Latest / Automated with Brian Heater / Bren Pierce on Why Humanoid Robots Are Overhyped and What Actually Works in Robotics

Transcript

Bren Pierce on Why Humanoid Robots Are Overhyped and What Actually Works in Robotics · Automated with Brian Heater

0:00
SPEAKER_01: So I didn't realise the CEO wanted a photo shoot.
0:02
SPEAKER_01: If you told me that out front, I could have just gone there with a robot and you could have charged me$5,000.
0:06
SPEAKER_00: In terms of just cool factor, have they progressed faster or more slowly?
0:13
SPEAKER_01: In three years, you're not gonna see that humanoid robot in your house and be like, make me a cup of coffee.
0:19
SPEAKER_01: You see that in videos, but that is because you've got 300 engineers just to make the robot take one step if you push it too hard.
0:27
SPEAKER_01: Nowadays you can do backflips and karate kicks.
0:30
SPEAKER_01: It's quite incredible.
0:31
SPEAKER_01: Like, how do you persuade them to come to the startup when they can go and work at Google, work at Nvidia, or some of these big well-funded companies that are gonna pay you half a million dollars a year, so it's quite hard to compete for talent there.
0:41
SPEAKER_01: It's scary but interesting.
0:43
SPEAKER_01: It's when you land and you're literally like, I have nowhere to live, I have a rucksack of my stuff, you'll rent a car, you go and start living for four weeks in a motel, and if you don't have that DNA to actually get out of your house, I was forced to literally knock on the door of 50 hotels and be like, Can I speak to the manager, please, to get it to be so generalistic, reliable, and just work.
1:06
SPEAKER_01: The classic 8020, we're probably 80% towards it, but that last 20%, no one knows.
1:15
SPEAKER_00: Hello everyone, welcome to Automated.
1:17
SPEAKER_00: I'm Brian Keater, the managing editor at the Association for Advancing Automation, and the guy who will be asking open-ended questions for the next 45 minutes or so.
1:26
SPEAKER_00: This week we are talking with Bren Pierce of Kineci Robotics.
1:30
SPEAKER_00: Bren is also uh co-founder of Bear Robotics, whose systems you might uh they might have brought you some ramen to your table at a busy restaurant.
1:38
SPEAKER_00: That said, I think we actually had our first extended talk on uh the occasion of Kinecy winning uh the startup competition at last year's Automate.
1:47
SPEAKER_00: Bren will also be back at this year's show, by the way, as part of our humanoid robot forum.
1:53
SPEAKER_00: If you've been enjoying the show, it would mean a lot if you liked and subscribed to our weekly newsletter over at automated.fm.
2:00
SPEAKER_00: And with that, please enjoy this conversation with Bren Pierce.
2:10
SPEAKER_00: You know, we talk a lot about what's coming up next in automation on the show, but if you really want to see the future motion, you've got to be there in person.
2:18
SPEAKER_00: Automate 2026 is where the world's leading innovators, builders, and dreamers come together to show you what's possible.
2:25
SPEAKER_00: Robots AI, machine vision, motion control, you name it, all automation under one roof.
2:31
SPEAKER_00: And as part of Automate this year, the Humanoid Robot Forum brings together leaders, engineers, and researchers for a two-day deep dive into the real world development, deployment, and commercialization of humanoid robotics.
2:44
SPEAKER_00: Register for free at automateshow.com to join us in Chicago June 22nd through the 25th.
2:51
SPEAKER_00: We will see you there.
2:52
SPEAKER_00: Does Kinesi identify as a US company or a UK company?
2:59
SPEAKER_01: Great question.
3:00
SPEAKER_01: I think it's more of a hybrid question.
3:02
SPEAKER_01: So we have our sales marketing uh Delaware corporation in America, but I'd say 95% of all our engineering talent is actually based in the UK.
3:11
SPEAKER_01: So it's uh best of both worlds, one might say.
3:15
SPEAKER_00: Yeah, I feel like you should lean into the UK thing, you know.
3:17
SPEAKER_00: It's a growing ecosystem right now.
3:19
SPEAKER_00: There's a few uh humanoid companies out of there that should be sort of part of your signature.
3:25
SPEAKER_01: I think I think it's been English as well.
3:27
SPEAKER_01: I think there's great talent here.
3:29
SPEAKER_01: You have like uh Edinburgh, Cambridge, London, top talent.
3:34
SPEAKER_01: Just historically, apart from a cardo, uh, there just hasn't been that big robotic companies that come out of it.
3:41
SPEAKER_01: I think that's a mixture of the VC ecosystem.
3:45
SPEAKER_01: You know, they're not really that geared towards deep tech uh like Silicon Valley is.
3:50
SPEAKER_01: What we've done is we've got to raise money in Silicon Valley, uh, then just have our tech talent in the UK, which I think is a is the best of both worlds.
3:58
SPEAKER_00: Yeah, it's interesting too, because you know, looking through your own CV, you're you've had a pretty diverse geographical journey of your own, I would say.
4:07
SPEAKER_00: You know, you spent time in uh at CMU, so obviously you were out here.
4:11
SPEAKER_00: I think Pittsburgh is one of those places that probably identifies a little bit more Midwest than East Coast, but um, and then you know, uh West Coast too, uh Bear Robotics for a number of years, and they're they're a Korean company.
4:24
SPEAKER_00: So you've kind of you've you've had your hand in a lot of different robotics ecosystems.
4:31
SPEAKER_01: Yeah, that's um primarily historically the um robotic community or place that did it was actually quite small.
4:39
SPEAKER_01: So you get Munich, you get Carnegie Mellon University, Bristol has the UK system, you have uh Keist in uh Korea.
4:47
SPEAKER_01: Up until about 15 years ago, I don't think there was that much robotics actually in the Bay Area.
4:52
SPEAKER_01: It's only about 10 years ago when the Ross community and Willow Garage actually got founded that I think the uh Bay Area really started to become a magnet to robotics.
5:03
SPEAKER_01: I think advice for founders as well, the Bay Area has such good talent.
5:07
SPEAKER_01: There's companies from Fetch, you know, all these companies that have actually built companies, you have to sort of go there, meet the ecosystem.
5:16
SPEAKER_01: Like obviously, uh Ross Foundation is based there, and the great companies have been built out of there.
5:21
SPEAKER_01: Like Bare Robotics uh was actually uh primarily based in the Bay Area.
5:26
SPEAKER_01: One downside and why uh Bare Robotics, my previous company, actually hired massively outside of it, primarily in Korea.
5:33
SPEAKER_01: You're just competing for talent in the Bay Area, which means the salaries are high.
5:38
SPEAKER_01: Uh, it's just harder to attract.
5:39
SPEAKER_01: Like, how do you persuade them to come to the startup when they can go and work at Google, work at Nvidia, or some of these big well-funded companies that are going to pay you half a million uh dollars a year.
5:49
SPEAKER_01: So it's quite hard to compete for talent there.
5:51
SPEAKER_01: That's also why we uh Khisi actually hired in the UK.
5:55
SPEAKER_01: I think there's only the humanoid company in London, all three, and Acado is really the big companies here that are actually hiring, let's say, the next wave of uh foundational AI uh graduates.
6:08
SPEAKER_00: Yeah, Bear in particular, I mean, it really does have that, you know, hearing you discuss the early days, it really does have that Bay Area garage startup feel connected to the Ross community of really going out, buying a turtle bot and kind of building building a company like on the back of this robot.
6:27
SPEAKER_01: A hundred percent.
6:28
SPEAKER_01: This is why I was part of that originally, like my two co-founders, Feng Wei and John, both ex-googlers, incredibly good computer scientists, they went and bought a turtle bot.
6:38
SPEAKER_01: Within a couple of days, they're like, all right, now what do we do?
6:41
SPEAKER_01: That's when I started to chat with them.
6:43
SPEAKER_01: And we actually built uh bear robotics out of uh my co-founder John's restaurant.
