Latest / Automated with Brian Heater / Clara Vu on Why General-Purpose Robots Are So Hard to Ship

Transcript

Clara Vu on Why General-Purpose Robots Are So Hard to Ship · Automated with Brian Heater

• 0:00
  SPEAKER_00: You have a prototype that works 80% of the time.
• 0:02
  SPEAKER_00: It looks cool.
• 0:03
  SPEAKER_00: It does the thing.
• 0:04
  SPEAKER_00: In robotics, that means you've done like 2% of the work.
• 0:07
  SPEAKER_00: People would see something that were like, oh look, this works.
• 0:10
  SPEAKER_00: I saw a demo.
• 0:11
  SPEAKER_00: And they'd be like, we're almost there.
• 0:13
  SPEAKER_00: I had a full-time job at iRobot before I was legally old enough to drink.
• 0:16
  SPEAKER_00: I believe that the most valuable applications for robotics are not in the home.
• 0:21
  SPEAKER_00: They're in places where challenging physical work gets done.
• 0:25
  SPEAKER_00: Anthropomorphizing is a huge challenge in robotics because people don't realize they're like, oh, I did this thing.
• 0:31
  SPEAKER_00: It was easy.
• 0:32
  SPEAKER_00: I can make a robot do that.
• 0:33
  SPEAKER_00: It's like.
• 0:34
  SPEAKER_02: Does it surprise you then, like now like everything is humanoid, that we kind of fell into that?
• 0:39
  SPEAKER_00: It doesn't surprise me because I feel like that trap has been there my whole robotics career.
• 0:45
  SPEAKER_01: But it is a trap.
• 0:46
  SPEAKER_00: In yeah, I think it is.
• 0:49
  SPEAKER_00: We don't need to invent any fundamental new technologies to automate a ton of stuff that is not automated today.
• 0:56
  SPEAKER_02: And the need is there, both in terms of job shortage, employment, but also all these conversations about decentralizing manufacturing.
• 1:03
  SPEAKER_02: Like, obviously, like people want to automate right now.
• 1:07
  SPEAKER_00: Right.
• 1:07
  SPEAKER_00: But not if you insist that you can build one thing that will solve all the problems.
• 1:12
  SPEAKER_02: Everybody I talk to in robotics thinks that it's generally positive that all there's all this investment going in.
• 1:18
  SPEAKER_02: But when things come crashing down to reality, like how negative of an impact does it have the industry at large.
• 1:46
  SPEAKER_02: If you watched the recent episode with Colin Engel, then you know that I have a uh borderline obsession with iRobots My Real Baby Doll.
• 1:56
  SPEAKER_02: I nearly lost it the time that Clara Vu told me that she had worked on the project a few years back.
• 2:02
  SPEAKER_02: Uh since then, she's become a mover and shaker in the industry, including uh co-founding the robotic safety firm VEO.
• 2:10
  SPEAKER_02: Uh it's a very fun chat.
• 2:12
  SPEAKER_02: If you've been enjoying the show, don't forget to like and subscribe.
• 2:15
  SPEAKER_02: And please enjoy this conversation with Clara Vu.
• 2:19
  SPEAKER_02: Victor and MAI are changing what's possible in industrial automation from quality inspection and predictive moments to robotics, logistics, medical devices, aerospace, and manufacturing.
• 2:30
  SPEAKER_02: At the advanced version and AI conference, you'll hear from industry leaders to see real-world AI applications connection, explore-related to machine version, computer vision, sensors, and computing and automation technology.
• 2:43
  SPEAKER_02: You'll also be able to connect with the people who are keeping what comes next.
• 2:47
  SPEAKER_02: Whether you're building, buying, or implementing these systems, this is a chance to stay ahead of the curve and see how vision and AI are moving from promise to practical impact.
• 2:57
  SPEAKER_02: Learn more and make sure to buy your ticket early at the Advanced Vision and AI conference September 23rd and 24th in Santa Clara.
• 3:08
  SPEAKER_02: So we were talking earlier and you uh were discussing, I don't even want to say your journey into becoming an engineer because the way you made it sound, you were just sort of like walking around one day and it just dawned on you like, oh, wait a minute, uh, mom and dad, I think I might be an engineer.
• 3:23
  SPEAKER_00: Yeah, here's what I said.
• 3:24
  SPEAKER_00: I discovered that I was an engineer.
• 3:26
  SPEAKER_00: Yeah, I I my family is uh very liberal arts.
• 3:29
  SPEAKER_00: Dad's a lawyer, mom's a librarian, um, and really just and when I was in school, you know, STEM hadn't been invented.
• 3:37
  SPEAKER_00: So I was just, I liked math class.
• 3:40
  SPEAKER_00: And my family was like, well, that's weird.
• 3:42
  SPEAKER_00: You like that stuff.
• 3:43
  SPEAKER_00: Um, and the same high school, I math and physics were my favorite classes.
• 3:46
  SPEAKER_00: So then I was like, well, I'm gonna go to college.
• 3:48
  SPEAKER_00: And I didn't even look at engineering schools because I didn't really know what that was.
• 3:52
  SPEAKER_00: And I was just like, so I went to college and I was like, well, I guess I'm gonna be a math major because that's my favorite class.
• 3:57
  SPEAKER_00: Um, and I actually, my summer after my freshman year, I I got a uh unpaid internship working for a lawyer, which taught me that I did not, in fact, want to be a lawyer.
• 4:06
  SPEAKER_00: Yeah.
• 4:07
  SPEAKER_00: Um, and then actually, my mom, the school where she worked, one of her um one of the parents was she he was a robotics engineer.
• 4:16
  SPEAKER_00: She did not know that.
• 4:17
  SPEAKER_00: She just knew he did something geeky like that stuff her daughter liked.
• 4:20
  SPEAKER_02: I I should say, by the way, like before we go any further, that like you are from the Boston area.
• 4:25
  SPEAKER_02: So, like at least ostensibly, like you're in this kind of hotspot for robotics.
• 4:30
  SPEAKER_00: Yep.
• 4:30
  SPEAKER_00: No, I went to, you know, Cambridge Public Schools, my high school Cambridge Engine Latin had one of the first robotics teams.
• 4:36
  SPEAKER_00: I had friends from my classes that were did those things.
• 4:39
  SPEAKER_00: I just sort of didn't really process that as like a career or like a thing you could do as an adult.
• 4:44
  SPEAKER_00: Sure.
• 4:44
  SPEAKER_00: So my mom introduced me to a guy named Bob Castles, who some folks in the industry may know.
• 4:51
  SPEAKER_00: Um, he sadly passed away actually about a year ago.
• 4:54
  SPEAKER_00: But she was like, this guy does nerdy stuff like you go talk to him.
• 4:58
  SPEAKER_00: He worked at iRobot when it was a 10-person startup doing DARPA grants in the attic of the Twin City shopping mall.
• 5:04
  SPEAKER_00: Um, and I went to visit him.
• 5:06
  SPEAKER_00: And I just, you know, it was just a like, hey, I'm a college student trying to figure out what to do with my life.
• 5:10
  SPEAKER_00: Can I like come meet up with you and see what you do?
• 5:14
  SPEAKER_00: Um and I, it was that was one of those life-changing experiences.
• 5:18
  SPEAKER_00: I I walked out and be like, oh my God, I can actually see myself doing this, you know, for a living.
• 5:25
  SPEAKER_00: And then I I he hired me as an intern to write documentation.
• 5:30
  SPEAKER_00: Because I was like, they're never gonna hire me because I don't know how to do any of that stuff.
• 5:33
  SPEAKER_00: But I didn't know how to write.
