Latest / Automated with Brian Heater / Colin Angle on Why Home Robots Failed Before and Why AI Changes Everything

Transcript

Colin Angle on Why Home Robots Failed Before and Why AI Changes Everything · Automated with Brian Heater

• 0:00
  SPEAKER_00: In the land of robotics, what AI's done is allowed us to do things that we could only dream of.
• 0:05
  SPEAKER_00: Now we're in a world where those things we could only imagine 10 years ago suddenly aren't just possible, they're practical.
• 0:13
  SPEAKER_00: What a cool moment in time to be a roboticist.
• 0:17
  SPEAKER_00: Necessity is the mother of an invention, and we hooked on to the AI bus at the right time, and the AI bus has only accelerated, not slowed down.
• 0:26
  SPEAKER_00: Building robots is really hard.
• 0:28
  SPEAKER_00: So we have to be more than just a watch mood toy.
• 0:32
  SPEAKER_00: What we're announcing is our first step in delivering a different kind of robot.
• 0:38
  SPEAKER_00: This is the greatest application of robots that in my entire career I could imagine.
• 0:44
  SPEAKER_00: I just couldn't build it until honestly six months ago.
• 0:49
  SPEAKER_01: The generations of attempts at home robots and home pets didn't last, and people do have expectations about what robots can and can't do in the home.
• 0:58
  SPEAKER_01: So, you know, how are you really going to express to people that this is something different?
• 1:05
  SPEAKER_00: Well, I think that's a good one.
• 1:21
  SPEAKER_01: I am the managing editor at the Association for Advancing Automation.
• 1:26
  SPEAKER_01: I am here to bring you our first ever returning guest.
• 1:30
  SPEAKER_01: As a matter of fact, it's only been, I think, a month or two since the last time we had Colin Engel on the show, although this time we are chatting about happier circumstances than recent iRobot News.
• 1:41
  SPEAKER_01: Colin is back in the CEO role at a brand new company, Familiar Machines and Magic.
• 1:48
  SPEAKER_01: He walks us through the decision to get back in the saddle and explains why this shot at home robotics is different than all of the others that came before it.
• 1:57
  SPEAKER_01: We also dig into the early days of iRobot and entertain my obsession with the company's My Real Baby Doll.
• 2:04
  SPEAKER_01: If you're enjoying the show, don't forget to like and subscribe, and please check out the newsletter over at automated.fm.
• 2:11
  SPEAKER_01: And with all of that, please enjoy this conversation with Colin Engel.
• 2:16
  SPEAKER_01: You know, we talk a lot about what's coming up next in automation on this show, but if you really want to see the future motion, you've got to be there in person.
• 2:24
  SPEAKER_01: Automate 2026 is where the world's leading innovators, builders, and dreamers come together to show you what's possible.
• 2:31
  SPEAKER_01: Robots, AI, machine vision, motion control, you name it, all automation under one roof.
• 2:37
  SPEAKER_01: And as part of Automate this year, the Humanoid Robot Forum brings together leaders, engineers, and researchers for a two-day deep dive into the real world development, deployment, and commercialization of humanoid robotics.
• 2:50
  SPEAKER_01: Register for free at automateshow.com to join us in Chicago June 22nd through the 25th.
• 2:57
  SPEAKER_01: We will see you there.
• 2:59
  SPEAKER_01: Was retirement ever an option?
• 3:03
  SPEAKER_00: How could you retire when the tools to create the coolest robots that we have ever seen were suddenly available?
• 3:12
  SPEAKER_00: I mean, it w uh you you I I guess I thought about it, but then um rejected it swiftly.
• 3:20
  SPEAKER_01: Yeah, so let let's let's break that down a bit.
• 3:23
  SPEAKER_01: Um when you say suddenly available, I you know, I assume to a certain extent you're probably talking about LLMs, you know, this these sort of breakthroughs that we've been seeing in in AI, some of the hardware breakthroughs.
• 3:37
  SPEAKER_00: Well, for a long time, the challenge of building a robot was situational awareness.
• 3:46
  SPEAKER_00: You could for a long time there's been speech recognition that could understand what you said, but robots had a really hard time understanding uh enough context to translate an understanding of go to the kitchen and get me a beer into going to your kitchen, opening your refrigerator, and actually bringing you back a beer or whatever it was.
• 4:13
  SPEAKER_00: And so the uh the power of what AI has brought to the robot industry uh actually is more profound in many ways than how what it's done to the uh chat world, because we can understand our environment, we can uh control higher degree of freedom systems in a dynamic and and graceful way.
• 4:38
  SPEAKER_00: Uh we can uh then take all that and start overlaying personality uh to what we're doing.
• 4:45
  SPEAKER_00: And so it's like it's not like we could used to be able to do spreadsheets.
• 4:51
  SPEAKER_00: It was just took forever and you had to be really good at math, and then suddenly spreadsheets made it fast and accessible.
• 5:00
  SPEAKER_00: In the land of robotics, what AI has done is allowed us to do things that we could only dream of.
• 5:09
  SPEAKER_00: There was no solution that was clunky but serviceable.
• 5:14
  SPEAKER_00: Now we're in a world where those things we could only imagine 10 years ago suddenly aren't just possible, they're practical to do.
• 5:23
  SPEAKER_00: And and so what a cool moment in time to be a roboticist.
