?

• 25:37Speaker 01

  So from a mobility standpoint, we have R and D going on that is actually focused on how would those Work from a both a platform and capability standpoint, but also from a character standpoint, where now like you have to be more careful. You can't fall over or hurt a pet or something like that. You have to be responsive to how people move around you and and move in a natural fashion. It's a it's a big, big enabler. When you think about the contextual understanding piece, this is where kind of the bottoms up approach comes into play. Everything we're doing in vector fits directly into all of these future product lines because we're using deep learning to classify humans and context in the environment. We're doing local slam. We're doing all these emotional interactions with people trying to understand the context of what's going on. We're mapping into into both utility,

• 26:21Speaker 01

  cloud functionality and smart home to integrate in all these ways. And so all of these learnings that we're going to get off of a smaller platform that is being driven by the more kind of companionship focus ends up extrapolating perfectly to the more serious applications where you're now doing deeper functionality in home, healthcare, or whatever the case might be. But those technologies actually scale well, everything down to the voice interface systems where you're speaking to to vector, you're giving, you know, you're asking him a question, are you giving him commands, where you say something mean and he responds in a way that like that right now, you know, a voice assistant can't respond to, but he can actually be a little bit upset, just like a friend would if you yelled at him. Right? So these are the sort of things that we can develop in vector that are very important for that application, but end up being core technologies that we can leverage elsewhere.

• 27:07Nilay Patel?

  Yeah. So do you think you're going to be able to solve like straight up the general AI problem? Is that something that you are working on where people are saying voice assistants are are always going to have a narrow domain? Or do you think over time you can actually build a truly capable assistant of this kind?

• 27:21Speaker 01

  Yeah. So I think like internally, like, no, we can't solve the generally I problem in the next few years.

• 27:25Nilay Patel?

  I figured that was the answer.

• 27:27Speaker 01

  Yeah. So and it's just it's totally empirical. And in a lot of ways, like, the interesting thing is that voice is a perfect example where. We kind of don't need to be focusing on solving the voice interface problem because it's becoming more and more commoditized with the amount of people working on it and the tools are becoming available. But how we use it is completely novel. And so that becomes a like, just like we're not going to invent new sensors, we're going to use those sensors in very new ways that that people hadn't thought of. That's how we think of something like voice, AI and understanding. But there are certain core problems that I think we may need to solve first before anybody else has had to solve in a mass market application. So really detecting contextual understanding of like, you know, humans and positions and

• 28:10Speaker 01

  roles in an environment with an intent to interface with them. There's research being done on this, but there's no product that's actually doing this at scale. And so in the case of Vector, we can develop a lot of this technology, even if his purpose is to look up at the person associated with the arm he saw and get their attention or go and like poke you because he wants your attention, just like a pet would, like a cat or a dog would. At the core of it, that's the exact same type of baseline technologies that will then scale to him getting your attention and delivering a personalized message or reminding you to do, you know, to do some medical step or something like that. So we're trying to tackle problems in a reasonable fashion that we can solve, but then keep our eyes out for how the external world of technology evolves every year and pull those in as fast as we can.

• 28:55Speaker 01

  And for something like general AI in terms of voice interface, we'll let Amazon and Google and others put in billions of dollars into that because the competition naturally creates that as a tool for others to be able to access.

• 29:07Nilay Patel?

  So let's end with Vector. We've talked about all these other stuff and we have not talked about the product you have out. Vector is a companion robot. Describe it to me.

• 29:16Speaker 01

  Yeah, so Vector is a, it's a robot that lives in your home. And what's special about this is that this isn't a robot that you turn on and start an app and play with for a session, but the goal is literally to have him be alive. And a year later, he's still alive and nobody turned him off.

• 29:29Nilay Patel?

  Wait, is that, is that the timeframe? Is that, is that success? You have it for a year and no one's ever turned it off?

• 29:34Speaker 01

  So that's one of our internal metrics is like, you know, a year later is everybody, like are people still actively engaged with it and using it and feel like it adds warmth into their, into their life? So it's funny, there's no, there's no. No comps for this because like mobile game has have comps and so forth. For us, this is like brand new territory.

• 29:49Nilay Patel?

  Yeah.