• 5:34
  SPEAKER_00: So they hired me to write documentation for the robot, little robots they were selling to um basically to labs to for academic research.
• 5:42
  SPEAKER_00: And then about halfway through the summer, yeah, he handed me a book um on Lisp programming.
• 5:48
  SPEAKER_00: And he said, Here, learn to code.
• 5:49
  SPEAKER_00: And I was like, Okay.
• 5:50
  SPEAKER_02: So wait, so that that was like your school for engineering was the book that he handed you?
• 5:54
  SPEAKER_00: Yeah, and then I went back, I had one more year of school.
• 5:56
  SPEAKER_00: I took one computer science class.
• 5:58
  SPEAKER_00: Um, and I will say that taking CS 101 as a senior math major is pretty funny.
• 6:04
  SPEAKER_00: It was like, I'm like, this is the stuff that they've been saying is so chill.
• 6:08
  SPEAKER_00: Um, and then I said, I asked them, I said, well, I'm graduate.
• 6:11
  SPEAKER_00: I said, Can I come back and be an intern, but can I write software this time?
• 6:14
  SPEAKER_00: And they said, sure.
• 6:16
  SPEAKER_00: And then halfway through the summer, they were like, Hey, you want a full-time job?
• 6:19
  SPEAKER_00: And I was like, sure.
• 6:20
  SPEAKER_00: So pretty everything I know about engineering robotics is learned on the job.
• 6:24
  SPEAKER_02: Yeah, I I I mean, we we we've talked about this before, and it was always it's always been fascinating to me that iRobot just was was like has been such a cornerstone to like just Boston in general, and that like everybody seems to have come through the ranks.
• 6:36
  SPEAKER_00: And I didn't realize that you were there like and that it was such a fundamental uh I was there when I was I I I had a full-time job at iRobot before I was legally old enough to drink.
• 6:47
  SPEAKER_00: So yeah.
• 6:49
  SPEAKER_00: It was it's pretty early.
• 6:50
  SPEAKER_02: So so so that that time when you said when you came back uh, you know, and then you were, I guess, sort of interning for the second time and then and then uh was sort of working there.
• 6:58
  SPEAKER_02: What was the state of the company like that?
• 7:00
  SPEAKER_00: So they had when I was an intern, the first year that I was an intern there, they were doing exclusively, they would do DARPA grants, and then they had these little uh six-legged bug robots that they would sell to other researchers.
• 7:11
  SPEAKER_00: Hexapod sort of yeah, it was that was a lot of the uh insect-inspired robot behavior-based robotics.
• 7:18
  SPEAKER_00: Um then when I came back the next summer, they had just started doing commercial stuff.
• 7:23
  SPEAKER_00: So um they and again, I was an intern, so I was not exactly clued into the business stuff at this at this time.
• 7:30
  SPEAKER_00: But um, yeah, so they had done a deal with uh Halliburton, which for an oil well exploration robot, which is the first thing I I worked on.
• 7:39
  SPEAKER_00: And I think they also had a deal with SC Johnson to build a commercial floor cleaner, which partially inspired Roomba, although it was completely different, you know, commercial and home appliances are completely different technology-wise.
• 7:53
  SPEAKER_00: Yeah, so I worked on that.
• 7:54
  SPEAKER_00: And then I also they did the next year the um the Furby came out.
• 7:59
  SPEAKER_00: And so Hasbro was like, we need to do interactive toys.
• 8:02
  SPEAKER_00: So Hasbro did a deal with iRobot to make interactive toys.
• 8:06
  SPEAKER_00: Yeah.
• 8:06
  SPEAKER_00: Um, and so I worked on the creepy baby doll.
• 8:08
  SPEAKER_02: This is when I found that out, um, I got so excited because I, you know, I knew about the baby doll.
• 8:14
  SPEAKER_02: And and um if you and I think to this day, I was informed by somebody recently that to this day, if you go to the iRobots headquarters in Bedford that they still have a museum up front and there's still the melted head baby doll up there.
• 8:28
  SPEAKER_00: They do have that museum.
• 8:29
  SPEAKER_00: I don't remember which instance of the baby doll is in it, but yeah, they're they have the oil well robot too, I think.
• 8:34
  SPEAKER_02: Okay.
• 8:35
  SPEAKER_02: I mean, my eyes just go straight to the baby doll.
• 8:37
  SPEAKER_00: I mean, it's kind of the outlier.
• 8:38
  SPEAKER_02: Yeah, uh the the the the face is kind of you know melted.
• 8:41
  SPEAKER_02: It really looks like a horror movie.
• 8:43
  SPEAKER_02: But what what I mean, what was that what was that experience like?
• 8:48
  SPEAKER_00: I mean, it was for me, it was uh it was amazing.
• 8:53
  SPEAKER_00: And um that is also where so it it's where iRobot learned low-cost manufacturing.
• 9:00
  SPEAKER_00: Um they actually, the original Roombas were made in a toy factory.
• 9:03
  SPEAKER_02: Okay.
• 9:03
  SPEAKER_02: Um and so they were sort of going back and forth to China, maybe at the end.
• 9:06
  SPEAKER_00: Yeah, yeah.
• 9:07
  SPEAKER_00: So the the factory, I think it was named Jeddah that they that they first worked with, um, had I don't know if it was the exact factory that we worked with with Hasbro, but it was a you know cheap plastic stuff with a few motors um factory, that that's what they did.
• 9:20
  SPEAKER_00: And and that's where because everything else iRobot had done before that was and you know, military, you know, the the Halliburton robot, I think this the one prototype was like a million dollars of cogs or something like that.
• 9:33
  SPEAKER_00: And so going from that to being like, hey, we need to make this thing for you know five bucks and half of that's the battery.
• 9:40
  SPEAKER_02: Yeah.
• 9:40
  SPEAKER_02: Um was and and we need to make 10,000.
• 9:43
  SPEAKER_00: And we need to make 10,000 of them.
• 9:44
  SPEAKER_00: And and um, so we were we were all just figuring everything out as we went along.
• 9:48
  SPEAKER_00: Although, of course, Hasbro had extensive experience with, I mean, they had this um what do they call it?
• 9:55
  SPEAKER_00: Model shop.
• 9:56
  SPEAKER_00: They could build anything like overnight.
• 9:59
  SPEAKER_00: And they had the first 3D printer I've ever seen.
• 10:01
  SPEAKER_00: At the time, it was like the size of a room, and I this had like was not a thing that anyone had I had heard of.
• 10:06
  SPEAKER_00: Um, but yeah, so they had, and also toys often have the most interesting mechanisms because any of like any product out there, because these days, if you want to have you know two different things happening, you just put two motors in.
• 10:21
  SPEAKER_00: But in the toy industry, motors cost money and plastic is free.
• 10:25
  SPEAKER_00: So what you do instead is you have one motor and an incredibly complex plot cam mechanism.
• 10:31
  SPEAKER_00: So you get the mechanical engineers have a lot of fun.
• 10:34
  SPEAKER_02: So every everything is just incredibly over engineered inside.
• 10:37
  SPEAKER_00: In yeah, in some ways.
• 10:38
  SPEAKER_00: Although on the quality side, it's very underengineered because most toys with batteries, I think they had some study that they they never get a second set of batteries, right?
• 10:46
  SPEAKER_00: So quality lifetimes.
• 10:47
  SPEAKER_00: Also, iRobot had some early early roombots had a lot of quality problems for probably similar reasons.
• 10:53
  SPEAKER_02: Yeah, it's I I uh so so so Joe Jones and one of your colleagues there and and at uh Harvest Automation, um, he he wrote a book.
• 11:02
  SPEAKER_00: Um Yes, yes, I it I read it.