6:48
SPEAKER_01: So me and Feng Wei, you know, would be sat there coding during the day, and then night shift, well, not night shift, but when the restaurant was uh serving customers in the evening, we'd actually run the robot to the customers.
7:01
SPEAKER_01: And our very first, let's say, product was literally just taking the uh bill and running it to someone's table and then bringing back their credit card back to the till.
7:10
SPEAKER_01: And that actually delighted the staff.
7:13
SPEAKER_01: They didn't have to now walk to that table when someone waves at you.
7:16
SPEAKER_01: They could just put the uh bill on the turtle bot, send it, and within about two, three weeks, we actually had a whole system up that was just working.
7:23
SPEAKER_01: And that sort of gave us the confidence to be like, all right, let's go raise some money, let's start building our custom hardware and let's start commercializing this product.
7:31
SPEAKER_00: To a certain extent, that's a product of having this direct connection to the restaurant industry.
7:36
SPEAKER_01: 100%.
7:36
SPEAKER_01: I think that is a big advantage we had there because John, the CEO, he had a restaurant, he realized firsthand it's actually very hard to hire staff, you know, they're quite unreliable, people just don't turn up to work, and you're also, you know, paying$20 an hour for staff.
7:52
SPEAKER_01: And he was like, How can I actually solve this?
7:55
SPEAKER_01: So he actually had the original idea to be like, let's actually go and buy a turtle bot.
7:59
SPEAKER_01: But then I came in as the roboticist with my 10 years' experience with Ross, with the navigation stack, understanding actually, you know, occupancy grids and things like that to actually help build out the robotic side of it.
8:11
SPEAKER_01: Oh, that's interesting.
8:12
SPEAKER_00: So, so he was just looking to like buy something just to help them in-house.
8:16
SPEAKER_00: They weren't looking to like commercialize outside of that specific restaurant, really.
8:20
SPEAKER_01: I think originally he just thought about his own restaurant, and then very soon he was just like, oh, this is uh a problem a lot of my uh people he knew in the restaurant industry actually faced.
8:31
SPEAKER_01: And that's where me and Feng Wei, uh, my other co-founder came in to help him actually build out the whole system.
8:37
SPEAKER_00: So you sort of started in software, SaaS, then went over to CMU and started studying robotics.
8:45
SPEAKER_00: Is that fair?
8:46
SPEAKER_01: Yeah, so all my life I'd be building robots as a hobby.
8:50
SPEAKER_01: Uh, you know, I've been obsessed with just Star Wars, I think, too many times, you know, watching it too many times.
8:55
SPEAKER_01: Yeah.
8:56
SPEAKER_01: So when I graduated originally uh with a computer science degree, there wasn't really a, at least in England, a robotic industry.
9:03
SPEAKER_01: It didn't even cross my mind to take what at the time was my hobby and commercialize it.
9:09
SPEAKER_01: So I actually went out, found customers, bootstrapped like a SaaS company, primarily actually for call centres to automate the whole call centres for charities, for all sorts of things.
9:21
SPEAKER_01: And then about four or five years into that, Bristol University actually opened the first robotics lab, Bristol Robotics Lab.
9:30
SPEAKER_01: I read about it in the paper.
9:31
SPEAKER_01: I'm like, wow, I could actually turn my hobby into a profession.
9:35
SPEAKER_01: So I luckily could sell my company to our biggest customer, and then I was like, all right, let's go in.
9:40
SPEAKER_01: I spent two years and did a master's in robotics.
9:43
SPEAKER_01: But yet again, when I graduated and you look around, this was probably about 2008, you look at the landscape, and there wasn't really a robotic industry in the world.
9:52
SPEAKER_01: There might be Couka, ABB doing like pick and place in factory, but at the time my passion was like humanoid robotics, control, and there just wasn't this uh industry to go and actually work out of.
10:04
SPEAKER_01: So, this is luckily I went to Carly Gamellan University, worked with Chris Axon there, working like the Sarkoze robot, worked with Samsung, worked with Honda on this early humanoid robot, and that allowed me to actually break into the professional space and get paid to do my obsession, which has been robotics now for 20 years, which I'm very grateful for.
10:25
SPEAKER_00: I wonder if that's one of those stories, though, that is like that that's obvious and that's easy to tell in hindsight.
10:32
SPEAKER_00: But at the time, although robotics was your passion and something that you had been interested in, as you said, your entire life, selling your company and going back to get your masters, like that's a big risk you're taking right there.
10:46
SPEAKER_01: Well, if you look at where I what I've done my whole life, everything's a risk, you know.
10:50
SPEAKER_01: Going to, you know, a great story, and I say this to some people graduating masters, like how I ended up at CMU, graduated with a master in robotics.
11:01
SPEAKER_01: I'd built for my uh master thesis like a 30 centimeter, one foot-high humanoid robot that I made walk.
11:09
SPEAKER_01: There just wasn't anything there to hire me.
11:11
SPEAKER_01: So I literally emailed all the professors around the world that does something interesting.
11:17
SPEAKER_01: And I'm like, I will come and work for you for six months for free.
11:20
SPEAKER_01: And Chris Atkinson from CMU said, Great, this sounds like a perfect opportunity.
11:25
SPEAKER_01: And this is where you have to take life in uh risks.
11:28
SPEAKER_01: This is before, you know, we didn't even do a Skype interview.
11:31
SPEAKER_01: I just sent him an email.
11:32
SPEAKER_01: He replied, like, yes, can you start on like I think it was like the 3rd of October?
11:37
SPEAKER_01: And I'm like, great, do I have to do anything?
11:39
SPEAKER_01: He's like, my EA will organize it.
11:41
SPEAKER_01: And I'm like, all right, within two emails, he's just like, come, here's an address.
11:44
SPEAKER_01: I'll see you at literally this room in Pittsburgh in like four weeks' time.
11:50
SPEAKER_01: I'm like, I great, I guess I'm moving to Pittsburgh.
11:54
SPEAKER_00: What was that ramp-up process like?
11:56
SPEAKER_00: You know, moving, moving to a different country, moving, you know, like halfway across the world like that.
12:01
SPEAKER_01: It's scary but interesting.
12:03
SPEAKER_01: It's when you land and you're literally like, I have nowhere to live, I have a rucksack of my stuff, you rent a car, you go and start living for four weeks in a motel four, you know, a motel six story on the side of a freeway, and you're just like, all right.
12:17
SPEAKER_01: So the EA, the EA didn't set up that part of it.
12:20
SPEAKER_01: No, that was me doing, I think she would have done it a bit better than just like a random uh motel on the side of a road.
12:25
SPEAKER_01: And you're like, all right, new place.
12:28
SPEAKER_01: It's also scary, let's be honest.
12:29
SPEAKER_01: You're like, you're walking around the campus, knocking on a door, and you're like, hello, I'm Brent.
12:33
SPEAKER_01: What do I do now?
12:34
SPEAKER_01: He's like, here's a uh a PhD student, here's a desk, have fun.
12:39
SPEAKER_01: Here's a, I think at the time, like a million-dollar humanoid robot sarcos, do something cool.
12:47
SPEAKER_00: Yeah, and and that's that's the good outcome because my big worry is I would I would, you know, do all this, I would land there, I would get the car, I would get the hotel room, I would knock on the door, and he would be like, Who?
12:59
SPEAKER_01: Yes.
13:01
SPEAKER_01: That is the big scare.
13:02
SPEAKER_01: You're like, you're on the play, you're like, I've never even spoken to him on a video call.
13:06
SPEAKER_01: This was like 2008, you know.
13:08
SPEAKER_01: I think Skype was a thing, but no one really did that.
13:10
SPEAKER_01: It was all over email.
13:12
SPEAKER_01: And I should say, in the intermediate time, after he's like the EA, I did actually then use Carnegie Mellon's department to get me a visa, so you had to get a level one visa.
13:20
SPEAKER_01: So there was like the university involved, but it was literally with Chris, it was like a couple of emails.
