Latest / The SME Stream / Robots and the new physical AI gold rush

Transcript

Robots and the new physical AI gold rush · The SME Stream

0:02
Speaker 1: Today on the Business of Tech powered by Two Degrees Business,
0:06
Speaker 1: we're talking to two kiwis who've gone from the bleeding
0:09
Speaker 1: edge of Silicon Valley to the frontline of the robotics revolution,
0:13
Speaker 1: and are both doing it through what's fast becoming known
0:16
Speaker 1: as physical AI. Harry Malsop was an autopilot engineer at Tesla.
0:22
Speaker 1: Adrian McNeil helped build out the self driving platform and
0:26
Speaker 1: data infrastructure at Cruise. Now they founded their own companies,
0:30
Speaker 1: Antioch and Foxglove to solve some of the hardest problems
0:33
Speaker 1: in taking robots safely from the lab into the real world. Foxglove,
0:38
Speaker 1: which Adrian co founded, has just raised forty million NEWS
0:43
Speaker 1: dollars in Series B funding led by Bessemer Venture Partners
0:47
Speaker 1: with New Zealand's own ice House Ventures. Also on the
0:50
Speaker 1: cap table. It's building the data and an observability platform
0:55
Speaker 1: that lets robotics teams capture, search, and visualize the ocean
1:00
Speaker 1: of censored data robots generate so they can understand how
1:04
Speaker 1: these systems actually behave, why they failed, and how to
1:07
Speaker 1: improve them before they're deployed at scale. Antioch Harry startup
1:13
Speaker 1: has pulled in four point two million US about seven
1:16
Speaker 1: point three million New Zealand dollars in pre seed capital,
1:19
Speaker 1: again with ice House Ventures involved, to build a cloud
1:24
Speaker 1: simulation platform that lets companies build, test, and validate robots
1:29
Speaker 1: entirely in software instead of on costly physical test rigs
1:34
Speaker 1: and fake warehouses. We hear so much about humanoid robots
1:38
Speaker 1: and optimists coming online from Tesla, the revolution in robots
1:43
Speaker 1: being able to work in unstructured environments. What is going
1:46
Speaker 1: to make it happen is this type of innovation that
1:49
Speaker 1: these two guys are involved in. Between them, Foxglove and
1:53
Speaker 1: Antioch sit on the critical plumbing of physical AI, the data, flywheels,
1:59
Speaker 1: the simulation, the tooling that let you move fast without
2:03
Speaker 1: breaking things in the real world. So we'll talk about
2:06
Speaker 1: why AI powered robots have taken so long to arrive,
2:10
Speaker 1: how simulation and data pipelines are slashing those development cycles,
2:15
Speaker 1: and why safely validating robots in software first is becoming
2:20
Speaker 1: non negotiable if you want to deploy fleets off autonomous
2:24
Speaker 1: machines into factories, farms, warehouses, and city streets. We'll dig
2:30
Speaker 1: into the very key we journeys that took them from
2:33
Speaker 1: Treasury Stanford coin based Teslat and Cruise to leading this
2:37
Speaker 1: new generation of physical AI infrastructure companies. Here's my interview,
2:42
Speaker 1: recorded just before the summer break with Harry Melsop and
2:46
Speaker 1: Adrian McNeil. Harry and Adrian, Welcome to the Business of Tech.
2:55
Speaker 1: How are you both doing great?
2:56
Speaker 2: Thanks for having me, Thanks having us doing well.
2:59
Speaker 3: Thank you well.
3:01
Speaker 1: It's real honor to have you on the podcast. I've
3:04
Speaker 1: been hearing so much about both of your companies and
3:07
Speaker 1: we're going to get into exactly what you're doing. But
3:09
Speaker 1: what I've found really interesting is the area that you've
3:12
Speaker 1: both ended up specializing in after a lot of different
3:17
Speaker 1: companies that you have worked at in software engineering, some
3:21
Speaker 1: of the big names of Silicon Valley. So I'm keen
3:23
Speaker 1: to drill into what you learn from those amazing roles,
3:27
Speaker 1: but sort of at the frontier of physical AI and robotics, which,
3:32
Speaker 1: as someone who's covered tech for twenty five years, probably
3:35
Speaker 1: the most difficult area to cover because there's so much
3:38
Speaker 1: hype there and we've been promised so much in robotics,
3:42
Speaker 1: but there's always been these hard barriers that have been
3:45
Speaker 1: really difficult to get over. So it's great that you've
3:47
Speaker 1: got too New Zealand founders with really incredible startups that
3:51
Speaker 1: are tackling some of those core issues that are preventing
3:54
Speaker 1: robotics from rolling out and giving us the productivity gains
3:58
Speaker 1: and making life easier sort of thing. So keen to
4:01
Speaker 1: get into exactly how you're tackling some of those key problems.
4:04
Speaker 1: But maybe backing up just a bit, keen to get
4:07
Speaker 1: your origin story. Maybe if we start with you first,
4:10
Speaker 1: Harry about how you got into tech, your pathway in
4:13
Speaker 1: New Zealand and then making that jump to the US.
4:17
Speaker 3: Yeah. Absolutely so.
4:19
Speaker 4: I grew up here in New Zealand, did high school
4:22
Speaker 4: here in Auckland. Always kind of enjoyed building things in
4:27
Speaker 4: the technology space. Had a couple of little commercial ventures
4:31
Speaker 4: based out of here in Auckland. One of them in
4:35
Speaker 4: particular got a wee bit of commercial traction and it
4:37
Speaker 4: got picked up by New Zealand Trade and Enterprise. Pam
4:42
Speaker 4: Forward from NZTA very kindly organized a trip for me
4:46
Speaker 4: through Silicon Valley. I went to a bunch of different
4:49
Speaker 4: companies there, also saw Stanford University and for me that
4:53
Speaker 4: was kind of a bit of an eye opening experience.
4:55
Speaker 4: I'd never really considered going further afield than maybe Australia
5:00
Speaker 4: was born in Australia. I've been over there a bit,
5:02
Speaker 4: but you know, I never really thought about living further
5:04
Speaker 4: abroad than that. But that really kind of changed my perspective.
5:07
Speaker 4: I thought the scale was extraordinary. I thought that the
5:11
Speaker 4: you know, work ethic and the innovation was just you know,
5:15
Speaker 4: fantastic and really inspiring. And so, you know, towards the
5:19
Speaker 4: last couple of years of high school, I think I
5:22
Speaker 4: probably did that trip when I was in year eleven,
5:25
Speaker 4: and so in year twelve I kind of pivoted and
5:26
Speaker 4: I was like, Okay, what do I need to do
5:28
Speaker 4: to get into one of these universities ever in the States,
5:31
Speaker 4: did a ton of research online about what I had
5:33
Speaker 4: to do to make that happen, and did my SATs
5:35
Speaker 4: and all that kind of stuff, and then was revery
5:39
Speaker 4: fortunate got accepted into Stanford, went across there, did my
5:43
Speaker 4: undergrad degree there, did my master's degree there in economics
5:47
Speaker 4: and artificial intelligence, and then that kind of spawned my
5:51
Speaker 4: career from there.
5:52
Speaker 1: Well, yeah, and what years did you study at Stanford?
5:56
Speaker 4: So for my undergrad I was there from twenty seventeen
5:59
Speaker 4: to twenty twenty one, and then I stuck around for
6:01
Speaker 4: another year after that and did did my master's degree.
6:04
Speaker 1: Yeah, obviously a legendary place to study, the legacy it
6:09
Speaker 1: has seeding innovation and talent into Silicon Valley. But I
6:13
Speaker 1: guess at that time, around twenty seventeen and that you
6:16
Speaker 1: were hearing a lot studying large language models, seeing the
6:19
Speaker 1: potential of all of that and generative AI, which spawned
6:23
Speaker 1: the open AI and those sorts of large language model services.
6:28
Speaker 3: Yeah, no, exactly.
6:29
Speaker 4: So I think, you know, I was really a student
6:32
Speaker 4: at the advent of some of these transformer models which
6:35
Speaker 4: have kind of formed the basis of open AI and
6:37
Speaker 4: some of these other big forefront labs. Now, so, I
6:40
Speaker 4: think when I first started doing kind of like language modeling,
6:43
Speaker 4: it was a technology called LSTMs and like other very
6:47
Speaker 4: comparatively primitive approaches at kind of solving these language problems.
6:52
Speaker 4: And then you know, we kind of saw GPT and
6:54
Speaker 4: then GPT two come out with these transformer architectures, and
6:59
Speaker 4: it was kind of more of a curiosity I think
7:01
Speaker 4: initially then anything it was like, okay, this, you know,
7:04
Speaker 4: this technology seems to be able to really accurately mimic
7:07
Speaker 4: a sort of human speech in a way that you
7:10
Speaker 4: sort of look at it, and it sort of looks
7:13
Speaker 4: at first glance like somebody could have actually written that,
7:16
Speaker 4: or somebody could have actually said that. But then I
7:19
Speaker 4: think it wasn't until a couple of years later with
7:21
Speaker 4: GPT three and then ultimately chant GPT that I think
7:25
Speaker 4: we started to see this is not only an intellectual curiosity,
7:29
Speaker 4: but actually this is a really useful tool and a
7:32
Speaker 4: multitude of kind of different areas.