• 11:04
  SPEAKER_02: Yeah.
• 11:06
  SPEAKER_02: And I spoke to him around that.
• 11:07
  SPEAKER_02: And uh when he was talking about I I guess sort of making the decision to leave, um he was discussing what I guess he thought was um his impression of what iRobot was going to be and what it ended up being.
• 11:21
  SPEAKER_02: And and you know, he really thought it was gonna be something more along the lines of I guess I would say maybe what a TRI is now, where it would just really just be just, you know, we'd be there to just sort of like build a lot of different things and and uh laboratories, like like you were early on, you know, building like oil and dolls and stuff like that.
• 11:38
  SPEAKER_00: Yeah, when he started.
• 11:39
  SPEAKER_00: Yeah.
• 11:39
  SPEAKER_00: Yeah, and he had been there for a while when I started.
• 11:41
  SPEAKER_00: So yeah, it really was just like, let's build cool stuff at that time.
• 11:44
  SPEAKER_00: And then yeah, when they had a hit.
• 11:46
  SPEAKER_00: Yeah.
• 11:47
  SPEAKER_00: Um and the funny thing was they actually had two hits at the sort of similar time.
• 11:50
  SPEAKER_00: They had Roomba, which of course everyone knows about, and then Packbot, which was their first, their first military project that wasn't just a DARPA grant.
• 11:57
  SPEAKER_00: It was like an actual product.
• 11:58
  SPEAKER_00: And so that the company had this sort of bizarre, uh, two there, there are not two businesses more different than consumer electronics.
• 12:06
  SPEAKER_02: You would go in, yeah.
• 12:07
  SPEAKER_02: You would you would drive in in the park in the parking lot in the front, there would just be that like military area outside area.
• 12:13
  SPEAKER_00: Yeah, it was, it was, it was, and so they I think they did eventually then spin off that military.
• 12:18
  SPEAKER_00: Yeah, that's right.
• 12:19
  SPEAKER_00: That's right.
• 12:19
  SPEAKER_00: That was well way after my time.
• 12:21
  SPEAKER_02: Yeah.
• 12:22
  SPEAKER_02: Um but you uh you you went along with him.
• 12:26
  SPEAKER_00: Yeah, so Joe, when when Joe left iRobot, yeah, um, yeah, I had been working.
• 12:31
  SPEAKER_00: So the the original Roomba team, Joe Jones, Paul Sandon, Phil Moss, Chris Casey, Elliot Mack, and there was uh Sarah, and I'm forgetting her last name.
• 12:38
  SPEAKER_00: Anyway, oh, and Winston Tau.
• 12:40
  SPEAKER_00: So they I was not on the original Roomba team, but they were actually my like best friends.
• 12:45
  SPEAKER_00: Like I would have lunch with them every day.
• 12:47
  SPEAKER_02: Um it was very high school.
• 12:48
  SPEAKER_00: Yeah, it was, I mean, it was tiny.
• 12:50
  SPEAKER_00: We were like this little bitty company.
• 12:52
  SPEAKER_02: I mean, to be fair, it sounds like you were basically in high school at that point anyway.
• 12:55
  SPEAKER_00: This is also true.
• 12:57
  SPEAKER_00: And uh Phil is a similar age to me.
• 12:59
  SPEAKER_00: The other one, everyone else is about 10 years older.
• 13:01
  SPEAKER_00: But um, yeah, so we were always buddies.
• 13:03
  SPEAKER_00: And then when Roomba took off, the only other people at iRobot that had experience with low-cost systems was the toy team.
• 13:13
  SPEAKER_00: Was the because that that deal with Hasbro had kind of fallen apart, or not had fallen apart, it was coming to an end.
• 13:18
  SPEAKER_00: I don't, again, I was not involved in any of the business side at this time.
• 13:21
  SPEAKER_02: I mean, it fell apart in that like they didn't ship the doll, right?
• 13:24
  SPEAKER_00: Well, the d they did.
• 13:25
  SPEAKER_00: The doll was the only thing they shipped.
• 13:26
  SPEAKER_00: Oh, they didn't ship the doll, okay.
• 13:27
  SPEAKER_00: But unfortunately, it was, it was, it totally didn't work in the marketplace because it came out only like a week before Christmas, it looked really creepy, and it was$100.
• 13:35
  SPEAKER_00: Okay.
• 13:35
  SPEAKER_00: And that's just like anyway, so they actually, the entire toy team then became part of the Roomba team.
• 13:44
  SPEAKER_00: And we actually we ported the Roomba code to, we had written our own programming language and framework for those toys because that's what we did then, and we ported all the Roomba software to that framework.
• 13:58
  SPEAKER_00: So um the Roomba we had to start calling it TL instead of toy language because we didn't want people to think that the Roomba code was written in toy language.
• 14:06
  SPEAKER_00: Um the base code for the Roomba the the very first Roomba, Phil Moss wrote all the code in C from scratch.
• 14:15
  SPEAKER_00: But then when we the toy team joined up, we actually re-rewrote it in um TL because that would give us we had a lot more flexibility.
• 14:25
  SPEAKER_00: There was a reason we written, I mean, these days it would be bonkers to DIY this stuff, but um your the chips they were running on, these were like I used to joke that they were chips that fell off the back of a truck somewhere in Hong Kong.
• 14:38
  SPEAKER_02: Like they were the cheap, cheap, cheap version of and and you mean because I mean we were talking about price points and and you know for them to keep it under 200 or certainly.
• 14:46
  SPEAKER_00: Exactly.
• 14:46
  SPEAKER_00: I mean, you and that was where we would there are periodic attempts to create synergies between like the military side and the and the and the consumer side, and they'd be we'd be like, well, we're trying to figure out how to sense this thing for under five under you know, components under five sets.
• 15:00
  SPEAKER_00: Yeah.
• 15:00
  SPEAKER_00: And they'd be like, well, our cheapest component is you know five hundred dollars.
• 15:04
  SPEAKER_02: Yes.
• 15:05
  SPEAKER_02: I I should say like in their not even in their offense, but it they they weren't like mounting like guns to these.
• 15:10
  SPEAKER_02: They were really like reconnaissance robots.
• 15:12
  SPEAKER_00: Yeah, it's a reconnaissance robot.
• 15:13
  SPEAKER_00: That's exactly right.
• 15:14
  SPEAKER_00: Um but you know, it just the the way that they're manufactured, the scale, the cost, it's just like completely different.
• 15:19
  SPEAKER_02: Yeah, but if you know, we we we've we uh you know we've got these deals in place and and we're manufacturing, might as well give it a shot.
• 15:26
  SPEAKER_00: They they both they it was it was uh it was it was fun.
• 15:28
  SPEAKER_00: We had some interesting conversations.
• 15:33
  SPEAKER_02: And one of them is just because there are so many things in there that are just like um I I I would say like ass backwards in the way that and you know, and and and it and it does seem like the it it it's kind of a minor miracle.
• 15:47
  SPEAKER_02: And and and that's part of the reason why, you know, when we talk now about how difficult it is to break into um home robotics, it's like, yeah, I mean, lightning in a bottle.
• 15:56
  SPEAKER_00: Well, yeah, I mean, and you were asking when we we left iRobot and then we started Harvest Automation, you know, one of the things is iRobot, because they had become a consumer products company, they kept trying to find other consumer products, right?
• 16:08
  SPEAKER_00: Because you want to sell to once you have a you know, when when we started, it was just a bunch of engineers.
• 16:12
  SPEAKER_00: And but it by this point, you know, now you have you have a business community, you know, you have business and marketing and sales and all these things.
• 16:19
  SPEAKER_00: So you want to sell other products to the same customers.