• 5:28
  SPEAKER_01: If I'm reading between the lines here, it sounds like you're talking um a bit about the difference between hard coding a robot and using machine learning.
• 5:36
  SPEAKER_01: Is that fair?
• 5:38
  SPEAKER_00: Yeah, but at different levels.
• 5:40
  SPEAKER_00: Sort of at the lowest level, just making a walking robot walk used to be incredibly hard.
• 5:47
  SPEAKER_00: Now that's table stakes.
• 5:49
  SPEAKER_00: And the question is how much dexterity and how much speed and uh can you get out of those legs?
• 5:57
  SPEAKER_00: Uh, how much expressiveness?
• 6:00
  SPEAKER_00: But what about facial expression?
• 6:02
  SPEAKER_00: What about manipulation?
• 6:03
  SPEAKER_00: What about full-body animation?
• 6:05
  SPEAKER_00: Things that as you add more and more degrees of freedom to your robot, it gets dramatically harder and harder using traditional techniques, but reinforcement learning suddenly allows you to train motion policies that are stylish and dynamic and amazing, and that's just the low-level stuff.
• 6:27
  SPEAKER_00: Then you add to that this idea that uh I can take audio and video streams coming in and translate them into a uh a rich understanding of the environment that I'm operating in.
• 6:42
  SPEAKER_00: Well, we couldn't do that particularly well, you know, two years ago.
• 6:49
  SPEAKER_00: We couldn't, you know, a year and a half ago.
• 6:51
  SPEAKER_00: Uh, and that's all changing incredibly quickly.
• 6:53
  SPEAKER_00: And then the ability to reason on that context and uh not just a reason reason in it on it from an intellectual perspective, but add an emotional perspective about what we're seeing, you can just it's cool.
• 7:09
  SPEAKER_00: It's uh a new toolkit of awesomeness that is unlocking the potential of uh, you know, even the word robot doesn't make sense anymore because, um at least in my mind, robot is associated with dull, dirty, dangerous tasks.
• 7:26
  SPEAKER_00: And when what we're really opening up is something much broader than what traditionally robots were thought to be able to do.
• 7:34
  SPEAKER_00: Um, and so using this new term physical AI is something that uh I think is a uh a great term because it's more expansive in scope.
• 7:45
  SPEAKER_01: Yeah, obviously you and I have talked many, many times over the years, and and we talked a lot uh during your time at irobot.
• 7:54
  SPEAKER_01: And towards the end, you know, it strikes me that you were laying the foundation or at least attempting to lay the foundation for this.
• 8:01
  SPEAKER_01: Um so much of the work that you were doing in uh your last several years came down to uh to mapping the home.
• 8:09
  SPEAKER_01: You know, I know that you had um the those sort of early partnerships with uh Amazon and Alexa.
• 8:15
  SPEAKER_01: So it was voice, it was smart home.
• 8:18
  SPEAKER_01: So, you know, it seemed like to me that you were sort of pulling together potentially the pieces or the uh maybe the infrastructure.
• 8:26
  SPEAKER_01: But now that now you're describing these um these breakthroughs that have happened in the last two years.
• 8:32
  SPEAKER_01: I'm wondering how much of it you would have had to have just sort of like scrapped and started from scratch anyway.
• 8:38
  SPEAKER_00: I mean, the ideas are translate over.
• 8:42
  SPEAKER_00: This I, you know, this idea when you're creating a a robot that is designed to live in a home, the iRobot journey brought with it just an appreciation of how important the relationship between your Roomba and the Roomba's owner was in actually having a positive experience.
• 9:05
  SPEAKER_00: How do you keep Roomba out of the closet?
• 9:08
  SPEAKER_00: And and what does it take for people to four years later be using the Roomba multiple times uh a week as opposed to once a week, then once every other week, and then it's oh yeah, I've got one of those.
• 9:21
  SPEAKER_00: And um the trajectory of engagement was some of the really interesting work we did toward the later years of iRobot.
• 9:32
  SPEAKER_00: Um, and you know, we had limited tools, but uh, you know, it came down to finding a way for the Roomba to be part of the family's routine.
• 9:44
  SPEAKER_00: And by far, largest failure mode of Roomba to clean your home was user cancel, someone turning it off because it tried to vacuum the living room while someone was watching television, so they just reached down and thinkingly turned it off because it was annoying, and it never got turned on again.
• 10:06
  SPEAKER_01: See, I thought you were gonna say poop.
• 10:08
  SPEAKER_00: Well, that's that's number two, so to speak.
• 10:12
  SPEAKER_00: Poop was was something that was actually much more solvable than getting into your routine.
• 10:18
  SPEAKER_00: I mean, that was that was a big, big deal because it didn't happen very often, but when it happened, moy, that was bad.
• 10:23
  SPEAKER_00: But uh computer vision actually worked really well in solving that problem.
• 10:29
  SPEAKER_00: But understanding a home environment, well, there's something that's uh a less easily tractable challenge.
• 10:39
  SPEAKER_01: Yeah, so as as you mentioned before we got on, I I I dug up a very old screenshot of you, and it was from a very old video of you.
• 10:48
  SPEAKER_01: And it's interesting to me in light of the conversation that we are currently having and in light of the news that you're just about to announce.
