Table of Contents

Episode

Start

[00:00:00] Audrow Nash: Like, I think the feeling is because of things like ChatGPT that there's no reason to go into something like computer science because you're just going to get in a thing and be automated right away for this kind of thing. I think that's the feeling. I wanted to hear your thoughts on it. Like, do you think it's a good time to be a roboticist, or,

[00:00:21] Melonee Wise: I think it's a great time to be a roboticist. This is the, probably the next 50 years are going to be like the heyday of robotics.

Episode introduction

[00:00:33] Audrow Nash: Are humanoids the next big thing? How long before they take our jobs? To get some perspective on this, I talk with Agility Robotics CTO, Melonee Wise. Her answers will probably surprise you.

I think you'll like this interview if you want to understand the current technology around humanoids, what's possible, and where the opportunities are. I also think you'll like it if you're curious about how LLMs like ChatGPT will impact robotics and jobs, or if you'd like to know about manufacturing in the US, especially the challenges and opportunities.

As always, I'm Audro Nash. This is the Audro Nash Podcast. After you listen, I'd love to know. on X or in the comments if you agree or disagree with Melonee's perspective on the timeline for humanoids and where we'll see them first. Also, if you want to talk about this interview or robotics in general, I host a weekly space on X on Thursdays at 9 p.

m. Eastern time or 6 p. m. Pacific time. It's free and has been a lot of fun. All right, here's the interview. I hope you find this conversation as enjoyable and enlightening as I did.

Alright. Hi, Melonee. Would you introduce yourself?

Introducing Melonee + Agility Robotics

[00:01:45] Melonee Wise: Hi, Audrow. I'm Melonee Wise, CTO of Agility Robotics.

[00:01:51] Audrow Nash: Now, last time we talked, you were at Fetch. tell me a bit about what's happened, like, your path, for the last, I don't know, year and a half since we've done an interview?

[00:02:01] Melonee Wise: Yeah, sure. well, last time we talked, Fetch had just been acquired.

I spent about a year and a half, working at Zebra Technologies and then I decided to take some time off. So about six months off, traveled the world, that was a lot of fun. Went all over Asia Pacific, went to Antarctica and then South America. Then I decided I should get back in the game. And, I decided to go join Agility Robotics.

[00:02:39] Audrow Nash: would you give a little bit of background on Agility Robotics?

[00:02:44] Melonee Wise: So, Agility Robotics is a mobile manipulation company. So, it has a humanoid ish Form Factor. That is targeting, tote manipulation for machine assisted operations. So, every time you order something or online, it potentially passed through a tote.

and Digit, our mobile manipulation robot, is one of the robots that might be handling a tote that has something you ordered in it. Or even parts for a thing that you might buy.

[00:03:21] Audrow Nash: Oh yeah, and so a tote, it's just a small bin? Is that what a tote is?

[00:03:24] Melonee Wise: Yeah, a tote is a plastic container that it can be, it can range in sizes quite a bit, so it can be anything from like 2 feet by 2 feet to like 10 inches by 10 inches or, so the range on the size of the container is very large.

[00:03:43] Audrow Nash: Mm hmm. Okay. And so you have Digit. Digit's a humanoid. Digit is helping with these TOTE related tasks. And you mentioned it's humanoid ish. Can you tell me what it looks like?

Introducing Digit

[00:03:56] Melonee Wise: Yeah, so Digit, is probably about five Feet two or three inches tall. it has two, six degree of freedom arms as a head with LED lights for a face.

and it has two legs. however, the leg architecture is a little bit reversed from the leg architecture that we have as people. then a lot of people would say the knees are backwards. However, Because it has more of an avian leg structure, it's actually the ankles are backwards. so it's, it's a little bit different, from that perspective.

[00:04:42] Audrow Nash: Yeah, I think of it like having ostrich legs or something like that. Because when it kind of, and it's cool because I was imagining while watching it that it's kind of pragmatic to have the legs go backwards, because then you can have it drop its and keep the arms in front of it, but you don't have the knees occluding whatever the arms may want to get into?

[00:05:02] Melonee Wise: Yeah. The knees don't get in the way. And it also biases is the center of gravity to drop straight down, which is nice.

[00:05:13] Audrow Nash: It's got a nice squat. Hell yeah. Okay. And so how did you pick to come to agility? Tell me a bit about that decision.

How’d Melonee to join Agility?

