Latest / What's Your Problem? / Building a Robot People Actually Want

Transcript

Building a Robot People Actually Want · What's Your Problem?

0:15
Speaker 1: Pushkin. If I close my eyes and picture a robot,
0:22
Speaker 1: I see basically a metal guy arms, legs ahead, maybe
0:28
Speaker 1: a couple of lights for eyes. And there's a reason
0:31
Speaker 1: I see this. This is the way people have been
0:34
Speaker 1: dreaming of robots for something like a hundred years now,
0:38
Speaker 1: and today people are doing more than dreaming. They're spending
0:41
Speaker 1: billions of dollars to build robots like this, to build humanoids,
0:46
Speaker 1: to build robots that look like people. But it's really
0:50
Speaker 1: hard to build a robot that looks like a guy.
0:53
Speaker 1: The physics are complicated. You need a lot of moving parts,
0:57
Speaker 1: which is why despite the amazing demos we keep seeing
1:01
Speaker 1: on social media, and despite all the smart people and
1:03
Speaker 1: all the money trying to make robots happen, almost nobody
1:07
Speaker 1: has a humanoid robot at their house. Maybe another approach
1:11
Speaker 1: would make more sense. Maybe instead of starting with the
1:14
Speaker 1: human body and saying how can we make a robot
1:17
Speaker 1: that looks like this, we should ask how can we
1:20
Speaker 1: build an affordable robot to help people solve real problems
1:24
Speaker 1: at home? Right now, I'm Jacob Goldstein, this is what's
1:40
Speaker 1: your problem? And my guest today is Aaron Edzinger. Aaron
1:43
Speaker 1: has been building robots for decades. A while back, he
1:46
Speaker 1: sold a company to Google, then he became the head
1:49
Speaker 1: of robotics at Google, and then he left to start
1:52
Speaker 1: a company called Hello Robot, where he is now the CEO.
1:57
Speaker 1: Aaron's problem is this, how do you build a robot
2:00
Speaker 1: that is useful and affordable for at least some people
2:04
Speaker 1: to buy and use at home. Hello Robot hasn't really
2:07
Speaker 1: solved that problem yet, but they're working on it. They
2:11
Speaker 1: have a robot called Stretch three. They've sold hundreds of them,
2:15
Speaker 1: and the robots are proving truly useful to some people
2:19
Speaker 1: in very difficult circumstances. Also, the robots don't look anything
2:24
Speaker 1: like metal people. They're not humanoids. In our conversation, Aaron
2:28
Speaker 1: and I talk about why home robots have barely progressed
2:32
Speaker 1: since the roomba, what it actually takes to get a
2:34
Speaker 1: robot to feed someone, and why the physical world is
2:38
Speaker 1: so much harder for AI than language. One of the
2:42
Speaker 1: interesting things to me about your approach about Stretch three,
2:45
Speaker 1: the robot that you have in the world now is, frankly,
2:48
Speaker 1: the things that it's not right. It's not autonomous, it's
2:51
Speaker 1: not a humanoid, it's a robot. Arm that's operated by
2:55
Speaker 1: remote control. And you know you've been in the field
2:58
Speaker 1: for a long time. You were the director of robotics
3:00
Speaker 1: at Google. So why did you leave Google to build this?
3:06
Speaker 2: Oh it's it yeah, complicated answer, it's a good question.
3:11
Speaker 2: In the end. You know, with envisioned robots being something
3:15
Speaker 2: different than what they were becoming at that time, one
3:20
Speaker 2: part of that is just about simplicity, right, So you know,
3:25
Speaker 2: more and more the robots that we're building at Google
3:29
Speaker 2: were becoming more complex, and you see that with the
3:32
Speaker 2: humanoids today. There's this push towards kind of human level
3:36
Speaker 2: dexterity capability. And our premise was, you know what, you
3:40
Speaker 2: could do a lot today if you push on simplifying
3:44
Speaker 2: the design, the approach. So the whole premise with Heller
3:48
Speaker 2: robot is how do we practically deliver a value with
3:52
Speaker 2: a robot to help people? And forget about should it
3:55
Speaker 2: look like a human Let's really focus on just that
3:58
Speaker 2: core question. And from that came to Toutch three, which
4:01
Speaker 2: is a very minimalist robot. We like to think of
4:05
Speaker 2: it almost like robotic cubism, right, So if you take
4:08
Speaker 2: a human form and decompose it, put it back together,
4:11
Speaker 2: in some other way. If you squint, it's recognizable as
4:15
Speaker 2: a human. Maybe it can do eighty percent of what
4:18
Speaker 2: a human or a humanoid could do. But it's much
4:20
Speaker 2: more practical, deployable from a business perspective, but also just
4:25
Speaker 2: from a usability and also autonomy. So the robot is autonomous.
4:29
Speaker 2: We don't highlight a lot of that, but the autonomy
4:33
Speaker 2: gets easier too when the robot gets simpler. So that's
4:37
Speaker 2: the basic premise of starting the company.
4:40
Speaker 1: Yeah, I mean it's interesting as you were saying that,
4:42
Speaker 1: I was thinking of sort of the classic software thing
4:45
Speaker 1: is the minimum viable product, right, Like, what's the simplest
4:48
Speaker 1: thing we can get out into the world and then
4:50
Speaker 1: learn what's going on with it, iterate, etc. It's like
4:54
Speaker 1: you're building the minimum viable robot exactly exactly, So you
4:59
Speaker 1: have built a robot and it is in the world.
5:03
Speaker 1: The current version is called Stretch three. Tell me about
5:06
Speaker 1: Stretch three, so.
5:08
Speaker 2: S Touch three a third generation. It is fundamentally think
5:12
Speaker 2: of it like a roomba with a big mast on it, right.
5:17
Speaker 2: Comes from that perspective, right, and then you add to
5:20
Speaker 2: that mast a telescoping arm, you know, telescoping the way
5:24
Speaker 2: a car Antenta might telescope a turned sideways and you
5:27
Speaker 2: put on the end of that telescoping arm a hand, so.