7:34
Speaker 1: And you quite quickly have to Stanford got into some
7:37
Speaker 1: really interesting companies and software engineering, the likes of Transpose
7:42
Speaker 1: which became Chainalysis, Stint to Amazon, and in Tesla, which
7:47
Speaker 1: is really interesting. But back at school, were you really
7:50
Speaker 1: interested in robotics? Who did that come later?
7:52
Speaker 4: Yeah, So we did a bit of work in embodied AI,
7:56
Speaker 4: so connecting some of these At the time it was
7:59
Speaker 4: particularly vision based neural networks, so convolutional neural networks and
8:04
Speaker 4: kind of imbuing.
8:05
Speaker 3: These systems with a visual understanding of the world.
8:08
Speaker 4: I did a lot of work there at school, but
8:12
Speaker 4: it was really I think at Tesla that started to
8:15
Speaker 4: kind of like professionalize and sort of solidify my particular
8:18
Speaker 4: interest in that physical AI world, working on models that
8:22
Speaker 4: went into the car. And then also at Tesla. At
8:25
Speaker 4: least in the early days, the models that were running
8:27
Speaker 4: on Optimists, the humanoid robots, were actually the same models
8:31
Speaker 4: as those running.
8:31
Speaker 1: On the vehicle Adrian. Similar sort of trajectory for you.
8:36
Speaker 1: Entrepreneurial from the start in New Zealand with Expresso used
8:40
Speaker 1: start up tell us about that, and then also a
8:42
Speaker 1: stint at New Zealand Treasury.
8:43
Speaker 3: Yeah.
8:44
Speaker 2: Yeah, I was at the Treasury for a couple of
8:46
Speaker 2: years when I first graduated, and then I spent a
8:51
Speaker 2: few years with a buddy of mine. We're running a
8:54
Speaker 2: web design company and building websites for people with this
8:56
Speaker 2: is back in the day. But yeah, pretty quickly was
9:00
Speaker 2: keen to get out out of New Zealand and make
9:03
Speaker 2: it over here to the States. I think similar to
9:05
Speaker 2: Harry when I first came out to San Francisco and
9:07
Speaker 2: so like a valley just kind of you know, your
9:09
Speaker 2: eyes are open to just how much was going on
9:11
Speaker 2: here and how many companies, and you know how much
9:14
Speaker 2: opportunity there was, and so when I moved out. Funny
9:18
Speaker 2: actually that we both started in the crypto space too
9:21
Speaker 2: when I moved out. Yeah too, too washed up crypto
9:26
Speaker 2: founders doing doing physically. I now, yeah, I came out
9:31
Speaker 2: to San Francisco in twenty fourteen and was at Coinbase
9:36
Speaker 2: for a couple of years and was leading the engineering
9:40
Speaker 2: team there for a little bit and then yeah, from
9:44
Speaker 2: there went to Cruz. So a friend of mine who's
9:48
Speaker 2: now my co founder, was at Coinbase. He bounced over
9:52
Speaker 2: to Cruz and got all excited about self driving cars.
9:55
Speaker 2: This is early twenty sixteen, so you know, it was
9:58
Speaker 2: pretty early for the self driving he you know, one
10:02
Speaker 2: thing led to another convinced me to come over there,
10:04
Speaker 2: and I spent about five years at Cruz leading various
10:08
Speaker 2: aspects of the infrastructure platform, develop a tooling, data, infrastructure,
10:12
Speaker 2: things like that, and so really got a first hand
10:15
Speaker 2: seats as that whole space evolved in how we kind
10:18
Speaker 2: of figured out from first principles what does it take
10:21
Speaker 2: to put a self driving car on the road. And
10:24
Speaker 2: you know, sure as Horry can attest from the Tesla
10:27
Speaker 2: side of the equation to you just a lot of
10:29
Speaker 2: these things. You know, there was no book for it, right,
10:32
Speaker 2: There was no there was no manual, There was no
10:33
Speaker 2: blog posting you go read about what other companies are doing.
10:36
Speaker 2: It was you know, it was a few companies are
10:38
Speaker 2: all kind of just figuring it out for ourselves and
10:41
Speaker 2: figuring it out as you go along. So that was
10:44
Speaker 2: that was really interesting, and then that led to twenty
10:48
Speaker 2: twenty one is when I founded my current company, box Log,
10:51
Speaker 2: which we're building a very similar thing, a data platform
10:55
Speaker 2: for a physical WAYI mirroring a lot of the lessons
10:58
Speaker 2: learned from the self driving industry, building a platform that
11:02
Speaker 2: lets people capture and learn from all of that data
11:05
Speaker 2: that they're gathering from robots in the real world.
11:07
Speaker 1: Yeah, and that company, Cruise, I think, was considered the
11:10
Speaker 1: real leader for many years at least in autonomous car development.
11:14
Speaker 1: I think it's been folded back into GM now and
11:16
Speaker 1: that technology has been dispersed across GM, but for a
11:19
Speaker 1: while it was you know, Cruise and then obviously Tesla
11:22
Speaker 1: where the real pioneers. And that incredible that another New
11:25
Speaker 1: Zealander is running Wave in the UK.
11:29
Speaker 2: Yeah, I like, yeah, he's doing great.
11:31
Speaker 1: Yeah, raised something crazy like a couple of years ago
11:34
Speaker 1: from Soft Bank of billion dollars or something. So they've
11:38
Speaker 1: got a lot of runway to make their technology successful.
11:42
Speaker 1: So that's brilliant. But just interested in you know, I'm
11:46
Speaker 1: constantly thinking about when I interview people. What is it
11:50
Speaker 1: about about us that gives New Zealand entrepreneurs a little
11:56
Speaker 1: bit of an edge. I'm just wondering if there's anything
11:57
Speaker 1: in your upbringing or your early education here in New
12:00
Speaker 1: Zealand that sort of shaped the way that you think
12:02
Speaker 1: about tackling some of these hard technology problems.
12:06
Speaker 2: Yeah, I actually, I actually think quite a bit about this.
12:08
Speaker 2: I think New Zealand is quite conducive to an entrepreneurial
12:12
Speaker 2: mindset because there is so much small business you actually,
12:16
Speaker 2: I mean in the States and Silicon Valley. I don't
12:19
Speaker 2: know if it's a majority, but certainly, like a large
12:22
Speaker 2: portion of the founders that you talk to were not
12:24
Speaker 2: born in the States and certainly didn't grow up in
12:26
Speaker 2: the Bay Area. You know, you have a lot of
12:29
Speaker 2: founders from all over the world here, and you know,
12:33
Speaker 2: I found, for example, growing up, like my parents rand businesses,
12:38
Speaker 2: various businesses over the time, growing up most of my
12:40
Speaker 2: friends parents run businesses. It's actually quite an unusual experience
12:45
Speaker 2: compared to the US, where most people are growing up
12:47
Speaker 2: with you know, parents with stable jobs, and then they
12:51
Speaker 2: get into a good college and then they're you know,
12:53
Speaker 2: coming out of college immediately looking for what it is
12:55
Speaker 2: a great job that I can get out of college.
12:57
Speaker 2: That's kind of the default experience. But I think, you know,
12:59
Speaker 2: at least my experience in New Zealand, and I think
13:01
Speaker 2: this is true in a lot of New Zealand's just
13:03
Speaker 2: there's a lot of small business and that has an
13:05
Speaker 2: impact on you from a young age.
13:07
Speaker 3: Yeah, I completely agree with that point.
13:09
Speaker 4: I think, you know, there are also some elements of
13:13
Speaker 4: being at the far end of the world where you
13:15
Speaker 4: just sort of have to be pragmatic and just have
13:17
Speaker 4: to figure it out. And I think that there are
13:19
Speaker 4: a lot of New Zealanders, if you kind of look
13:21
Speaker 4: across history, have sort of embodied that. And maybe it's
13:24
Speaker 4: a little bit of a New Zealand cultural trope, but
13:26
Speaker 4: I think there's also a lot of truth to that
13:29
Speaker 4: as well. And I think when you're tackling really hard problems,
13:34
Speaker 4: you have to be extremely resourceful and you have to
13:37
Speaker 4: sort of look at that problem from a distance and
13:40
Speaker 4: figure out you know, I've got a ton of constraints.
13:43
Speaker 4: How many of them are often financial, But if you're
13:46
Speaker 4: attacking this really large problem, even with a ton of money,
13:49
Speaker 4: you're still going to be running up very quickly against
13:52
Speaker 4: that and so I think, you know, there's a lot
13:54
Speaker 4: of New Zealand history that's that's kind of been written
13:56
Speaker 4: around people kind of overcoming a lot of that, and
13:59
Speaker 4: I think that layers into culture there too.
14:00
Speaker 1: Yeah, and that's something that Sir Peter Beck I think
14:03
Speaker 1: is still part of his DNA and the DNA of
14:06
Speaker 1: Rocket Lab is that famous quote that Sir Ernest Rutherford said,
14:10
Speaker 1: you know, we don't have the money, so we need
14:12
Speaker 1: to think I'm paraphrasing it there, and that's that inspired
14:15
Speaker 1: Sir Peter, which is why he named the you know,
14:17
Speaker 1: the Rutherford Engine after Ernest Rutherford. And I think it's
14:22
Speaker 1: still evident in the approach in that company today. You know,
14:25
Speaker 1: with the new rocket, it's actually a relatively modest budget
14:28
Speaker 1: they've developed that on. So that frugality, I think is
14:31
Speaker 1: a benefit that we have. We can do more with less.