13:25
SPEAKER_01: I see it, and then when you land, you're like, What do you want to me to do, Chris?
13:28
SPEAKER_01: He's like, Here's the Sarko's robot, do something cool with it.
13:31
SPEAKER_01: So then I got to work with some great talent, do some controllers, get it to work.
13:36
SPEAKER_01: There was also a project with like Honda and Sarkozy, where it's like, all right, here's some other robot companies, get it to do something interesting.
13:44
SPEAKER_00: Well, okay, so you know, I you you you get Honda coming along, and that gives you some some kind of parameters or or a goal to work towards.
13:52
SPEAKER_00: But, you know, day one, somebody gives you the keys to a very expensive robot and says, do something cool with it.
14:00
SPEAKER_00: Like, how do you how do you begin that process?
14:03
SPEAKER_01: Yeah, look, well, luckily, or they don't exactly give you the keys to the expensive robot on day one.
14:09
SPEAKER_01: They do have a nice simulation program.
14:11
SPEAKER_01: So you're just like, all right, let's ramp up, install the software, figure out how their software stack works, figure out the controller, and just start like writing controllers.
14:20
SPEAKER_01: Back then, wasn't it doing anything like Boston Dynamics and backflip?
14:24
SPEAKER_01: It was very much like, can I just get this robot to balance?
14:26
SPEAKER_01: So when you push it, you know, it doesn't fall over.
14:29
SPEAKER_01: It's more like a two-free link pendulum, just trying to get that balancing working.
14:34
SPEAKER_01: That's why I primarily spent a year working on like putting a balancing board on the floor, pushing the robots around, and really going from let's call it a toy humanoid robot uh that just has like servo motors you do, precision control, to a fancy robot with force and torque, and actually, like, can you actually write a proper controller using all the force and information available to these uh expensive robots that actually just get it to balance?
15:00
SPEAKER_00: That's funny.
15:01
SPEAKER_00: So, you know, cool by uh 2008 standards would have been don't fall over.
15:07
SPEAKER_01: Exactly.
15:08
SPEAKER_01: Like um Ben, who I think he's one of the head of uh Bosso Dynamics Control now.
15:14
SPEAKER_01: Even he's been he was there for like three years, and he his PhD work was just like push it with a uh a pole and make sure the robot, I think his big breakthrough was the robot would make one step forward uh if you push it too much.
15:25
SPEAKER_01: And I think he spent like five, six years with his PhD just to make the robot take one step if you push it too hard.
15:32
SPEAKER_01: How and nowadays you can do backflips and karate kicks.
15:35
SPEAKER_01: So how the control has uh evolved in the last 20 years is quite incredible.
15:39
SPEAKER_00: Yeah, when uh when you look back on a time period, like especially comparing, I mean, this is interesting, you know, as we're recording this, it's the tail end of CES.
15:47
SPEAKER_00: So obviously, like this was the big, you know, we saw the new Atlas, you know, we saw a bunch of um humanoid robots created out on stage.
15:56
SPEAKER_00: In terms of like capabilities, in terms of just let's not even talk about scaling and doing stuff in the real world, but in terms of just cool factor, have they progressed faster or more slowly in the long run than you expected them to in that time period?
16:13
SPEAKER_01: I think the last three years it's just been incredible.
16:18
SPEAKER_01: For about the first 15 years, it was just like slowly incrementalism.
16:22
SPEAKER_01: They, you know, you're walking.
16:24
SPEAKER_01: I think you've got um Jonathan Hearst, the agility, looking at like putting in springs in it, and a lot of my PhD work was like serialastic activators and all of this, and it was like slowly incrementalism.
16:38
SPEAKER_01: And then three years ago, when you're suddenly like, we're just gonna use reinforcement learning, we're gonna put a robot in simulation, we're gonna have 20,000 of them that spend the next million years in simulation to learn to control.
16:51
SPEAKER_01: It suddenly was that's a big unlock.
16:54
SPEAKER_01: It's like before then, you'd get incredibly complex algorithms where people are like, yeah, you know, 20, 50 uh different algorithms trying to work together.
17:03
SPEAKER_01: You have a balancer, you're trying to work out the center of mass, inertials, all of this sort of like complexity, just to balance and make one or two steps, and then suddenly you put that all into an AI system that has no real understanding of physics, but you just brute force incredible controllers that could do backflips.
17:23
SPEAKER_00: That's just so it's it's not just like a marketing thing, it's not just people's imagination, like something has really been unlocked in the last few years.
17:32
SPEAKER_01: Oh, a hundred percent.
17:33
SPEAKER_01: It's I think this reinforcement, you have to break it down.
17:36
SPEAKER_01: I think there's controls.
17:37
SPEAKER_01: So can you walk?
17:39
SPEAKER_01: Can you move your arms?
17:41
SPEAKER_01: That is basically solved now with reinforcement learning.
17:44
SPEAKER_01: The next frontier is gonna be manipulation, because obviously now, you know, it it's in a way, it's like these RL learn policies, they just manage to brute force.
17:55
SPEAKER_01: If I put my arm here, I'm gonna fall over.
17:57
SPEAKER_01: You know, you do that a million times, you learn, oh, if I put my arm here, I don't fall over.
18:02
SPEAKER_01: Whereas before, people would actually spend years trying to figure out the algorithm.
18:05
SPEAKER_01: If I come here, then you know, what is all the ways?
18:07
SPEAKER_01: What is all, you know, how to construct that in code and it'll work together.
18:12
SPEAKER_01: And now you're just like, all right, the robot can't do anything, let's just put a supercomputer in there, spend a week, and it spits out a controller that just works and could do a backflip.
18:20
SPEAKER_01: There is caveats in this though.
18:22
SPEAKER_01: When you see uh Unit's robot, it's always on flat ground, so it's very much feet forwards when it does a backflip.
18:29
SPEAKER_01: If you put it on a slope of uh five, ten degrees and it does a backflip, it's gonna fall over.
18:34
SPEAKER_01: It doesn't really have that knowledge.
18:36
SPEAKER_01: Unless you train it on all these different terrains, it does break.
18:39
SPEAKER_01: But that just comes down to brute force again.
18:41
SPEAKER_01: You just learn on all the different terrains, and you know, and that's what they're doing.
18:45
SPEAKER_00: Uh it strikes me that uh a lot, much, if not even like all of the industrial, certainly industrial, and certainly like whatever home humanoid robot industry may or may not exist in the near term.
19:02
SPEAKER_00: To a certain extent, like hinges on there having to be a somewhat similar breakthrough happening in manipulation.
19:11
SPEAKER_00: And people obviously that's a really difficult problem to solve.
19:15
SPEAKER_00: That the data isn't there.
19:18
SPEAKER_00: And a lot of people that I've spoken to think it's an impossibility to really, really crack manipulation in say the next three years.
19:27
SPEAKER_01: I would I probably agree.
19:29
SPEAKER_01: In three years, you're not going to see that humanoid robot in your house and be like, make me a cup of coffee.
19:36
SPEAKER_01: It walks around, it knows what's inside your cupboard.
19:41
SPEAKER_01: You see that nowadays in videos, but that is because you've got 300 engineers in this fake kitchen, they've modeled everything, they've not hard-coded, but they're like, that cupboard there has got a tea mug in it, and they're all the same tea mugs.
19:57
SPEAKER_01: You know, they know the coffee machine.
19:59
SPEAKER_01: If you want to drop it into any house and it just knows, like, for example, it walks around, it opens all the cupboards, it's like, oh, now I know where all the teacups are.
20:08
SPEAKER_01: Now I know where the dishwasher is.
20:10
SPEAKER_01: Now I'm gonna search online to figure out how the dishwasher works.
20:14
SPEAKER_01: That is the dream.
20:15
SPEAKER_01: I think that's still probably five to ten.
20:17
SPEAKER_01: And let's be realistic.
20:18
SPEAKER_01: If I say five years, it's probably 10 years away.