• 16:23
  SPEAKER_00: Yeah.
• 16:23
  SPEAKER_00: And um, you know, Joe and I, and like the we we felt like there aren't other home robots that you can make practically at this level.
• 16:33
  SPEAKER_00: Like I remember a product person once saying, like, can you make a robot that cleans all the windows?
• 16:38
  unknown: Yeah.
• 16:38
  SPEAKER_00: And I was like, and how much are people willing to pay for it?
• 16:41
  SPEAKER_00: 60 bucks.
• 16:42
  SPEAKER_00: Yeah.
• 16:42
  SPEAKER_00: And it's just like, no, no, we cannot.
• 16:44
  SPEAKER_00: Yeah.
• 16:45
  SPEAKER_00: So we really wanted to when we left iRobot.
• 16:48
  SPEAKER_02: Um we didn't get the luge though.
• 16:50
  SPEAKER_02: Yeah, that was the gutter cleaning robot.
• 16:52
  SPEAKER_00: Yeah, the guy who came up with that idea was the electrical engineer on the toy, Jim Lynch T.
• 16:56
  SPEAKER_01: Okay.
• 16:56
  SPEAKER_01: Yeah.
• 16:57
  SPEAKER_01: Yeah.
• 16:57
  SPEAKER_00: So they tried, you know, they did the pool cleaning robot.
• 16:59
  SPEAKER_00: There were a few, but there's just, it's just not.
• 17:01
  SPEAKER_00: They they did a bunch of attempts at doing a lawn mowing robot.
• 17:05
  SPEAKER_01: And scuba.
• 17:06
  SPEAKER_01: Yeah.
• 17:06
  SPEAKER_00: Yeah.
• 17:07
  SPEAKER_00: So I and I think that we really felt like, and I still to this day believe this, that the most valuable applications for robotics are not in the home.
• 17:16
  SPEAKER_00: They're in places where really, you know, challenging physical work gets done.
• 17:22
  SPEAKER_00: Um, and so when we left, we didn't actually know what we were gonna do.
• 17:25
  SPEAKER_00: We just were gonna start a mystery robotics company, but we knew that we didn't want to be in consumer anymore.
• 17:30
  SPEAKER_00: Oh, no consumer, no military.
• 17:32
  SPEAKER_00: We wanted to be an industrial application.
• 17:34
  SPEAKER_02: So, from that standpoint, I mean it was similar to iRobot in that you know, you're like you've got a bunch of talented engineers together and you know, you've got some backing and you you don't really know what you're building.
• 17:45
  SPEAKER_00: Yes.
• 17:45
  SPEAKER_00: Yes.
• 17:46
  SPEAKER_00: I think the difference was that, yeah, I mean, obviously it was I think when iRobot started, the only place you could get money to build robots was DARPA.
• 17:53
  SPEAKER_00: So that's what they did, right?
• 17:54
  SPEAKER_00: And and we were starting, you know, much later and there were more options.
• 17:58
  SPEAKER_01: Yeah.
• 17:58
  SPEAKER_00: But yeah, so and then I think the other difference is that we did want to be a product company from the get-go.
• 18:04
  SPEAKER_00: We were looking for a specific application rather than like a research project.
• 18:08
  SPEAKER_02: Aaron Powell So Harvest is kind of like I mean, it's interesting in terms of like the actual products, right?
• 18:13
  SPEAKER_02: Because it's greenhouse robots, but also kind of logistics.
• 18:17
  SPEAKER_00: Yeah, it was moving potted plants.
• 18:19
  SPEAKER_00: Um and I will still say that as far as I know, that is the most autonomous system that does both mo manipulation and mobility in an unconcerned outdoor environment.
• 18:29
  SPEAKER_00: Um it was and and that was really Joe, you know, Joe and Paul, who were the two guys that thought up the Roomba originally.
• 18:36
  SPEAKER_00: Yeah.
• 18:36
  SPEAKER_00: Their real genius was being able to look at a task and saying, how how can this be done, you know, in an autonomous but incredibly simple way.
• 18:46
  SPEAKER_02: And this is almost like 20 years ago now.
• 18:48
  SPEAKER_00: Oh yeah.
• 18:48
  SPEAKER_00: Yeah.
• 18:48
  SPEAKER_00: I mean, so the these were out there doing, like I said, autonomous manipulation and mobility in, you know, 2008, right?
• 18:56
  SPEAKER_00: Or to by the time we got from the field, it was in 2010.
• 18:58
  SPEAKER_00: Um, but yeah, like way ahead of their time from that perspective.
• 19:02
  SPEAKER_00: Um and uh yeah, so we started saying we think that there are and there are industrial applications that we can serve with the capabilities that we know we can build.
• 19:12
  SPEAKER_00: So let's go look for them.
• 19:13
  SPEAKER_00: Um and then that was the one that we we we kind of targeted because there were a lot of people saying, like, well, why don't you pick fruit?
• 19:21
  SPEAKER_02: And we were like That was a big thing for a long time.
• 19:24
  SPEAKER_00: Well, and we both knew that the the techn we were what Roomba did for all of us is we shipped an autonomous product in real life, and very few people have actually done that.
• 19:38
  SPEAKER_00: And you learn how hard, you know, what I the thing I will typically say is if you want to commercialize a robotic product in three to five years, you need to pick something your average robotics grad student thinks they can build in three to five weeks, because it's just all about uh reliability and edge cases and the infinitely variable real world.
• 20:00
  SPEAKER_00: And so we knew that if you tried to do something like pick fruit with the technology of that era, no way.
• 20:05
  SPEAKER_00: It just it's so c that's so complicated and so variable.
• 20:09
  SPEAKER_00: Whereas when we stuck with you picking up potted plants, they're plastic and they're round and they're a fixed shape and they're on the ground.
• 20:18
  SPEAKER_00: And also the nursery and greenhouse industry is way bigger than any particular fruit industry.
• 20:23
  SPEAKER_02: Aaron Powell So you've got a solvable problem, you've got a large addressable market.
• 20:27
  SPEAKER_00: And now, of course, that didn't pan out.
• 20:29
  SPEAKER_00: Yeah.
• 20:29
  SPEAKER_00: Um, I mean actually the company did it didn't pan out on the VC side.
• 20:33
  SPEAKER_00: The company actually still exists and sells the product to um to customers in the nursery and greenhouse industry as sort of a small business.
• 20:39
  SPEAKER_00: I think the scale, you know, and again, I I was at this point, I'm a software engineer.
• 20:44
  SPEAKER_02: But but also uh, you know, one of the things you pointed out to me when we were talking uh prior to this conversation is, you know, if you kind of map out your story, you mentioned Halliburton before, and now we're talking 2008, and it was probably starting uh starting to get difficult maybe to get a lot of funding for Box Company 2008.
• 21:02
  SPEAKER_00: That we we had a term sheet um in September of 2008 when Lehman Brothers happened.
• 21:07
  SPEAKER_00: Yeah that went away.
• 21:09
  SPEAKER_00: Um and uh I actually we we did end up raising a series A, but it took us a lot longer.
• 21:15
  SPEAKER_00: And in fact, we had even had the possibility we'd been talking to people in the industry about having funding from greenhouse operators, um, which would have been awesome.
• 21:24
  SPEAKER_00: Um, and then their business tanked into because a lot of their, a lot of more they would sell to is you know, you build a new housing development, you put a lot of new plants in there and things like that.
• 21:32
  SPEAKER_00: So so that the housing crash really imp impacted them negatively as well.
• 21:36
  SPEAKER_00: So um, yeah, we it took us a while.
• 21:38
  SPEAKER_00: It was much harder to raise money at that point.