14:35
Speaker 1: We're used to doing that. And I think Adrian, you know,
14:38
Speaker 1: you were interested in robotics all the way back right
14:40
Speaker 1: like growing up in rural New Zealand looking at kiwi
14:43
Speaker 1: fruit processing and that sort of stuff.
14:45
Speaker 2: Yeah. Yeah, I think that definitely had an impact on me.
14:49
Speaker 2: I mean, I wasn't so much thinking about robotics at
14:51
Speaker 2: the time, although I didn't spend a fair amount of
14:53
Speaker 2: time in packhouses doing various jobs and you know, in
14:57
Speaker 2: high school summers and things. But once I sort of
15:03
Speaker 2: started working in the self driving industry and spending time,
15:06
Speaker 2: you know, around self driving vehicles, I reached a point
15:09
Speaker 2: where I was like, so, you know, we made a
15:13
Speaker 2: car drive around San Francisco, and yet look at what
15:16
Speaker 2: we have not why are there no self driving tractors
15:19
Speaker 2: in the fields for example, right? And so I reached
15:22
Speaker 2: this point where you know, you're sort of living in
15:25
Speaker 2: a bit of a bubble at the bleeding edge of
15:27
Speaker 2: some technology, right, Like you know, every every week I'm
15:30
Speaker 2: going for drives in a self driving car and thinking about,
15:33
Speaker 2: you know, how this technology was put together and what
15:35
Speaker 2: it took to get there. And then you know, looking
15:38
Speaker 2: around and seeing all of the things that would be
15:42
Speaker 2: way easier to automate that have not yet been automated.
15:44
Speaker 2: And you reach this point you kind of reached us
15:46
Speaker 2: like technology overhang, I guess where you're like, look, we
15:48
Speaker 2: have this technology, it's just it's not widely distributed in
15:50
Speaker 2: the world. Get and so those kind of things always
15:53
Speaker 2: really excite me. Where there's kind of an opportunity to
15:56
Speaker 2: push the ball forward.
15:57
Speaker 1: You're both at that point in your career. You're both
16:00
Speaker 1: in really good companies, well paid jobs. You could probably
16:05
Speaker 1: have carried on, you know, in big tech companies, gaining
16:08
Speaker 1: more experience than that, but you obviously both made a
16:10
Speaker 1: decision to go out on your own. What was it
16:13
Speaker 1: for you, Harry, that convinced you to leave that, you
16:15
Speaker 1: know that the safety of a Silicon Valley job and
16:19
Speaker 1: a great company, to forge out on your own. What
16:21
Speaker 1: was the painful problem that you really wanted to tackle
16:24
Speaker 1: that you saw you couldn't do as part of the
16:27
Speaker 1: current system.
16:27
Speaker 3: I think it's a really good question.
16:29
Speaker 4: I think for me it was kind of a confluence
16:32
Speaker 4: of different factors coming together.
16:35
Speaker 3: I think I've.
16:35
Speaker 4: Always liked the process of building something from a more
16:39
Speaker 4: holistic perspective, and so the engineering is a is a
16:43
Speaker 4: key component of that, but it's far from the only piece.
16:46
Speaker 4: There's the interpersonal piece of it. There's connecting that to
16:49
Speaker 4: some kind of concrete business outcome. There's everything else that
16:52
Speaker 4: kind of sits around what it takes to actually commercialize something.
16:56
Speaker 4: And I really enjoy all of that personally, and so
17:00
Speaker 4: I think for me, going to Tesla initially was me
17:05
Speaker 4: thinking to myself, Okay, I know that ultimately I kind
17:08
Speaker 4: of want to go and I want to build technology companies,
17:12
Speaker 4: But first of all, I want to kind of surround
17:13
Speaker 4: myself with, you know, as many kind of brilliant minds
17:17
Speaker 4: who challenged me, who are much better than me in
17:20
Speaker 4: a multitude of kind of different ways, to kind of
17:23
Speaker 4: optimize for that learning gradient as quickly as possible, and
17:26
Speaker 4: just sort of optimize for talent density so I can
17:28
Speaker 4: kind of, by process of osmosis of absorb as much
17:32
Speaker 4: of that as possible. So I think for me it
17:34
Speaker 4: was less of a push to get out of Tesla
17:37
Speaker 4: and more of you know that I kind of always
17:40
Speaker 4: saw that experience at some of those larger companies as
17:43
Speaker 4: a bit of a means to an end to sort
17:45
Speaker 4: of take that out and build something myself.
17:47
Speaker 1: Yeah, and take us through Antioch. What you're trying to
17:51
Speaker 1: achieve there you've said, I think integration testing for atoms,
17:57
Speaker 1: which is a really interesting phrase that probably you keep
18:01
Speaker 1: talking about, and simulation. Tell us what Antiocha is actually
18:05
Speaker 1: setting out to achieve when it comes to robotics, AI
18:09
Speaker 1: and simulation.
18:10
Speaker 4: Yeah, so today if you're building a physical AI company,
18:14
Speaker 4: if that's self driving vehicle or drones or robots that
18:18
Speaker 4: are in factories. The general kind of development process that
18:22
Speaker 4: you're taking is you're buying all of these physical components,
18:26
Speaker 4: you're getting them on a workbench in front of you,
18:28
Speaker 4: you're writing the software for them, and you're kind of
18:30
Speaker 4: testing them manually out there in the real world. And
18:34
Speaker 4: there's a multitude of kind of problems with that. But
18:37
Speaker 4: the biggest problem comes when you kind of hit scale
18:39
Speaker 4: and you start thinking about the tail cases of when
18:43
Speaker 4: I'm deploying this out in reality, what are all the
18:45
Speaker 4: edge cases that I'm running into that I didn't foresee
18:48
Speaker 4: when I was developing, or that are actually really difficult
18:51
Speaker 4: and really prohibitive to kind of replicate in my local
18:54
Speaker 4: kind of development and kind of testing environment. And so
18:58
Speaker 4: you know, at anti K what we're doing is we're
19:01
Speaker 4: shifting as much of that development and that evaluation process
19:04
Speaker 4: as possible into software simulation, where you don't have the
19:08
Speaker 4: constraints of the real world. You know, getting actors, getting scenes,
19:13
Speaker 4: getting objects, and a multitude of kind of differentch cases
19:16
Speaker 4: is effectively free. And so you know, we talk to
19:19
Speaker 4: a lot of customers now who spend you know, tens
19:22
Speaker 4: to hundreds of millions of dollars every year on their
19:26
Speaker 4: kind of physical testing infrastructure. I kind of saw this
19:29
Speaker 4: firsthand when I was at Tesla, Right, we would have
19:32
Speaker 4: hundreds of cars on the road, hundreds of drivers collecting
19:35
Speaker 4: these real driving miles, and it's a huge problem for
19:39
Speaker 4: other companies as well. So by kind of shifting that
19:41
Speaker 4: into simulation as much as possible, you're slashing a lot
19:45
Speaker 4: of cost. But you're also kind of democratizing the access
19:48
Speaker 4: to building something in the physical AI world, not just
19:51
Speaker 4: to a company like Tesla who can afford to spend
19:53
Speaker 4: one hundred million plus dollars a year on that testing infrastructure,
19:57
Speaker 4: but because you can do it in software, it's really approachable.
20:00
Speaker 4: Anyone could sign up for an account and get started there.
20:03
Speaker 1: Yeah. And is there any particular type of robotics or
20:07
Speaker 1: physical systems that you're focusing on at the moment or
20:10
Speaker 1: is this the whole point of it. It's sort of
20:11
Speaker 1: democratizing simulation. So it could be a robot in a
20:15
Speaker 1: factory doing manufacturing, it could be an agritech robot, or
20:18
Speaker 1: it could be an autonomous vehicle.
20:20
Speaker 4: Yeah, that's a great question. I think there's sort of
20:24
Speaker 4: two answers to that. So there's an answer from the
20:26
Speaker 4: platform capability, like what can we do? And the answer
20:29
Speaker 4: to that is it's actually very generally capable. You know,
20:32
Speaker 4: we've got high quality sensor simulation, so cameras, radars, lighters,
20:37
Speaker 4: all of these different kinds of senses that you have
20:39
Speaker 4: on your robots, and those are kind of actually common
20:42
Speaker 4: across you know, all sorts of different morphologies of robots
20:45
Speaker 4: generally speaking. And the sort of second answer to that
20:50
Speaker 4: is from a pragmatic perspective, what are we going after first?
20:53
Speaker 4: Because you know, you chase two rabbits, you catch either.
20:55
Speaker 4: We sort of have to focus on a couple of
20:57
Speaker 4: particular industries to just think our teeth too, and they've
21:01
Speaker 4: really been two so far.
21:04
Speaker 2: So there's a lot of.
21:04
Speaker 4: Manufacturing applications that that we've kind of seen, and that
21:08
Speaker 4: takes a lot of different forms, everything from like bringing
21:11
Speaker 4: together packaged meals using robots to assembling hardware components using robots.