• 10:59
  SPEAKER_01: Um far in a way, the most interesting thing about the early days of iRobot is this kind of bifurcation in the projects that the team was doing, right?
• 11:09
  SPEAKER_01: Um so much of the funding was coming from DARPA.
• 11:12
  SPEAKER_01: So there were a lot of these sort of like military-based robots, uh search and rescue, things like that.
• 11:18
  SPEAKER_01: And then at the same time, you know, you were going after the home.
• 11:21
  SPEAKER_01: Obviously, Roomba is the big one that everybody knows.
• 11:24
  SPEAKER_01: And then the other one that makes an appearance in here that I think is, you know, maybe in a sense a precursor is my real baby.
• 11:34
  SPEAKER_00: My real baby was actually the culmination of a lot of work we did on can a robot effectively emote.
• 11:42
  SPEAKER_00: You know, there have was uh back then, and the body of evidence has only grown, uh showing that many interactions on on the screen are less impactful than a similar interaction, where instead of it being a screen, it's an embodied physical thing.
• 12:06
  SPEAKER_00: Uh the really cool research into brain MRI scanning, where you have you look at someone doing a chat bot and their brain activates a teeny tiny bit.
• 12:18
  SPEAKER_00: Whereas if they're having the same interaction with a physical thing, the brain is lighting up, oxytocin is being uh uh emitted, and and the the the impact is greater.
• 12:32
  SPEAKER_00: And so that even in early iRobot days, we are seeing, we are thinking, you know, if you build a robot, it's hard.
• 12:41
  SPEAKER_00: I mean, building robots is really hard.
• 12:43
  SPEAKER_00: So we have to be more than just a watch me toy.
• 12:49
  SPEAKER_00: We have to be more than something that heck, I could put a plate of glass between me and the robot and the interaction wouldn't change.
• 12:57
  SPEAKER_00: If if that's what you're doing, well, just make it a computer program that runs on a screen.
• 13:04
  SPEAKER_00: But if there's actually an opportunity to do something physical, whether that is something as mundane as Roomba and vacuuming the floor, or something that is more physically interactive, where you're actually holding something like Myelreal Baby and rocking it, um, you're going to have a much more powerful experience.
• 13:25
  SPEAKER_00: And uh, we were trying to figure out how to um really create significant value with physical interaction in the early days, but the technology just wasn't there.
• 13:41
  SPEAKER_00: Um, it was too hard to understand what was going on.
• 13:44
  SPEAKER_00: The um uh and we had you know a few interesting successes, but none of it really scaled.
• 13:57
  SPEAKER_01: Yeah, the reason why I stumbled on that video specifically is because on the uh about page of your new company in your it there's a you have a small bio and it mentions Genghis.
• 14:11
  SPEAKER_01: Genghis was a very, very early project that you were doing.
• 14:15
  SPEAKER_01: And I I'd love it if you could really kind of walk me through this connection because you know it's it says that there's a connection there to, or that it was maybe an early precursor to Ibo to Sony's robot dog.
• 14:27
  SPEAKER_01: What what what is the link there?
• 14:29
  SPEAKER_00: Uh Genghis was actually my undergraduate thesis at MIT.
• 14:34
  SPEAKER_00: And we were thinking about you know what our robots are good for, and and at the time um that we had conversations around uh, I think NASA had talked about its Mars rover program.
• 14:48
  SPEAKER_00: And and at the time, the plan was to send something about the size of a Hum V to Mars or the Moon, you know, as the next generation of rovers, and it needed to be big, that big, because in order to roll over a half-meter boulder, you needed a one-meter tire, and so that set the whole logic behind the scale.
• 15:12
  SPEAKER_00: And yet, insects are really good at climbing over things that are much bigger than they are, and so that this idea of walking robots uh is uh a really interesting place.
• 15:24
  SPEAKER_00: And at the time, this is 1989, walking robots were research topics that had multimillion dollar hardware budgets and so forth.
• 15:36
  SPEAKER_00: And Rodney Brooks, my professor, and I said, you know, let's build a bug.
• 15:41
  SPEAKER_00: And Genghis was a six-legged walking robot using hobby servos to control the legs and a very simple set of behaviors, at least computationally simple, that interacted with one another to allow Genghis to uh walk over very rough terrain.
• 16:03
  SPEAKER_00: And Genghis indirectly led to uh the micro rover program at NASA and led to uh the Sojourner Rover being added to the Measure Pathfinder mission, and so that it had a really interesting trajectory in space exploration, but we had this group from Sony come in, look at it and say, wait, you can do that?
• 16:27
  SPEAKER_00: Oh, that's interesting, and led directly to the Sony iBow Dog being created, and that that's fascinating and is relevant because when we founded iRobot, the actual first name of iRobot was Artificial Creatures Inc.
• 16:44
  SPEAKER_00: This uh this idea that we could use the AI that we were creating way back 35 years ago, um, to uh try to build the robots they were promised.
• 16:59
  SPEAKER_00: And these were actually sentient creatures that lived alongside us and had interesting reasons to exist and purpose.
• 17:07
  SPEAKER_00: That, you know, we did work on facial expression, we did work on, you know, my real baby to try to figure out how do you make a baby that actually understands how it's being played with.
• 17:17
  SPEAKER_00: Uh, and uh ultimately that got put down and then started to be picked up late in the game as we realized we needed to move from building robot vacuums that uh sort of operated independent of people and moved toward building robot vacuums that really worked with you in close partnership to understand what you wanted it to do and uh knew itself whether it was doing a good job or not.
