Latest / Automated with Brian Heater / Melonee Wise on Humanoid Robots, Safety Tradeoffs, and Real-World Deployment

Transcript

Melonee Wise on Humanoid Robots, Safety Tradeoffs, and Real-World Deployment · Automated with Brian Heater

• 0:00
  SPEAKER_00: Forklift operation causes by humans causes a death approximately every three days.
• 0:07
  SPEAKER_00: So every three days a forklift is being operated, and that forklift kills someone.
• 0:15
  SPEAKER_00: If I made an autonomous technology and I said to you, every three days this thing's gonna kill something.
• 0:22
  SPEAKER_00: How would you feel about that?
• 0:36
  SPEAKER_01: My name is Brian Heater.
• 0:37
  SPEAKER_01: I am the managing editor at A3.
• 0:40
  SPEAKER_01: It is always an adventure speaking with international woman of mystery, Melanie Wise.
• 0:45
  SPEAKER_01: In fact, I don't think that I have had a remotely dull conversation with her in the past dozen or so times that we've chatted over the years.
• 0:55
  SPEAKER_01: The elephant in the room here, of course, is that Melanie is no longer at agility as of this recording.
• 1:01
  SPEAKER_01: We're not entirely sure what's next, aside from hopefully a boat trip around the world, but it is a fascinating conversation nonetheless, and I am sure you will get a lot out of it like I did.
• 1:14
  SPEAKER_01: So I'm gonna date this interview, the the timeline immediately coming right at the gate.
• 1:20
  SPEAKER_01: But I have to ask you if you saw the race that happened over the weekend, the half marathon.
• 1:25
  SPEAKER_00: Oh yeah.
• 1:26
  SPEAKER_00: I didn't watch it, but I I I heard about it, you know, the half marathon with the the humanoid and with the teleoperator walking very awkwardly behind it the entire time.
• 1:38
  SPEAKER_00: Yeah.
• 1:39
  SPEAKER_01: Yeah, there were 21 humanoids, I think.
• 1:42
  SPEAKER_01: Um I I think a few of them actually completed it.
• 1:45
  SPEAKER_01: This was happening in Beijing, and it was really like I don't know.
• 1:49
  SPEAKER_01: I get the sense it was China just saying, hey, like, look at all these humanoid companies that we have right now.
• 1:54
  SPEAKER_01: Ultimately, something like that, obviously it's a big novelty, but ultimately something like that you feel like good, positive or negative for the industry?
• 2:02
  SPEAKER_01: Indifferent?
• 2:04
  SPEAKER_00: I mean, cool, flashy, but I I don't think it it demonstrates any kind of autonomy competency.
• 2:14
  SPEAKER_00: Um and we've been past the point of hardware working for a while now, so I mean, great, you can do a half marathon, but I mean that's nothing in the grand scheme of real operating hours, you know what I mean?
• 2:32
  SPEAKER_00: And it now, if someone had said they did a half marathon with a fully functionally safe robot, then maybe that would be interesting.
• 2:41
  SPEAKER_00: But I mean, yay, hardware walked continuously.
• 2:46
  SPEAKER_00: It's like I took a joystick out and I I teleoped a mobile robot.
• 2:51
  SPEAKER_00: I mean, it I don't know.
• 2:54
  SPEAKER_00: I I guess I I guess it I I don't see it any more interesting than maybe a novel research paper that was done one time.
• 3:05
  SPEAKER_01: Yeah, actually I was thinking of you as I was reading up on it because they actually, and I think this was probably the right move, they separated the humanoids from the humans.
• 3:13
  SPEAKER_01: They were they weren't actually running in the same lane.
• 3:15
  SPEAKER_01: Yeah.
• 3:16
  SPEAKER_01: One of them I think actually ended up like taking out one of their teleoperators.
• 3:20
  SPEAKER_01: Um But yeah, I mean I I asked partially because obviously this is uh this is a moment of of of buzz right now.
• 3:27
  SPEAKER_01: This is or you know, we're at the point right now where like my parents unprompted are asking me questions specifically about humanoids, and it seems like it's probably pretty easy to get lost or to get kind of caught up in the noise.
• 3:42
  SPEAKER_00: I think that a lot of people are excited about the prospect of humanoids in their home or in their daily lives.
• 3:49
  SPEAKER_00: But I think we're still struggling with the notion of autonomous cars in our daily life in terms of actually making that a practical reality.
• 3:58
  SPEAKER_00: And humanoids are an even more complex activity than I would say uh autonomous driving.
• 4:05
  SPEAKER_00: And I know there will be people who will absolutely disagree with me, but um the idea that you could have a baby free roaming around this device is terrifying, for example.
• 4:20
  SPEAKER_00: Um much less an adult human being.
• 4:23
  SPEAKER_00: And, you know, like when you look at like autonomous cars, you don't, yes, you have a lot of chaotic actors, but you still don't have an infant crawling around in front of it.
• 4:34
  SPEAKER_00: Like that's just not a use case that you would have to account for in normal operation.
• 4:40
  SPEAKER_00: Whereas any kind of home robot would have to account for those types of activities.
• 4:46
  SPEAKER_00: And I mean, most of these companies that are um doing these kinds of activities have kind of stated aspirationally that they are aiming at the home.
• 5:01
  SPEAKER_01: Uh when you say most, is is agility not counted in that, or is that like somewhere way down the roadmap for you?
• 5:08
  SPEAKER_00: Yeah, it's way down the roadmap for us.
• 5:09
  SPEAKER_00: I I think that when you look at it, when you look at um, you know, the the the competitive landscape for humanoids right now, you have like the pragmatic, and there's a group of companies that fall into that agility is one of them.
• 5:27
  SPEAKER_00: I would say Boston Dynamics is another that have a very long horizon roadmap, but their first approach into market, their beachhead into the market is through an industrial space, which is pretty pragmatic.
• 5:42
  SPEAKER_00: You have the aspirational, you have the people that are like, well, we're just gonna go straight into the home, which I would say is extremely aspirational.
• 5:51
  SPEAKER_00: Um, then you just have a lot of hardware-focused people where they aren't really focused on autonomy, they're not really focused on a use case or anything, they're just focused on making really capable hardware.