5:31
Speaker 1: It's basically an arm. There's a room b with a
5:33
Speaker 1: vertical pole going up what three feet or something about
5:36
Speaker 1: four feet yep, four feet, and then basically a robot
5:40
Speaker 1: arm what we think of as a robot arm on
5:41
Speaker 1: top of that pole. Exactly a room but that can
5:43
Speaker 1: drive a robot arm around.
5:44
Speaker 2: Your house, drive around, and it can reach up to
5:47
Speaker 2: the countertop underneath the couch. That type of thing. Very
5:51
Speaker 2: very simple design, and what we saw early on is
5:54
Speaker 2: just by remote controlling it, you can do quite a
5:57
Speaker 2: few things around a house that you care about. And
5:59
Speaker 2: so our first launch video showed it picking up toys
6:03
Speaker 2: tele operated, but picking up toys, wiping down countertops, putting
6:07
Speaker 2: laundry in a laundry machine with a very simple device.
6:11
Speaker 2: You know, I just come from Google, where we're building
6:13
Speaker 2: a very complex device and to do the same type
6:16
Speaker 2: of thing. And I just thought, you know, if these
6:18
Speaker 2: robots are ever really going to get out to the
6:20
Speaker 2: world as real products, it's going to be and look
6:23
Speaker 2: and feel more like this.
6:25
Speaker 1: How much does it cost?
6:27
Speaker 2: Right now? It's twenty about twenty five thousand dollars. And
6:31
Speaker 2: that's that's because we run the company off of sales.
6:34
Speaker 2: We're not it's trying to you know, it's low volume.
6:37
Speaker 1: So does that mean your your marginal cost is about
6:40
Speaker 1: twenty five thousand dollars, you're selling it for about what
6:42
Speaker 1: it costs.
6:43
Speaker 2: No, so the opposite. So what I'm saying is it's
6:45
Speaker 2: actually can be very low cost, but because we're funding
6:49
Speaker 2: R and D and engineering and production.
6:51
Speaker 1: You have a margin on it.
6:53
Speaker 2: We have a real, operational, sustainable business selling robots today.
6:57
Speaker 1: So you're a real company.
6:58
Speaker 2: We are a real company selling robots to people. Yes,
7:02
Speaker 2: And that was part of the strategy from the get gos.
7:05
Speaker 2: You know, I've been to robotics for twenty five years
7:08
Speaker 2: now and seen a lot of companies over promise what's
7:13
Speaker 2: possible with the technology, and then you know, their capital,
7:19
Speaker 2: their investors getting patient, and they end up closing up shop.
7:22
Speaker 2: And we see this time and time again.
7:24
Speaker 1: We're selling to Google and a happier version.
7:27
Speaker 2: Yeah, well you could do that too, which yeah, but
7:30
Speaker 2: you know, really we wanted to have a viable business,
7:33
Speaker 2: built a product that people liked and loved, and then
7:37
Speaker 2: from there then you have permission to take the time
7:40
Speaker 2: that it needs to really to do it correctly.
7:43
Speaker 1: How many have you sold?
7:45
Speaker 2: Yes, we have about three hundred robots in the world
7:49
Speaker 2: across twenty three countries, and about forty five to different institutions,
7:54
Speaker 2: from top tier research labs to small startups, to people
7:58
Speaker 2: looking at vertical farming, to people looking at healthcare, you know,
8:01
Speaker 2: a whole range of things.
8:04
Speaker 1: I saw a video about a man named Henry Vince
8:09
Speaker 1: who uses one of your robots, and I found it
8:13
Speaker 1: quite compelling. Yeah, tell me about Henry Evans, right.
8:18
Speaker 2: Henry and his care partner and wife, Jane. You know,
8:22
Speaker 2: they're in California, not too far from us. They've collaborated
8:26
Speaker 2: with my co founder, Charlie camp for at least I
8:30
Speaker 2: think fifteen years now, so it's been a long relationship.
8:33
Speaker 2: Henry is quadriplegic from a stroke. Jane is the primary caregiver.
8:39
Speaker 2: But Henry is, you know, an early adopter. He's really
8:42
Speaker 2: excited to find ways that technology can give him agency,
8:47
Speaker 2: not just that, but make life easier on chain and
8:51
Speaker 2: so you know, he's been looking at robotics as a
8:54
Speaker 2: solution partial solution for that for quite some time. We've
8:57
Speaker 2: collaborated with him, and really when we found it, Hello robot,
9:01
Speaker 2: you know, he was one of the key well i'd
9:05
Speaker 2: say inspirations and users end users that we had in mind.
9:09
Speaker 2: So through some some work that was funded by the NIH,
9:11
Speaker 2: we've spent it last three years working with him and
9:14
Speaker 2: some other people in similar situations looking at how Stretch
9:19
Speaker 2: can can benefit him. And so, you know, Henry isn't
9:22
Speaker 2: able to communicate, to talk except through I gays, so
9:26
Speaker 2: he uses I gaze on a letter board to spill
9:29
Speaker 2: out words and then he can also use I gays
9:32
Speaker 2: to control a computer mouse, and through that he's able
9:36
Speaker 2: to control Stretch to do simple tasks, things that might
9:39
Speaker 2: seem you know, trivial, are you not so exciting to
9:43
Speaker 2: other people, but for him, it's it's it's it's really
9:46
Speaker 2: life changing. And so a really really nice story with
9:49
Speaker 2: with Henry is, you know, we came in thinking like, well,
9:54
Speaker 2: you know, Stretch will do very practical tasks very quickly.
9:58
Speaker 2: What we learned is there's a whole range of things
10:01
Speaker 2: that are much more social, emotional about connection that we
10:05
Speaker 2: just didn't anticipate with the device like this. And so
10:08
Speaker 2: one example is his granddaughter at the time was probably
10:11
Speaker 2: about three or four. She was a little bit scared
10:16
Speaker 2: of him because you know, he's immobile, he's in a
10:20
Speaker 2: powerchair in bed, can't talk, and so they didn't really
10:23
Speaker 2: have a relationship. And what we found is through Stretch,
10:27
Speaker 2: because he could control Stretch, he could start to play
10:29
Speaker 2: with her, interact with her, and in the physical world.