• 17:54
  SPEAKER_00: It was sort of a journey of focus on human connection toward more uh simpler uh uh solutions that required less human connection, you know, turn it on by pushing the button, and then um later years, you know, if this thing is going to take its next step forward, we got to bring the human connection back into the equation.
• 18:21
  SPEAKER_01: One of the things you mentioned in the in that video with regards to my real baby is um uh there's a there's a ball inside the robot, which you know it it seems to essentially function like a gyroscope, you know, as far as that the baby knows that it's being it's being moved and it and and it and it reacts to that.
• 18:38
  SPEAKER_01: And obviously it, you know, these are hard-coded reactions we're talking about, that there are like a set number of reactions that this baby has based on certain ways that you uh that you interact with them.
• 18:50
  SPEAKER_01: Now, as we're talking about this next step and these things that AI can do, you know, what does it mean for a more complex, uh more sophisticated robot to actually learn from human behavior?
• 19:07
  SPEAKER_00: If you're trying to build a uh a machine that that uh is designed to design for human connection, then somehow you need to teach that machine social norms, or at least uh some level of social norms so that it knows when it behaves in a certain way that you're going to interpret that behavior in a certain way.
• 19:38
  SPEAKER_00: And when you do certain things, it it should be able to get a sense of of what's going on, because I should react in this in an appropriate fashion.
• 19:50
  SPEAKER_00: And so that the uh, you know, if you try to sort of break the robot AI revolution into three parts, one part is the motion part, which we've talked about already, one part is the perception part, this idea of of taking audio and and video streaming information and turn that into uh, you know, your your um uh your someone named John that you live with just came in the door and he looks upset and the room is messy and there's the following things going on, and that this tableau of information is then sent to a reason model, which has a job of taking that tableau of information coming to it and deciding what to tell the body of the machine what should it do.
• 20:48
  SPEAKER_00: And it's not hard-coded anymore, it's trained.
• 20:53
  SPEAKER_00: And in fact, there are ways of training models now where the way we do it uh at my new company is we write stories.
• 21:05
  SPEAKER_00: We write stories about how we want our familiar.
• 21:10
  SPEAKER_00: This is my name for an emotionally sophisticated, emotionally intelligent um robot machine that's focused on uh human connection.
• 21:19
  SPEAKER_00: So my familiar has a personality that is built from uh, you know, a few dozen stories about how it should react and behave.
• 21:35
  SPEAKER_00: And then those stories are taken by AI and turned into thousands of stories, and those thousands of stories are used to train models uh that connect information about what's going on in the world to um uh to actual familiar behavior.
• 21:58
  SPEAKER_01: Yeah, I Uh I you know I I bring this up uh every so often when we're kind of t talking in this space, but um you know years ago I was interviewing the the founders of Anki, and remember at the time that they they raised a lot of money, and one of the things that they did with that funding was go and hire a bunch of animators from DreamWorks and Pixar.
• 22:22
  SPEAKER_01: And looking through um, you know, some of the staff with Familiar Machines in Magic, uh, there are some Imagineers on there, there are some Disney folks on there.
• 22:33
  SPEAKER_01: You know, what what does it mean to pull people in to do storytelling?
• 22:38
  SPEAKER_01: And, you know, what are those sort of sorts of lessons that you can learn from like animation, for example?
• 22:44
  SPEAKER_00: It might be helpful to take a step back and sort of talk about uh the company's mission, and then we can get into the tactics because the company is this really exciting blend of advances that iRobot made and just blowing out of the water the price performance uh expectations of what a robot could be.
• 23:10
  SPEAKER_00: We couple that with uh a strong dose of Boston dynamics, where uh which taught us that machines don't have to move like machines, or at least they don't have to move like traditional concepts of machines.
• 23:31
  SPEAKER_00: Uh and then you add a third uh dimension of Hollywood, um and you get a very interesting set of capabilities.
• 23:43
  SPEAKER_00: And my stepping back is why are you doing this?
• 23:47
  SPEAKER_00: And it it gets at the fact that this new toolkit that we've talked about is revolutionizing what machines can do.
• 24:00
  SPEAKER_00: And and so there's this new name, physical AI, of uh, that represents an opportunity for machines uh to enter our world and create tremendous value and expectations of these of the value that physical AI can bring to the world, you know, five trillion dollars over the next 20, 30 years.
• 24:21
  SPEAKER_00: And um uh and typically the thought process is okay, we're really talking about humanoids and factories, right?
• 24:31
  SPEAKER_00: And the answer is, well, yes, but only about half of that five trillion dollars represents manufacturing and and and uh uh inspection and warehouse automation types of types of uses.
• 24:46
  SPEAKER_00: The other half has to do with machines that interact with people that are doing things like elder care, that are doing things like providing emotional support or helping parents get through the day or providing entertainment or office security or finally making the smart home work.
• 25:05
  SPEAKER_00: Things where these robots are not operating in a lights out factory on the other side, they're actually creating very useful work in our homes, and we need to figure out how to interact with them.
• 25:23
  SPEAKER_00: And so, okay, that's a really hard, really interesting problem.