20:22
SPEAKER_00: Oh no, I keep having this conversation because I I keep thinking back to like conversations that I was having with um with like Gil Pratt on stage like five years ago, and we kept saying five to ten and five to ten.
20:33
SPEAKER_00: And the nice thing about saying five to ten is you can always just kind of get moved five to ten back.
20:37
SPEAKER_01: It's it's like a fishing reaction.
20:40
SPEAKER_00: Yeah, yeah, exactly.
20:42
SPEAKER_01: I I think because maybe I've been in robotics too long, but I remember about 11, 12 years ago, one of my friends, the founder of Apex, they do autonomous car software.
20:55
SPEAKER_01: We went around Palo Alto completely autonomously, about four blocks, about I don't know, 10, 10 years ago, something like that.
21:03
SPEAKER_01: And I was just blown away by, you know, he found the company.
21:05
SPEAKER_01: I think there's about 40 people.
21:07
SPEAKER_01: I've been in an autonomous car.
21:08
SPEAKER_01: I'm like, wow, this is incredible.
21:11
SPEAKER_01: Jump forward 10 years' time, you know, we're only just starting to get out of geofency in San Francisco.
21:17
SPEAKER_01: So I think that is the same where I see now the robotics, let's say the home, uh, the humanoid-esque measure like, great, we could do, we could have a lot of engineers figure this one small problem out and get it working, but then to get it to be so generalistic, reliable, and just work, you know, it would just take so much time to iron out all those corner cases.
21:39
SPEAKER_01: The classic 80 20, we're probably 80% towards it, but that last 20%, no one knows.
21:44
SPEAKER_01: And I'm not gonna say I sit here and be like, I'm the world expert in AI, but I just came back like two, three months ago from Coral, which is the top AI conference for academics.
21:55
SPEAKER_01: There was about six or seven professors on stage, world famous, be doing this.
22:00
SPEAKER_01: For years, and they couldn't even settle on what is the technique we're going to need.
22:04
SPEAKER_01: Is that reinforcement learning, imitation learning?
22:07
SPEAKER_01: They couldn't even all decide as a group, is that the solution, or do we need a new architecture?
22:12
SPEAKER_01: They all had a completely different idea of how to get to the promised land of AGI robotics.
22:18
SPEAKER_01: And if they can't decide of this is just the way it goes, then it shows that we haven't really got a concrete idea of what to or how to solve this problem.
22:27
SPEAKER_00: Yeah, yeah.
22:28
SPEAKER_00: It yeah, and I I think there are there are differences too.
22:32
SPEAKER_00: I don't know, maybe I'm misremembering this, and maybe it's because robotics is more in my coverage wheelhouse than self-driving cars was at the time, but it it does seem like this is really being positioned to consumers as a this is this is happening right now, and this is happening in in the next year.
22:51
SPEAKER_00: And it's hard not to feel like that's ultimately doing a disservice to maybe the industry at large, is setting these expectations and these completely unrealistic timelines.
23:01
SPEAKER_01: Yeah, and I I think this is where I see the analog with the autonomous cars, because I remember like Luke's, I think 10 years ago, they had like a cool video of the the wheel steering, no steering wheel, and they were making things like we're designing ball bearings for the wheels that last a million miles because we need all this, because AI is solved for autonomous cars.
23:24
SPEAKER_01: And you saw the videos, and they're driving around like their test facility, and it just works.
23:29
SPEAKER_01: So you're like, wow, this is this is the future.
23:31
SPEAKER_01: Then 10 years later, you're like, Well, A, that company sort of doesn't exist anymore, and B, like, where where is this?
23:37
SPEAKER_01: And you see all these autonomous trucks where even I'm 10 years ago was like, this makes sense.
23:42
SPEAKER_01: I they're like, we will have trucks that go from LA to San Francisco on a highway very easy.
23:50
SPEAKER_01: So a driver would get out at LA just on the 101 when it's just flight, uh, you know, straight shot straight.
23:57
SPEAKER_01: Yeah, the truck truck driver will take the very straight road, and when it comes near San Francisco, uh a driver gets in to do the hard bit about the city.
24:06
SPEAKER_01: Ten years later, you're like, this makes sense.
24:09
SPEAKER_01: No one's achieved it.
24:11
SPEAKER_00: Yeah, yeah, you know, and so I mean, if if if that's the parallel, and and um obviously these things are never one-to-one, but you know, a lot of people are suggesting that it that it is a good parallel, and it's a lot of the same people in in these cases, then that means that these companies that we're seeing right now, that means probably again, the lion's share of companies that even we saw at CES, you know, a few years down the road, aren't gonna exist or at least won't exist in the form that they exist right now.
24:41
SPEAKER_00: When you're building a company, like how do you build for that long term?
24:46
SPEAKER_00: And and how do you make sure that it outlasts that initial, what is probably right now, to a certain extent, a bubble?
24:53
SPEAKER_01: I think I did this well at uh bear, and I'm trying to do it now.
24:56
SPEAKER_01: Instead of betting on foundational models, like I need a foundational model and my hope my robot will work in the house.
25:05
SPEAKER_01: You know, I I personally think household robotic companies are dead in the water because they, unless your LG or some big company like Samsung or even Boston have that money just to work on this for the next five, 10 years, and can easily be like, all right, what is the technology we have now that sort of works, that just needs 20, 30, 40 really good engineers to take academic or proof of concept and actually commercialize it.
25:33
SPEAKER_01: So we're not betting on like a foundational AGI robot that can learn to do everything.
25:38
SPEAKER_01: It's very much more pragmatic of warehouse factory automation where you drive to something using uh a classical occupancy AMR base because that works, and then you're like using imitation learning and RL to pick up objects, drive and put it down.
25:56
SPEAKER_01: You know, no one's really doing that at scale.
25:58
SPEAKER_01: And the technology actually exists there.
26:00
SPEAKER_01: We had to just put in a lot of different packages, uh, different methods, and now just thinking about how do you actually get this system that a company can set it up themselves that doesn't require my engineers to spend four weeks, you know, and that's I think the key that I learned is like you have to have a system that your end user can actually teach, train, or configure without having to use a PhD roboticists.
26:25
SPEAKER_01: And I think that's where a lot of robotic companies fail.
26:28
SPEAKER_01: They build really good tech and then they forget, like, oh no, a forklift driver has to actually operate this robot.
26:34
SPEAKER_00: Yeah, so so to a certain extent, maybe not even to a certain extent, like the AMR problem.
26:39
SPEAKER_00: I don't know, is there an AMR problem?
26:41
SPEAKER_00: The AMR problem is solved, right?
26:42
SPEAKER_00: In that like these are deployed, large scale, these are deployed.
26:46
SPEAKER_00: They're good at doing traffic, it's good with a lot of them around, they can operate alongside people.
26:52
SPEAKER_00: So the unsolved problems, the the big challenge is uh by dexterous manipulation, right?
26:59
SPEAKER_00: I mean, that's gotta be the difficult thing for you to deploy at scale.
27:03
SPEAKER_01: Perfect.
27:04
SPEAKER_01: Yeah, perfect.
27:05
SPEAKER_01: And I I I think the you know, tell people what we're working on and they they shouldn't compete with us.
27:11
SPEAKER_01: But I, you know, the big unlook for at least for Kennisi, and I think there are some other companies going after this.
27:18
SPEAKER_01: You go to a warehouse that's doing drop shipping or just distribution, they have miles of racking, it's very cluttered, they have a box there, and a human walks that box, they have to get a knife, they have to open that box to get out an ink cartridge and come back to the pick area.
27:36
SPEAKER_01: You could, in theory, do a Geek Plus or Cardo or something like that, where you spend like 50 million on an automated solution where you have totes and it's all automated, but I'd be around so many warehouses where it's just like racking and racking and racking of just shelves with just like a label that just says A12, A13, A14.
27:56
SPEAKER_01: So the goal for my company is like, how can you set up a system where the human operator or the system's like go to get the ink cartridges of A12?