21:17
Speaker 4: But also I think what's kind of surprised US has
21:20
Speaker 4: been actually a second market, which is kind of like
21:23
Speaker 4: a smart security system market, and so you don't tend
21:27
Speaker 4: to think of that when you think of robotics, because
21:29
Speaker 4: these things aren't often moving around, but the problems are
21:32
Speaker 4: all kind of the same. You've got cameras mounted to
21:35
Speaker 4: you know, warehouses or construction sites or things like that,
21:38
Speaker 4: and you need to sort of understand track progress of construction.
21:42
Speaker 4: You need to track, you know, where agents are in
21:44
Speaker 4: the scene, if they're in places that they shouldn't be,
21:46
Speaker 4: So people have kind of come into the property and
21:48
Speaker 4: they shouldn't be there, for example. So the kind of
21:51
Speaker 4: testing problem is actually very very analogous. So we're sort
21:54
Speaker 4: of starting with those two markets, but broadly the product
21:57
Speaker 4: is capable of going a lot wider.
22:00
Speaker 1: Yeah, that's a hugely promising market I think. I mean
22:02
Speaker 1: several years ago in Silicon Valley, I think it was
22:05
Speaker 1: US Robotics. I saw their little dog they were demoing,
22:08
Speaker 1: and they were saying, instead of having a big security
22:12
Speaker 1: guard force, you can send out these dogs. They've got
22:15
Speaker 1: covered in sensors are constantly giving you real time intelligence,
22:19
Speaker 1: night vision and all of that back. So if you
22:21
Speaker 1: have a big sort of manufacturing site or retail premises
22:26
Speaker 1: or something, but it's always just taken so long to
22:29
Speaker 1: get to fruition, and you know they're expensive robots, but
22:32
Speaker 1: I'm surprised it's one area that has taken so long.
22:37
Speaker 1: So it sees as some of the barriers that for
22:40
Speaker 1: a safety point of view, being able to simulate if
22:43
Speaker 1: we put these into an unstructured environment, what is actually
22:47
Speaker 1: going to happen? Is it going to be safe to
22:48
Speaker 1: unleash these in an industrial or even in an environment
22:51
Speaker 1: where civilians are Yeah, there.
22:53
Speaker 3: Have been so many different barriers.
22:54
Speaker 4: I mean, I think the cost of hardware has historically
22:57
Speaker 4: been a huge one. You know, even some of the
23:01
Speaker 4: sort of like table stakes, perception and planning algorithms that
23:05
Speaker 4: go into this as well have started to be more
23:08
Speaker 4: commoditized in recent years too. And then I think, you know,
23:11
Speaker 4: you bring up the safety point, which is an absolutely
23:14
Speaker 4: critical one, and that's one that I think that we're
23:16
Speaker 4: thinking a lot about because if you can kind of
23:19
Speaker 4: you know, go through a lot of that safety and
23:21
Speaker 4: that validation process in simulation for really really cheap you're
23:25
Speaker 4: able to slash a lot of development cost out of
23:27
Speaker 4: what you would otherwise be validating in the real world
23:31
Speaker 4: shutting down a real construction site in this case, right
23:33
Speaker 4: to test that your robot dog is not going to
23:36
Speaker 4: flip itself over or not not going to run into
23:38
Speaker 4: a person or get itself in front of a car
23:40
Speaker 4: or something like that. And so I think lots of
23:43
Speaker 4: different barriers, but there are lots of different companies kind
23:45
Speaker 4: of across the ecosystem targeting each of those, and so
23:48
Speaker 4: I'm pretty excited about what the next few years look
23:51
Speaker 4: like as those start to get eroded.
23:52
Speaker 1: So you're looking at simulation Adrian fox gloves. It's sort
23:57
Speaker 1: of at the data and observe ability layer of robotics.
24:01
Speaker 1: For listeners who aren't really familiar with what that actually means.
24:05
Speaker 1: Take us through what that means day to day for
24:07
Speaker 1: a robotics engineering team trying to put one of these
24:09
Speaker 1: systems together.
24:11
Speaker 2: Yeah, So fox gott is a platform for managing data
24:14
Speaker 2: that you're getting off of these robots or physical AI systems.
24:18
Speaker 2: You can think about most robotics company have some kind
24:21
Speaker 2: of data flywheel, right. They're logging data in the real world,
24:25
Speaker 2: they're bringing their data back, they're analyzing that data to
24:27
Speaker 2: understand what was the performance of the robot that using
24:29
Speaker 2: that analysis to drive you know, what do we need
24:32
Speaker 2: to improve in the robot simulations? Do we need to
24:34
Speaker 2: create you know what what do we need to work on?
24:37
Speaker 3: That?
24:38
Speaker 2: Quite often training on and learning from their data. Now
24:40
Speaker 2: as the AI is getting better and better, they're actually
24:42
Speaker 2: like learning directly from recordings that we've got from the
24:45
Speaker 2: real world. They're needing to build up these data sets.
24:49
Speaker 2: They're needing to search across data and find like, hey,
24:51
Speaker 2: we recorded all of this, but we're at specific individual
24:54
Speaker 2: points in time where we're you know, the analogy the
24:58
Speaker 2: example with self driving writer is you can have millions
25:01
Speaker 2: and millions of miles and a lot of it is
25:03
Speaker 2: very useless. A lot of it is okay, we have
25:04
Speaker 2: plenty of data of people driving along the straight road
25:07
Speaker 2: with a nice clear, sunning sky or whatever, but we
25:10
Speaker 2: need more examples of us trying to make you know,
25:13
Speaker 2: a left turn across a divided highway with like pedestrians
25:16
Speaker 2: in the way while it's raining or whatever right, And so,
25:18
Speaker 2: you know, how do you navigate through all of this data,
25:22
Speaker 2: find the interesting examples to build out their data set
25:24
Speaker 2: and really really understand what they robot is doing. So
25:28
Speaker 2: we're fundamentally a platform for collecting, searching, storing, and visualizing
25:33
Speaker 2: and debugging all the data that you're getting off these systems.
25:36
Speaker 1: And similarly to Harry's approach, you're sort of horizontally integrated.
25:40
Speaker 1: It's not as say you've chosen one particularly promising vertical
25:45
Speaker 1: or like manufacturing that is ahead of others. You're actually saying,
25:49
Speaker 1: we can apply this and we can probably improve the
25:51
Speaker 1: product and learn from deploying it as widely as possible.
25:55
Speaker 2: Yeah, exactly, we have hundreds of paying customers across autonomous vehicles, drones,
26:00
Speaker 2: across agriculture, across ocean, robots, manufacturing, logistics, constructions. It's a
26:08
Speaker 2: very generally applicable problem, as is I think you know,
26:11
Speaker 2: Harry's Harry's company. We're trying to solve these horizontal plays.
26:15
Speaker 2: A lot of this infrastructure needs to exist because if
26:17
Speaker 2: you look at a large, a large mature self driving
26:20
Speaker 2: program like today at a Tesla or Awaimo, there are
26:24
Speaker 2: you know, a thousand plus engineers and the bottom of
26:26
Speaker 2: the iceberg is building all of this tooling and infrastructure.
26:29
Speaker 2: Right like, you have a small number of people who
26:31
Speaker 2: are really working on like novel model architecture, and then
26:35
Speaker 2: you have all the people that are dealing with.
26:36
Speaker 5: The oceans of data that you get with the complexity.
26:39
Speaker 2: Around simulation, with the training pipelines, with the monitoring and
26:42
Speaker 2: observability of production and understanding like you know, how is
26:45
Speaker 2: the fleet managing production and like building this flow. Also
26:48
Speaker 2: that you're constantly learning from all of the driving that
26:50
Speaker 2: you're doing out in the world. This is how we
26:53
Speaker 2: have been able to build self driving cars in twenty
26:56
Speaker 2: twenty five. But if we want all of these other
26:58
Speaker 2: things to exist, if we want there to be a
26:59
Speaker 2: lot more autonomous robots and manufacturing or logistics or any
27:03
Speaker 2: of these other areas, that platform needs to be available
27:06
Speaker 2: off the shelf. We can't have thousands of robotics. Companies
27:09
Speaker 2: and startups cannot be successful unless they can buy the
27:12
Speaker 2: bulk of this kind of tooling and infrastructure off the shelf.
27:14
Speaker 2: And that's how every other industry works, right, Like in
27:16
Speaker 2: the as SASS industry, you want to go build a
27:18
Speaker 2: you want to launch an e commerce store, you don't
27:20
Speaker 2: start like writing code about how your checkout's going to work. Right,
27:24
Speaker 2: you have all of those things available off the shelf
27:25
Speaker 2: so that you can focus on what's unique to your business.
27:28
Speaker 1: And what was there anything really that crystallized it for
27:32
Speaker 1: you at cruise where you thought, wow that if I
27:36
Speaker 1: ever do my own startup, which I will do. I
27:39
Speaker 1: want to encapsulate that in my approach. Was there anything
27:42
Speaker 1: we thought, Wow, even if it was a thousand person team,
27:45
Speaker 1: you thought I can on a smaller scale replicate that.