10:33
Speaker 2: She would decorate stretch, put stickers on it. They would
10:36
Speaker 2: have a tablet on it, so you know, his face
10:39
Speaker 2: would appear, and soon they started to play. So they'd
10:44
Speaker 2: played basketball, pick up, a little nerve, basketball, play different games,
10:49
Speaker 2: and you know, she started to really kind of see
10:52
Speaker 2: him in a different light and former relationship with him,
10:55
Speaker 2: and so now she calls him Papa Wheeli because he's
10:58
Speaker 2: in his chair and and she, you know, can be
11:01
Speaker 2: excited to see him. So it really bridge that gap
11:05
Speaker 2: between the two of them, and and the type of
11:08
Speaker 2: thing that with robot that there's so much potential and
11:12
Speaker 2: I think we really don't quite know how far we
11:17
Speaker 2: can go on the sort of the social side of it.
11:19
Speaker 2: You know, really this can be this empathetic technology that
11:23
Speaker 2: brings so much to people, and that's that's kind of
11:25
Speaker 2: what we're excited to explore.
11:29
Speaker 1: As you were talking about the robot playing with a
11:32
Speaker 1: you know, if you're a young child, the question of
11:35
Speaker 1: safety crossed my mind, Like what do you have to
11:38
Speaker 1: worry about in that setting? How do you deal with it?
11:41
Speaker 2: Like?
11:41
Speaker 1: How much is the robot weigh? Yeah, it's a great question.
11:44
Speaker 2: So this robot weighs about fifty pounds, a little over
11:47
Speaker 2: fifty fifty pounds most that weighs down in the base, right,
11:52
Speaker 2: And it's really it's an important question because there's so
11:56
Speaker 2: much happening in robotics right now where we're seeing humanoids
12:00
Speaker 2: in the home doing dishes right, and these are scary machines.
12:05
Speaker 2: And really the fundamental thing you have to think about
12:07
Speaker 2: is just that physics of it. Right, when you have
12:12
Speaker 2: a humanoid that's balancing, particularly a human scale one, the
12:16
Speaker 2: amount of weight that's up high, the potential energy that's
12:19
Speaker 2: there is significant, and when things go.
12:23
Speaker 1: Wrong, if it falls over, you could smush things. That's fundamentally,
12:27
Speaker 1: that's fundamentally it.
12:28
Speaker 2: And so just the pure physics physics of that are
12:31
Speaker 2: fundamentally you know, we can't cheat it, right, you can't,
12:35
Speaker 2: and things things will fail. So really, with from the
12:38
Speaker 2: get go for for hello, robot and stretch. We thought
12:41
Speaker 2: about safety from just the physical properties of it. How
12:46
Speaker 2: do we make it so it's very hard for this
12:49
Speaker 2: device to cause harm? So first off, it should be
12:52
Speaker 2: light weight, that potential energy, the mass should be down
12:57
Speaker 2: low so that if if it falls over, it's not
12:59
Speaker 2: gonna you know, smash and break something. And then you
13:03
Speaker 2: think about the robot joints themselves.
13:06
Speaker 1: And is there is there a pincer We haven't actually
13:08
Speaker 1: talked about the sort of yes, gripper is it? Is
13:10
Speaker 1: it like a two piece pincer.
13:12
Speaker 2: It's a it's basically like a grabber that you buy
13:14
Speaker 2: an Amazon, you know, like to pick up trash. Yeah,
13:17
Speaker 2: it's one of those that we've attached to motor to essentially.
13:21
Speaker 2: But if you think about, you know, the robot kind
13:24
Speaker 2: of motors. If if you have to hold your arm
13:27
Speaker 2: out against gravity all day long, you know you're gonna
13:31
Speaker 2: need a big motor there, right, because that's a lot
13:33
Speaker 2: of energy to support that. If you don't, if you
13:37
Speaker 2: can design the robot in a minimal way where you're
13:39
Speaker 2: not spending all of your time fighting energy or first off,
13:43
Speaker 2: you're more energy efficient. Second you have a smaller motor
13:46
Speaker 2: which will be less dangerous. And so the arm moves
13:50
Speaker 2: in a way, it is designed in a way that
13:51
Speaker 2: it's n have to fight gravity, and that just makes
13:53
Speaker 2: it inherently safer. Right, And so when you're working around
13:58
Speaker 2: older adults, people with disabilities, kids, pets, these things really
14:03
Speaker 2: really matter. And so that's that's just the starting point
14:06
Speaker 2: from the design. The intrinsic safety has to be there.
14:10
Speaker 2: The physics of it has has to be right.
14:12
Speaker 1: So people with quadriplegia, people with extremely limited mobility are
14:18
Speaker 1: an obvious early use case. What else, like, what else
14:23
Speaker 1: are people using this for? What else do you have
14:25
Speaker 1: in mind for early use case?
14:26
Speaker 2: Yeah, one of the most interesting use cases is in
14:30
Speaker 2: the home around caregiving. It's a very challenging one. It
14:34
Speaker 2: is further out, and so you know, there's a lot
14:36
Speaker 2: of use cases nearer term which I think will be
14:40
Speaker 2: in service, you know, back of retail logistics. Uh, And
14:45
Speaker 2: we're seeing you know, data centers, a lot of stuff
14:47
Speaker 2: like that that we are seeing interest in and people exploring.
14:50
Speaker 1: Like picking and pecking. I mean there are specialized robots
14:53
Speaker 1: for those things.
14:54
Speaker 2: Right. Well, yeah, there's some places where there's not yet, right, So,
14:59
Speaker 2: for example, micro fulfillment little fulfillment, you know, like even
15:02
Speaker 2: like a you know, stocking a seven to eleven. There's
15:05
Speaker 2: lots of things where a robot like this could play
15:08
Speaker 2: a role.
15:09
Speaker 1: Options And just to bit clear, my understanding is that
15:13
Speaker 1: your robot is mainly not autonomous, that it's mainly remote
15:17
Speaker 1: controlled tteley operated. Is that right? And what does that
15:19
Speaker 1: mean for that use case?