• 25:27
  SPEAKER_00: And the uh, you know, depending on which YouTube videos or ex uh videos you you you are uh what your favorite channel is, there's a lot of uh energy out there sort of showing us humanoids in the home, uh which certainly could happen eventually, but that's not how it's gonna start.
• 25:52
  SPEAKER_00: You know, this this idea that you're gonna have a humanoid rub in your home pushing an upright vacuum cleaner, it's like, come on, guys, uh, we solved this one already.
• 25:59
  SPEAKER_00: We don't we don't need a$20,000 humanoid robot pushing in an uh uh an upright vacuum cleaner.
• 26:07
  SPEAKER_00: What we do need is maybe a little bit of help in the home to reinforce a healthy routine.
• 26:15
  SPEAKER_00: You know, it's not so much robot pets, but maybe it's in the same genre of creatures as opposed to humanoids that we want to physically interact with, but that have a little bit more intelligence than um at least or at least a little more purpose and reliability beyond staying fed and getting the right amount of sleep, that could help us uh get some more exercise, that uh could help reduce isolation by saying, you know, how about we go for a walk?
• 26:52
  SPEAKER_00: Or uh, you know, this doom scrolling thing you've been doing for the last hour and 10 minutes isn't really probably how you want to be spending your time.
• 27:00
  SPEAKER_00: Why don't you go to bed and tomorrow make a different decision?
• 27:05
  SPEAKER_00: Um there's a very interesting entry point as long as we can solve some important challenges.
• 27:20
  SPEAKER_00: Um, you know, does the this new familiar um does it look like something that I want with me when I take a walk?
• 27:32
  SPEAKER_00: Um can I trust it?
• 27:36
  SPEAKER_00: Obviously there's lots of stories in the media about um AIs that are overstepping, you know, plush toys with an AI chip giving dating advice and things like that.
• 27:46
  SPEAKER_00: And how do I make sure that this isn't something that ends up in a closet like so many robots after their minutes of uh initial minutes of glorious wow, look how cool this thing is.
• 28:02
  SPEAKER_00: Uh and uh at Familiar Machines and Magic, we uh really tried to take on these challenges head-on and create something that could, you know, one of my go-tos is I gotta build a robot that delivers more value than it costs to create.
• 28:21
  SPEAKER_00: I need to make it delight, I need it to be trustworthy, I need it to uh deliver long-term engagement.
• 28:30
  SPEAKER_00: And uh what we're announcing uh is our first step in delivering a different kind of robot, which we think meets these criteria.
• 28:45
  SPEAKER_00: So it is um more pet-like in appearance.
• 28:50
  SPEAKER_00: It's not a dog, it's it's it's not a cat, it's a furry quadruped, it's because we want you to pet it.
• 28:56
  SPEAKER_00: I mean, people spend on on average an hour and a quarter a day petting their pet.
• 29:01
  SPEAKER_00: If you could do something like that with a familiar, oh my god, um you need to trust it.
• 29:08
  SPEAKER_00: So let's not stream any video or audio to the cloud.
• 29:12
  SPEAKER_00: Let's do it all on the edge.
• 29:14
  SPEAKER_00: Let's sort of use Roomba rules where if you want to connect it to the cloud, you can.
• 29:22
  SPEAKER_00: It'll make its memories a little bit better and it will uh evolve over time if you do, but you don't have to.
• 29:30
  SPEAKER_00: And if you do, we'll be very transparent as to what information is being shared.
• 29:35
  SPEAKER_00: Let's make it not talk.
• 29:38
  SPEAKER_00: It'll make it'll make sounds, it'll it'll make pet-like noises, so you'll know if it's happy or sad or excited and so forth.
• 29:45
  SPEAKER_00: But we just take off the table any idea of this thing giving you advice.
• 29:49
  SPEAKER_00: Uh because you think about your pet can communicate really, really well and it doesn't talk.
• 29:55
  SPEAKER_01: So this is that chat GPT problem that you're getting at before people using it as a therapist.
• 30:00
  SPEAKER_01: Yeah.
• 30:01
  SPEAKER_00: Use of human language is not necessary for a wide variety of home applications.
• 30:10
  SPEAKER_00: And then uh let's do this all at a similar price point to pet ownership.
• 30:17
  SPEAKER_00: And uh suddenly this this thing that we're talking about, this familiar, um feels kind of cool.
• 30:30
  SPEAKER_00: And uh, you know, why wouldn't you want this?
• 30:34
  SPEAKER_01: I I wanted to walk through some kind of I guess through some some lessons learned through your time at irobot and and how that informed this company and this I know this isn't technically a product announcement, uh, but you know, the the the the the kind of the portfolio that you're working on.
• 30:54
  SPEAKER_01: Uh the the first the the first and most obvious one, you know, when we've been having this conversation for 20 years of of what's what's after the robot vacuum, right?
• 31:04
  SPEAKER_01: And and and you know by and large for iRobot, it was the question was, okay, what's the next value prop that we can have?
• 31:13
  SPEAKER_01: What's the next thing in the home that we can do?
• 31:16
  SPEAKER_01: Cleaning pools, you know, cleaning gutters, um mowing the lawn, for example.
• 31:21
  SPEAKER_01: Uh this is a very this is a different approach, right?
• 31:24
  SPEAKER_01: I mean you're you're talking about a sort of a different, at least to start with, a different theory of what constitutes value.
• 31:31
  SPEAKER_00: Not really.