[00:05:24] Melonee Wise: so when I, when I, you know, looked out in the robotic space, there were a lot of things that I was interested in.

I thought very briefly about starting my own company, another one, and I decided I wanted a little bit of a break. and so I thought I'd take a less stressful role, as like a CTO, not a CEO, cause that's a very stressful role. And I spoke with a couple of different companies, um, and the thing that really excited me about agility were, was a couple of things.

One, it was very much in a market that was similar to Fetch, so I knew a lot of the customers, I knew a lot of the, the customer challenges, and why people buy. Robots. so it helped me really understand whether I believe that the product had a product market fit and whether people would even want to buy the thing.

Two, Agility's product is relatively mature. They have physical robots. They have been working with customers. I didn't want to start with a company that was still in the does the technology even work, you know, past the first prototype stage. and so that was another aspect of why I was very interested in joining Agility.

And then the, the third thing was, is You know, one of the, one of the things I've always been interested in is mobile manipulation. I mean, at Fetch we had a mobile manipulator, but also as we worked in the AMR market, it was very clear that there was, there's a whole set of tasks that are better suited towards mobile manipulation and not just mobile.

And I thought that that would be an interesting, Direction to go. And if you're wondering why I didn't go and do more mobile, well, I had a non compete. So I couldn't just go off and do another AMR company.

[00:07:36] Audrow Nash: That's very funny. How long does, I mean, I'm just curious about the non competes. How long does that last?

Is it like for five years you can't be in an AMR company?

[00:07:44] Melonee Wise: Yeah, so my non compete, because I was a key person in a material transaction, as they call it, my non compete is three years from the date of acquisition.

[00:07:56] Audrow Nash: Gotcha. That's pretty interesting as a constraint for how to pick the next robotics company.

[00:08:02] Melonee Wise: Well, it It's, it is a material constraint, you know,

[00:08:07] Audrow Nash: yeah, for sure. Okay. So then because of this, I guess, because of the constraint, then digit having legs makes it so that it's a different product in a sense. And so the non compete is not valid. for this kind of thing?

[00:08:24] Melonee Wise: Yeah, because my non compete was very specific to autonomous mobile robots, so AMRs, and didn't apply to other types of mobile manipulation technology.

[00:08:37] Audrow Nash: Okay, and so I interviewed Jonathan Hurst a number of years ago, who is one of, is he one of the founders of Agility? And so he's a professor at Oregon State, right? Who was deeply involved. Yeah, one of the founders and everything. and so when we talked, Agility Robotics was really about the last. My last last hundred feet of delivery this kind of thing so like going from the FedEx truck to the curb I guess curb to doorstep or something.

And I understand from talking with you earlier, I think at Roscon or some other time that it's, pivoted just a little bit. Can you tell me kind of what market you guys are going into and also why not? Last little bit of delivery, this kind of thing.

Pivoting from curb to doorstep delivery

[00:09:31] Melonee Wise: So, like all startups, our product intentions have evolved.

and when you look at last mile delivery or the last hundred feet of delivery, let's call it, there's a lot of challenges that go beyond the robotics technology. Safety, compliance, Weather, many of the, many of the challenges that you see, even in the autonomous car market, Digit would have to face if it was going to do the last hundred feet.

And you're seeing today that, you know, a lot of companies have been successful in indoor semi structured environments. And before I even got to Agility, they were already, converging on product market fit and product alignment that was more geared towards indoor Material Handling. And since I joined, I've been working with the team to really get us focused down on a set of solutions that, really play to Digit's strengths and are scalable and repeatable within our customer sites.

And this is, this is something that, you know, many startups have done in the AMR space, Locus, Fetch, Automotors, we all ended up getting into kind of a set of repeatable workflows that our customers really were excited about. And we're doing the same thing at Agility.

[00:11:14] Audrow Nash: Hell yeah, and so that workflow for Agility is the moving totes.

Is that correct?

[00:11:20] Melonee Wise: Yeah. And there's tons of workflows in the warehouse that involve moving totes. Oh,

[00:11:26] Audrow Nash: hell yeah. And then, so, I mean, like just thinking about humanoids, why is a, like, so a thing that strikes me is that perhaps. And I'd love to hear kind of the, why, the, why Digit is a very good fit for this, but like I imagine a humanoid is going to be more expensive than a robot with a wheeled base, walking is a challenge, probably limits the payload, um, why a humanoid Form factor for this kind of task.