15:21
Speaker 2: Right? It can be autonomous. There are autonomous it's completely
15:25
Speaker 2: open source as a whole ecosystem of open source software
15:28
Speaker 2: people have built for it, including autonomy, and the generation
15:33
Speaker 2: of robot we're working on now will really be autonomy powered, right,
15:40
Speaker 2: so in terms of the sensing of the software. But
15:42
Speaker 2: you know, back to that sort of really minimal approach. Yeah,
15:46
Speaker 2: we're trying to say what can we do today that
15:48
Speaker 2: has value? And for someone like Henry Evans, he controls
15:53
Speaker 2: the robot, it has autonomy mixed in there, so he's
15:56
Speaker 2: not controlling every motion necessarily, and so what we see
16:01
Speaker 2: is as a steady progression from direct teleoperation to mixed
16:06
Speaker 2: autonomy to full autonomy, but doing that really deployed, right,
16:10
Speaker 2: So this is not something in a research lab. It's
16:13
Speaker 2: actually in the product. And you know, we also see
16:18
Speaker 2: a path from people with really severe mobility impairments like Henry.
16:23
Speaker 2: They're interesting because they have a high motivation to use
16:27
Speaker 2: the device, and it's been the time to control it, right,
16:30
Speaker 2: They've got time and motivation. And then you can layer
16:34
Speaker 2: in the economy over time, and as you do that,
16:37
Speaker 2: now you can broaden the set of people who can
16:40
Speaker 2: who benefit from it. I think from people like Henry.
16:45
Speaker 2: The next step in terms of in home care is
16:49
Speaker 2: a remote operation by a family member a care provider.
16:55
Speaker 2: You know, for example, I'm in New Orleans right now,
16:57
Speaker 2: my mother is back in California, and you know, I'd
17:00
Speaker 2: love to hop on and check in, and you know,
17:03
Speaker 2: of course you can zoom, but sometimes you need to
17:06
Speaker 2: go into the other room, open the cabinet door, and
17:09
Speaker 2: you know, find the medication or something like that. And
17:12
Speaker 2: so again there's a human in the loop, but they're
17:15
Speaker 2: not going to be controlling every motion. They're going to
17:17
Speaker 2: be clicking on a map and saying go to the bedroom, open,
17:21
Speaker 2: open the drawer, that type of thing. Yeah, And then
17:24
Speaker 2: from there, of course, you can bring in more autonomy,
17:27
Speaker 2: and suddenly you have in home autonomous care and eventually
17:32
Speaker 2: in home robots that are doing the laundry and dishes
17:34
Speaker 2: and everything that you know, people are excited about. But
17:38
Speaker 2: for us, again, we're taking this very pragmatic approach about
17:41
Speaker 2: how do we stair step, how do we kind of
17:43
Speaker 2: climb our way up to that future?
17:47
Speaker 1: Stairstep is an interesting choice of words for a robot
17:50
Speaker 1: on wheels. But you know, one of the things people
17:56
Speaker 1: always talk about in the context of robots in the
18:00
Speaker 1: last few years is getting the data to train the
18:04
Speaker 1: model for autonomy, right, And the obvious analog is large
18:08
Speaker 1: language models got so good because the Internet is full
18:12
Speaker 1: of text, right, and there is no Internet of the
18:15
Speaker 1: physical world. There is no giant data set that you
18:17
Speaker 1: can just train a model on so that robots can
18:22
Speaker 1: understand how physics works in the real world. Right, how
18:24
Speaker 1: are you dealing with that?
18:28
Speaker 2: You know, I would say, in terms of AI physical
18:32
Speaker 2: AI as it's called around robotics, we're taking, you know,
18:37
Speaker 2: an approach that's really about taking things that are mature
18:40
Speaker 2: and integrating them into our autonomy, versus what's called end
18:45
Speaker 2: to end learning, where you know, you might collect a
18:48
Speaker 2: huge data set, right, you might train the robot tele
18:51
Speaker 2: operated to open the dishwasher a thousand times, collect that
18:54
Speaker 2: data and train a deep learned model to do that task,
19:00
Speaker 2: and you see very exciting results from that, and I
19:02
Speaker 2: do think that will continue to progress, but in terms
19:06
Speaker 2: of being ready to be deployed, it'll be take some time.
19:10
Speaker 2: And the amount of data that's required beyond just a dishwasher, Well,
19:15
Speaker 2: now it's a different dishwasher. Now it's a dishwashering different
19:18
Speaker 2: lighting conditions. Oh there's something in front of it. That
19:21
Speaker 2: complexity gets so large so quickly.
19:23
Speaker 1: And it doesn't generalize, like there are not models where
19:26
Speaker 1: you can teach it one dishwasher and it can make
19:29
Speaker 1: inferences about other dishwashers.
19:31
Speaker 2: There's inklings of that happening, right, but we're not there yet,
19:34
Speaker 2: and resourcers are still really exploring what's possible. You know,
19:38
Speaker 2: some are saying it's going to learn from YouTube, some
19:41
Speaker 2: are saying it's going to be all on simulation. Some
19:44
Speaker 2: are our hiring teams of operators to collect that data. Right.
19:49
Speaker 2: Our view is, you know that that world is progressing
19:52
Speaker 2: quickly and it'll sort itself out soon. You know, we
19:58
Speaker 2: will be able to train the robot to open that dishwasher.
20:02
Speaker 2: I think what's more interesting is how do you get
20:04
Speaker 2: it to do you know the five or six important
20:07
Speaker 2: tasks you care about and actually do it in a
20:10
Speaker 2: robot that people feel is trusted in the home, it's
20:12
Speaker 2: safe in the home, is affordable, you know, all the
20:15
Speaker 2: things that actually to have a business.
20:17
Speaker 1: Yeah, so you're basically it's again the like the physical
20:21
Speaker 1: intelligence is the bright shiny object that you are not
20:24
Speaker 1: chasing so that you can build a robot that works exactly.
20:27
Speaker 2: And and you know, think about uh, you know, you know,
20:31
Speaker 2: is Apple building a big AI team? Well, no, it's
20:34
Speaker 2: partnering with the people who are doing the AI. And
20:37
Speaker 2: I think you know, for us, we're more on that
20:40
Speaker 2: side of the fence.
20:42
Speaker 1: Are some of those AI gains that you are sort
20:44
Speaker 1: of piggybacking on coming from autonomous vehicles? I mean when
20:48
Speaker 1: you talk about saying go to the other room and
20:51
Speaker 1: have it go there, I think of that, That's where
20:53
Speaker 1: my mind goes, right that those are the robots out
20:55
Speaker 1: in the world right now driving around San Francisco.