• 6:03
  SPEAKER_00: I would say a large segment of the humanoid um companies fall into that, especially out of Asia-Pacific markets.
• 6:14
  SPEAKER_00: And then there's the wield.
• 6:17
  SPEAKER_00: And that's a whole category that we we don't really talk much about because when we talk about the humanoid form factor, we assume it's bipedal, but there's a fair number of people coming out with wield, uh, humanoid or I mean this is the problem with the term humanoid in general, is really what we're all talking about are mobile manipulation platforms and some variation of that, MMRs.
• 6:43
  SPEAKER_00: Um, but humanoids has become this like lexiconical thing that we throw around because it's like, well, you know, it looks like a human, but there's many morphologies.
• 6:54
  SPEAKER_00: And if you if you look at like the progression of like industrial arms and AMRs, like the next logical thing should have been mobile manipulation platforms, but the hype of humanoids kind of outpaced the practical evolution of mobile and arm to mobile plus ARM.
• 7:14
  SPEAKER_00: You went along with it, right?
• 7:16
  SPEAKER_01: I mean, you jumped from uh from fetch and zero, so these these AMRs into Yeah, but I had a non-compete.
• 7:24
  SPEAKER_01: We never talk about the kinds of innovations that actually come from uh signing a legal document.
• 7:29
  SPEAKER_00: Yeah, yeah.
• 7:30
  SPEAKER_00: Like let let's let's be real.
• 7:32
  SPEAKER_00: Like, why didn't I continue working on mobile robots like uh AMRs?
• 7:38
  SPEAKER_00: I couldn't.
• 7:39
  SPEAKER_00: I I literally had to non-compete.
• 7:41
  SPEAKER_01: So this, I mean, in that in that case, it's good that that the that we're defining that category differently, so you know they can't make that argument.
• 7:48
  SPEAKER_00: Yeah, no, I I I'm I'm just being very like on the nose about it, right?
• 7:53
  SPEAKER_00: Like I think that is some of the innovators' dilemma, right?
• 7:57
  SPEAKER_00: Like once you uh hit some success point, there are a lot of innovators who get taken off the market for three to four years just based on their outcomes.
• 8:07
  SPEAKER_00: So I I think you know, when we when we look at um at some of the the irony of success is every time an innovator has a success or an entrepreneur has a success, many times it comes along with um uh lockup that basically prevents them from continuing to innovate in many cases for years at a time.
• 8:34
  SPEAKER_00: I mean, if you look retrospectively at like 2013 when Google bought all of those robotics companies, all those great entrepreneurs, it almost created for a four or five-year period kind of this dampening of the robotics industry.
• 8:52
  SPEAKER_00: And you you can actually see it.
• 8:53
  SPEAKER_00: Like suddenly all of the innovator entrepreneur people started getting sucked out in that brief period of time.
• 9:02
  SPEAKER_00: And so you you saw this kind of dip in robotics entrepreneurship in the 2013 time period because there was that big acquisition of seven or eight significant robotics companies at the time.
• 9:16
  SPEAKER_01: Yeah, obviously the the dampening or the slowing of that to a certain extent um is due to the fact that like Google didn't really do a lot with it, like certainly not at the time.
• 9:26
  SPEAKER_01: We're finally starting to see some some things from it, and then like other um, let's say office-related things that I I'm not gonna go into on the podcast as well.
• 9:37
  SPEAKER_01: But to a certain extent, is is there an inevitability?
• 9:40
  SPEAKER_01: I mean, this is I'm sure something that you thought a lot about on selling fetch to to zebra.
• 9:46
  SPEAKER_01: Is is there an inevitability that once you go from startup mode to you know being owned by a large corporation, that there's gonna be a certain amount of slowdown that happens?
• 9:55
  SPEAKER_00: I think it depends.
• 9:56
  SPEAKER_00: Um I think it can have that inevitable outcome.
• 10:01
  SPEAKER_00: I think it just depends on how the business is run post-acquisition.
• 10:06
  SPEAKER_00: Um, I think that there's really good examples of uh companies being run well post-acquisition and some that aren't.
• 10:14
  SPEAKER_00: Um and a lot of it has to do with how independent that company uh maintains, is maintained post-acquisition.
• 10:24
  SPEAKER_00: And I also think about how knowledgeable the acquiring company is in the technology and in selling the technology, actually.
• 10:36
  SPEAKER_00: Um I think that that if you have, let's say, a solution-based selling process for your technology, and then you're acquired by a company that doesn't do solution-based selling, I mean, yeah, that's going to have a significant impact in the way the company is run and how successful it is long term.
• 11:00
  SPEAKER_00: And I think that that is true for a lot of robotics companies, is that they have very particular ways that they need to go to market.
• 11:11
  SPEAKER_00: And I think it's one of the more nuanced um things that acquires don't think about.
• 11:19
  SPEAKER_00: They they think about, you know, how are we going to keep the technology people?
• 11:24
  SPEAKER_00: How are we going to uh keep them motivated and incentivized to keep developing cool and interesting technology?
• 11:32
  SPEAKER_00: But a lot of times they forget about the long tail.
• 11:36
  SPEAKER_00: And I've I've talked with like some of my other entrepreneur friends who have sold their robotics companies, and and they say that it uh that sometimes that's what gets missed in in the mix is you know, it's one thing to have the technology, it's another thing to keep to sell it and to sell it well and to to keep the momentum going there from a sales and channel perspective.
• 11:59
  SPEAKER_01: Obviously from the outside looking in, um Google was almost buying a lot of these companies as as just the prestige of like owning Boss Dynamics was enough in and of itself.
• 12:09
  SPEAKER_01: And and there was no, you know, there there was no expectation that these things were going to be on the market in the near term.
• 12:16
  SPEAKER_01: Like they didn't really start, as far as I know, they didn't really start uh the process of commercializing, at least not maybe in a in a major way until kind of after that had occurred.
• 12:27
  SPEAKER_00: I think that depends on the company.
• 12:28
  SPEAKER_00: I think that that some of the companies they did, but Oh, I just met Boston Dynamics specifically.