• 29:49Speaker 01

  But it'll be the first time where there's a robot that's always on. He recognizes you, he gets excited to see you. There's all these like kind of emotional and companionship driven interactions that are modeled almost like how you would interact with a cat or a dog. But he also has this ability to create applications in that are utility driven, but with a really unique spin around it where you have this emotional and characterful wrapper around it driven by Vector as your actual assistant who's actually helping with those things. And so what we're doing is we're tackling some of the most commonly used elements of voice assistants that we can wrap personality and character around and create kind of a deeper warmth to it. And then we're also working on ones that you physically would not be able to do with a traditional voice assistant because of the lack of awareness of the environment and

• 30:34Speaker 01

  ability to initiate interactions. And at the core, I think the hook of what makes people love Vector and it's and this is why it's such an experiential product because there is no nothing to compare it to in terms of other products today. But it's the same reason why you fall in love with a pet that's happy to see when you come home, that is responsive to touch, to voice, that hangs out with you both in like an intensive interaction when you might be playing with him, but also just being close by if you're working or watching TV or something like that. And it starts to really vet out both for us and for customers, how important and powerful can this emotional layer of robots be? And from all of our testing and our kind of conversations with with Dieter and others,

• 31:16Speaker 01

  it's very surprising when you see him on how immediately kind of like alive he feels and how different that is from anything else that you've experienced in a home.

• 31:25Nilay Patel?

  So who should be interested in buying Vector? I mean, you're obviously doing great on Kickstarter, but who's who's like your target customer for Vector?

• 31:32Speaker 01

  Yeah, so our term for the target audience is tech immersive. So it's, you know, this isn't a product that fits squarely into a replacement for something else because it's a very new category. But it's basically everybody who's interested in the future of home technology that is interested in kind of the fun aspect of having almost a pet like living robot in their home. And a lot of it will be families where it's kind of the tech savvy parent that buys. And the kids love it and the parents love it and it kind of creates kind of a warmth in the home. But effectively, it's, at this point, it's still because you want it, not because you absolutely have to have it. In the same way that people want a dog or a cat, even though when you like list it out on paper, both like financially and in terms of restrictiveness, it doesn't make sense.

• 32:12Speaker 01

  But you want it because of all the emotional cues. And for us, there's a lot of people that even back in Cosmo, they're like, look, I want this guy as a desk pet, but ABC's missing. Or I would love this guy to give this to my grandmother who lives alone, or just to have this around the house. So it's effectively kind of this new category of home robotics that for us is very kind of companionship driven at its core, which is something that I think we do in a very unique fashion compared to anybody else.

• 32:37Nilay Patel?

  Are you doing software updates over time? Is Vector going to get more capabilities over the air? How does that work?

• 32:42Speaker 01

  Oh, without a doubt. So Cosmo's already had like 23 updates at this point, maybe 24. And so, and massive ones at that. So Vector's going to increase tremendously. So the benefit of having a product that's, you know, he's got a quad-core CPU, he's got all these sensors, cloud connectivity, all these sort of things. The hardware is meant to be as general as can be for all the things we're going to do in the future. And so we're going to be aggressively pushing on a huge amount of updates in software across the board on all these capabilities. And so I'm going to give you one example, like for next year, one of the things we're thinking about is, on the kind of utility side, think of like kind of his ability to be kind of a home camera or home monitor. Where you have all these like kind of cameras today, they're static and have a fixed vantage point and they're pretty much a commodity.

• 33:23Speaker 01

  With just a software update, with Vector, you have a 360-degree field of view, high-definition camera. He can look up and down, and when there's a noise, he can look in that direction and he can recognize people. And so now you have a superior home camera that's based on this platform, that's a pure software update that we'll release somewhere down the road. Same thing goes for other utility functions, same thing goes for all of these kind of characterful interactions where we want him to have all of these, hard to like quantify as features, but there are all of these like deep capabilities that make him feel more and more alive. As always, we wish we had the next six months of roadmap we could shove into the launch. And you always end up kind of launching a product and having a gigantic roadmap in front of you. We're going to be releasing multiple updates before Christmas and then probably on the order of every one month or so, we release an update as we've been doing on Cosmo.

• 34:09Nilay Patel?

  The idea of a mobile security camera with emotions is both very powerful and interesting. And do you ever worry it's ever like a tiny little bit invasive or creepy?

• 34:17Speaker 01

  Well, essentially, it's kind of a first of all, it's an opt in. So, you know, from a privacy standpoint, we're really careful about this where everything is on the edge, like on the device, so that there's no worry, like no images or anything get, you know, get sent up, except for a voice after the trigger word, just like you would with an Alexa for voice classification. With something like this, it would be an opt in where like I would imagine this being an option, where if you want to tap into your vector, you can, just like you would your Nest cam or something like that. Or if you want it to be always on, maybe there's a cloud feature that we can enable that, you know, is a small fee or something like that, that is always storing. But at the end of the day, it's no different than the huge market that's there for home security cameras, except that you have far more capability to cover the entire room you're in versus just a fixed vantage point with no ability to actually understand context beyond that.