• 21:40
  SPEAKER_00: And uh that is actually where why I ended up doing some consulting at Rethink.
• 21:44
  SPEAKER_01: Yeah.
• 21:44
  SPEAKER_00: Um, which is really where I met Patrick Solvaro, who I founded Vail with and and Michael Taylor, who we were talking about earlier.
• 21:50
  SPEAKER_02: And yeah, okay.
• 21:52
  SPEAKER_00: So it was gonna it was a cool, cool little diversion.
• 21:54
  SPEAKER_02: So, yeah, so I mean, so let's so let's talk about that because okay, so you're you know, technically you're kind of like Back in the orbit of one of the iRobot co-founders, right?
• 22:04
  SPEAKER_02: So you're back with somebody who does know how to ship a lot of robots.
• 22:08
  SPEAKER_00: Yeah, I mean, I think the iRobot, so one thing that was that was interesting there is that, you know, Rod and Helen and Colin founded iRobot.
• 22:16
  SPEAKER_00: You know, Rod was only tangentially like involved in the data.
• 22:19
  SPEAKER_00: No, he I think he was their advisor.
• 22:21
  SPEAKER_00: He was on the board.
• 22:22
  SPEAKER_00: It's just all the stuff he was doing that I know nothing about, because as I said, I was writing code.
• 22:26
  SPEAKER_00: Um, but in terms of like the data day like manufacturing, there was a you know, that they was not, it was really that team that I mentioned earlier that was doing that work and then hiring people who knew.
• 22:38
  SPEAKER_00: And so actually, rethink, I felt to a certain extent, I was like, wait a minute, no, we we we've learned to do things differently.
• 22:45
  SPEAKER_00: Like, you know, they sort of were falling into a bit of that um that same that that that that same um trap whereas it's actually I find it now I you see things that Rod Brooks writes, and you know, with the state of the robotics world, especially the West Coast robotics world, he's the one who's like practical, down to earth.
• 23:02
  SPEAKER_00: Like this is this is what really makes sense, folks.
• 23:05
  SPEAKER_00: Um he was much more on the dreamer end.
• 23:07
  SPEAKER_02: That's interesting, yeah.
• 23:08
  SPEAKER_00: So he did, so he kind of, you know, a lot of it was like learning it the hard way, it it sounds like I mean it's yeah, and and and you know, and so we were saying uh we're talking earlier about um self-driving cars and how the people who were looking at the because that was when we were starting VEO, um, it was in that peak self-driving car hype.
• 23:27
  SPEAKER_00: And you had all the people who were doing self-driving cars, many of whom were actually new to robotics and autonomy.
• 23:34
  SPEAKER_00: Whereas you had us old disgruntled, you know, people who worked on the Roomba who'd shipped autonomous product.
• 23:39
  SPEAKER_02: These were like people coming over from automotive.
• 23:42
  SPEAKER_00: I think Arch Computer Vision are out of labs, a lot of them academics, you know.
• 23:46
  SPEAKER_00: It was, you know, because it's a really interesting factor.
• 23:48
  SPEAKER_00: Sure, like CMU was sort of spinning out a lot of people.
• 23:51
  SPEAKER_00: Right.
• 23:51
  SPEAKER_00: And that's an iRobot had sort of been through that academia to production learning curve, you know, with actually some of it with the creepy baby doll, at least the manufacturing side.
• 24:01
  SPEAKER_02: And then So the thing that all at the end of the day, the thing that self-driving cars was lacking was that creepy baby doll to really bridge the well to realize, I mean, I think it all has to do with you know, you have a prototype that works 80% of the time.
• 24:14
  SPEAKER_00: It looks cool, it does the thing.
• 24:16
  SPEAKER_00: In robotics, that means you've done like 2% of the work.
• 24:20
  SPEAKER_00: The that all of the difficulty, all of the challenges is in the edge cases and the variability of the environments and the exception handling.
• 24:29
  SPEAKER_00: And and so I think people would see something that were like, oh, look, this works.
• 24:34
  SPEAKER_00: I saw a demo.
• 24:35
  SPEAKER_00: Yeah.
• 24:35
  SPEAKER_00: And they'd be like, we're almost there.
• 24:37
  SPEAKER_00: And and then having that, and I think I'm sure this is true in other industries as well, but I think with autonomous systems, it's especially true.
• 24:44
  SPEAKER_02: Yeah, I mean, it's the complications, you know, like obviously, like cars, cars are a death machine.
• 24:52
  SPEAKER_02: Like that's a that's a big thing, right?
• 24:53
  SPEAKER_02: I mean, the consequences are so much higher if if a car messes up.
• 24:57
  SPEAKER_00: And I think the other thing I I say sometimes about if you want to know how difficult something is to automate, don't even talk about how you solve the problem.
• 25:06
  SPEAKER_00: Just describe the problem.
• 25:08
  SPEAKER_00: Um, and the ex one of the things, you know, they used to say if you know computers can play chess, they'll be as you know, if they can beat us at chess, they'll be smarter than us.
• 25:15
  SPEAKER_00: Yeah.
• 25:15
  SPEAKER_00: And the thing with chess is like being good at chess, I I am terrible at chess.
• 25:20
  SPEAKER_00: I have no idea how you do that.
• 25:21
  SPEAKER_00: It's really, really, really hard.
• 25:23
  SPEAKER_00: But how to play, like what are the rules?
• 25:26
  SPEAKER_00: Any first year computer science student could write that program.
• 25:29
  SPEAKER_00: So describing the problem is trivial as opposed to what do you do when you drive?
• 25:35
  SPEAKER_02: What do I do when I drive?
• 25:37
  SPEAKER_00: Yeah.
• 25:37
  SPEAKER_00: What what problem are you solving?
• 25:38
  SPEAKER_00: What are all the things you do as a driver to decide where the car goes?
• 25:41
  SPEAKER_02: I mean, the problem solving is getting from point A to point B.
• 25:44
  SPEAKER_00: But like, you know, describe the problem.
• 25:46
  SPEAKER_02: Yeah.
• 25:47
  SPEAKER_00: Right?
• 25:47
  SPEAKER_00: Oh, uh, there's an intersection, uh, there's cars here, cars there.
• 25:51
  SPEAKER_00: Well, I'm using my brain and I'm also using, you know, I'm I'm visualizing things and I'm predicting the future motion of these things.
• 25:56
  SPEAKER_00: And then if there's a if there's a uh you know, an emergency vehicle, I'm gonna do something completely different.
• 26:01
  SPEAKER_00: And if there's a guy with a hat, you know, with a it's just even describing the problem.
• 26:05
  SPEAKER_02: And that and something that the the and and this is very relatable to me because as I mentioned to you, uh I was living in New York City for 20 years and not driving, and then I moved out of town and started driving again, and it was like learning to drive all over again, and you forget how terrifying the experience of learning to driving really is.
• 26:24
  SPEAKER_00: My latest theory is that when we learn to drive, we actually sort of map our um brains to a different physical embodiment.
• 26:34
  SPEAKER_02: You like rewire?
• 26:35
  SPEAKER_00: Yeah, basically that you learn to because you don't think when you're turning, once you get good at driving, you just think like I'm turning.
• 26:42
  SPEAKER_00: You don't think I need to press on this accelerator so that it goes slower and then faster, and I need to, I need to turn this wheel, you know, this many degrees so that my turning radius will be X.
• 26:52
  SPEAKER_00: And you don't do that.
• 26:53
  SPEAKER_00: You just we sort of take the the part of our body, a part of our brain that can move our body and like map it to a car.