20:58
Speaker 2: Yeah, I mean it's slightly different in terms of the
21:02
Speaker 2: requirements the sensors, but it's certainly certainly related. I mean,
21:06
Speaker 2: we've all benefited from he's wrapping gains and sensing and
21:10
Speaker 2: compute and and you know even ll MS is they're
21:14
Speaker 2: being used to help their robots plan and think and replan. Right.
21:19
Speaker 2: So you know, when I was started in robotics twenty
21:22
Speaker 2: twenty years ago, I couldn't imagine being able to say
21:26
Speaker 2: to the robot, hey, go to the other room and
21:28
Speaker 2: find the blue mug that's next to the microwave. Right,
21:32
Speaker 2: Like the context of that was just like, how are
21:34
Speaker 2: we going to solve that? Are we going to write
21:36
Speaker 2: every rule into the robot and that's.
21:38
Speaker 1: Like commodified like those are That's that's the easy part
21:41
Speaker 1: now at some.
21:43
Speaker 2: Level, yes, I mean certainly there's more to be done.
21:47
Speaker 2: I'd say, you know, the real challenges today are going
21:49
Speaker 2: to be around desk to manipulation and being able to
21:54
Speaker 2: do things that are really in dynamic, you know, kind
21:58
Speaker 2: of home type environments.
22:00
Speaker 1: So when you when you talk about you know, dexter
22:03
Speaker 1: sort of fine motor skill type things, what are you
22:05
Speaker 1: thinking of? What do you want the robot to be
22:07
Speaker 1: able to do in the relatively near term that it
22:09
Speaker 1: can't do now?
22:11
Speaker 2: I think just broaden the range of tasks. Right. So
22:14
Speaker 2: one thing that's really exciting to this user group is feeding, right,
22:18
Speaker 2: I think feeding. Feeding is one of the things where
22:22
Speaker 2: right now, so your partner care partner has to come
22:25
Speaker 2: and feed you, and that there's a number of reasons
22:27
Speaker 2: why that's just they would prefer to do it themselves.
22:29
Speaker 2: Of course.
22:30
Speaker 1: The basic it's a basic function of adult life is
22:35
Speaker 1: you feed yourself.
22:36
Speaker 2: You feed yourself, and depending on someone changes that relationship
22:39
Speaker 2: with that person, right, And and so feeding is one
22:43
Speaker 2: that we've looked at a fair amount. It is challenging.
22:46
Speaker 2: There are safety considerations, so we're careful about what utensils
22:49
Speaker 2: and what type of food. But I think if it
22:53
Speaker 2: can feed, if you can feed yourself, if you can
22:58
Speaker 2: get yourself a glass of water, you can do basic
23:00
Speaker 2: you know, hygiene maintenance things for yourself. I think that's
23:03
Speaker 2: kind of the base line of regaining a sense of agency, right,
23:08
Speaker 2: And what do you have.
23:08
Speaker 1: To figure out to make feeding really work that you
23:11
Speaker 1: haven't figured out yet?
23:12
Speaker 2: Yeah, it's a mixture of we have. Food is kind
23:16
Speaker 2: of messy, mushy, right, it's not like a rigid object.
23:20
Speaker 1: Well, and it's it's super heterogeneous, right, Food is many
23:24
Speaker 1: different kinds of.
23:25
Speaker 2: Ways exactly, and so but there is you know, one
23:29
Speaker 2: thing we've seen is you know, the caregiver is willing
23:31
Speaker 2: to prepare the food, maybe put it in a special tray,
23:34
Speaker 2: certain types of food that we agree it can work with.
23:37
Speaker 2: So but of course you'd love that to get broader.
23:40
Speaker 2: But you know, food in itself is challenging. Then there's
23:43
Speaker 2: the safety challenges. But also, you know, a really interesting
23:46
Speaker 2: part of this, which kind of relates to you know,
23:48
Speaker 2: where AI is in general, is understanding the person, right,
23:54
Speaker 2: are they ready for another bite? You know, are they
23:58
Speaker 2: you know, what side would they like to be fed from.
24:01
Speaker 2: That's all these little things that are about that human
24:05
Speaker 2: interaction that make the product useful, exciting, acceptable in the home.
24:11
Speaker 2: And so much of what's happening in ai AI right
24:14
Speaker 2: now doesn't really get at the human side of the story.
24:18
Speaker 2: And I think that's one thing where we would love
24:21
Speaker 2: to see a advancements is really having better understanding of
24:26
Speaker 2: these sort of more subtle cues and interactions that the
24:29
Speaker 2: robot will need in order to be accepted.
24:35
Speaker 1: We'll be back in just a minute. So I'm curious
24:48
Speaker 1: because you have been working on robots for more than
24:53
Speaker 1: twenty years, right, I mean, so much has happened in software, yeah, right,
24:58
Speaker 1: and even in other kinds of hardware, Like you were
25:01
Speaker 1: working on robots before there was an iPhone, right, and
25:04
Speaker 1: so so much has changed, and it feels like, at
25:07
Speaker 1: least as a as a layperson, certainly the home like
25:10
Speaker 1: robots are not around the way iPhones are around. And
25:13
Speaker 1: even now large language models are around. And so has
25:17
Speaker 1: less happened with robots than you thought.
25:20
Speaker 2: I would say less in terms of deployed robots, right,
25:24
Speaker 2: Like the rumba is still one of the most successful robots,
25:27
Speaker 2: but there aren't many other examples like it, right, and
25:30
Speaker 2: so why is that?
25:32
Speaker 1: Yes?
25:32
Speaker 2: What?
25:33
Speaker 1: Like that's a great way to ask it, and that
25:34
Speaker 1: I mean, you worked with Rodney Brooks, right, one of
25:37
Speaker 1: the rumbook guys at MIT, and that's like the only
25:41
Speaker 1: robot anyone can name, and there are lots of failed
25:44
Speaker 1: companies that were rumba.
25:45
Speaker 2: Like right, yeah, yeah, and so yeah, So I did
25:49
Speaker 2: my PhD with rod Brooks at MIT and at the
25:52
Speaker 2: time he was spinning up a robot, inventing the rumba.