• 12:33
  SPEAKER_00: Oh yeah.
• 12:34
  SPEAKER_00: Boston Dynamics, well, I think that's a little bit of a complication because you know, if you look historically, I mean I I would say that from an outside perspective and from everything I know about the company, uh they they have their vision and no no acquirer is going to be able to change their vision.
• 12:54
  SPEAKER_00: And you can see that that being true through each of the successive acquisitions that Boston Dynamics has gone through.
• 13:03
  SPEAKER_01: Yeah, but I mean there was there obviously like there was a change that occurred when they were part of SoftBank in that like you know, I think it was that time period when they announced that they were going to be selling spot at some point.
• 13:13
  SPEAKER_01: And you know, it was 30 years of them essentially being a research company making really cool robots uh than to moving something where they're actually bringing these things to the market.
• 13:22
  SPEAKER_00: I think but that was more of an outcome of DARPA dollars and the way funding was done.
• 13:28
  SPEAKER_00: Because actually I would have described Boston dynamics for the first like 15 years as a DARPA, like an arm of DARPA, right?
• 13:39
  SPEAKER_00: Like they developed Patman basically solely for the US government for testing uh like hazmat suits, right?
• 13:47
  SPEAKER_01: Like so Yeah, something I think I think people forget too is uh I think that I think this is a story.
• 13:53
  SPEAKER_01: Maybe, maybe you obviously you were in At PySynamics, maybe you can confirm this that um obviously they're making these like pack robots, right, to transport things around.
• 14:02
  SPEAKER_01: And the hydraulic systems were just way too loud to have those things on a battlefield.
• 14:07
  SPEAKER_00: Yeah, I I heard the same thing.
• 14:08
  SPEAKER_00: I mean, I I've never worked there, but uh I've heard that anecdotally from people is that they I mean, if you've ever seen even listened to the videos, it's like you know, so it was one of the things that motivated them to build electric versions and and things like that.
• 14:26
  SPEAKER_01: Aaron Powell Obviously this predates your time at agility, but it again from the outside looking in, uh the research that Jonathan was doing specifically, it seemed to me flew pretty quickly from from Cassie to Digit to that commercialization.
• 14:43
  SPEAKER_00: Yeah, I mean, I think in the in the in the sense that it takes ten years to have an overnight success, yes.
• 14:51
  SPEAKER_00: Um uh I you know I think that that they've been Jonathan and the team had been working on um this technology for a long time.
• 15:03
  SPEAKER_00: Um so I I mean I joined the company eight years in, six or eight years in.
• 15:09
  SPEAKER_00: So um and if you were to ask Jonathan, it's been his lifelong passion.
• 15:14
  SPEAKER_00: Yeah, of course.
• 15:15
  SPEAKER_01: Well, his lifelong passion was you know it was like legged robots, but was his lifelong passion selling legged robots?
• 15:22
  SPEAKER_00: I think his lifelong passion is to have his technology out there in the world having impact, right?
• 15:29
  SPEAKER_00: I I think if you if you get to the end of it for for most roboticists, many of us start as researchers and we do research for a long time.
• 15:41
  SPEAKER_00: Um we end up making cool videos, but one day it hits us like we're we're making technology for the sake of technology instead of the sake of having impact.
• 15:54
  SPEAKER_00: And most of us end up doing something to try and have impact.
• 15:58
  SPEAKER_00: And I think this was Jonathan's push for having impact.
• 16:03
  SPEAKER_00: My personal push was leaving Willow Garage and and starting my own company.
• 16:08
  SPEAKER_00: Um, and so I I think in many ways, Jonathan and I are very alike in that is that we both uh wanted to have impact in the world.
• 16:19
  SPEAKER_00: And, you know, uh both of us, I I guess it seems, decided that the best way to do that was to go create our own robotics companies.
• 16:28
  SPEAKER_01: Yeah, I actually wanted to talk to you a bit about Willow Garage because you were very, a very early employee.
• 16:34
  SPEAKER_01: What was your number?
• 16:35
  SPEAKER_01: Second.
• 16:36
  SPEAKER_00: I actually started on the same day as the first employee, January 20th, 2007.
• 16:42
  SPEAKER_00: You just clocked in a little bit later that day?
• 16:44
  SPEAKER_00: Yeah, basically.
• 16:45
  SPEAKER_00: I I think I just I, you know, I think um if I recall, it was just random how the I think what's funny is the founder actually, I think his employee number was actually number nine or something.
• 16:57
  SPEAKER_00: That's funny.
• 16:57
  SPEAKER_00: He just never gave himself an employee ID.
• 17:02
  SPEAKER_01: Yeah, I mean obviously to a certain extent, um, Will Garage's fate was there were a lot of ways in which they were probably a little bit too early or a lot too early to a lot of things.
• 17:12
  SPEAKER_01: But uh was an aspect of that you think at the same time that they weren't really actively trying to commercialize things the same way Fetch was?
• 17:20
  SPEAKER_00: I think historically that's complicated because when I met Scott in December 2006, Scott pitched me on Willow Garage being a company that was, as he put it, run by engineers who want to become entrepreneurs to basically build cool things and get to cool outcomes.
• 17:45
  SPEAKER_00: Now, about four or five months after Willow Garage was like, you know, had a couple employees, myself, Jonathan Stark, Kurt Myers, John Sue, those were the first four employees of Willow Garage.
• 18:00
  SPEAKER_00: Um a couple months later, maybe in May or June, I can't remember exactly when you would have to ask Steve Cousins, but Scott hired Steve Cousins, who at the time was um a manager, I think, at Xerox Park, um, I think, or was it IBM?
• 18:19
  SPEAKER_00: I can't remember.
• 18:21
  SPEAKER_00: Um, but he I think he was at Park, but he he came in and he decided to focus Willow on more research.
• 18:33
  SPEAKER_00: And I think that there was some contention between Scott and and Steve on that viewpoint of like, what is the ultimate goal of Willow Garage?
• 18:46
  SPEAKER_00: And so for the first five-ish years of Willow, um, four-ish years, let's call it four years of Willow, we were focused very heavily on research uh and creating open source software.