• 27:00
  SPEAKER_02: Oh no, but I it's funny, we uh as far as like remapping, because uh and I don't know if this is apocryphal or not, but I heard that um the the Cordy keyboard was developed to actually slow I I that's I believe I don't again, I I have heard that a lot, and I think as far as I know, it's true, but because it physically keeps your fingers.
• 27:17
  SPEAKER_02: But it's not built, it's not like the most ideal layout for a keyboard.
• 27:21
  SPEAKER_02: And like clearly the human brain has to remap itself in order to use it.
• 27:26
  SPEAKER_00: And then and that this is okay, this is getting into crazy ideas here.
• 27:30
  SPEAKER_02: Okay, let's all right, let's go.
• 27:31
  SPEAKER_02: Let's get into it.
• 27:32
  SPEAKER_02: Are we at that are we at that point?
• 27:33
  SPEAKER_02: Okay, good.
• 27:34
  SPEAKER_00: One of the things that I've heard as an advantage of a humanoid robot is that you can teleop it very easily because its form factor is the same as your form factor.
• 27:42
  SPEAKER_02: One to one.
• 27:43
  SPEAKER_00: But what I actually wonder is I think especially the younger generation kids could telly up any form factor if you gave it a video game controller.
• 27:52
  unknown: Right.
• 27:52
  SPEAKER_00: Because they know how to, they they map, yeah.
• 27:56
  SPEAKER_00: They know it's it's like humans are incredibly good at adapting, right?
• 28:01
  SPEAKER_00: So you map your brain to whatever the form factor is, whatever the sensing capability is, and whatever controls you have.
• 28:09
  SPEAKER_02: You're describing the plot of Avatar.
• 28:13
  SPEAKER_00: Yeah, I guess basically.
• 28:14
  SPEAKER_00: But I think you know, that is a thing that was I always say 25, what at 30, it's a long time.
• 28:20
  SPEAKER_00: Building robots is you learn that humans are completely amazing.
• 28:24
  SPEAKER_00: Yeah.
• 28:24
  SPEAKER_00: And that we don't understand anywhere near the sophistication of all the things that we do.
• 28:30
  SPEAKER_00: Um, like at Vail, one of the things we saw is people, we had we had a 3D system, which meant we have, you know, two-point depth cameras, and then we had a uh 3D reconstruction that we have to do with the depth cameras.
• 28:43
  SPEAKER_00: People, when you look around, you say I see in 3D.
• 28:46
  SPEAKER_00: You don't see in 3D, you see in 2.5D, and then you do a 3D reconstruction where you make use all kinds of information.
• 28:53
  SPEAKER_00: Like, how do I know there's no one behind your chair right now?
• 28:55
  SPEAKER_00: I can't see that.
• 28:56
  SPEAKER_00: But I know things about chairs and people and where things are, and I've been in this room for a while, and so I have like all this stuff that I think I'm just looking, but I'm not.
• 29:06
  SPEAKER_00: I'm doing this complex reconstruction.
• 29:08
  SPEAKER_00: And so, okay, we move things, we use our hands, we pick stuff up, we do all these things, but then we also have that ability to remap all that processing onto a completely different mechanism like a car.
• 29:21
  SPEAKER_00: And, you know, we learn that pretty fast compared to like a lot of other stuff.
• 29:26
  SPEAKER_02: So the the the these are the peripheral pieces of information that um uh you know a like a binary machine is not taking into account when it's trying to solve a problem.
• 29:38
  SPEAKER_00: Well, I think these things are so natural to us that we see a machine when we see, and and this is the classic thing, you see a machine doing something that looks like something a person does.
• 29:47
  SPEAKER_00: And I think this applies to like LLMs too, right?
• 29:50
  SPEAKER_00: It does a thing that looks like a thing that people do.
• 29:54
  SPEAKER_00: So you assume that it is doing all the things that you do in your head or with your body to accomplish and and you and those are things that you do subconsciously.
• 30:03
  SPEAKER_00: Like two-year-olds learn how to do them.
• 30:05
  SPEAKER_02: So it's not something you're like, it can do a backflip there before it could do every like motor thing that somebody can do leading up to a backflip.
• 30:12
  SPEAKER_00: And you just don't realize that that's the fact that the thing it does looks like a thing you do does not mean that it has all of the capabilities that you have, right?
• 30:21
  SPEAKER_00: And and I think that's one of the the I mean anthropomorphizing is a huge challenge in in robotics because people don't realize they're like, oh, I did this thing, it was easy.
• 30:30
  SPEAKER_00: I can make a robot do that.
• 30:31
  SPEAKER_02: It's like Does it I so I mean, uh you know, does it surprise you then, like you know, fast forward to now and now like everything is humanoid that we kind of fell into that?
• 30:41
  SPEAKER_00: It doesn't surprise me because I feel like that trap that trap has been there my whole robotics career.
• 30:49
  SPEAKER_01: But you but it is a trap.
• 30:50
  SPEAKER_00: In uh yeah, I think it is.
• 30:52
  SPEAKER_00: I think it is because, well, a bunch of different reasons, but here's one.
• 30:57
  SPEAKER_00: So we are made what our our fundamental actuators are like what tendons that that that move like this?
• 31:03
  SPEAKER_01: Elastic.
• 31:04
  SPEAKER_00: The fundamental actuator in uh a robot is a motor that spins.
• 31:08
  SPEAKER_00: So when you say you have to make this thing out of motors and make it look the same as something that was designed for tendons, you're just putting a completely unnecessary set of aesthetic constraints on your underlying problem.
• 31:20
  SPEAKER_00: So even if you want to build a machine that can go in all the places that a human goes, it doesn't necessarily make sense that that machine should resemble a human in its outward effect.
• 31:32
  SPEAKER_00: But but if you do make it resemble a human, you get uh people think it will because people look at it and they see, you know, I think I think Michael talked about this as well.
• 31:42
  SPEAKER_00: All you know, they see commander data and the terminator and you know, all these, you know, science fiction characters, and and so they attribute all this capability just because of the form factor.
• 31:54
  SPEAKER_02: Yeah, it's it's gonna be really interesting couple of years, especially with all the videos that we're seeing now of especially robots in the home.
• 32:02
  SPEAKER_02: I think probably even the like the most invested people will uh admit that there's a bit of a bubble going on right now.
• 32:08
  SPEAKER_02: But largely, you know, and and everybody I talk to in robotics thinks, you know, uh thinks that it's generally positive that all there's all this investment going in.
• 32:17
  SPEAKER_02: But yeah, when things come crashing down to reality, like how negative of an impact does it have the industry at large?
• 32:25
  SPEAKER_00: Yeah, I really don't know.
• 32:26
  SPEAKER_00: Um, and I think because I've been sort of noodling on uh an alternate path for like because I think that a lot of what a lot of us see in robotics is we have so much amazing technology capabilities, and there are definitely there is work out there, especially in industrial and commercial settings.
• 32:43
  SPEAKER_02: And and it really has like accelerated the last two years.
• 32:47
  SPEAKER_00: But there's a ton of things that we could be automating that we with modern technology that we are not, and you know, why not?
• 32:53
  SPEAKER_00: Um and and one thing I do think is that there's essentially a fundamental mismatch between the typical VC model and um some fundamental truths about automation and robotics.
• 33:04
  SPEAKER_00: Um and I think that because you know, the I mean, I'm vastly oversimplifying, but the VC model is, you know, we invest in 10 things, one of them goes to the moon, the other ones crash.
• 33:14
  SPEAKER_00: So one of the in order to make that work, you have to be confident that the thing that does succeed will succeed, you know, trillions of dollars.
• 33:22
  SPEAKER_00: And so you have to have, when you're pitching a VC, you have to have, you know, an enormous total addressable market, right?