27:48
Speaker 2: Yeah. For me, that was the visualization piece. And I
27:51
Speaker 2: think maybe even backing up, I agree with the comment
27:53
Speaker 2: Harry made earlier, which is that I think if you
27:57
Speaker 2: have a sort of startup founded mentality a lot of
28:00
Speaker 2: the time, at least for me, I had, you know,
28:03
Speaker 2: I had that from a very early age, and most
28:06
Speaker 2: of the companies that I joined was intention was with
28:10
Speaker 2: the intention of learning and getting a front receipt to
28:12
Speaker 2: a high growth startup and seeing what the environment was saying,
28:15
Speaker 2: understanding what it took to make a company successful, and
28:18
Speaker 2: getting to work really closely with these founders. So in
28:22
Speaker 2: that respect, the entire time, you know, you're always on
28:25
Speaker 2: the lockout for what a great sort of business opportunities,
28:28
Speaker 2: And that is a very common thing when you're working
28:32
Speaker 2: full time in a new growth industry. There are a
28:35
Speaker 2: lot of opportunities that haven't yet been kind of productized
28:37
Speaker 2: and turned into companies yet. So that was always on
28:40
Speaker 2: my mind. I think for me, you know, really the
28:44
Speaker 2: thinking really crystallized around the data, and initially gio entry
28:47
Speaker 2: point more specifically even than the data, was this aspect
28:51
Speaker 2: of visualization and debugging. So at Cruz we built very
28:54
Speaker 2: sophisticated tools for doing multimode also sort of three D
28:58
Speaker 2: and two D visualization of the vehicle and of the data,
29:01
Speaker 2: being able to see the cameras alongside the three D
29:04
Speaker 2: visualization of where the car thought it was on the
29:06
Speaker 2: road and you know, understanding things like hey, why did
29:10
Speaker 2: we slam on the brakes here? Well, it was because
29:12
Speaker 2: we we thought that this other car was going to
29:14
Speaker 2: pull out in front of us, or you know, like
29:16
Speaker 2: how do you understanding what the AI on the car
29:18
Speaker 2: is sort of thinking you need to do that. It's
29:21
Speaker 2: inherently a kind of a three D problem, and so
29:23
Speaker 2: that was one of the things that we built our
29:24
Speaker 2: crews that from very early on I saw, as you know,
29:30
Speaker 2: it was, it was a differentiator at Cruise. We had
29:33
Speaker 2: we invested a lot in that. But then you know,
29:36
Speaker 2: later on talking to I learned that from talking to
29:39
Speaker 2: people at Tesla and Weymo that very similar systems had
29:41
Speaker 2: been kind of invented. Right, we weren't talking to each other,
29:44
Speaker 2: but we compared notes and realized that we're all built
29:46
Speaker 2: the same thing. It's kind of a clear, clear, a
29:49
Speaker 2: clear signal that there's an opportunity.
29:58
Speaker 1: Similarly, Harry, for you that stint you did at Tesla
30:02
Speaker 1: and we all know from the outside, you know what
30:04
Speaker 1: an intense environment that is and led by you know,
30:08
Speaker 1: a visionary engineer. Any key takeaway that's infused your approach
30:15
Speaker 1: with Antioch.
30:16
Speaker 4: Yeah, I mean, I think similar to what Adrianne was
30:18
Speaker 4: talking about, it's really been I was really impressed by
30:21
Speaker 4: the tooling inside. And I think particularly what struck me
30:26
Speaker 4: was you know, in software we talk about shifting left
30:30
Speaker 4: and that's used in the context of security, and it's
30:32
Speaker 4: used in the context of testing. But essentially what that
30:34
Speaker 4: means is as early as possible in the process of
30:39
Speaker 4: developing something, you want to be getting a good signal
30:42
Speaker 4: that it is doing what you think it's doing, safe, secure,
30:46
Speaker 4: And I think Tesla had invested a huge amount in
30:49
Speaker 4: that kind of infrastructure. So to put that concretely, you know,
30:52
Speaker 4: when I was training a new neural network, there are
30:56
Speaker 4: a ton of software tools that my network could be
30:59
Speaker 4: running against me quick feedback on was this better?
31:02
Speaker 3: Was this worse than the last thing that we had
31:04
Speaker 3: been using.
31:05
Speaker 4: So as a developer, I'm immediately getting that sense of okay,
31:08
Speaker 4: like I'm doing the right thing or I'm doing the
31:10
Speaker 4: wrong thing. And then also, you know, there were simulation
31:13
Speaker 4: tools that were kind of built internally there as well
31:16
Speaker 4: as kind of like the second phase of how these
31:18
Speaker 4: things were evaluated. So some software first, then you put
31:21
Speaker 4: it into the virtual car, see how that goes, and
31:24
Speaker 4: only is this kind of tertiary step that you're taking
31:26
Speaker 4: that actually out to the field. This is a bit
31:28
Speaker 4: of an oversimplification of what it's sort of like, but
31:31
Speaker 4: I think it's the right kind of way to think
31:32
Speaker 4: about it, where you want to bring your evaluation as
31:35
Speaker 4: close as possible to developers. And I think, you know,
31:39
Speaker 4: outside of Tesla, when we talk to prospective customers, that
31:43
Speaker 4: that kind of tooling just doesn't exist. You know, we
31:46
Speaker 4: talked to folks in construction robotics, you know who they
31:49
Speaker 4: train some new machine learning model for an excavator or
31:53
Speaker 4: something like that, and they're sending it like three states
31:56
Speaker 4: over to some guy who's actually sitting on the excavator
31:58
Speaker 4: and he goes and digs a hole using this neural
32:01
Speaker 4: network and then he's kind of sending that back to
32:04
Speaker 4: to base fundamentally, and that's the evaluation loop there, and
32:08
Speaker 4: that's so latent. So like as a developer, you make
32:11
Speaker 4: a change, it's going to take you maybe days to
32:13
Speaker 4: kind of get insight into you know, was that actually
32:17
Speaker 4: the right change to make, And that's slowing you down
32:20
Speaker 4: several orders of magnitude. And I think Tesla had done
32:23
Speaker 4: a great job of building tooling to reduce that as
32:25
Speaker 4: much as possible.
32:26
Speaker 1: So both of what you're doing is speeding up development ultimately,
32:30
Speaker 1: so hopefully you can get a prototype into a production
32:35
Speaker 1: grade robot. Quicker Adrian. From your perspective, what sectors do
32:40
Speaker 1: you expect to see sort of real world deployment if
32:44
Speaker 1: robots accelerate, say in the next five to ten years.
32:47
Speaker 1: We've got warehouses, you know, they're talking about dark warehouses
32:51
Speaker 1: where robots are just working continuously. We've seen some really
32:54
Speaker 1: cool stuff in agritech. The whole drone thing is a
32:57
Speaker 1: whole other world. Any particular areas where you're saying, wow,
33:01
Speaker 1: this is actually racing ahead. Now.
33:04
Speaker 2: The robotics industry is really broad and there are different
33:08
Speaker 2: sectors that are at quite different stages of maturity. So
33:11
Speaker 2: I think it's really interesting when you're working in these
33:14
Speaker 2: horizontal areas and getting a cross section of where people
33:18
Speaker 2: are at and where robots are succeeding. So, for example,
33:22
Speaker 2: the warehouse space is one where it's a little bit
33:24
Speaker 2: further along that maturity curve, right, I mean, Amazon has
33:27
Speaker 2: a million robots in production or more at this point.
33:30
Speaker 2: A lot of warehouses are looking at autonomous material movement
33:34
Speaker 2: and little mbile robots that can move things around warehouses
33:37
Speaker 2: and things. That sector is at a point where they're
33:41
Speaker 2: really starting to scale. Their autonomy problems are a little
33:44
Speaker 2: bit simpler. They're not trivial, but they're simpler than like
33:46
Speaker 2: bolding laundry or making beds or some of these things
33:49
Speaker 2: that people are trying to do. So you're starting to
33:53
Speaker 2: see a lot of real scale there. Similarly, the German
33:56
Speaker 2: industry is really starting to hit scale. Self driving car
34:00
Speaker 2: industry is slowly really starting to scale. Although it's very
34:04
Speaker 2: very capital intensive building cars, it's compared to a lot
34:07
Speaker 2: of the cost of the vehicles is a lot higher
34:09
Speaker 2: than the cost of small drones or small warehouse robots.
34:13
Speaker 2: And then at the other end of the spectrum you
34:14
Speaker 2: have areas where there's significant amount of R and D,
34:18
Speaker 2: you know, significant amount of investment and PC capital, and
34:23
Speaker 2: areas like just pure robot foundation model plays where they're
34:26
Speaker 2: trying to just build general purpose models or humanoids for example.
34:30
Speaker 2: I think these areas we're seeing a ton of investment into.
34:33
Speaker 2: They're further out from commercialization, right, they're still very early.
34:36
Speaker 2: It's kind of humanoids today is where you know, I
34:40
Speaker 2: think about where self driving was around twenty seventeen or so,
34:44
Speaker 2: where you know, it's very promising technology. You can see
34:48
Speaker 2: the opportunity that it's going to bring, but it's going
34:50
Speaker 2: to take quite a while before they get to sort
34:52
Speaker 2: of a useful enough level of ability that they're going
34:54
Speaker 2: to go into real production US cases and be replacing
34:58
Speaker 2: any existing people of systems.