• 19:03
  SPEAKER_00: And then there was this point in which we suddenly became more focused on maybe commercial activities or spin-outs, let's call it.
• 19:13
  SPEAKER_00: That's when Willow Garage created spin-outs.
• 19:15
  SPEAKER_00: I think that's around the time that um the founder Scott created suitable technologies and spun the company out.
• 19:22
  SPEAKER_00: And that caused a real like ground shift at Willow.
• 19:27
  SPEAKER_00: Like we we suddenly started working on these projects um where it was like, what can we build?
• 19:35
  SPEAKER_00: And there's a huge internal battle uh about what industry we're gonna focus on first, the home or industrial settings.
• 19:45
  SPEAKER_00: And we ended up focusing first on the home, and we spent a year looking at like stuff for the home.
• 19:52
  SPEAKER_00: There, we actually have like these project books that we like are really nicely published.
• 19:57
  SPEAKER_00: If you can ever get your hands on one, I have some.
• 20:00
  SPEAKER_00: Um you you should because they're super interesting.
• 20:04
  SPEAKER_00: Like uh everyone who worked on the projects at Willow has them, but the catalog, like everything we did at Willow, um uh pretty extensively related to.
• 20:16
  SPEAKER_00: So we did one on home automation, we did one on connecting with with people for like the home.
• 20:24
  SPEAKER_00: And then at the end of that, we kind of decided that the home, all the stuff that Willow would be best at for the home was just too far out, like building a robot to help you inside of your home.
• 20:37
  SPEAKER_00: And so then we started focusing on industrial applications.
• 20:40
  SPEAKER_00: So we looked at hotels, uh, grocery retail, warehousing, logistics.
• 20:46
  SPEAKER_00: And that's actually what kind of got me excited, and it's why we ended up uh going off and starting our own company, and is because it became very clear that there were real problems to be solved for mobile robots in in logistics and mobile manipulation robots.
• 21:06
  SPEAKER_00: It's it's actually one of the reasons I continued my journey after fetch into agility, because they were focused very similarly on the same market.
• 21:16
  SPEAKER_00: It's a great beachhead market for robotics and automation.
• 21:20
  SPEAKER_00: If you look at Willow, like I think the problem why Willow, I think, eventually ended up shutting down was largely because of a philosophical disagreement between the founder and Steve about what Willow Garage was supposed to be at the end of the day.
• 21:43
  SPEAKER_00: Um ironically enough.
• 21:46
  SPEAKER_01: Yeah, and obviously, you know, being a Silicon Valley company and being funded that way, like at a certain point, if you don't at least have on your roadmap some kind of profitability or some kind of you know market outlook that.
• 22:00
  SPEAKER_01: And the money's going to stop trickling in.
• 22:02
  SPEAKER_00: I mean, Willow was always complicated because it was funded by a single source.
• 22:08
  SPEAKER_00: Um, you know, it was funded by a single founder.
• 22:12
  SPEAKER_00: And if it's what they wanted it, if it was doing what they wanted to achieve, they continued to fund it.
• 22:17
  SPEAKER_00: If not, they stopped funding it.
• 22:19
  SPEAKER_00: And that has its upsides and its downsides.
• 22:24
  SPEAKER_01: I'm gonna say we, and by we I mean most people, not everybody, but most people in the industry, it seems, take for granted the reasons why the home is that much harder to break into than industrial.
• 22:37
  SPEAKER_01: But at the beginning of that journey, um, would you have thought that 10-15 years from now it would be more realistic that we would have home robots or like humanoid robots in the in industrial settings?
• 22:54
  SPEAKER_00: I still think it would have been more realistic to have uh industrial setting robots.
• 23:00
  SPEAKER_00: Just even at the time, I although I wasn't a safety expert, I understood the semantic complexity enough of the home that irrespective of the safety problems that get layered on top of it, the uniqueness challenges of our home and people's unwillingness to compromise on on certain aspects of their home lend itself to being a very complicated problem.
• 23:30
  SPEAKER_00: Like, for example, I think that if you if you could find some some sad subcategory of people who are willing to switch out all their dishes and switch out their furniture and a whole bunch of other things to make the robot that to make the home completely compatible with robotics technology, yeah, you probably could have had home robots a while ago.
• 23:55
  SPEAKER_00: But that's a bridge too far, right?
• 23:58
  SPEAKER_00: Like standardization in the home, I think, would would help a lot to making it go faster for certain applications.
• 24:05
  SPEAKER_00: But it's that and the other problem that that I quickly found out while working at Willow on these projects was the value proposition that people hold in their heads around what technology or or items in their home bring are very well calibrated.
• 24:25
  SPEAKER_00: How much does your dishwasher cost?
• 24:28
  SPEAKER_00: Like somewhere between three and eight hundred dollars, right?
• 24:31
  SPEAKER_01: Sure.
• 24:32
  SPEAKER_01: I I it was with the house when I bought it, but I'll take your word for it.
• 24:36
  SPEAKER_00: When you look at like if I'm going to spend$500, then it has to have the value to me of a dishwasher.
• 24:44
  SPEAKER_00: If I'm going to spend$1,000, let's say, you know, uh, it has to have the value of, let's say, a very high-end TV to me, or or um the capability or value of a washer dryer.
• 25:01
  SPEAKER_00: If I'm going to spend$10,000, it has to have the value of a car to me.
• 25:06
  SPEAKER_00: You know, like that radiation is very, very well established for people.
• 25:13
  SPEAKER_00: And so when you go and you talk to people and you're like, okay, let's talk about uh your robotic vacuum cleaner.
• 25:21
  SPEAKER_00: And this is this is super hilarious.
• 25:23
  SPEAKER_00: We we did this whole user study to try and understand like how people assess the value of their robotic vacuum cleaner.
• 25:28
  SPEAKER_01: And this was at Willow?
• 25:30
  SPEAKER_00: Yeah, yeah.
• 25:30
  SPEAKER_00: And yes, I'm like, so does it do a good job?
• 25:34
  SPEAKER_00: And they're like, it does a good enough job.
• 25:36
  SPEAKER_00: And we were like, did it replace your vacuum cleaner?