25:56
Speaker 2: And around the time I finished, he had just sold
25:58
Speaker 2: his million rumba and he's he inspired me to go
26:01
Speaker 2: off to my own company.
26:03
Speaker 1: I mean, it must have seemed like they're here home
26:05
Speaker 1: robots are here, like you would think at that moment,
26:08
Speaker 1: it's like, oh, hey, here we go.
26:10
Speaker 2: Yeah yeah, well but things seemed very hard, right, so
26:13
Speaker 2: it's like, okay, a vacuum cleaner that bounces around off walls,
26:18
Speaker 2: like we can do that, right, Yeah. And I think
26:21
Speaker 2: the insight from Rumba that I still carry is, you know,
26:25
Speaker 2: they thought really carefully about the price, the complexity of
26:29
Speaker 2: the device. They weren't trying to There was a competitor
26:32
Speaker 2: call the Electrial Lucks at the time that was about
26:34
Speaker 2: three thousand dollars had all sorts of cool features and
26:37
Speaker 2: computer and sensing. But with the Rumbo, what they did
26:41
Speaker 2: is they went to the mall and they asked people,
26:45
Speaker 2: how much can you spend on something without having to
26:47
Speaker 2: ask your significant other for permission?
26:49
Speaker 1: Oh?
26:50
Speaker 2: And the answer was about two hundred dollars.
26:53
Speaker 1: I was just talking to a friend who basically does
26:56
Speaker 1: a B to B business. That's business a business, and
26:59
Speaker 1: apparently that's a sales strategy in business as well. Like
27:01
Speaker 1: the person who you're selling to they have some spending
27:05
Speaker 1: authority that they can spend on their own. Yes, of course,
27:07
Speaker 1: it's about a thing times greater. Right, It's like they
27:10
Speaker 1: can spend one hundred thousand dollars or whatever without asking
27:13
Speaker 1: somebody else, So you really want to price it like that,
27:16
Speaker 1: otherwise it's much harder.
27:17
Speaker 2: So anyway, yeah, exactly, But you know, I think the
27:21
Speaker 2: room to success in part was they made a minimal
27:24
Speaker 2: product within that spending, you know, authority and kind of
27:29
Speaker 2: made it all work. And I think the reason we
27:31
Speaker 2: haven't seen a lot beyond that is it's hard. It's
27:34
Speaker 2: been hard traditionally to find the things that match all
27:37
Speaker 2: those constraints. That technology can be there, the use case
27:41
Speaker 2: can be there, right clearly in home care, there's so
27:44
Speaker 2: much need for help in the home, but can we
27:48
Speaker 2: do it at a price? And is it technology there?
27:50
Speaker 2: You know, five thousand dollars would be a lot of
27:52
Speaker 2: money for most families.
27:54
Speaker 1: Yes, although for a caregiver, right like, if you narrow
27:57
Speaker 1: it not to like doing the dishes.
27:59
Speaker 2: But yeah, I.
28:01
Speaker 1: Mean, you know, many people have been bankrupted by just
28:07
Speaker 1: needing to hire a simple caregiver, right Like, people could
28:11
Speaker 1: definitely figure out how to spend five thousand dollars on
28:13
Speaker 1: a robot that could significantly replace a caregiver. One would think,
28:18
Speaker 1: I guess that's your bet, that's your.
28:20
Speaker 2: That's partial partially our bet. Yeah, and you know, I
28:23
Speaker 2: think it's not replacing, it's it's augmenting, you know, I
28:25
Speaker 2: think think what makes sense, yes, because we're not going
28:28
Speaker 2: to do everything, and nor should we, you know, but
28:31
Speaker 2: there's just aren't enough people ours caregiving hours in the
28:35
Speaker 2: in the world today to take care of everyone who
28:37
Speaker 2: needs care. That's just the truth of it.
28:39
Speaker 1: Yeah, zooming out from from Hello robot, from your company
28:43
Speaker 1: to the field more generally, like what's the landscape look
28:46
Speaker 1: to you? How does the landscape look to you right now?
28:48
Speaker 2: Yeah, it's fascinating there. There's a lot of first off,
28:52
Speaker 2: a lot of capital flowing into it. Into two areas really,
28:56
Speaker 2: one is robot foundation models. You know, this is sort
28:58
Speaker 2: of deep learning the alms of robots. And then the
29:02
Speaker 2: other would be humanoids, right, and sometimes there's both, right,
29:06
Speaker 2: but there's a lot of excitement around that, and would
29:09
Speaker 2: say generally they both have a little ways to go
29:11
Speaker 2: before they get deployed and really applied.
29:15
Speaker 1: I mean the foundation model one, you know, an ll
29:19
Speaker 1: M for robots sounds good to me, It seems interesting
29:22
Speaker 1: to me. Tell me about what's happening there.
29:24
Speaker 2: Yeah, Well, it's first off, it's remarkable what you see
29:29
Speaker 2: the robots doing with these these models, right, so peeling
29:33
Speaker 2: an egg or tying shoelaces or folding laundry. And they're
29:38
Speaker 2: getting better, right, They're they're getting they're having long I think.
29:41
Speaker 1: The peeling an egg was teleoperated.
29:43
Speaker 2: Okay, fair enough, but they are doing things like that
29:45
Speaker 2: that you think, wow, I just could not imagine this.
29:50
Speaker 2: And and they're getting better, and that they're having longer
29:53
Speaker 2: memory so they can do stuff over a longer period
29:55
Speaker 2: of time. They're getting better at recovering from from issues.
30:00
Speaker 2: But also they haven't yet learned to generalize. Right, So
30:03
Speaker 2: if you think about what they're learning, that's such as
30:07
Speaker 2: we were saying, such richness to diversity in the environments
30:10
Speaker 2: they meant to operate, and they're learning on one kind
30:14
Speaker 2: of little island and the whole sea of data that
30:16
Speaker 2: they need access to to really be generally useful. And
30:21
Speaker 2: so that's kind of a real challenge that that field
30:23
Speaker 2: is trying to work through, is how do you get
30:26
Speaker 2: to that generality and that robustness to that diversity. And
30:30
Speaker 2: it's very much an unsolved problem. I think they're making progress,
30:34
Speaker 2: but it also means to me that this is not
30:36
Speaker 2: something that we're going to be deploying into a robot
30:40
Speaker 2: product anytime soon.