35:00
Speaker 1: Yeah, I mean we've seen a lot of pr attention
35:04
Speaker 1: on optimists out of Tiesler US robotics and the you know,
35:09
Speaker 1: robots doing backflips and all that sort of thing. Apparently.
35:12
Speaker 2: I mean, there's an a tunnel investment happening in the space,
35:15
Speaker 2: and there's a lot of players, a lot of Chinese
35:17
Speaker 2: companies building humanoids, there's a number in the US. There's
35:21
Speaker 2: a huge range of models and skills too. I mean,
35:24
Speaker 2: you've got everything from very cheaply made sort of five
35:27
Speaker 2: thousand dollars things that you can push over if you
35:30
Speaker 2: accidentally bumped into them, compared to optimists and bigger, you know,
35:34
Speaker 2: pretty very expensive advanced pieces of machinery basically at kind
35:41
Speaker 2: of the other end of that. It'd be really interesting
35:43
Speaker 2: to see how this plays out. I think I'm extremely
35:45
Speaker 2: fullish long term on humanoids, but right now some of
35:49
Speaker 2: the things that you're hearing about how soon they're arriving
35:52
Speaker 2: are at a little inflated, shall we say.
35:56
Speaker 1: Yeah, it's obviously a hot category for investment, and you've
36:00
Speaker 1: both recently raised significant money at different stages.
36:04
Speaker 2: So we just raised forty actually for Fox Love. Yeah,
36:07
Speaker 2: we've raised about sixty million total now in the in
36:11
Speaker 2: the foreign RP years who have been running and you
36:14
Speaker 2: really looking to scale teams about fifty at Fox dot
36:17
Speaker 2: Manware Hir hiring pretty aggressively.
36:19
Speaker 1: And Harry's raised three or four million.
36:22
Speaker 3: Yeah, that's right.
36:23
Speaker 1: Icehouse Ventures involved in your rays.
36:25
Speaker 4: Yeah, so it was primarily US based venture capitalists, but
36:29
Speaker 4: super happy to have you know, the ice House there.
36:32
Speaker 4: They're on board as with as with Adrian, so it's
36:35
Speaker 4: really nice to have the New Zealand connection there, and
36:39
Speaker 4: I think as we start to think about deploying that
36:41
Speaker 4: it's primarily going to be deployed in the US, but
36:45
Speaker 4: we are wanting to sort of take advantage of some
36:47
Speaker 4: of the amazing kind of like three D artistry talent
36:50
Speaker 4: here in New Zealand as well, So looking to kind
36:52
Speaker 4: of maintain that connection.
36:54
Speaker 1: And Adrian and that's a pretty decent chunk of money.
36:57
Speaker 1: What sort of that runway? What does that enable you
37:00
Speaker 1: now to do.
37:01
Speaker 2: Really expanding a product team for us, where at this
37:04
Speaker 2: point you know that you hope to get to as
37:06
Speaker 2: a startup where you've got enough, You've demonstrated that you
37:09
Speaker 2: have a viable product, you have product like a fit,
37:11
Speaker 2: you have a lot of customers out there, and the
37:14
Speaker 2: customers want more. Essentially, there's a lot, you know, there's
37:18
Speaker 2: a lot to build here, There's an endless list of
37:20
Speaker 2: feature requests and so really for us, we could have
37:24
Speaker 2: kept scaling without raising money, but it would have really
37:28
Speaker 2: slowed down our product process. Right then you're really spending
37:30
Speaker 2: a lot more time selling than you are actually building
37:32
Speaker 2: the product. So for us, this leads us massively expand
37:36
Speaker 2: the engineering team and build a lot more of the
37:39
Speaker 2: roadmap that everyone keeps asking for.
37:41
Speaker 1: A bigger picture. We hear a lot I follow sort
37:45
Speaker 1: of geo strategists like Peter Zion, who's informing us at
37:48
Speaker 1: you declaring birth rate around where we're really going to
37:51
Speaker 1: need robotics if we want to have the standard of
37:55
Speaker 1: living that we currently do now, because literally we're not
37:57
Speaker 1: going to have the sort of work that we've enjoyed
38:01
Speaker 1: in the past. This is very apparent in places like
38:03
Speaker 1: China where the birth rate has plunged. Japan has already
38:06
Speaker 1: experienced this there massively into robotics. So yeah, interested in
38:13
Speaker 1: your thoughts about I guess the sort of the social
38:16
Speaker 1: contract around robotics. We're going to need more robotics. It's
38:19
Speaker 1: a practical necessity, but we want to have good robotics.
38:23
Speaker 1: We want to have a future of robotics that's good
38:26
Speaker 1: for society, good for workers.
38:29
Speaker 2: Friendly robots, Yeah, don't leave robots.
38:32
Speaker 1: Yeah, obviously we don't want terminator robots. But in terms
38:36
Speaker 1: of you know, the impact on the workforce and that,
38:38
Speaker 1: do you guys think about this when you're designing, I guess, Harry,
38:42
Speaker 1: particularly for simulation, you know this is a really valuable
38:45
Speaker 1: thing that you're adding to that you can actually talent scale.
38:48
Speaker 1: What's going to happen when you have hundreds or thousands
38:51
Speaker 1: of robots out there?
38:52
Speaker 4: Yeah, No, I think it's I think it's very forefront
38:56
Speaker 4: and kind of how we think about a lot of
38:58
Speaker 4: this stuff. I think we think about, you know, safety
39:02
Speaker 4: at a couple of angles. There's, of course, the practical
39:05
Speaker 4: safety of you don't want the robot to hurt or
39:08
Speaker 4: to kill somebody, and so for that, the more that
39:11
Speaker 4: you can do in terms of testing and evaluation is
39:14
Speaker 4: really critical. But there's also safety in the sort of
39:16
Speaker 4: more macro sense of like, you know, what does the
39:19
Speaker 4: sort of you know, revolution of this technology coming into
39:22
Speaker 4: society do for jobs and for the economy as a
39:27
Speaker 4: whole and so on that you know, I think it's
39:30
Speaker 4: going to take a sort of very carefully thought out
39:34
Speaker 4: rollout between the private sector and also the public sector
39:38
Speaker 4: to make sure that that's done in a way that
39:40
Speaker 4: is safe in that sense. I think I'm particularly excited
39:44
Speaker 4: about a couple of things. One is that it makes
39:47
Speaker 4: industries that were previously understaffed or uneconomic a lot more robust.
39:53
Speaker 4: And so that's everything from how do you think about
39:56
Speaker 4: scalably sorting recycling Like it's really depressing when you look
40:00
Speaker 4: at those sorts of metrics of all the stuff you
40:02
Speaker 4: put in the recycling, but how much is actually getting
40:04
Speaker 4: recycled versus thrown out because it's just so expensive and
40:07
Speaker 4: difficult to sort that stuff through to how we care
40:12
Speaker 4: for people with sort of like medical conditions or older
40:15
Speaker 4: people industries where we're really struggling to get the sort
40:19
Speaker 4: of volume in the workforce that we need there. But
40:22
Speaker 4: also I think you know, over the last fifty or
40:25
Speaker 4: so years in New Zealand and also in the US
40:28
Speaker 4: where Adrian and I are based, we've seen a lot
40:31
Speaker 4: of jobs and industries moving overseas for cost reasons, and
40:35
Speaker 4: that's really been I think quite a sort of a
40:38
Speaker 4: difficult trend from a social standpoint. It's leading to this
40:42
Speaker 4: eroding of the middle class in our countries, and I
40:45
Speaker 4: think we're kind of seeing a little bit of that
40:47
Speaker 4: play into the political environments as well in a way
40:50
Speaker 4: that that's not great. And I think that autonomy kind
40:53
Speaker 4: of gives us the opportunity to move some more of
40:55
Speaker 4: those primary industries back home, things like manufacturing, scaling agric
41:00
Speaker 4: culture again that sort of stuff.
41:01
Speaker 2: Yeah, this is one of the big sort of definitely
41:04
Speaker 2: one of the big tailwinds behind the robotics at the
41:06
Speaker 2: moment is macroeconomic and geopolitical factors. You've got a lot
41:11
Speaker 2: of this push to resure manufacturing things like that, You've
41:15
Speaker 2: got a lot of, as you mentioned, kind of aging workforce,
41:17
Speaker 2: and so you have these these driving factors that are
41:20
Speaker 2: encouraging the use of more automation, more autonomy. I think
41:25
Speaker 2: that when any new piece of technology comes out, there's
41:30
Speaker 2: always a group of people who are sort of saying,
41:32
Speaker 2: this time it's different, and you know, this is going
41:34
Speaker 2: to be the one piece of technology that suddenly gives
41:38
Speaker 2: humans normal purpose in life. And people have been saying,
41:41
Speaker 2: you mean, you go back to the early nineteen hundreds, right,
41:43
Speaker 2: you had like more than fifty percent of the population
41:45
Speaker 2: and employed in agriculture, and now there numbers like two percent.
41:49
Speaker 2: You go back and there are people, you know, as
41:51
Speaker 2: absolutely freaking out about every step along the way every
41:54
Speaker 2: piece of technology. People are like, oh my god, like
41:56
Speaker 2: nobody's going to have a job anymore. That has never
41:58
Speaker 2: been true. I fundamentally do not believe that it's going
42:01
Speaker 2: to be true with robotics, there were a long long
42:04
Speaker 2: way from the robots being able to manufacture and maintain themselves.