• 25:39
  SPEAKER_00: And they were like, no, it doesn't replace my vacuum cleaner, but it's good enough that I don't have to vacuum a lot.
• 25:47
  SPEAKER_00: And then we asked, well, you know, how much value did that provide to you?
• 25:52
  SPEAKER_00: And they were like, well, I don't know.
• 25:53
  SPEAKER_00: And we were like, well, you you bought it for some amount of money.
• 25:57
  SPEAKER_00: You know, how much did you pay for it?
• 25:59
  SPEAKER_00: And at the time, I think this was in 2012, approximately, 2011, 2012.
• 26:07
  SPEAKER_00: Everyone was telling us that they bought their robot vacuum cleaner for$199.
• 26:12
  SPEAKER_00: And at the time, you literally could not go and buy a brand new robot vacuum cleaner for$199.
• 26:18
  SPEAKER_00: You literally couldn't.
• 26:20
  SPEAKER_00: But so everyone had earmarked or pegged the value of what this service was providing at about$200.
• 26:29
  SPEAKER_00: But at some point, they had bought this thing or got one of these things.
• 26:33
  SPEAKER_00: Don't know how.
• 26:34
  SPEAKER_00: Either they bought it for$199 or they misremembered what they bought it for.
• 26:39
  SPEAKER_01: There's some black market going on time that we were aware of.
• 26:42
  SPEAKER_00: Yeah, yeah.
• 26:43
  SPEAKER_00: Or just a lot of deep discount uh bins.
• 26:47
  SPEAKER_00: But but pretty people pretty consistently could tell you that they earmarked the value of their robot going around and cleaning their home for$199.
• 26:57
  SPEAKER_00: And so when you when you go to them and you're like, look, I can give you a robot that might be able to kind of do your dishes for$10,000 to$15,000.
• 27:11
  SPEAKER_00: They're like, no, thank you.
• 27:12
  SPEAKER_01: But also, like, let's say hypothetically that uh nobody had invented a dishwashing robot until right now, that would be different, right?
• 27:20
  SPEAKER_01: People have an established value on what dishwashers cost because they know what dishwashers are and they've been around for a long time.
• 27:27
  SPEAKER_00: But the problem is that you you start running into these like weird jumps because we're not even close to the price point that that would make it even reasonable for them to try, right?
• 27:42
  SPEAKER_00: Um and I think that's the the problem is like there, and if you were to ask anyone, there's almost no value you can provide in someone's home that is like$20,000 worth of value.
• 27:58
  SPEAKER_00: If you look at the cost of like um having a service worker come to your home, like a cleaning person or something like that, or a gardener.
• 28:08
  SPEAKER_00: Um, all of those people are low cost enough that even at$30,000, like you basically can pay for 10 years of cleaning service at your home.
• 28:20
  SPEAKER_01: And$30,000 is like the low, low, low end of what a humanoid would cost.
• 28:25
  SPEAKER_00: Right.
• 28:25
  SPEAKER_00: And so that's and and well, of a bipedal humanoid.
• 28:29
  SPEAKER_00: For for a for a wheeled uh mobile manipulation robot, a wheeled robot, you could probably get into the$10,000 range.
• 28:39
  SPEAKER_00: I don't think that that's unachievable.
• 28:42
  SPEAKER_00: Um and but then you have like stairs and other problems, and and none of this contemplates safety, right?
• 28:48
  SPEAKER_00: Like, like we haven't even touched the safety problem here.
• 28:51
  SPEAKER_00: Like, no matter what the form factor of the robot, the good luck right now.
• 28:56
  SPEAKER_00: Um and so I I I think that's kind of some of the gap.
• 29:03
  SPEAKER_01: That's interesting.
• 29:04
  SPEAKER_01: I may you know, maybe there's an extent to which legs are actually even more necessary in the home than the factory because you know they're not controlled environments in the same way.
• 29:15
  SPEAKER_00: Legs are more necessary, I believe, unless everyone lives in a um studio.
• 29:22
  SPEAKER_00: A flat flat level home.
• 29:25
  SPEAKER_00: I mean, right now I'm in in a house that has three stories.
• 29:29
  SPEAKER_01: So I it it's funny though, because I think part of that math that people are doing, obviously it seems like a lot of it is kind of um is retroactive, right?
• 29:38
  SPEAKER_01: I like okay, I bought this thing, I invested in this thing, and now I kind of have to justify it to myself.
• 29:44
  SPEAKER_01: But but on the same note, obviously people didn't get rid of their couches so they could get a Roomba, but people for a long time and probably continue to do so, were like, you know, moving things around in order to accommodate this robot.
• 29:58
  SPEAKER_00: They still do.
• 29:58
  SPEAKER_00: They still do.
• 29:59
  SPEAKER_00: And remember, it wasn't until a couple of years ago that Roomba finally got a poop sensor, right?
• 30:06
  SPEAKER_00: Like, so it's not like smearing dog feces all over your home.
• 30:10
  SPEAKER_00: I mean, I think that people did accommodate, do accommodate their robot vacuum cleaners quite a bit.
• 30:16
  SPEAKER_00: But if you look at the product overall over the last what, 20-ish years, the features have evolved, but they haven't made giant leaps in terms of capability.
• 30:29
  SPEAKER_00: Like I said, in in the last couple of years, they finally added a poop sensor.
• 30:34
  SPEAKER_00: Now they have um automatic bin emptying, you know.
• 30:40
  SPEAKER_00: But a lot of that is because they're constantly against the the constraint of cost or value proposition, right?
• 30:48
  SPEAKER_00: I mean, really think about think about the the value proposition that trade-off that they had to make for like the robot to like build this complicated mechanism for the robot to empty its own bin.
• 31:03
  SPEAKER_00: That's how like how niche the like value proposition challenges are, where it's like, okay, you bought a robot vacuum cleaner that cleans your own floor, but now you just don't want to empty its goddamn bin.
• 31:19
  SPEAKER_00: Sorry.
• 31:22
  SPEAKER_00: But you don't want to empty its bin, right?
• 31:25
  SPEAKER_00: Like, like that's that's the problem.