• 33:29
  SPEAKER_02: So it has to be a general purpose solution.
• 33:30
  SPEAKER_00: It has to be a general purpose solution.
• 33:32
  SPEAKER_00: And so humanoids are of course brilliant for that.
• 33:34
  SPEAKER_00: You can say this thing can do anything a person can do.
• 33:36
  SPEAKER_00: So your market is essentially infinite.
• 33:38
  SPEAKER_00: Um, whereas if you actually look back to the last time that robotic technology kind of escaped the lab, um you look at the, I think, I think it's like the early 80s when people, the robotics, you know, I think it was Stanford, developed the six degree of freedom arm that now builds every single car and appliance and everything, you know, in the world.
• 33:58
  SPEAKER_00: But the way that that got rolled out was through this mechanism of systems integration, where you take some general purpose components, but then you do custom engineering for the specific application that you're doing.
• 34:11
  SPEAKER_00: And so it's this mixture of sort of components and customization.
• 34:16
  SPEAKER_00: And that, I think that's how you you get out of the trap where, like we were talking about earlier, the hard part in robotics is all about the variability, right?
• 34:24
  SPEAKER_00: And the edge cases.
• 34:25
  SPEAKER_00: And that means that the more general purpose your robotic system is, the more impossible it is to actually ship.
• 34:33
  SPEAKER_00: So you need to have this ability to say, okay, yeah, we have some general purpose components, but then we can specialize them for this environment, for this application.
• 34:41
  SPEAKER_00: We can do custom engineering work there.
• 34:43
  SPEAKER_00: Um and I think that that model was critical in making that generation of industrial robotics ultimately widespread and successful.
• 34:54
  SPEAKER_00: And so I think I think we need a new model of that.
• 34:57
  SPEAKER_00: I think we need something where we can take, you know, AI capabilities, all the insanely wonderful new sensors we have now, many of which due to self-driving cars.
• 35:07
  SPEAKER_00: Thank you very much.
• 35:07
  SPEAKER_00: Um that's great that they that's what, you know, like you're saying, we human the prices came down.
• 35:12
  SPEAKER_00: There will be, yeah, absolutely.
• 35:13
  SPEAKER_00: Like that was a huge benefit.
• 35:15
  SPEAKER_00: But we need to be able to have a uh a business model that can take those components, but then do some amount of custom engineering so that you could say, hey, we can automate this thing, and it would be a$10 million a year market, which that's really valuable.
• 35:28
  SPEAKER_00: But no VC will fund that as a private as a single company.
• 35:31
  SPEAKER_02: And and and obviously, like and and the need is there both in terms of um, you know, like job job shortage, employment, but also all these conversations that like for better or for worse that we're having right now about decentralizing manufacturing, like obviously like people want to automate right now.
• 35:48
  SPEAKER_00: Right.
• 35:48
  SPEAKER_00: And and and one of the biggest things that and we have the technology to do it, but not if you insist that you can build one thing that will solve all the problems.
• 35:58
  SPEAKER_00: You have to have the ability to you know profitably customize.
• 36:04
  SPEAKER_00: Um, and I don't know, I've been talking to uh a couple different folks um uh about how this how this might actually like work out, because I think you need a new funding model as well as a new kind of business model.
• 36:16
  SPEAKER_00: Um, but I I think that that could be a a big game changer for for it's you know, it's not sexy, it's not gonna be, you know, the the latest, biggest thing, it's not gonna be on the front of all the because most of the places where people use automation are never on the front of the magazine, right?
• 36:31
  SPEAKER_00: They're in a farm, they're on an oil rig, they're they're in a factory, they're in these places that we don't see, but that's where the hard work gets done.
• 36:38
  SPEAKER_00: But you need to be able to customize um for that application.
• 36:42
  SPEAKER_00: But you do have a set of transferable skills, maybe a software stack, knowledge of components, all of these other pieces that you can make it easier to do once you've done one application, makes it easier to do the next one.
• 36:55
  SPEAKER_02: Yeah, I think I think a big part of the reason why uh VC was able to sort of slip in there so comfortably is because between the 80s and VC is, and you had mentioned this earlier, is DARPA, is the closest thing you have to VC, which is a uh a massive unit that is just gonna throw money at things and accept that only a small percentage of them are gonna succeed.
• 37:17
  SPEAKER_00: Actually, I should go back.
• 37:17
  SPEAKER_00: I don't actually know a lot of the details of the history of that first wave of you know how robot arms got into all the car factories and and and how that all worked out and what kind of you know entities made that made that possible.
• 37:28
  SPEAKER_00: But um yeah, I think there's there's just we're kind of stuck in one model that we need, you know, the one solution that solves all the problems, and that that's kind of fundamentally at odds with the true nature of the work of automating stuff.
• 37:42
  SPEAKER_02: But you've been at but as a free agent, you've been sort of like stewing in this.
• 37:46
  SPEAKER_02: You feel like that there is an alternate.
• 37:48
  SPEAKER_00: Yeah, I mean, I I so I um I left the Valeo got acquired by Symbodic about a year and a half ago, and then I was at Symbiotic for a while.
• 37:56
  SPEAKER_00: And then I did this actually before I started Valeo.
• 37:59
  SPEAKER_00: That time was kind of accidental, but I was basically unemployed for six months and it was the best because I just got to talk to people and think about things and just take a step back and you can kind of process.
• 38:09
  SPEAKER_00: And so even back then I said, all right, next time, next time I'm I'm between jobs, I'm gonna do the same thing.
• 38:15
  SPEAKER_00: And so it's very this is just the way I operate.
• 38:18
  SPEAKER_00: I'm not super strategic.
• 38:19
  SPEAKER_00: I'm just like, I'm gonna talk about whatever I'm thinking to all the other interesting people that I know and see what happens.
• 38:24
  SPEAKER_02: As long as you're not starving to death.
• 38:25
  SPEAKER_00: And yeah, I'm super lucky.
• 38:27
  SPEAKER_00: My my husband's salary as a physician assistant at Boston Medical Center is paying the bills.
• 38:31
  SPEAKER_00: Thank you very much.
• 38:32
  SPEAKER_00: Um and uh yeah, so you know, you talk to people and then you see what catches on.
• 38:39
  SPEAKER_00: Um, and there have been a number of folks that I I don't know if this, you know, I'm not like I have that we haven't started a company yet or anything like that.
• 38:47
  SPEAKER_00: But this is an idea that I found that a lot of people are when I talk to them about it, are either saying, like, oh yeah, I've I've thought about this too, or you know, yeah, that really makes sense.
• 38:56
  SPEAKER_00: I wonder how that would work.
• 38:58
  SPEAKER_00: Um so I don't know.
• 38:59
  SPEAKER_02: No, these these are, you know, these are the um, you know, we were we were talking about uh I mean this is gonna be way old by the time this errors, but like uh we're you know, we're doing a we're doing a happy hour tomorrow.
• 39:08
  SPEAKER_02: And you were talking about um, you know, eat like eating eating lunch with uh with with Joe and Paul and Iruba.
• 39:16
  SPEAKER_02: These are these are the things that like you know, years from now that you reflect on and you just like that was That was where that came from.
• 39:21
  SPEAKER_00: I mean, so I met Patrick at at Rethink.
• 39:24
  SPEAKER_00: I worked for him for six months, and then when I was unemployed before Veo, I had a chat with him and he was talking about this idea that he had about essentially making um large industrial robots collaborative, which I thought was you know, again, because it comes to this instead of trying to make the robots do the things that the people are already good at, take the things that the robots are good at, the things that the people are good at, and figure out the right way to combine them, um, which I still think is is you know the right long-term direction.