42:08
Speaker 2: And so in the meantime, like you know, the opportunities
42:12
Speaker 2: out there in the world are constantly changing. We have
42:14
Speaker 2: many jobs today that didn't exist one or two decades ago.
42:19
Speaker 2: Jobs are currently changed, They are always changing. It's always
42:22
Speaker 2: kind of opportunities for things like training and that. But yeah, overall,
42:28
Speaker 2: I mean it's technology has historically been a benefits for society.
42:32
Speaker 1: Yeah, that's every wave of technology has created more jobs
42:36
Speaker 1: than it has destroyed. I think that's proven to be true.
42:40
Speaker 1: But you know, autonomous vehicles is a good example of
42:43
Speaker 1: where California in particular was very proactive around regulation to
42:47
Speaker 1: allow all of those companies to test in real world
42:51
Speaker 1: environments and that accelerated the technology. When it comes to
42:54
Speaker 1: something like humanoid robots moving in unstructured sort of environments,
43:00
Speaker 1: is there going to need to be a lot of
43:02
Speaker 1: changes happening in the US around regulation and even I
43:06
Speaker 1: guess labored dynamics. You've got strong unions in some industries
43:10
Speaker 1: over there, is there a lot of non tech sort
43:13
Speaker 1: of blockers that are going to have to be removed
43:15
Speaker 1: to really allow the number of robots that are deployed
43:19
Speaker 1: to really increase the difference.
43:21
Speaker 2: The big difference distinction I would draw with self driving
43:24
Speaker 2: vehicles is the fundamentally putting a lot of other people's
43:26
Speaker 2: lives at risk with them. And so when you think
43:29
Speaker 2: about robots that are being deployed into manufacturing facilities or
43:33
Speaker 2: things like this, these are controlled environments, right, They know
43:35
Speaker 2: every single person that's checking in and out, They train
43:38
Speaker 2: those people, and so it's not you know, regulation needs
43:43
Speaker 2: to step in when you're sort of putting other people's laser. Certainly,
43:46
Speaker 2: unions are concerned, but again that's really a concern between
43:50
Speaker 2: these companies and the unions and the technology that they're
43:54
Speaker 2: looking to deploy. But I don't see certainly and some
43:58
Speaker 2: of these things like manufactur during and logistics and things,
44:01
Speaker 2: regulation being a huge, you know, kind of block or anything.
44:06
Speaker 2: Certainly in the way that it was, and I'm not
44:08
Speaker 2: even saying regulation was a blocker for self driving, but
44:11
Speaker 2: it was certainly a fact that there was a lot
44:12
Speaker 2: of education that needed to be done across the industry.
44:17
Speaker 2: In the US, it's quite lucky that there's a lot
44:20
Speaker 2: of different states and we could let these kind of
44:22
Speaker 2: approaches play out and see how I worked in some
44:24
Speaker 2: states that were willing to be more more pro testing.
44:28
Speaker 2: But that is findamentally something where you're driving out in
44:31
Speaker 2: public roads and you're putting public lives at risk, which
44:34
Speaker 2: which those people crossing the street or with their kids
44:37
Speaker 2: playing on the street didn't sign up for.
44:40
Speaker 4: Yeah, I think that distinction is spot on, and I
44:43
Speaker 4: think there's also sort of a symbiosis between you know,
44:46
Speaker 4: regulations enacted in response to some sort of thing going wrong,
44:50
Speaker 4: and obviously things go wrong with cars all the time,
44:52
Speaker 4: so we have a sort of well defined regulatory framework
44:56
Speaker 4: generally speaking for how we kind of think about mitigating
44:58
Speaker 4: and managing and license for those sorts of risks. And
45:01
Speaker 4: so I think I totally agree with Adrian that I
45:05
Speaker 4: don't sort of see that happening in a lot of
45:07
Speaker 4: these other industries where we're kind of seeing the adoption
45:12
Speaker 4: of autonomy. But I think it's also critical that we
45:15
Speaker 4: sort of take it upon ourselves to make sure that
45:17
Speaker 4: that rollout is safe less things go wrong, and then
45:20
Speaker 4: you kind of start to see more of that response
45:22
Speaker 4: come in there as well.
45:23
Speaker 1: Harry, you know, if we played this episode back in
45:25
Speaker 1: twenty thirty five, what do you think when it comes
45:29
Speaker 1: to robotics, we're going to see that was sort of inevitable.
45:33
Speaker 1: But also conversely and interest in your views as well,
45:36
Speaker 1: Adrian on stuff that gets a lot of hype, but
45:39
Speaker 1: it's just going to be so much slower to roll out.
45:41
Speaker 2: Everything takes a lot longer to roll out than people realize,
45:45
Speaker 2: especially or more so in the physical world. So everyone,
45:49
Speaker 2: there's a lot of talk right now about the chat
45:51
Speaker 2: GPT moment for robots. Basically right, we hit this point,
45:55
Speaker 2: as Harry said, back in twenty twenty three where chat
45:57
Speaker 2: Gibt came out and it changed everything. Literally the day
46:00
Speaker 2: came out, I spent like twenty four hours gillingto my computer,
46:03
Speaker 2: like chatting to this thing and just being like, what
46:05
Speaker 2: the hell did they just invent? But then this is
46:07
Speaker 2: a purely digital product, right, Like scaling it was challenging
46:10
Speaker 2: and throwing together data centers and you know, nuclear actors
46:14
Speaker 2: and things as fast as possible to scale it. But
46:16
Speaker 2: like digital things are very much a lot lot easier
46:20
Speaker 2: to scale in the physical world. It just you know,
46:25
Speaker 2: there are real challenges manufacturing and operating. My example here
46:28
Speaker 2: would be Weimo, right, like what the constraint on Weimo
46:31
Speaker 2: scaling now is not the autonomy SEFWMO has software that
46:34
Speaker 2: can drive around cities, that can drive on freeways. Now,
46:38
Speaker 2: even if Wemo, even if even if you made Wave
46:41
Speaker 2: the magic one I said, Weymo's driving is absolutely perfect
46:43
Speaker 2: and and available and it never makes a single mistake.
46:46
Speaker 2: It would still be an enormous challenge to roll that
46:48
Speaker 2: out to you know, enormous capital expense and almost operational
46:52
Speaker 2: challenge of rolling that out to millions and millions of
46:54
Speaker 2: vehicles across one hundred cities, to have cleaning crews that
46:57
Speaker 2: are servicing those vehicles every night, because that's not automated yet.
47:01
Speaker 2: Just you know, the logistical challenge of rolling out millions
47:04
Speaker 2: of anything is difficult. And so I think the thing
47:08
Speaker 2: that we're going to see over the next few years
47:09
Speaker 2: of Eurotics system incredible breakthroughs made from a technology perspective,
47:13
Speaker 2: and they're going to see robots solving things that previously
47:15
Speaker 2: were way way out of reach of robots. But that
47:18
Speaker 2: doesn't mean that they're going to be deployed by the
47:20
Speaker 2: millions anytime in the next few years, right, it's going
47:23
Speaker 2: to take That's the thing that by twenty thirty five,
47:26
Speaker 2: we will absolutely be into those numbers and we'll start
47:29
Speaker 2: seeing millions and then tens of millions and hundreds of
47:31
Speaker 2: millions of robots. We have billions of smartphones today, so
47:34
Speaker 2: certainly like humanity knows how to manufacture billions of things,
47:38
Speaker 2: but it takes a while to like ramp up those
47:40
Speaker 2: production chains and to even like operate that many of anything.
47:43
Speaker 4: I completely agree with that. I also think that I
47:47
Speaker 4: think a little bit about morphology and ownership, and so
47:51
Speaker 4: I guess what I mean by that is, you know,
47:54
Speaker 4: I think right now there's a ton of excitement, particularly
47:56
Speaker 4: around humanoid form factors, and I think this also excitement
48:00
Speaker 4: around this idea that like we might all have humanoids
48:03
Speaker 4: in our homes kind of performing all these tasks, we
48:06
Speaker 4: sort of own these these sorts of machines. I think
48:10
Speaker 4: it's early days. I think it's hard to say with
48:12
Speaker 4: with with too much conviction, but generally speaking, I think
48:16
Speaker 4: my view is that there's going to be a ton
48:18
Speaker 4: of autonomy out there. I think most of it is
48:20
Speaker 4: not going to look human, even in our kind of
48:23
Speaker 4: home environments. So smart vacuum cleaners, you know, smart lawn mowers,
48:28
Speaker 4: all of these sorts of things that that kind of exist.
48:31
Speaker 4: I think, you know, there are a ton of returns
48:33
Speaker 4: to specialization. We've kind of seen that since Henry Ford
48:37
Speaker 4: kind of invented there the production line, and so I
48:40
Speaker 4: think as these starts of these types of things that
48:43
Speaker 4: have start to roll out in the world, my bed
48:45
Speaker 4: is that that's probably what we're going to see. First's
48:48
Speaker 4: like a lot more point solutions solving very very specific problems.
48:53
Speaker 3: And they may not all be.