• 31:28
  SPEAKER_00: Um instead of, you know, uh, like, okay, now I want it to like pick up my room or something like that.
• 31:38
  SPEAKER_00: That that's kind of where we're at because we can't get to more complicated tasks very easily.
• 31:45
  SPEAKER_01: Aaron Powell This isn't a one-to-one.
• 31:47
  SPEAKER_01: Obviously, uh corporations have a lot more resources, but to a certain extent, they have to also be doing that math, right?
• 31:54
  SPEAKER_01: If we're gonna pay this much for these many humanoid robots, like at a certain point it has to pay off for us.
• 32:01
  SPEAKER_00: I think it's just really hard to make them safe.
• 32:03
  SPEAKER_00: It's really hard to make them capable, it's hard to deal with like the semantic differences between home A and home B.
• 32:11
  SPEAKER_00: The cultural differences are gonna be a huge problem when these robots first start getting into the home and when they start getting into the home.
• 32:22
  SPEAKER_00: It'll there'll be some really weird cultural artifacts, things like that.
• 32:26
  SPEAKER_00: There'll be, you know, other like real concerning problems about pets, I'm sure.
• 32:35
  SPEAKER_01: Um Well, yeah, and the and the other thing is, you know, you mentioned babies before, but I always think about how many of these robots are like um the Labrador.
• 32:43
  SPEAKER_01: Obviously, it's not a dangerous robot in that it's a cart, but so many of these companies are going after we call it aged tech now, what we used to call elder care.
• 32:52
  SPEAKER_01: And those people are like every bit as fragile as the babies are.
• 32:56
  SPEAKER_00: One of the reasons why I uh some people asked me why we didn't um go into like elder care service with our mobile robots when I was at Fetch.
• 33:06
  SPEAKER_00: And I said, well, the problem was is that when you look at the um like survival rate of some of an older adult after a hip break, it's very low, right?
• 33:18
  SPEAKER_00: They their like life expectancy after a hip break, I think, is under four years.
• 33:23
  SPEAKER_00: And one of the problems is that they have a lot of stability issues.
• 33:28
  SPEAKER_00: And so the last situation that you want is the robot is stopped, an older adult leans on it for stability, but because it doesn't have appropriate sensing, it drives off.
• 33:40
  SPEAKER_00: And so that in itself is like a that most likely will result in a hip break.
• 33:46
  SPEAKER_00: It drives off while someone's leaning on it.
• 33:48
  SPEAKER_00: Um and you don't even know whether because it's because the person's leaning on it, they cause the robot then to fall on them as well.
• 33:57
  SPEAKER_00: Um and that just compounds the problem.
• 34:02
  SPEAKER_00: And so in order to then you know make sure that that doesn't happen, you know, how do you do that?
• 34:08
  SPEAKER_00: Do you do you assume when anyone is too close to you, you just don't drive?
• 34:14
  SPEAKER_00: Do you put tactile sensing all over the robot to basically sense that?
• 34:19
  SPEAKER_00: But both of those, because they can result in what's called an S3 injury, where you have permanent debilitating damage that could lead to maiming or or death, uh, those have to be safety-rated functions.
• 34:37
  SPEAKER_00: Um and so, you know, how do you implement that uh in a in a safe way?
• 34:43
  SPEAKER_00: And then if no one can ever get too close to your robot, then what is the value of it, right?
• 34:49
  SPEAKER_00: And so you end up in these like paradoxical situations where if a person gets too close to a robot, the robot has to stop and not move until the person goes away, then when can the robot be productive?
• 35:00
  SPEAKER_00: Um and so you like these are the the problems that you end up having to solve.
• 35:07
  SPEAKER_00: And what I think many, many companies in the space right now that are looking at these problems um are doing a lot of things that, like I said, I think they're adding tactile sensors, I think they are adding extra sensing, I think they are adding a lot of capabilities.
• 35:27
  SPEAKER_00: I don't know if they're meeting all the safety requirements um for those products, um, but they are adding the capabilities.
• 35:37
  SPEAKER_00: But it, you know, we the right now the population of testing is not large enough to see the efficacy and the um, I guess, the long-term outcomes of the safety technology that they're applying.
• 35:53
  SPEAKER_01: Aaron Powell The first time I think you and I really started talking about safety was uh at Automate last year when we did the humanoids panel.
• 36:00
  SPEAKER_01: And and your your sense at the time was that there weren't nearly enough people doing nearly enough to actually address the issue when it comes to humanoid robots.
• 36:08
  SPEAKER_01: Has how much has that conversation changed over the past year?
• 36:12
  SPEAKER_00: Aaron Powell I think that it's bifurcated pretty heavily.
• 36:15
  SPEAKER_00: Um I think there's the the group of companies that are just putting their hands over their ears and closing their eyes and being like la la la la la la la.
• 36:26
  SPEAKER_00: I am not focused on the real world anyway, so I don't care about safety.
• 36:31
  SPEAKER_00: Um and then there's the people who, like I said, are in the pragmatic category who are starting to come together as part of a coalition to look at safety in a real practical way.
• 36:42
  SPEAKER_00: Uh currently there's a dynamically stable industrial mobile robot standard that's being proposed for the ISO realm uh to look at this problem, to start putting more permissibility out there in the market, most likely, and better requirements around what it means to have a functionally safe, uh dynamically stable system.
• 37:10
  SPEAKER_00: Agility's been working on that in conjunction with Boston Dynamics team.
• 37:15
  SPEAKER_00: Um, and so we put a proposal forward uh earlier this year.
• 37:21
  SPEAKER_00: Um we're also looking to potentially update um R1508, um, which I think we'll be proposing that in the near future.
• 37:33
  SPEAKER_00: Um so there there is work there for for creating more guidance for dynamically stable robots.
• 37:40
  SPEAKER_00: Um but I think that the big gap is that there's just not enough market education.
• 37:47
  SPEAKER_00: Like, for example, uh I was with a customer the other day, and they told me that as long as someone's teleoping it, even if it's a remote teleop, it's safe, and that's a practical safety mitigation.
• 38:00
  SPEAKER_00: I said, no, it's not.