28:05
SPEAKER_01: You go to the shelf, you pick up the object, or you open the box and take one object out and bring it back.
28:11
SPEAKER_01: No one's really done that.
28:13
SPEAKER_01: And that is a really hard problem.
28:15
SPEAKER_01: So then going back to the household, if no one could solve this simplistic, i.e., drive to a shelf, bring me back an object, how are you really going to do a household robotic where you like go get me a cup of coffee?
28:27
SPEAKER_01: Ha! That's not all hard-coded with waypoints and uh prior knowledge.
28:32
SPEAKER_00: Yeah, and one of the, I don't want to say cheap codes, um, but you know, one of the things that you've done, and and obviously, like, I'm far from the first person to comment on this, and this is far from the first time that I've commented on this specifically, but it's a path of much less resistance to not even have to like worry about legs in order to get there, right?
28:50
SPEAKER_00: That just gets you straight to the market with the wheel base.
28:53
SPEAKER_00: It's even better than that.
28:55
SPEAKER_01: It helps I founded Bear Robotics, but we have a partnership.
28:58
SPEAKER_01: I actually just buy their base, it comes with their software stack.
29:02
SPEAKER_01: So we use our example there of driving to A12 shelf and picking off an item, going from the pack area to the pick area, that is solved.
29:11
SPEAKER_01: I could just say Bear Base, take me to A12.
29:14
SPEAKER_01: They have beautiful cloud infrastructure for doing mapping, for doing all of that.
29:18
SPEAKER_01: So that's a solved.
29:19
SPEAKER_01: So then what we could just focus on, how to actually open a box, take an item off it, and then for your listeners, the big prick is it's a two uh factor problem.
29:30
SPEAKER_01: A, it's just the manipulation, how to randomly pick things, how to take it out of boxes.
29:36
SPEAKER_01: If it was all in a tote, so you just go there, open a tote, pick an item out, that's a classic pick and place.
29:42
SPEAKER_01: That's not too hard.
29:43
SPEAKER_01: The hard part is like, how do you open up a box that already that's got 20 items in there?
29:47
SPEAKER_01: You have to actually cut it open, it's just cluttered.
29:51
SPEAKER_01: And the second part is a lot of these shelves that aren't all uniform, how do you know that you're picking out the, you know, Hewlich Packard, print cartridges, XYZ?
30:03
SPEAKER_01: A human could just see that.
30:04
SPEAKER_01: They read the document, they can read a box, they can verify it without spending years or months tying into a warehouse operating system.
30:13
SPEAKER_01: How do you recognize 60,000 SKUs and make sure you are picking an ink cartridge and not just a USB drive or something else random?
30:21
SPEAKER_00: So, like right now, as we're recording this, and and you know, and and I have to mention that Kinesia is a really young company.
30:29
SPEAKER_00: Kinesu was a young company when you know you you first came on my radar.
30:33
SPEAKER_00: Kinesi was a young company when you won our our startup battle at Automate.
30:39
SPEAKER_00: But right now, you do have systems, you are shipping systems.
30:43
SPEAKER_00: Um, if one of your customers buys and deploys a system, like what can that robot actually do on a factory floor right now?
30:49
SPEAKER_01: Yeah, very simple.
30:51
SPEAKER_01: Like we have conveyor belts, a pot comes out at the end of the conveyor belt, our robot, we just trade it, it takes about one or two days set up.
30:58
SPEAKER_01: We can just pick up that item, drive, and put it into racking.
31:02
SPEAKER_01: Very simple, nothing, you know, groundbreaking, but it shows that the system just works.
31:08
SPEAKER_00: Okay, so so I mean, in a sense, is it is it really just kind of proof of concept right now?
31:14
SPEAKER_01: So the hardware, the main focus over the last year and a half was getting a whole system working.
31:19
SPEAKER_01: So you have the, you know, it took it to about three months ago.
31:23
SPEAKER_01: You could just press a button, the robot just boots on, everything works.
31:27
SPEAKER_01: So now, luckily for my software team, the hardware's near enough solved, you're not debugging like driver issues.
31:33
SPEAKER_01: So now that's there.
31:35
SPEAKER_01: You have all the integration with the bare base, you have like the safety controllers for the arms just do what you want, everything's calibrated.
31:41
SPEAKER_01: So now we can much more focus on how to train the robot.
31:45
SPEAKER_01: So we use imitation learning, so we have a nice teleop system where you can collect the data, you send it up to AWS or Google Cloud, train a model and deploy it.
31:55
SPEAKER_01: That has taken us about a year, year and a half to get all of that working.
31:59
SPEAKER_01: So now we can focus the team on we could do imitation learning, we could do a bit of RL, how to actually start getting that flywheel and data collection so seamless that we can give it to our customers so they can collect the data, so we don't have to go out there.
32:13
SPEAKER_01: And that's the big unlock that we want to spend the next six months working on.
32:16
SPEAKER_01: That's interesting.
32:17
SPEAKER_00: So I think I think what you're getting at is right now the big value add for them is that these systems are out there effectively collecting a lot of data.
32:25
SPEAKER_00: Because I'm always curious, you know, I understand why, like, you know, Mercedes is like, hey, let's let's pilot these aptronic robots because it looks like, you know, hey, we're you know, we're looking towards the future right now.
32:36
SPEAKER_00: It doesn't matter that they can't perform at a human level.
32:39
SPEAKER_00: An early customer of yours when these robots are again like really pilot phase, early phase, aren't aren't really outperforming humans?
32:48
SPEAKER_00: Like, what's what's the sales pitch?
32:51
SPEAKER_00: What's the value add for them to buy one of your robots?
32:54
SPEAKER_01: Reliability and just like it just works.
32:57
SPEAKER_01: So a lot of the tasks that we're talking to customers, they're just boring.
33:01
SPEAKER_01: You know, do you want to stand at the end of a conveyor belt, picking up a pot, walking two meters, putting it in a racking for eight hours straight?
33:09
SPEAKER_01: No one wakes up in the morning and be like, I'm looking forward to my job.
33:12
SPEAKER_01: And after a couple of months, they just don't turn up to work.
33:15
SPEAKER_01: So they're looking at that like reliability and cost is always a factor.
33:19
SPEAKER_01: You know, let's be honest, the robots is about a third of the price.
33:23
SPEAKER_01: I would say though, to give your listeners a bit more perspective of reality, at the moment, our system is probably like half the speed of a person.
33:33
SPEAKER_01: So we can't go into you say Mercedes, so we couldn't go into their high production line and you know compete with a cougaron that's doing doing this 60 hours a day.
33:42
SPEAKER_01: So it's very much that low volume people that go and do it for four or five hours a day and then go and do something else.
33:48
SPEAKER_01: We're like, well, we could just do it all day.
33:50
SPEAKER_01: Um, but for us as a company, it it's two-factor.
33:54
SPEAKER_01: A, it proves out the technology and it just gets that internal ability to be like how to deploy it.
34:00
SPEAKER_01: Because one of the key learnings I had is a lot of robot companies get to the stage where they're like, we have like a couple of pilots, we have five robots, but then they end up in this situation where you need roboticists and their own team to always go and deploy it.
34:14
SPEAKER_01: And it's so hard to scale.
34:16
SPEAKER_01: They're having to go into the code base and write custom, you know, interfaces, writing customer code.
34:21
SPEAKER_01: And we're literally at that stage now.
34:23
SPEAKER_01: So we're focused on like how do you actually scale it?
34:26
SPEAKER_01: How do you get it so a big customer, like hypothetically, Mercedes can take our robot and their you know, warehouse workers can actually set this robot up in a couple of hours like they would with a UR5?
34:37
SPEAKER_00: Yeah, I mean, I uh obviously like I completely understand, you know, what's what's in these pilots for you, but um, you know, if if the ROI is isn't there and is maybe a little ways off, like if I'm a potential customer, why not just wait until you get to a point when it is there?
34:56
SPEAKER_00: Yeah, so that's a great question.