30:42
Speaker 1: You know, it's interesting because it seems like one of
30:44
Speaker 1: the frankly big surprises about large language models is how
30:50
Speaker 1: good they are at generalizing, right, how they often are
30:54
Speaker 1: much better than specialist models.
30:57
Speaker 2: Right.
30:57
Speaker 1: Just take a frontier model and throw anything at it
30:59
Speaker 1: and it will be really good. Not quite, but kind of.
31:02
Speaker 1: And so it's interesting to hear that that's not at
31:05
Speaker 1: all the case with robot foundation models. Why do you
31:09
Speaker 1: pose that.
31:09
Speaker 2: Is, Well, you know, I think one thing is just
31:12
Speaker 2: the amount of data that's available, Right, So if you
31:17
Speaker 2: think of a ll M, one simple way to think
31:20
Speaker 2: of it, it's going to predict the next word that
31:21
Speaker 2: you say, and it has the whole corpus of the
31:24
Speaker 2: Internet to make that prediction. Right, for a robot to
31:28
Speaker 2: predict the next thing that's going to happen when it's
31:30
Speaker 2: peeling the egg, there's not a lot of data out
31:32
Speaker 2: there for it to understand that. And some people are saying, well,
31:38
Speaker 2: I can do that in simulation, but very hard to
31:42
Speaker 2: imagine simulating a lot of that kind of complexity that's
31:46
Speaker 2: actually in the world, just.
31:47
Speaker 1: Because like the world is so diverse. What do you
31:49
Speaker 1: even simulate or how.
31:50
Speaker 2: Do you even simulate an egg being peeled? And so,
31:55
Speaker 2: you know, I think it's a challenge because robots are
31:59
Speaker 2: in the physical world.
32:00
Speaker 1: Is there something as I understand it, the basic model,
32:06
Speaker 1: the transformer model that powers all of the frontier elements
32:09
Speaker 1: today was developed based on the nature of language itself, right, Like,
32:16
Speaker 1: is there a thought that that's maybe just not the
32:18
Speaker 1: best fundamental model for understanding the physical world for training robots.
32:23
Speaker 1: Might it be the case that some very different approach
32:26
Speaker 1: ends up working much better.
32:28
Speaker 2: Yeah, I know less about that. I do know there
32:30
Speaker 2: is a large interest in world models today that can
32:33
Speaker 2: kind of infer the physics of the world. And I
32:37
Speaker 2: think one thing that that is beneficial is there is
32:41
Speaker 2: so much structure in the world, both from physics.
32:44
Speaker 1: Like laws of physics, laws of physics, yep, but also.
32:47
Speaker 2: In the way that we design our homes, our appliances
32:51
Speaker 2: are you know, there is a lot of structure to
32:53
Speaker 2: be gained and learned, and so you can imagine these
32:58
Speaker 2: models become very good at predicting what your home is
33:02
Speaker 2: going to look like because it's seen a lot of
33:03
Speaker 2: homes and they don't vary that much fundamentally, so so
33:07
Speaker 2: that there is there is an enough structure to learn
33:10
Speaker 2: from what that model and how that transpires. I don't know,
33:14
Speaker 2: but people are exploring that.
33:16
Speaker 1: I mean, one of the one of the obvious challenges
33:19
Speaker 1: is we're just making the robots come live in our world, right,
33:22
Speaker 1: And I could imagine, for a case like someone with quadriplegia,
33:26
Speaker 1: it might I could imagine a universe where it's like, well,
33:29
Speaker 1: if you set up your room this way, not you
33:32
Speaker 1: have to remodel your house, but you have to do
33:33
Speaker 1: some set of things, then our robot can do a lot.
33:36
Speaker 1: And I can imagine that being worthwhile in many settings.
33:40
Speaker 1: Is that something you are doing or thinking about? Sort
33:42
Speaker 1: of a compromised solution like that?
33:44
Speaker 2: Absolutely? And you know, as an example, a Stretch three
33:48
Speaker 2: as it is today, has a hard time opening some refrigerators, right,
33:51
Speaker 2: It's just the magnetic force is too high. But you
33:55
Speaker 2: can attach a little you know, grabbing thing that breaks
33:59
Speaker 2: the seal and makes it easy to open it. Right,
34:02
Speaker 2: And so it's a three dollars apart, you stick it
34:04
Speaker 2: onto your fridge. Now Stretch can open their fridge for
34:07
Speaker 2: you and for for these users, you know, those accommodations
34:12
Speaker 2: make a lot of sense, and uh, you know, I
34:15
Speaker 2: do think you know, maybe over time you'll have special
34:18
Speaker 2: plates at the robot nose it cannot you know, grab
34:20
Speaker 2: and not break, and you know, there will be accommodations
34:23
Speaker 2: that people make for it. But you know, for for now,
34:29
Speaker 2: we are using and counting on those types of accommodations.
34:33
Speaker 1: So I want to do two kind of forward looking
34:37
Speaker 1: things first, the sad one and the happy one. Basically okay,
34:41
Speaker 1: first is how might it not work out? And then
34:44
Speaker 1: the second is what does it look like if it
34:45
Speaker 1: does work out for the low robots, So like what
34:49
Speaker 1: might go wrong for for your company in a you know,
34:53
Speaker 1: significant way, right.
34:57
Speaker 2: Well, there's there's always the risk of just the business
34:59
Speaker 2: doesn't work out. I think, I think from from the
35:01
Speaker 2: technology and product side, it's really about acceptance, right, So
35:05
Speaker 2: I think implicit to bringing this technology and your life
35:09
Speaker 2: into your home, there has to be trust, right, and
35:12
Speaker 2: people have to trust and want this technology around them,
35:16
Speaker 2: and so I think, you know, there there's always a
35:19
Speaker 2: risk that people like you know, what we're going to
35:21
Speaker 2: we're going to find different ways to addrust the caregiving problem,
35:25
Speaker 2: you know, which you know, I should. There's lots of
35:28
Speaker 2: going to be lots of solutions to that, but robots
35:31
Speaker 2: may may not turn out to be part of that story.