• 38:01
  SPEAKER_00: Um it's actually not allowed under the standard.
• 38:05
  SPEAKER_00: Um it doesn't mitigate, it doesn't create safety.
• 38:08
  SPEAKER_00: The only thing that's allowed under the standard is line of signed operation.
• 38:12
  SPEAKER_01: So as as a backup, any kind of teleop, even as a backup, is prohibited by this.
• 38:17
  SPEAKER_00: Yeah, because it's not functionally safe.
• 38:19
  SPEAKER_00: What if there's latency issues?
• 38:21
  SPEAKER_00: Let's okay, let's say you have your murderous teleop robot that has a knife in its hand, right?
• 38:27
  SPEAKER_00: Um, and I don't mean murderous, but it has a knife in its hand, right?
• 38:32
  SPEAKER_00: It's teleoping, there's a person nearby, and you have low latency, you have like really high latency, right, in your teleop.
• 38:41
  SPEAKER_00: So if you have no safety regime on board the robot, that robot, while you're waiting for the next like you know, update from your teleop loop, which is super high latency, let's say you had tons of packet loss, its arms flailing around with this knife, like you wouldn't even know.
• 39:03
  SPEAKER_00: Like, like that, for example, is like a perfect example where remote teleoperation, and I know that's an extreme example, but that's an example.
• 39:13
  SPEAKER_00: The minute the arm does any kind of uncontrolled behavior that you as a teleoperator have no control over, and that essentially will happen during any non-safety-rated remote teleoperation, you basically have this problem.
• 39:29
  SPEAKER_00: As far as I know, there is no um uh safety-rated teleoperation scheme that I've seen um for operating these robots.
• 39:39
  SPEAKER_00: And and if there was, I'm guessing that it would not be uh productive enough or meet the productivity rates that customers would care about.
• 39:51
  SPEAKER_00: And so what a lot of customers are seeing today are like really like nice demos that don't meet any of the requirements for market.
• 40:04
  SPEAKER_00: And that kind of misinformation, that kind of uh misunderstanding of the capability of the technology and what's allowed, like is is going to cause a problem in the in in the short to near term when when eventually someone gets hurt.
• 40:22
  SPEAKER_00: And the biggest problem is is that many of these robots were talking about putting mass in the hands in the end effectors of these robots that are in the plus five kilogram range, and they can move around at upwards of a meter per second.
• 40:39
  SPEAKER_00: That is definitely enough to hit you in the head or the throat and kill you.
• 40:44
  SPEAKER_00: And that's all it's gonna take, is one of those incidents.
• 40:48
  SPEAKER_01: You can mitigate them to a certain extent, but is there you know, peep people uh working in warehouses and factories, they bump into each other, you know, or or like obviously there's a lot of um infamously there's a lot of forklift accidents that happen there as well.
• 41:06
  SPEAKER_01: To a certain extent, is it an inevitability that if humanoid robots get rolled out in large numbers that somebody's going to get injured on the job by a humanoid robot?
• 41:16
  SPEAKER_00: All of that is an inevitability.
• 41:18
  SPEAKER_00: The problem is, is do you want it to be greater or less than that of what it is when people do it?
• 41:26
  SPEAKER_00: And like forklift operation is a good example of that.
• 41:29
  SPEAKER_00: Like forklift oper forklift operation causes by humans causes a death approximately every three days.
• 41:39
  SPEAKER_00: So every three days, a forklift up a forklift is being operated, and that forklift kills someone every three days.
• 41:48
  SPEAKER_01: Yeah.
• 41:49
  SPEAKER_00: True facts.
• 41:50
  SPEAKER_00: If I made an autonomous technology and I said to you, every three days this thing's gonna kill something.
• 41:58
  SPEAKER_00: How would you feel about it?
• 42:00
  SPEAKER_01: Depends on the person.
• 42:04
  SPEAKER_01: I'm just saying, like, No, I know, I I know.
• 42:06
  SPEAKER_01: It's the same with it's the same with autonomous cars.
• 42:08
  SPEAKER_00: It's a very simple thing.
• 42:09
  SPEAKER_00: Right, it's the same with autonomous cars, right?
• 42:11
  SPEAKER_00: Like, I mean, we deploy and and when you talk to customers, people see autonomous forklifts as the route to having a safer warehouse, to making sure that, you know, like in all of forklift operation in North America, for example, we don't kill someone every three days.
• 42:29
  SPEAKER_00: Um, and the injury rate is even higher.
• 42:32
  SPEAKER_00: I mean, the the injury rate for forklift operation is is insane.
• 42:36
  SPEAKER_00: It's it's even worse.
• 42:37
  SPEAKER_00: Like, yes, someone may not die every three days, but someone's definitely getting maimed or injured in a pretty serious way due to forklift operation.
• 42:47
  SPEAKER_00: And so now let's like take it away from the death toll.
• 42:51
  SPEAKER_00: Let's just look at like, okay, you know, we're gonna injure, you know, across all of the United States 100 people a day, you know, like with a forklift, autonomous forklift.
• 43:01
  SPEAKER_00: Now, how do you feel about it?
• 43:03
  SPEAKER_00: I've heard this argument before, right?
• 43:05
  SPEAKER_00: Like, people, people hurt people.
• 43:08
  SPEAKER_00: Why can't machines hurt people?
• 43:10
  SPEAKER_00: Well, we created machine directives, we created machinery safety standards so that they wouldn't hurt people, because the bigger problem typically when you put industrial machinery into play is, and this is even true with autonomous cars, is when it's good, it's good, but when it's bad, it's really bad.
• 43:32
  SPEAKER_00: Like uh it's you know, like the the outlying cases are like full death.
• 43:40
  SPEAKER_00: Um whereas like a lot of the times when you look at fender benders between like people, they don't result in fatality.
• 43:49
  SPEAKER_00: Um but if you look at when autonomous cars mess up, they tend to mess up in a more fatal way in instead of a less fatal.
• 44:02
  SPEAKER_01: There's a reason why like so many industrial robots are are still to this day are caged off.