• 31:32
  SPEAKER_01: No.
• 31:33
  SPEAKER_00: Meaning that when you go and you look at a product for the home, uh, you're competing for wallet share, right?
• 31:41
  SPEAKER_00: A family has only so much money it's willing to spend, and it chooses um uh very uh deliberately how it spends money.
• 31:52
  SPEAKER_00: How much do you spend on your cable bill?
• 31:55
  SPEAKER_00: How much do you spend on your large screen TV?
• 31:58
  SPEAKER_00: How much do you spend on vacuuming and mopping?
• 32:01
  SPEAKER_00: Um and the challenge at iRobot was our logical expansions, you know, went from vacuuming to mopping to air purification to pool cleaning, and and you know, there was adjacencies, but each of those adjacencies were uh you know cast in the uh you know, against the what does the average uh household spend on these different things?
• 32:38
  SPEAKER_00: And so a robot vacuum is a problem that people do frequently, they dislike doing it and they spend a lot of money on it.
• 32:49
  SPEAKER_00: So boom, boom, boom, you got three wins.
• 32:52
  SPEAKER_00: Mopping, people hate doing it so much they actually don't do it very often, which works against you.
• 33:01
  SPEAKER_00: And they do it with a mop that they're used to spending, you know, eight dollars on, and so that well, buying a robot vacuum you know, it's almost price parity versus a really, you know, a nice robot vacuum and a nice regular vacuum are similarly priced.
• 33:20
  SPEAKER_00: A robot mop versus a manual mop has a huge price discrepancy between those two things, and that's why ultimately an independent scuba robot failed, but it was a nice add-on to your your vacuum.
• 33:42
  SPEAKER_00: Now, people spend a lot on pets.
• 33:47
  SPEAKER_00: There is an epidemic level crisis uh on isolation, feelings of isolation and loneliness among adults in the United States, and it's even more severe outside of the United States in some regions.
• 34:02
  SPEAKER_00: The pet industry is taking off uh or uh growing very rapidly.
• 34:09
  SPEAKER_00: And so that there is both need and significant spend around trying to solve challenges of isolation and loneliness and uh looking at wellness and looking at um uh you know parents saying, for the love of God, uh I feel so guilty giving my child an iPad to you know watch all afternoon long, but I don't have another choice.
• 34:38
  SPEAKER_00: And so that there's these very concrete needs that a technology that can now potentially exist, uh, if it's well executed, if it's properly priced, and if it's amazing, could actually be even more attractive than Roomba because of the acuteness of the need that many households face.
• 35:02
  SPEAKER_00: And so this isn't, gee, Colin, you used to be very pragmatic, and now you're doing this weird soft stuff.
• 35:10
  SPEAKER_00: Uh, it's this is the greatest application of robots that in my entire career I could imagine.
• 35:21
  SPEAKER_00: I just couldn't build it until honestly six months ago.
• 35:26
  SPEAKER_00: In fact, the company was started uh when the technology didn't yet exist to do what we're doing now.
• 35:34
  SPEAKER_00: But it felt like maybe it did, or maybe it would soon, and we're actually every day growing into a world where building familiars is possible.
• 35:46
  SPEAKER_01: What does it mean, you know, in those early days for the company to exist prior to the technology that's really going to uh you know unlock it?
• 35:56
  SPEAKER_01: Obviously, it has to sort of exist in a in a different form than it is right now or it is product focused.
• 36:02
  SPEAKER_00: Well, it's you know, you have the idea and you, you know, I think coming out of iRobot, I realized that I shouldn't be afraid of having a lot of degrees of freedom.
• 36:16
  SPEAKER_01: Aaron Powell That's a very roboticist of you, Colin.
• 36:18
  SPEAKER_00: Yeah, yeah, yeah, yeah.
• 36:19
  SPEAKER_00: I mean, that's a totally geeky and nerdy thing to say.
• 36:22
  SPEAKER_00: But, you know, it's uh if every motor you add to your robot costs you$250,000 if you want it to be dynamically controlled, you're gonna build a robot with four motors.
• 36:38
  SPEAKER_00: You know, it's it's just not going to be very complicated.
• 36:41
  SPEAKER_00: But when you start saying, well, maybe it's a couple tens of dollars, uh suddenly it's like, well, to build a quadruped, you need 12 motors, uh, three for each three degrees or for each leg if you don't want to cheat, and we don't want to cheat.
• 37:00
  SPEAKER_00: Okay, that's not so bad.
• 37:02
  SPEAKER_00: We can do that.
• 37:03
  SPEAKER_00: And and so that the company sort of started getting, you know, getting getting off the ground with this idea of what we wanted to build, and then starting to sort of do the math.
• 37:14
  SPEAKER_00: Could we build this thing that isn't fake?
• 37:18
  SPEAKER_00: It's not an illusion, it's real, it is dynamically controlled, it could move in a graceful, lifelike fashion, it could carry enough compute on board to uh actually run AI models, uh, you know, and have enough budget for stereo sensing and and array microphones, and uh and you you say, wait, that that could that could actually work.
• 37:46
  SPEAKER_00: And so we had confidence that um, you know, because my team included some folks who helped set up the logistics and operations that iRobot benefited from, we we started off with a really amazing reality around uh uh you know, target for hardware costs that we convince ourselves was achievable.