• 35:02Speaker 01

  But it's funny you say that because like it becomes an extra feature of this character living in your home. But we like one of our challenges is, you know, we want to be focused in the message. We don't want this to be distracting from the fact that this guy's a character first. And it's about living with a robot for the first time in a way you see in a sci fi film. But then we see these opportunities to start to layer in some of these capabilities, which frankly, then also become probably subsets of deeper functionality down the road that other robots might be useful for. But everything like this, we'll be talking about would just be a feature that is an opt in if you want it.

• 35:34Nilay Patel?

  Yeah. What do you think the timeline is for that next big robot? Are you two years away? Are you five years away? When do you imagine we're going to have the full on home companion robot?

• 35:45Speaker 01

  Yeah, so I think that's, you know, similar to like a mobile phones or, you know, state of computers. It's never done. They're always improving. And robotics is even more susceptible to the external ecosystem because of like sensors, computation, like state of AI. So there'll never be like a checkpoint where like we're done checkbox. Here's a home robot, but you're going to start seeing increasing stages of functionality, where I think we're probably only a handful of years away from really robust and cost reasonable functionality with mobility in both consumer and enterprise environments, where you truly find a good parity between. Value and function. But the next big leap that's going to be missing is manipulation. Where if you think about all the things that you do in the home, a vast majority of them are not just informational,

• 36:29Speaker 01

  but they're actually physically involving some sort of labor. That's still probably, I would say, at least like five, six years away. It's almost like where autonomous driving was like five years ago or six years ago or so, where the costs are much higher because of the cost of the manipulation required for a mobile arm or home robot. The AI is much more complex. And the combination of those means the bar is very high for a higher priced robot than to become functionally meeting the bar in a home environment. And so right now that's still research, but when you fast forward 15 years, robotics will be in the home and will have manipulation as part of it. Because both the cost structures and the AI capabilities will create enough value to where the costs will be worth it.

• 37:13Nilay Patel?

  It's 15 years. I'm holding you. 15 years. I'm setting a calendar reminder right now. August 28th.

• 37:18Speaker 01

  I don't think that's unreasonable. We're almost there from an AI standpoint for very robust mobility and obstacle avoidance and so forth. We're making huge progress towards like interface, voice, emotional, the things that we're doing and so forth. Manipulation has some amazing research going on that still has to advance and then there's cost structures involved. And deep learning has opened up incredible abilities in planning and perception and understanding the environment. A decade is a lifetime when it comes to fast evolution on that front. I mean, I was using a comparison autonomous driving like I was, you know, back in 10 years ago. We were finishing up graduate school and we're kind of deep into graduate school. And autonomous driving was a huge research focus, but like nobody even paid attention to it outside of research.

• 38:01Speaker 01

  I mean, it was like pulling teeth to get like a million dollars of funding from GM. And then you fast forward 10 years and you have companies putting in half a billion dollars at a time into these autonomous car companies where it's not there yet, but you can tell that it's kind of inevitable that autonomous driving is going to happen. So a lot can change in a pretty short amount of time where on a 10, 15 year window, that's a pretty long period of time to transform the sort of human robot interactions that are possible in a home.

• 38:25Nilay Patel?

  August 29th, 2033, I'm calling you. We're gonna see, we're gonna see what kind of robot you have out.

• 38:31Speaker 01

  We'll have a chat on the anniversary.

• 38:33Nilay Patel?

  Excellent. Well, Boris, thank you so much for joining us today. Vector is, you can get it. When can you just buy it regular?

• 38:39Speaker 01

  Vector launches mid-October and it'll be $249 at retail.

• 38:42Nilay Patel?

  Excellent. Well, thank you so much for joining us, Boris. A fascinating conversation. We'll talk to you again soon.

• 38:46Speaker 01

  Wonderful. Thank you so much for having me.

• 38:48Nilay Patel?

  All right, Boris Softin, CEO of Anki, a great conversation, cool dude. Vector is, I think the Kickstarter is done, but you can go preorder it right now if you're interested in it. We'll be back later this week with a normal, crazy Vergecast and on Tuesday again with another interview, I'm not gonna give it away, but I do want to know what you think of these interview episodes, so tweet at me, I'm at Reckless, we'll see you soon. This episode of Vergecast is brought to you by the Audi e-tron. Despite all this technology, there's a lot that the all-new Audi e-tron doesn't offer. For example, it has no tailpipe emissions and no need to fill up at the gas station, you just plug it in at home. The quattro all-wheel drive system offers no reason not to tackle roads and almost any weather, and long-range capabilities and high-speed charging e-tron is a new way to think about electric mobility, which makes sense, it's the first fully electric vehicle from Audi.

• 39:33Nilay Patel?

  E-tron was built to defy the elements and upend the conventional wisdom, so in truth, it's not lacking anything, a twist that you did not see coming. After all, it isn't just an electric car, it's an electric Audi. E-tron is here and the future is electric. Visit Audi.com/e-tron to learn more and stay informed.