• 44:08
  SPEAKER_01: Um and I wonder how much of the this this human to robot encounter, these safety concerns, are are somewhat short-term concerns because ultimately is the goal to just have a fully automated factory.
• 44:25
  SPEAKER_00: Yeah, I I think that is the goal.
• 44:29
  SPEAKER_00: I think that the problem is is the cost of retooling is too high.
• 44:36
  SPEAKER_00: Yeah.
• 44:37
  SPEAKER_00: Um although, I mean, if I'm being really honest, I think China is proving that that they're willing to throw away all infrastructure to to go after that dream.
• 44:52
  SPEAKER_00: Um But if you if you look at it, it's it's very wasteful, right?
• 44:58
  SPEAKER_00: Like um you you basically have to generationally like basically wipe the slate clean and then build it up from the ground up.
• 45:08
  SPEAKER_00: But the problem is is it's it's really hard to predict you know your throughput and it's highly um sensitive to to like product lifetime, machinery lifetime, things like that.
• 45:23
  SPEAKER_00: But I I I mean, yes, that is the goal, but I don't see it coming in a short-term horizon.
• 45:29
  SPEAKER_00: I think that for humanoids to get deployed in industry, they're going to have to go through a lit warehouse to get to a dark warehouse.
• 45:39
  SPEAKER_01: So we're we're uh coming up on time, but I'll close on this.
• 45:43
  SPEAKER_01: You know, and speaking of of throughput, what does success look like for agility and the humanoid robotics market five years from now?
• 45:56
  SPEAKER_00: Success for the humanoid market five years from now would to be having would be to be having the same traction that the AMR market had five years in.
• 46:08
  SPEAKER_00: I think we're about there.
• 46:09
  SPEAKER_00: I think that that's really where if we're doing that, it means that the industry is going to be very successful.
• 46:17
  SPEAKER_00: I think that it's gonna be challenging to do that because there's a lot of safety hurdles to get over.
• 46:22
  SPEAKER_00: Because remember, like if you look at the AMR market, they came out of the gate fully collaborative.
• 46:27
  SPEAKER_00: Like, not even a problem.
• 46:30
  SPEAKER_00: Um, but humanoids are coming out of the gate basically back where industrial arms came out in work cells.
• 46:37
  SPEAKER_00: And so we have to climb not only the commercialization fence, but the safety fence at the same time.
• 46:44
  SPEAKER_00: But I think that when you look at, you know, like for agility in five years, I think being fully cooperatively safe outside the work cell and deployed in multiple countries, I think will be a big success.
• 47:01
  SPEAKER_00: We plan on having our CE mark in in two years with a functionally safe cooperative robot.
• 47:07
  SPEAKER_00: And I think that is going to be a big determiner on our on our um success.
• 47:14
  SPEAKER_00: And I I think we're gonna hit it, done it before, have faith that we're gonna do it.
• 47:19
  SPEAKER_01: Yeah, yeah, it's funny.
• 47:20
  SPEAKER_01: I I've never thought of it this way, but you know, as comparing it to uh AMRs versus in in industrial arms, it kind of feels like we're letting, at least in some instances, we're letting a lot of babies out into a factory and expecting them to kind of like grow up on the job.
• 47:38
  SPEAKER_00: What's really interesting about the the categories I told you about, the pragmatic, the aspirational, the hardware and the wield, is you'll notice that all the veteran roboticists who've deployed this stuff before are in the pragmatic category because we all learned how hard it is not to build the technology, but to deploy it.
• 48:02
  SPEAKER_00: Um it's actually funny.
• 48:04
  SPEAKER_00: If you if you ask most roboticists what they found was the hardest thing in getting their product or their technology out into the world, almost none of them would say that it was solving the robotics problem.
• 48:20
  SPEAKER_00: Almost all of them would say it was like, oh, building the fleet management or building the IT security platform or dealing with like compliance and um SOC2.
• 48:36
  SPEAKER_00: And but very few would say, oh man, the hard thing was getting slammed to work.
• 48:41
  SPEAKER_00: Like no one, no one literally is gonna say that.
• 48:45
  SPEAKER_00: Like it's gonna be the long tail.
• 48:48
  SPEAKER_00: And so when you when you look at humanoids, like the people that I personally believe, the people that are gonna be successful are the people who get out there in the world and stop making videos.
• 48:59
  SPEAKER_00: Like the longer you make videos, the longer you do all these demos, the further you are getting behind in the realities of deploying the technology and dealing with the edge cases, the crappy problems that happen with reliability, the problems where uh you you can't, you know, you just don't know about when you go talk to a customer and they're like, Look, all I need you to do is move the tote from there to there.
• 49:27
  SPEAKER_00: And then you get on site and there to there is not the problem.
• 49:32
  SPEAKER_00: It's the fact that packout doesn't move fast enough and you got to put labels on the bin.
• 49:36
  SPEAKER_00: And oh, by the way, this person actually part of their job is rearranging everything in the bin to make sure that when it gets the automated process downstream, it all works.
• 49:47
  SPEAKER_00: Um and and that kind of stuff, no one is going to like find in a robotics textbook, you know, and and I think I think and and AI isn't gonna AI it away, you know, because that's business logic.
• 50:04
  SPEAKER_00: That's not technology logic.
• 50:07
  SPEAKER_00: And I think people conflate the two things.
• 50:10
  SPEAKER_00: They they believe that because you can teach a robot to like um manipulate something very complex, that that's going to solve the problem of a large organization that built a workflow to do a process to achieve an outcome.
• 50:29
  SPEAKER_00: And and those two are totally independent of each other.
• 50:33
  SPEAKER_00: And like all the technology in the world can't solve those process problems.
• 50:40
  SPEAKER_01: Thank you so much, Melanie.
• 50:41
  SPEAKER_01: We'll have to get her on again soon to discuss the boat trip that I made up at the beginning of the podcast.
• 50:46
  SPEAKER_01: Thanks again for tuning in to automated.
• 50:48
  SPEAKER_01: If you made it this far into the episode, please like and subscribe and check out the automated newsletter over at automated.org.
• 50:57
  SPEAKER_01: Thank you for tuning in.
• 50:58
  SPEAKER_01: We will catch you next time.