• 38:09
  SPEAKER_00: And then I think that where I say, gee, it wasn't actually possible until six months ago.
• 38:14
  SPEAKER_00: What I'm really talking about is can we put a full stack AI system, including reinforcement learning, including a behavior engine, including a reasoning engine, including a perceptual engine, any memory system on the edge, you know, yeah, with an embedded solution.
• 38:34
  SPEAKER_00: And the answer was well, maybe we can on day one of the company, and now we're sitting around having working prototypes of it uh uh you know functioning on the edge.
• 38:47
  SPEAKER_00: And that wasn't actually possible when we started, but is possible today and will continue to improve as the weeks pass, meaning the world is changing that rapidly.
• 39:03
  SPEAKER_01: But so there was a sense that you know that like, okay, we can't really do this until this thing is unlocked, and they had to anticipate that those things were coming, you know, in a certain timeline.
• 39:14
  SPEAKER_00: Well, I mean, we kind of did the official entrepreneur thing where we benefited from uh a little bit of wishful thinking.
• 39:22
  SPEAKER_01: Yeah.
• 39:22
  SPEAKER_00: You know, necessity is the mother of invention, and we uh sort of hooked on to the AI bus at the right time, and the AI bus has only accelerated, not slowed down.
• 39:35
  SPEAKER_00: And so that it was a uh it wasn't that we knew it didn't exist, we knew that it might exist and then benefited from progress.
• 39:49
  SPEAKER_01: Yeah, so and so and then the other the other iRobot learning that I wanted to talk about, and and I think you were kind of in a roundabout way getting at it is the privacy question, because that is effectively why it's really important to run this thing on on edge.
• 40:04
  SPEAKER_01: And you know, that was something that like I suspect that you kind of had to learn a few times over the years with iRobot, like as the sensors were becoming more and more sophisticated on the systems.
• 40:15
  SPEAKER_00: Right.
• 40:16
  SPEAKER_00: And you know, the the cloud was always promised to be infinite-free compute, and yet compute in the cloud is neither free nor infinite.
• 40:26
  SPEAKER_00: And so that if you're doing humanoids for industrial settings, the cloud is a great resource.
• 40:33
  SPEAKER_00: But if you're in the home, just having a solid Wi-Fi connection is the exception, not the rule.
• 40:42
  SPEAKER_00: So latency is going to be a problem.
• 40:45
  SPEAKER_00: If you're trying to run a real-time system over the cloud, it's completely unaffordable from the perspective of a consumer.
• 40:54
  SPEAKER_00: So cost is a problem, and then streaming video and audio up to the cloud is a privacy non-starter.
• 41:02
  SPEAKER_00: Uh, you know, and so that uh people don't want that either.
• 41:07
  SPEAKER_00: So, you know, latency, privacy, cost issues, um all drive you to having to have an edge solution.
• 41:17
  SPEAKER_01: This is something that I've never asked you, and and and it's just occurring to me now as as we're talking.
• 41:22
  SPEAKER_01: Um throughout the you know, your very long time working at iRobot and working on the Roomba Um Was there ever a conversation or was it ever pitched to you to actually give Roomba a voice?
• 41:39
  SPEAKER_00: You know, um very controversial.
• 41:43
  SPEAKER_00: You know, uh the um we gave it its beeps and boops back at the beginning because uh there weren't any good affordable speech chips, and so uh making it more R2D2 um was sort of the only option given the tools that we had.
• 42:03
  SPEAKER_00: The um we ultimately gave it a utility voice to give error messages so that it would, you know, your Roomba will say, you know, Wi-Fi not connected, or you know, please check your your rollers, they're they're clogged.
• 42:22
  SPEAKER_00: Because doing that by beeping just was a miserable consumer experience, but we kept the the use of that voice to be uh used in a way that it was clear it wasn't Roomba talking to you, it was something, uh you know, a maintenance system or or something.
• 42:45
  SPEAKER_00: So um, and that um that made it easier um because people did personify Roomba, but every Roomba owner sort of has a different mental image as to what their Roomba is, and giving it a voice would probably work very rapidly against us because you'd start assuming that if you talk to Roomba, well, it'll understand.
• 43:13
  SPEAKER_00: And yeah, even today, natural language-based conversations that uh are meant to be coherent and consistent over long periods of time are extraordinarily difficult to do.
• 43:28
  SPEAKER_00: And back, you know, even just two, three years ago, it it wasn't even close, so it wasn't considered.
• 43:34
  SPEAKER_00: And I don't think the benefits even today with today's technology make sense for Rumba to have a real personality.
• 43:43
  SPEAKER_01: Yeah, the one uh uh one of the sort of big selling points um with the familiar is this idea of it uh you know adapting over time, adapting to your your your schedules and your wants and needs.
• 43:57
  SPEAKER_01: And I'm I'm wondering if part of that is.
• 44:00
  SPEAKER_01: The ability to um have each one of these develop, in a certain sense, their own individual personality from one familiar to another.
• 44:14
  SPEAKER_00: Absolutely.
• 44:15
  SPEAKER_00: I mean, there will be no two familiars that are the same.
• 44:19
  SPEAKER_00: Um, these are learning and adapting uh machines, both from you know working to make sure that everyone looks a little different to uh different personalities, to different learnings.