34:57
SPEAKER_01: I think a lot of these companies want to be on the forefront.
35:02
SPEAKER_01: So they're like, we know we have problems, so some of our customers have like 600,000 employees.
35:09
SPEAKER_01: They don't want to wait for it to become uh a product they can go and buy because they have to learn themselves where to deploy it, and they use us as the way to actually educate themselves.
35:21
SPEAKER_01: What is the state of the art?
35:23
SPEAKER_01: How is it useful?
35:24
SPEAKER_01: Where can we deploy it?
35:26
SPEAKER_01: And they can also cut through some of the BS that they get in the sales channel.
35:30
SPEAKER_01: Because if you start to probably myself or anyone else, of course this robot's going to solve your problems.
35:36
SPEAKER_01: You know, everyone's going to be if you talk to Figure, Apptronic, X1, all of these, they're like, yeah, we can go and solve all your problems.
35:43
SPEAKER_01: And I think this is why a lot of these bigger companies, you might say they're wasting money, but they're educated themselves.
35:49
SPEAKER_01: What is the capability?
35:50
SPEAKER_01: And then they can reaffirm internally what is the timeline they think is realistically to deploy these robots.
35:56
SPEAKER_00: Yeah, I mean, I think it's helpful, and probably your your company's success is that you're you're a good you're a good pitchman.
36:02
SPEAKER_00: I'm sure I'm not the first person to tell you that.
36:05
SPEAKER_01: Yeah, and I think also, and this is luckily for the robotic industry, a couple, well, not even a couple, a lot of companies have come to us and said, the CEO has seen CES, has seen uh, you know, unitary robot doing a backflip.
36:21
SPEAKER_01: We want 10,000 humanoids deployed next year.
36:25
SPEAKER_01: Make it happen.
36:26
SPEAKER_01: And then someone in their C-suite has to now go and buy humanoid robots because the CEO is like, I want 10,000 next year.
36:34
SPEAKER_00: That is one of the most relatable things.
36:37
SPEAKER_00: And it's not necessarily humanoid robots, but like I think probably anybody who works in a corporate environment can relate to that from like the AI standpoint.
36:48
SPEAKER_00: Like I heard somebody recently talking about their company having a contest for what to name their like internal AI agent.
36:55
SPEAKER_00: It's because you're seeing AI everywhere, AI is the next thing.
36:59
SPEAKER_00: So somebody, somebody in a corporate suite who doesn't even necessarily know like what the value add is, but knows that you know, in order to keep shareholders happy and or and in order to kind of future-proof, you do have to embrace new technologies.
37:15
SPEAKER_00: They're looking towards that.
37:16
SPEAKER_00: So you're trying to figure out a way to, as somebody who's buying and integrating systems, you're looking for a way to actually get in on the ground floor.
37:25
SPEAKER_00: And a Kinesia robot is an easier way to get in on the ground floor with a humanoid robot, certainly than uh like a figure or an apptronic.
37:35
SPEAKER_01: Yeah, and I think this is also lesson, painful lessons learned.
37:39
SPEAKER_01: As a founder or a uh robot company, you have to go into some of these deals with your eyes open.
37:47
SPEAKER_01: So lots of these companies just want it because they're bust.
37:52
SPEAKER_01: So they can go out and they can do a LinkedIn post or they can say, you know, we are XYZ company, we have humanoid robots, and they go into it.
38:00
SPEAKER_01: Like, I can spend$50,000,$100,000 on a proof-of-concept pilot with a humanoid company.
38:07
SPEAKER_01: I'm gonna just take some photos with the CEO.
38:10
SPEAKER_01: And then the problem is the company, you're like, well, I've just invested two, three hundred thousand dollars into this, all my engineers.
38:17
SPEAKER_01: I thought you said you're gonna buy a thousand robots.
38:19
SPEAKER_01: I didn't realize the CEO wanted a photo shoot.
38:21
SPEAKER_01: We could have just, if you told me that out front, I could have just gone there with a robot and you could have charged me$5,000.
38:26
SPEAKER_01: Or just give me a POC, you know, letter intent, and I we would have done it for free.
38:31
SPEAKER_01: Uh, so you have to go in there with a two-factor one.
38:34
SPEAKER_01: There isn't, in a certain way, as long as you know this is what you're getting into, it's not too bad because it's a two-way relationship.
38:40
SPEAKER_01: You go into a multinational corporation, get your robot taking a foto with a famous CEO.
38:45
SPEAKER_01: You can also use that like, oh, look, we're an ex-company for raising money.
38:50
SPEAKER_01: But you also have to pay attention to we go to companies that have real problems and we can actually solve them.
38:56
SPEAKER_01: And you're not just in that innovation hell where you're like, oh, they're actually they actually want to solve a problem.
39:02
SPEAKER_01: And you know, if we do well, we have clear KPIs, and if we hit them, is realistically they're by the robot.
39:10
SPEAKER_00: Yeah, yeah.
39:11
SPEAKER_00: I mean, if you wanted to do that, you could just start like uh a photo booth company, right?
39:15
SPEAKER_00: I mean, if they if that's if the end goal was to just get people to take pictures with your robot, like that's a much easier path to make an income.
39:24
SPEAKER_01: I literally heard about this about two months ago, where a big logistics company got in a robot company just so the CEO could take a photo and be like, we're deploying humanoids next year.
39:35
SPEAKER_01: And then you hear your back channel, you're like, oh, how is that POC going?
39:38
SPEAKER_01: They're like, Yeah, it was just for a photo shoot.
39:41
SPEAKER_00: That's like the really slimy thing side of things, but but I also but like kind of middle ground there too is a company, you know, genuinely having interest in these technologies, seeing them, as you say said, do backflips at CES, deploying them and realizing that they can't get there.
39:58
SPEAKER_00: And that's why they're abandoning their contracts.
40:00
SPEAKER_00: And then that means that like, oh, we tried it once, we spent all that money, and years from now, you know, we're we're not gonna go back to that well.
40:08
SPEAKER_01: Yeah, I think this is where you have to go in with expectations.
40:12
SPEAKER_01: One of the things that I think I've done well is we have pre-screened a lot of companies.
40:17
SPEAKER_01: They send us the way we work at Kennisi is we speak to these companies, we get so much input, we haven't really done any marketing, but everyone just seems to want to do humanoids.
40:26
SPEAKER_01: We ask them, can you just send us five to ten videos of problems you have?
40:30
SPEAKER_01: And we will pre-screen it.
40:32
SPEAKER_01: And about 90% of them, we're like, we like the problem, we understand it, but realistically, we can't do it for two years.
40:39
SPEAKER_01: And, you know, fact get back to us because a lot of this is just like so complex.
40:43
SPEAKER_01: You're like, we could probably do it, but it's gonna be such a custom installation for you.
40:48
SPEAKER_01: It's gonna cost us million, and you've got to buy two robots.
40:51
SPEAKER_00: I don't know how you've managed this or even like how intentional this was, but your timing was like really perfect as far as you know, really getting on people's radar, you know, within the past year, two years, at a point when the pendulum was sort of starting to swing back.
41:09
SPEAKER_00: Because I was, you know, years and years ago, I was having this conversation of, okay, AMRs, this is great.
41:16
SPEAKER_00: Obviously, the next step is we're gonna stick some manipulators on these.
41:19
SPEAKER_00: These are actually gonna go and do the picking themselves.
41:21
SPEAKER_00: And then it seems like we skipped a couple chapters, and now we're talking about you know, robots with two legs.
41:27
SPEAKER_00: People are seeing to a certain extent how difficult they are.
41:30
SPEAKER_00: The pendulum swung back a little bit to like, okay, you know, 70, 80% of the problems could be solved with a wheeled base.
41:39
SPEAKER_00: And then a company like Kinesi and, you know, other companies with wheeled base, or at least, you know, you mentioned humanoid company or or aptronic companies that like have a modular approach, you know, are pushing that as well.