48:54
Speaker 4: Owned by us at our homes in the way that
48:56
Speaker 4: we kind of expect. They might also be owned by
48:59
Speaker 4: companies who are us in these things to you know,
49:01
Speaker 4: like a few you know, have a gardener or something
49:04
Speaker 4: like that. Maybe they're kind of bringing a smart lawn
49:07
Speaker 4: mower to you, or a smart hedge trim or to
49:09
Speaker 4: you or something like that, rather than us necessarily all
49:12
Speaker 4: having all of these different robots in our homes there.
49:16
Speaker 1: Yeah, it's interesting. I've just been traveling recently in Europe
49:19
Speaker 1: in the US and just you know, just seeing industrial
49:23
Speaker 1: vacuum cleaner robots starting to wander around airports, seeing robots
49:29
Speaker 1: picking up plates from around restaurants. So it's it's just
49:32
Speaker 1: becoming and diners are just ignoring them. As they zoom around,
49:35
Speaker 1: you know. So it's it's interesting how just that sort
49:38
Speaker 1: of stuff was a novelty just a few years ago.
49:40
Speaker 1: Now it's just part of the fabric of everyday life,
49:43
Speaker 1: and that will that will eventually be the same for
49:46
Speaker 1: humanoids as well.
49:46
Speaker 3: Totally crue.
49:47
Speaker 2: Yeah, the expectation has changed very quickly, right yeah, I
49:50
Speaker 2: mean it's just a complete non event now seeing a
49:53
Speaker 2: way mow. I mean you go for a walk around
49:56
Speaker 2: San Francisco, I mean just you probably see ten waymos
49:59
Speaker 2: over the space of three minutes. It's they're just everywhere,
50:01
Speaker 2: and so it's a car with no one driving. It
50:04
Speaker 2: is now not even worth looking out from your phone
50:06
Speaker 2: for and that can change in the span in five years. Right,
50:11
Speaker 2: so we'll definitely agreeing with Harry. We'll see some of
50:14
Speaker 2: these use cases. We see a lot of robots to
50:16
Speaker 2: plod largely you know, initially probably point solutions, and within
50:21
Speaker 2: the span of a year or two, people are going
50:22
Speaker 2: from thinking it's normal to just completely ignoring it and
50:25
Speaker 2: moving on with their life.
50:26
Speaker 4: I was shocked by you know, writing in these way
50:29
Speaker 4: more cars and also in like Tesla and cruise powered
50:33
Speaker 4: kind of vehicles too, Like I think you sort of
50:36
Speaker 4: assume that it's going to be this like crazy experience
50:39
Speaker 4: when you're inside this thing, and it's going to be
50:41
Speaker 4: this constant sense of novelty. But I was just struck
50:44
Speaker 4: by her. After about forty five seconds, I was completely
50:46
Speaker 4: bored of the fact that I was in an autonomous vehicle.
50:48
Speaker 4: I was back on my phone, you know, tapping away
50:50
Speaker 4: on Reddit or whatever it was. I think that sort
50:54
Speaker 4: of normalcy. Yeah, I think I completely agree with that point.
50:58
Speaker 1: I had exactly the same experience. Have been to San
51:00
Speaker 1: Francisco two years in a row, so I've had about
51:02
Speaker 1: maybe a dozen rides and waymos and it's just second
51:06
Speaker 1: nature now, So you know that that is the way
51:10
Speaker 1: it is going to go. Just finally, guys, you know
51:13
Speaker 1: you've found your way into physical AI and robotics via
51:19
Speaker 1: you know, a really interesting journey through software development and
51:22
Speaker 1: the like. If you know, if a young Kiwi engineer
51:26
Speaker 1: came to you and said, I really want to get
51:28
Speaker 1: into robotics, what's the pathway that you would recommend. Not
51:32
Speaker 1: a lot of that going on in New Zealand. We
51:34
Speaker 1: do have a couple of interesting companies, but would you
51:37
Speaker 1: would your recommendation be follow that path that you did,
51:40
Speaker 1: try and get to a larger market as soon as possible,
51:42
Speaker 1: and try and work for one of those great companies
51:47
Speaker 1: that employs a lot of robotics engineers.
51:49
Speaker 2: I would say that even putting aside of robotics, if
51:54
Speaker 2: you're if you're young and ambitious and trying to break
51:58
Speaker 2: into anything, you need to get to working at a
52:01
Speaker 2: high growth, a high growth tech company that is at
52:06
Speaker 2: the front of their industry. And so that's definitely true
52:09
Speaker 2: in robotics, it's true elsewhere, and I really, I really
52:12
Speaker 2: encourage people to if there are, like you said, a
52:15
Speaker 2: few companies in New Zealand that you can get to,
52:18
Speaker 2: but you need to be at a company that's growing quickly.
52:21
Speaker 2: You know, this is like ideally that's more than doubling
52:24
Speaker 2: year of a year of their revenue or their headcount
52:28
Speaker 2: that is growing fast, and you can get a front
52:30
Speaker 2: receipt to that. So there are very few companies like
52:33
Speaker 2: that in New Zealand. There are a lot in the States.
52:35
Speaker 2: It's quite hard to get a visa for the States,
52:37
Speaker 2: but if you care about it, make it happen. There
52:39
Speaker 2: are ways within robotics specifically. The one thing I will
52:45
Speaker 2: add is that people underestimate how broad the field of
52:48
Speaker 2: robotics is, and so at every robotics company, like you
52:51
Speaker 2: don't need a robotics PhD. Yeah, there are mechanical engineers,
52:55
Speaker 2: there are electric engineers, there are AI engineers, there are
52:57
Speaker 2: data engineers. There are front and engine is building interfaces
53:01
Speaker 2: for the things. There are legal teams, there are operations teams,
53:04
Speaker 2: there are product managers. You know, there are so many
53:07
Speaker 2: roles at these companies, and so I think people that
53:10
Speaker 2: I talk to sometimes get too hung up on thinking
53:13
Speaker 2: I wish I had studied robotics in college. It doesn't matter.
53:16
Speaker 2: I didn't study robotics in college. Now I've been doing
53:18
Speaker 2: it for ten years and I know a lot. But
53:21
Speaker 2: you know, find an entry point at those companies where
53:23
Speaker 2: the skill set that you do have, these companies are
53:25
Speaker 2: hiring for dozens or hundreds of different roles. That would
53:29
Speaker 2: be my main kind of robotic specific point.
53:31
Speaker 5: Yeah.
53:31
Speaker 4: I don't have too much to add to that because
53:33
Speaker 4: I think that's a great answer. But generally speaking, I
53:36
Speaker 4: think you want to see what great looks like up
53:40
Speaker 4: close and personal as much as possible. By growth companies
53:45
Speaker 4: are exactly one fantastic way to do that. But I
53:48
Speaker 4: think at every step in the journey, right, whether it
53:51
Speaker 4: is the kinds of classes that you take, the things
53:54
Speaker 4: that you're learning, who you're kind of spending your time with,
53:57
Speaker 4: the sort of bigger, more competitive, scarier the rumors that
54:01
Speaker 4: you put yourself in, that kind of thing compounds very quickly,
54:05
Speaker 4: and so that would kind of be my two cents.
54:08
Speaker 2: Yeah, if you're if you're if you're fifteen in a
54:11
Speaker 2: high school, follow Harry's path, not mine, and try and
54:13
Speaker 2: get into Stanford, because well massive university was great, but
54:17
Speaker 2: it did not you know, this is coming back to
54:20
Speaker 2: Harry's point about put yourself in the room with the
54:22
Speaker 2: smartest people in the world.
54:23
Speaker 1: Great advice. Well, thanks so much, guys. It's been a
54:26
Speaker 1: fascinating conversation. You're working an incredible dynamic area of tech development.
54:33
Speaker 1: Good luck for things, and twenty twenty six is going
54:36
Speaker 1: to be a big year. I think for AI continue
54:39
Speaker 1: to be a big year. Twenty twenty five was, but
54:41
Speaker 1: marrying that with some of the you know, the advances
54:45
Speaker 1: in robotics is going to be really interesting to watch.
54:47
Speaker 1: So we'll keep a close eye on Fox, Glove and Antioch.
54:50
Speaker 1: And thanks so much for coming onto Business of Tech.
54:53
Speaker 5: Thanks Peter, thanks very much for having us. Fantastic So
55:06
Speaker 5: there you go.
55:06
Speaker 1: Really decent conversation with Harry Malsop from Antioch and Adrian
55:10
Speaker 1: McNeil from Foxglove two. Kiwis quietly building this scaffolding that
55:14
Speaker 1: will let robots move into the mainstream. Take away from
55:18
Speaker 1: me from this conversation, it's just how central data simulation
55:22
Speaker 1: and observability to deploying robots safely, From simulating thousands of
55:28
Speaker 1: edge cases and software to building the data flywheels that
55:32
Speaker 1: show you not just what a robot did, but what
55:35
Speaker 1: it thought it was doing. So I'll leave it there.
55:38
Speaker 1: It's been a longer episode. Thanks so much for listening.
55:41
Speaker 1: Thanks to two degrees for sponsoring the Business off Tech show.
55:45
Speaker 1: Notes are on the Business Desk website. Just go to
55:48
Speaker 1: Businessdesk dot co dot nz and I'll see you back
55:50
Speaker 1: here next week with another episode of the Business of Tech.
55:54
Speaker 1: Catch you then,