• 44:35
  SPEAKER_00: Every familiar has some unique connection between the familiar's owner, which we we call the owner the guardian, and the familiar itself.
• 44:45
  SPEAKER_00: So you see the robot in the first seconds, you're like, oh my god, that's the coolest thing I've ever seen.
• 44:50
  SPEAKER_00: And then within a minute, you see that it is behaving in a way that is purposeful, that is illustrative of a behavior.
• 44:59
  SPEAKER_00: And then our mission is once we get home, what is our role?
• 45:05
  SPEAKER_00: Where can we help out?
• 45:07
  SPEAKER_00: Are we greeting you when you come home?
• 45:10
  SPEAKER_00: Are we uh uh watching television with you after dinner?
• 45:14
  SPEAKER_00: Are we hanging out with you while you make breakfast?
• 45:17
  SPEAKER_00: The uh establishment of routine is uh a critical part of the family's learning.
• 45:25
  SPEAKER_00: And and um, you know, for the analogy to roomba was if you scheduled your roomba, you were super happy and you had years of of engagement.
• 45:36
  SPEAKER_00: If you didn't schedule your roomba, after a few months your utilization went way, way, way down, and maybe you were in a closet.
• 45:43
  SPEAKER_00: And so that this idea of uh within the first few days, do you have a place?
• 45:50
  SPEAKER_00: And then long term, what does the familiar do for you and what do you do for it?
• 45:59
  SPEAKER_00: This idea of bi-directional caring that evolves, and you will be able to teach your familiar new tricks, you will be able to learn new things from your familiar, and uh that relationship will grow and become deeper over time.
• 46:15
  SPEAKER_00: And by getting each of these four time periods right, we believe we can create a little bit of happiness in people's lives that give them some helpful nudges to reinforce a positive routine, um, be this hyper-loyal presence in your world, and you know, with an affordable cost of ownership.
• 46:43
  SPEAKER_00: And and you know, the uh the other wonderful lesson that we took away from my robot is how do we test all these hypotheses and how we were able to bring some of the um the thought leaders around consumer testing to make sure that these philosophies that we uh have talked about are actually real, such that this next robot um is never described as soft in its benefit, but uh instead this is a bit of joy that I would uh that I care for and is one of my favorite purchases.
• 47:31
  SPEAKER_01: So we we're we're just about out of time, so we can close on this, but I'm wondering if to a certain extent you feel like there's going to be a bit of an uphill battle because of all of the generations of attempts at home robots and home pets, or there are these like beloved, you know, we mentioned Ibo, uh, you know, there's there's there's PLIO and and Jibo, and these are robots that like people loved, but for your you know, uh whatever reasons, like, you know, didn't didn't didn't last.
• 47:58
  SPEAKER_01: And I people do have expectations about what robots can and can't do in the home.
• 48:03
  SPEAKER_01: So, you know, how are you really going to express to people that this is something different?
• 48:10
  SPEAKER_00: Well, I think that ultimately people will try and realize that this is different.
• 48:17
  SPEAKER_00: I mean, this is not a watch me toy.
• 48:20
  SPEAKER_00: This is something that if you're not motivated to touch and pet it, then boy, we did something wrong.
• 48:31
  SPEAKER_00: This is something that doesn't live in a four by four box in front of you.
• 48:35
  SPEAKER_00: This is something that can take you for a walk outside.
• 48:42
  SPEAKER_00: There's never been a consumer robot that was able to do that.
• 48:45
  SPEAKER_00: It this actually lives in your space.
• 48:49
  SPEAKER_00: You know, you get up and you go to the kitchen, it can follow you and hang out with you.
• 48:55
  SPEAKER_00: No robot has ever been able to do that before.
• 48:58
  SPEAKER_00: Right?
• 48:58
  SPEAKER_00: So uh what we're doing is so miles beyond what has gone before.
• 49:09
  SPEAKER_00: You know, I appreciate, uh, and I think it's unavoidable for people to initially think, gee, why is this not like that?
• 49:19
  SPEAKER_00: But the reality of what we have created is so different that um it's not even a useful comparison.
• 49:32
  SPEAKER_00: And just like Roomba, you know, I no one believed Roomba was gonna work until people tried Roomba.
• 49:38
  SPEAKER_00: Uh, I think that uh this is going to be a new product that word of mouth that uh reviewers um bring home, and their voice is the voice that matters, and we're very confident just in the experience that we're creating.
• 50:03
  SPEAKER_01: Well, we are out of time.
• 50:05
  SPEAKER_01: Colin, always a pleasure.
• 50:07
  SPEAKER_01: Thank you so much.
• 50:08
  SPEAKER_00: My pleasure.
• 50:10
  SPEAKER_01: Thanks to Colin.
• 50:11
  SPEAKER_01: Thanks to Dan for setting up the conversation.
• 50:13
  SPEAKER_01: I will be bothering you very soon to test drive one of those new robots.
• 50:18
  SPEAKER_01: Uh thanks as always for tuning in.
• 50:20
  SPEAKER_01: Please like and subscribe and rate the show.
• 50:23
  SPEAKER_01: Tell some friends, conveniently forget to mention it to your enemies.
• 50:27
  SPEAKER_01: Check out the automated newsletter that hits inboxes on Thursday and LinkedIn on Friday.
• 50:34
  SPEAKER_01: And with that, we will see you next week with another episode of Automated.