41:53
SPEAKER_00: Seems like you came like along right as people were starting to have that conversation again.
41:57
SPEAKER_01: It was luck, but you create your own luck.
42:01
SPEAKER_01: Because about two years ago, I was at the uh humanoid and iros conference, that's the academic big conference, and I was just like sitting in to AI uh talks.
42:14
SPEAKER_01: I've I've been doing for 20 years, and you suddenly see a lot of these demos of imitation learning, diffusion, act models.
42:20
SPEAKER_01: I'm like blown away, like, wow, AI can actually do stuff.
42:24
SPEAKER_01: You know, before it was just like very, you know, big in the industry, you'd fake it, you'd like, oh, you put an X on the table and you're hard-coded trajectories, and then you suddenly see like the Gemini, the Grot, all of these things about two years ago, and you're like, wow, the robots actually work, the robots that actually do stuff.
42:42
SPEAKER_01: And I saw a very clear parallel to when I founded uh Bear, where you're like, all right, they have these open source AI stacks that sort of work that's written by two or three PhD students.
42:54
SPEAKER_01: You're like, great, they have pretty good functionality.
42:57
SPEAKER_01: What it takes is taking these open source stacks, putting uh 20 engineers on it, and actually turning it into a product.
43:06
SPEAKER_01: I think it was also maybe lucky that I also had uh found an AMR company and I started Kidizi sitting in uh Bear Robotics headquarters, got myself a desk for free as a uh cheap WiiWorks, um, borrowed an AMR base uh from Bear and just started building a robot on top of it.
43:26
SPEAKER_01: I think if I didn't have uh bear robotics already and my PhD was in legged locomotion, I think I would have just built a normal legged humanoid.
43:35
SPEAKER_00: Yeah, yeah.
43:36
SPEAKER_00: That that's something I was again wanted to ask you back, start backing up a bit and and and going down your CV.
43:43
SPEAKER_00: Um so you know, I guess you officially leave Bear um at some point in in 2021, and then Kinesia is really launched towards the end of 2023, and then in that period between, you're doing some consulting.
43:58
SPEAKER_00: Um when you left.
44:00
SPEAKER_00: Bear in that official capacity, though, were you already looking for the next company, looking for uh, you know, the new way to commercialize robots?
44:11
SPEAKER_01: No, to be fair, you know, I'd been doing PhD, multiple startups for like 20 years.
44:18
SPEAKER_01: I was a bit like, oh, I just need a bit more like time off, doing some consulting.
44:23
SPEAKER_01: I actually went six months and just uh I spent six months in Whistler where I don't think I even wrote a line of code just snowboarding every day.
44:32
SPEAKER_01: And I, you know, and it it was really that impetus of just that six months off, nothing but snowboarding, partying, and relaxing.
44:42
SPEAKER_01: I basically got bored.
44:44
SPEAKER_01: And then I was just like, all right, I guess I should do stuff.
44:47
SPEAKER_01: And in that time, I'd be going to Iros, and I just like, and before I even uh founded Kinesi, I spent about three, four months coming back from IROS.
44:58
SPEAKER_01: I was like, wow, foundational models.
44:59
SPEAKER_01: So I actually spent four months just programming for the first time in years, no management, not even working for a different company, just like four months programming uh genetic systems, learning how these foundational act diffusion policy worked to really retool myself, going from historically a control engineer.
45:21
SPEAKER_01: My PhD work was like MPC, Jacobians, control, bare robotics.
45:27
SPEAKER_01: Back there was classical, like occupancy grids and like your A-star, Dexter, that sort of uh navigation, to suddenly like wow, you you somehow magically just give AI uh foundational models data and suddenly it spits out a controller.
45:41
SPEAKER_01: How does this work?
45:42
SPEAKER_01: And then actually going through, like understanding like what the transformer actually is, how it works, how you can actually use it and actually how to program a robot.
45:51
SPEAKER_01: And I think one of the earlier videos I released for Cadisi that I did on my own was like getting uh chat GPT to control a robot using like MoveIt, using like Sam, using all this vision-based system just to control the end effector where you can actually chat to the robot and ask it, and then also just like how do you actually uh install like open VLA and these sorts of software?
46:13
SPEAKER_00: Well, you know, I think I think once I get off the call with you, my producer Jana is gonna have a good pitch put together about why she needs six months to snowboard in order to like take the podcast to its next level.
46:26
SPEAKER_00: But um, so it's it sounds like Kenisi early on for a certain period of time is really it was just pretty much you.
46:34
SPEAKER_01: Yeah, so um it was early on, it was I basically like seen these foundational models, wanted to, I didn't even start with the idea of let's start a new company.
46:44
SPEAKER_01: It was more like wow, look at these uh new agentic systems, let's and this was also when uh Chat GPT came out.
46:51
SPEAKER_01: I was like, wow, let's try and learn how to use uh an LOM to actually program.
46:56
SPEAKER_01: And that just snowballed it until I'm like, I think I've got to buy a robot arm, and then six months later, you're like, I guess I should just start a company.
47:04
SPEAKER_01: And this is about the time I actually met my co-founder Ed.
47:07
SPEAKER_01: I've known him for a while, and we just started chatting about business case, and he takes care of like a lot of the business, uh like going actually talking to customers, figuring out what they want.
47:16
SPEAKER_00: In a way, is there an advantage to almost like backing into starting a company in in that you have no expectations going into it?
47:25
SPEAKER_01: Yes and no.
47:27
SPEAKER_01: I think the advantages I think I had instead of being like, let's go and raise money and figure it out, is like I had time myself to actually learn the technology.
47:37
SPEAKER_01: Me and Ed went out for about six months chatting from everyone from Wendy's to Walmart to anything to figure out where you can actually use this newfangled uh AI and robotics.
47:50
SPEAKER_01: The original idea was actually to be a shelf-stacking robot.
47:53
SPEAKER_01: We ended up spending a lot of time on like Walmart chatting to them about inventory.
47:57
SPEAKER_01: Inventory in store, but then this is where you back into like the idea.
48:01
SPEAKER_01: They're like, why are you starting in our store?
48:04
SPEAKER_01: We, you know, we put Walmarks right in the middle of a city that have loads of staff.
48:08
SPEAKER_01: Where we really have a problem is we put our distribution center in the middle of nowhere because it's cheap, but we can't get staff.
48:15
SPEAKER_01: And then we went round uh like their warehouse, seen these problems, and I then went to like 20 warehouses uh and looking at the same problem where you walk, you go in there, there's no automation, there's not even an AMR, there's no coupon, there's no arms, there's zero, it's just lots of people, and you're like, why don't you have um AMRs and stuff?
48:36
SPEAKER_01: They're like, well, it's too much money.
48:37
SPEAKER_01: We, you know, we don't want to do the cap eggs.
48:40
SPEAKER_01: How do we integrate it into our uh operating system?
48:42
SPEAKER_01: Because we all just use paper.
48:44
SPEAKER_01: They're like, if you just had a prop in replacement that I could just say, unload that van, put it there, do this, do that.
48:52
SPEAKER_01: We'd buy 50 of them.
48:53
SPEAKER_01: You're like, oh, okay.
48:54
SPEAKER_01: And how many warehouses do you have?
48:56
SPEAKER_01: They're like 600.
48:57
SPEAKER_01: So you're like, oh, you would take 3,000 of these humanoids.
48:59
SPEAKER_01: They're like, yes, you're like, okay, I think there's a business.
49:02
SPEAKER_01: I guess I should found a company and raise some money.
49:05
SPEAKER_00: Are currently available AMRs like like there to an extent?
49:09
SPEAKER_00: I mean, they they they are they're fairly like brownfield solutions, right?
49:12
SPEAKER_01: You can drop them in to aMRs uh they they just work because if you think about all you all you could do is you could say a waypoint, go go to here to load your bay two, or you could could or uh AI stack can actually just correct it and just give it like a velocity and a turn, and we could just could uh drive it like a um like just a base.