• 39:52
  SPEAKER_02: Yeah, I I I I guess my concern, and I don't know, like I I tend to be well, I tend to just be pessimistic, just like as a as a human being, but um uh especially as somebody who's covered like VC for so long, um, is that I feel like you know maybe things might have to come crashing down a little bit in order for it to start to recorrect.
• 40:12
  SPEAKER_00: I mean, I honestly, yep, that could perfectly well be true.
• 40:16
  SPEAKER_00: I do think though that this this sort of the world of traditional automation that could really benefit from some of these new technologies is almost a peril.
• 40:26
  SPEAKER_00: You know, I used to describe it as a Berlin wall between the world of industrial robotics and the world of VC, AI, you know, MIT robotics.
• 40:34
  SPEAKER_00: Like I had before I started working at VAO, I had never heard of a PLC programmable logic controller.
• 40:40
  SPEAKER_00: These are the computers that run every single factory in the world.
• 40:45
  SPEAKER_00: And I had never heard of them.
• 40:46
  SPEAKER_00: So there's just this huge gap.
• 40:48
  SPEAKER_00: And when you can take some of the, when you can cross that, there's a huge amount of value there.
• 40:53
  SPEAKER_02: Okay, so this is interesting because this is something, because you you and I briefly spoke about this before, and this is something I think I didn't quite pick up on before, which which I'm getting now, is what you're what you're saying is that all of the pieces to do this exist.
• 41:07
  SPEAKER_02: You know, they're disparate in much the same way that self-driving cars came along, dropped the price down, and a lot of people are picking up the pieces.
• 41:15
  SPEAKER_02: My my parallel that I use a lot, having come over from consumer electronics, is smartphone industry just like launched like all these other and so you're so so what you're saying is uh the pieces are there.
• 41:25
  SPEAKER_00: Somebody needs to sort of come and like gather them together and and then be able to kill get the right balance of of you know development that can be that repeats across projects and customization.
• 41:38
  SPEAKER_01: Yeah.
• 41:38
  SPEAKER_00: I think that's the as opposed to trying to take those components and put them into a solution that is a one one solves everything.
• 41:45
  SPEAKER_00: You want to take those pieces and and allow for customization, um, at at least at a certain to a certain level.
• 41:52
  SPEAKER_00: Um because yeah, the all you know, I remember we couldn't have gyroscopes in the original Roomba because they cost like, I don't know,$50,$500, something.
• 42:00
  SPEAKER_00: And then they put them in the whatever the Nintendo Wii.
• 42:03
  SPEAKER_00: Yeah, yeah.
• 42:04
  SPEAKER_00: And now you can get, you know, nine.
• 42:06
  SPEAKER_02: The Kinect was the same.
• 42:07
  SPEAKER_02: The Kinect completely changed stereoscopically.
• 42:10
  SPEAKER_00: Yeah.
• 42:10
  SPEAKER_00: So I think the all those um all the components are there.
• 42:15
  SPEAKER_01: Yeah.
• 42:15
  SPEAKER_00: That's the question is how do you bring them together?
• 42:17
  SPEAKER_00: And do you try to do that as a one size fits all solution?
• 42:20
  SPEAKER_00: Or do you do that as an entity that becomes skilled at customization and application?
• 42:27
  SPEAKER_02: Maybe there's a world where there are uh and continue to be mid-sized startups and companies and and companies that are actually able to like deliver ROI within the first couple of years.
• 42:38
  SPEAKER_00: Aaron Powell Well, right.
• 42:39
  SPEAKER_00: I think that's a big part of it, is that when you do this, because again, if you have to build the general purpose humanoid robot, that's a massive research project, right?
• 42:47
  SPEAKER_00: Whereas if you say, hey, we're gonna take an industrial robot arm that exists, and then we're gonna take a sensor that exists, and we're gonna take software techniques that may not be familiar in the industrial space, but that are, you know, that that exist, right?
• 43:01
  SPEAKER_00: You have deep learning techniques, you have all these other computer vision techniques, you have all these new sensors, and we'll say, hey, we're gonna add this sensing to this robot arm and we're gonna ship something in, I don't know, six months or 12 months.
• 43:13
  SPEAKER_00: You don't need, you know, the theory is you can sort of bootstrap it a little bit better.
• 43:16
  SPEAKER_00: Um, you probably do need some kind of external funding because I think one of the challenges that traditional systems integrators face is they're pretty much fee for service.
• 43:25
  SPEAKER_00: And that means you don't really have time to build out your, you don't have the capacity to build out your technological capabilities beyond what that what you can get a particular customer to pay for at any one time.
• 43:35
  SPEAKER_00: And I think you are gonna need some sort of infrastructure, um, especially on the software side.
• 43:42
  SPEAKER_00: Um, but it's it's not a it's not a traditional VC model.
• 43:47
  SPEAKER_02: There's a lot of reason to be excited, and it seems like you are excited.
• 43:51
  SPEAKER_00: Yeah, there's, I mean, I think there's it's been it's it's it's very interesting being in the robotics industry because everyone, you know, everyone's like, oh, we're you know, we're months away from the giant transformation of all the things.
• 44:01
  SPEAKER_00: And you get to a point where we're like, no, no, no, we're not.
• 44:03
  SPEAKER_00: Like when we started VAO, it was everyone's gonna be driving, all the cars will be driving themselves by 24.
• 44:08
  SPEAKER_02: I will say, I will say, like, obviously, you know, and and and you know, for years it's always been like the five to ten conversation.
• 44:14
  SPEAKER_02: Um, but I will say that like it does feel like a lot of progress has been made in the last five to ten years.
• 44:20
  SPEAKER_00: Yeah, and and there's so much technology that when I I look at robotics, I'm like, we don't need to invent any fundamental new technologies to automate a ton of stuff that is not automated today.
• 44:32
  SPEAKER_00: We just need to take existing technologies and focus them on the right problems and and use them in effective ways and have a business model that allows us to do that.
• 44:44
  SPEAKER_00: Um, because I don't think there's going to be any general purpose solutions anytime in the next 10, 20, 30 years.
• 44:51
  SPEAKER_00: I mean, it's because that curve, when when the when the environment gets more chaotic and when the tasks get more varied, it's not linear, it's exponential, right?
• 45:00
  SPEAKER_00: It's it's a million times harder to build that thing than this thing.
• 45:04
  SPEAKER_00: So I'm like, well, let's build the easy thing now, yeah, with the stuff we have now, and then get, you know, versus building the complex thing and then making it go back to fix the problem later.
• 45:13
  SPEAKER_00: Because the vast majority of jobs don't need a full, or as um and say, Michael pointed out in the other podcast, if you're if you've got a if you're gonna use a pallet jack to pick up a pallet, why don't you make that pallet jack intelligent rather than having a humanoid robot controlling a pallet jack?
• 45:30
  SPEAKER_00: Like it just it just makes no sense.
• 45:32
  SPEAKER_02: Well, Clara, thank you so much for taking the time.
• 45:34
  SPEAKER_00: It was a part of the thank you for letting me talk about all my crazy ideas about the world of robotics.
• 45:41
  SPEAKER_02: All right, thank you, Clara and everyone at Mass Robotics for helping to facilitate that conversation.
• 45:47
  SPEAKER_02: If you've been enjoying the show, please like and subscribe.
• 45:50
  SPEAKER_02: Please check out our newsletter every Thursday.
• 45:53
  SPEAKER_02: You can subscribe over at automatedit.fm.
• 45:56
  SPEAKER_02: Uh, that's about all we got for this week.
• 45:59
  SPEAKER_02: Stick around.
• 45:59
  SPEAKER_02: We'll be back this time next week with another episode of Automated.