35:34
Speaker 2: We're bullish that it will, but there it's a real
35:37
Speaker 2: kind of complex dynamic between the people and the technology.
35:41
Speaker 2: So that's that's sort of the the bare case. The
35:44
Speaker 2: bowl case would be, you know, ten years out. Call
35:47
Speaker 2: it a five thousand dollars device that's in your home
35:51
Speaker 2: that can provide caregiving support in the times when there's
35:55
Speaker 2: not care present. So basically it's just kind of facilitating
35:59
Speaker 2: quality of life at home, agency at home as you
36:02
Speaker 2: want to kind of age independently without having so when
36:05
Speaker 2: you're when your grandkids come over, you're not they're not
36:09
Speaker 2: doing chores for you. You're interacting, right because of the
36:12
Speaker 2: robots helping fill that gap.
36:17
Speaker 1: We'll be back in a minute with the lightning round.
36:33
Speaker 1: Let's finish with the lightning round.
36:35
Speaker 2: Okay, do you.
36:37
Speaker 1: Still make art?
36:38
Speaker 2: Oh? I not for a while, But that is how
36:42
Speaker 2: I got into robotics, was doing robotic sculpture, and then
36:47
Speaker 2: that led one thing led to another. But yeah, really
36:52
Speaker 2: fascinated by well, love art but also love you know,
36:55
Speaker 2: kind of robotics and how it relates to art.
36:58
Speaker 1: What's the least practical robot you've ever built?
37:02
Speaker 2: Wow? Uh? So let's see in the in the nineteen
37:06
Speaker 2: nineties in San Francisco, we had a robotic art performance
37:09
Speaker 2: troupe and it was a theatrical multimedia performance of robots
37:14
Speaker 2: on stage acting performing. You know, this was a kind
37:19
Speaker 2: of robotic gallery of kind of characters from the from
37:25
Speaker 2: the local neighborhood. So these were a street musician and
37:31
Speaker 2: you know, someone who was got of causing trouble and
37:35
Speaker 2: so they were kind of robotic denizens of the Tenderloin
37:39
Speaker 2: of nineteen nineties, San Francisco put onto stage.
37:43
Speaker 1: What was the first robot you ever built?
37:46
Speaker 2: It was actually the Robotic Street Preacher. So it was
37:49
Speaker 2: a little robot that would go around and big spinning
37:53
Speaker 2: head of a speaker and it would be on stage
37:57
Speaker 2: and perform and we would sit there and tell operate it.
38:01
Speaker 2: And it had, you know, no real autonomy. This is
38:04
Speaker 2: you know, everything with the hands salted, handbuilt and their
38:06
Speaker 2: computers were super tiny, are super a small capability, And
38:12
Speaker 2: we put that on stage and performed it.
38:15
Speaker 1: Did you ever try and build robots when you were
38:17
Speaker 1: a kid?
38:19
Speaker 2: No, you know, Strangely, I never had a fascination with
38:23
Speaker 2: robusts as a kid. I came out of grew up
38:27
Speaker 2: in Washington State on a farm, and uh, you know,
38:31
Speaker 2: I think my interest in robots came from having to
38:33
Speaker 2: do farm work and learning about automation.
38:36
Speaker 1: Like Henry Ford. You know, Henry Ford was the same
38:39
Speaker 1: story about a farm and thought that was too much
38:41
Speaker 1: labor exactly, and I just thought, you know, there's got
38:43
Speaker 1: to be a better way than this.
38:45
Speaker 2: My my mother's family, which is Mexican, had a history
38:48
Speaker 2: and kind of uh kind of working in the fields
38:52
Speaker 2: and kind of migrant labor, and so kind of the
38:55
Speaker 2: relationship to labor and work kind of inspired my interest
38:59
Speaker 2: in automation and then roproducts.
39:02
Speaker 1: Did you ever try and build a machine to do
39:04
Speaker 1: your chores for you?
39:05
Speaker 2: I absolutely did. It never worked, but it was fun,
39:09
Speaker 2: you know.
39:10
Speaker 1: I mean, I guess that's still what you're working on
39:12
Speaker 1: at some level.
39:13
Speaker 2: Exactly.
39:15
Speaker 1: What's one thing you've learned working with Frank Gary?
39:17
Speaker 2: Oh? Right, So we had a project looking at this
39:23
Speaker 2: conceptual but a large robotic kind of botanical garden and
39:28
Speaker 2: this sole thing in Singapore, so that we were looking
39:31
Speaker 2: at developing that with him and Uh, you know, I'd say, actually,
39:36
Speaker 2: the thing that was most interesting is just he really
39:41
Speaker 2: put a lot of trust into the young people in
39:43
Speaker 2: the studio and really was generous in that way. And
39:48
Speaker 2: you know, it's been an inspiration for me.
39:50
Speaker 1: That's a lovely surprise one. This is I guess some
39:54
Speaker 1: bias that I have. I imagine a famous architect as
39:56
Speaker 1: not being like that.
39:58
Speaker 2: So to hear that so lovely, very very generous in
40:01
Speaker 2: a way that can be inspiring to to the to
40:04
Speaker 2: people in their career. So I would try to carry
40:07
Speaker 2: that forward in my own work.
40:09
Speaker 1: Uh, thank you for your time.
40:11
Speaker 2: Yeah, it's been a pleasure. Thank you.
40:20
Speaker 1: Aaron Edzinger is the co founder and CEO of Hello Robot.
40:25
Speaker 1: Today's show was produced by Gabriel Hunter Chang and edited
40:28
Speaker 1: by Lydia Jane Kott. Our engineer this week was Hansdale Sheet.
40:33
Speaker 1: We're always looking for ideas for who to talk to
40:35
Speaker 1: and what to cover on the show. You can email
40:38
Speaker 1: us at problem at Pushkin dot fm. You can find
40:42
Speaker 1: me on x at Jacob Goldstein. You can find me
40:44
Speaker 1: on LinkedIn. I'm Jacob Goldstein. Thank you very much for
40:48
Speaker 1: listening to the show, and we'll be back next week
40:50
Speaker 1: with another episode