49:32
unknown: Yeah.
49:33
SPEAKER_01: So you get the best of both worlds.
49:35
SPEAKER_00: It's funny.
49:35
SPEAKER_00: I was reading uh I was reading an interview that I I think it might have been with the the robot report that you did it like um, you know, maybe even like a couple of years ago, and it sounded like you were still in that phase of like really trying to figure out what Kinesis uh business model is gonna do.
49:52
SPEAKER_00: You know, maybe this was the bear influence because you were looking at like Miso, so you know, you're maybe not straying like too far from the restaurant side of things.
50:00
SPEAKER_00: But it's been interesting to watch a lot of these humanoid companies evolve.
50:05
SPEAKER_00: I I'm thinking of agility because they were, you know, so early, but also again, we're sitting here during CES week, and I was at the CES where they uh rolled out uh digit in the back of a Ford, right?
50:19
SPEAKER_00: That was their initial attempt to commercialize.
50:21
SPEAKER_00: And it it does to a certain extent feel like there is a lot of like solutions in search of problems kind of still happening in robotics.
50:31
SPEAKER_01: And I think uh we I never really talk about it because in a way it was a failure.
50:36
SPEAKER_01: My first robotics company could uh robotize.
50:39
SPEAKER_01: We did it 100% the wrong thing.
50:41
SPEAKER_01: It was like just graduated with a PhD, uh Technical University of Munich had an incubator where you take PhD students and you're like, let's commercialize, and then you just literally sat in front of a whiteboard and they're like, Cool, what did you do for the last six years?
50:58
SPEAKER_01: Let's make a robot, let's just artificially sit inside an office and come up with a solution.
51:04
SPEAKER_01: And they're and you know, a mythic because I know uh the people now like Steve and um other people work there, we're like, oh, relay robotics, they look successful, let's just clone them.
51:18
SPEAKER_01: So you're just like uh all right, let's go to hotels, ask them what they do, let's just make a room service robot.
51:24
SPEAKER_01: We're like, oh, with a twist, we're like, let's let's do a robot that goes to your room and it's a mini bar.
51:30
SPEAKER_01: So you can replace the mini bar with a robot that goes to your room.
51:33
SPEAKER_01: You're like, you go to talk to every single hotel.
51:36
SPEAKER_01: They're like, we love it, we're buy it.
51:38
SPEAKER_01: So you're like, so I spent two years building this, and then you suddenly realize the Unique economics do not work because they're like, we're paying you$1,500 a month.
51:47
SPEAKER_01: You're like, that sounds a lot.
51:48
SPEAKER_01: You suddenly realize it takes you four days to map the place, it takes you about two weeks to do the installation, you just spent about 20,000 installing it, and you're like, great, it's gonna take us a year and a half just to pay back the installation time.
52:02
SPEAKER_01: Uh, and then you're like, this does not scale.
52:04
SPEAKER_01: And I think that is the problem where you just have a technology and you just go around randomly and be like, you know, let's just find a problem to solve without actually spending a lot of time deep diving, figuring out what is the business case, what is the development cycle, how many people it's gonna take.
52:21
SPEAKER_01: And I think I think this is where I see all the time in robotics.
52:24
SPEAKER_01: No matter what robot you want, there seems to be a golden rule, you need about a hundred engineers, and then people solve very small, like we are going to automate, I don't know, drug discovery for pharmaceuticals.
52:37
SPEAKER_01: Sounds great on paper, but you're like, well, there's only probably 20 companies in the world that wants it.
52:41
SPEAKER_01: If you do it really well, you're gonna sell a thousand robots.
52:44
SPEAKER_01: Um, whereas you look at Bear or somewhere like Locust, 25,000, you and you have the same amount of engineers.
52:51
SPEAKER_00: Yeah, so we're we're running up against time right now.
52:54
SPEAKER_00: Um, so what is your obviously, you know, unfortunately not everybody has the luxury of you know clearing their minds with six months of snowboarding.
53:04
SPEAKER_00: But for those people who, you know, I don't know, maybe maybe they're recent grads, maybe maybe they're getting their PhD, you know, maybe maybe they're at a big company, maybe like unfortunately, you know, you mentioned like Fetch, obviously.
53:18
SPEAKER_00: There's that Ross, like uh wheel garage connection.
53:21
SPEAKER_00: They unfortunately had through Zebra, had some layoffs, um, irobot had some layoffs.
53:26
SPEAKER_00: So those people who are looking maybe to really start a thing to commercialize technology.
53:32
SPEAKER_00: Where do you even start?
53:34
SPEAKER_01: I I was actually talking uh to some uh people at the weekend actually about this.
53:39
SPEAKER_01: I think if you have the advantage that you could just go and get a normal job, get paid good salary, you could then at in your spare time build up some sort of uh proof of concept, you know, go and start talking to people, go and see, do you even have that DNA to start cold calling?
53:58
SPEAKER_01: You know, go back to Khadizi.
54:00
SPEAKER_01: I would like phone up random people at Walmart and be like, hello there, can I speak to you about robotics?
54:06
SPEAKER_01: And if you don't have that DNA to actually get out of your house, I was forced by one of the uh go-to-market people at Toomb to literally knock on the door of 50 hotels and be like, Can I speak to the manager, please, to figure out what is there, you know, but if you don't have that like uh drive, you either have to go and find a CEO or someone to do that for you.
54:30
SPEAKER_01: But you know, you you have to get outside your comfort zone.
54:33
SPEAKER_01: So you could take the the leap and just jump in with two legs, or it could be better for most people to be like, let's do it the weekend, and do I still have that energy and drive at 8 p.m.
54:45
SPEAKER_01: after a uh working all day to do my startup because I give people advice for free.
54:51
SPEAKER_01: I get tired at 8 p.m.
54:53
SPEAKER_01: after working all day in my startup, 8 p.m.
54:56
SPEAKER_01: I then have to be like, great, I got about 30 email backlog that I need to clear because I didn't get it done during the day.
55:02
SPEAKER_01: So it's not like you people say, like, oh, but I gotta work my nine tool five.
55:05
SPEAKER_01: It's like I have to do 9 to 5, and then in the evening, then clear my backlogs or speak to people in America or anything else.
55:13
SPEAKER_01: Uh yes.
55:14
SPEAKER_00: It's so funny because, like, you know, I I I uh for a long time, you know, I've I've been doing a podcast where I interview like musicians and artists, and it's the same.
55:23
SPEAKER_00: I think under a capitalist system, it's always the same advice.
55:26
SPEAKER_00: It's like figure out how to live, and then you can sort of start from there to do the other thing.
55:32
SPEAKER_01: Or or or if you're lucky enough, because I I'm doing something wrong in life, because I speak to people that are professors or PhDs coming out UCL or Stanford that seem to raise 20 million by just saying I have a PhD in AI and robotics at the moment, maybe that's the approach if you could do it.
55:48
SPEAKER_00: Okay, absolutely.
55:50
SPEAKER_00: Well, uh, Brenn, uh always a pleasure.
55:52
SPEAKER_00: Thank you so much for taking the time.
55:54
SPEAKER_00: Perfect.
55:55
SPEAKER_00: It's always great to be on your show.
56:02
SPEAKER_00: Thanks so much for joining us for another episode of Automated.
56:05
SPEAKER_00: Thanks to Brent Pierce.
56:07
SPEAKER_00: If you enjoyed that conversation, you can check him out live at the Humanoid Robot Forum.
56:11
SPEAKER_00: That is uh June 23rd and 24th in Chicago.
56:15
SPEAKER_00: It's part of the automated show.
56:17
SPEAKER_00: If you've been enjoying this show, the easiest way to support us is to like, subscribe, and leave a comment.
56:22
SPEAKER_00: You can tell a friend.
56:24
SPEAKER_00: Uh, you will find more info about this show and our relevant show notes, along with our weekly newsletter over at automated.fm.
56:31
SPEAKER_00: And with that, we will see you next week for another episode of Automated.