Why not just a mobile base with a big pinch, like, like a two foot pinch gripper that could just grab the totes and move on.

Why Humanoids?

[00:12:07] Melonee Wise: So as someone who spent a lot of time building mobile robots, one of the things that you'll find is there's a lot of companies, including Fetch and Automotors who built base platforms and then got into the business of building lots of accessories, right?

You can see that with Fetch, Geek Automotors, and even, Locus now with, with their acquisition of, Waypoint, right? Whatever. but, but the thing is, is that Is that when you look at, that the accessories start to eat into the payload, right? So like, when you add, so like if you look at a lot of autonomous mobile robots, right?

Like, everyone wants to put a hundred kilograms or more on top of these things, but every bit of shelving or cart or whatever you put on there reduces the payload that you can put on. And then the next thing is, so say you want to put a hundred kilograms of You know, e commerce goods on there. Now it usually gets to some conveyor end point and they typically want you to push the tote off onto the conveyor.

But now you have to build the conveyor, uh, you know, tool, and typically they want to put in multi bins. And so now you need like multi layer conveyors.

And then you end up in this situation where you're building all of this mechanism to do these like really complicated things. And. all at the same time, you're trying to battle the other problem of having a small footprint because most warehouses are not meant to have, you know, 10 foot aisles, they're meant to have relatively narrow aisles because people are relatively narrow, and you're also fighting against stability.

So, as you put mass higher on a very small platform, it wants to tip a lot more. And so one of the nice things about legged or dynamically balancing systems is as you reach higher, you can do other things with your stability platform to enable you to reach and pull weight from, you know, kind of outstretched positions back into your footprint.

And so. One of the advantages of having a dynamically balancing system or a bipedal robot is you can have a relatively small footprint, you can reach relatively high, and you can carry a relatively competitive payload.

[00:14:56] Audrow Nash: Very cool. Yeah, it's an interesting I guess the alternative of an AMR has a lot of trade offs that you run into and so a humanoid is very flexible in what it can do.

You can change the dynamics. It's, it's almost like these environments were made for people, which is very interesting and so you can kind of leverage that. What do you, what do you think about that perspective where it's like one of the big benefits of a humanoid form factor is that the Majority of the world and a lot of the infrastructure was built around people.

So the aisles and the warehouses are thin so that a people can, a person can walk through, not a big robot carrier, like with a huge footprint. what do you think of that idea?

[00:15:43] Melonee Wise: Yeah. I mean, that's just the reality of the world, right? Like. And the thing is, is that it's very hard and costly to change over these, these environments for the infrastructure, right?

And our customers are very targeted at return on investment in under two years. And so remodeling an entire warehouse. For a robot, typically puts you in the 5 10 year return on investment time frame. But if you can drop a robot right in, your return on investment is pretty rapid.

How does Digit compare to people?

[00:16:18] Audrow Nash: That's awesome. And so, our how does Digit compare to people in terms of speed of doing tasks or this kind of thing?

Or what's, maybe it's not even a fair comparison, but how do you think about that?

[00:16:32] Melonee Wise: Yeah, so, typically it's not about Direct Speed Comparison, it's about throughput comparison. Because what you'll see when people do tasks, they do a lot of compound, rapid tasks, and then they wait around a lot of time. A lot of time

They do. people are really good.

[00:16:55] Audrow Nash: So it's like the tortoise and the hare Yeah. Kind of thing. So person goes and does like six things at once. Yeah. And then they just like take a 15 minute break kind of thing, I guess. Whereas the robot can be doing like one every three minutes or something like that.

Yeah.

[00:17:08] Melonee Wise: You find, you find if you go and watch the way people do. A standard activity is it's very bursty, where they have lots of like activity and then no activity. and so what our customers really care about is total, total activity over some timeframe. The other thing that you see is, is people take breaks, right?

Like people over the course of a, an eight hour day. Take at least one hour of breaks. And so that's a lot of time for a robot to catch up in as well, um, in terms of total utilization and total throughput.

[00:17:53] Audrow Nash: What, on the order of how does it compare? So if you consider a long period of hours, what, what are, I guess, long enough so that the breaks and everything average out, what's kind of the ratio of throughput, for digit versus say a human doing a similar role?

[00:18:12] Melonee Wise: Yeah, so we don't have I would say a large enough data set to make any like claims on the direct claims on that

[00:18:23] Audrow Nash: Give any any guesses like is it 1 to 1 is it?

[00:18:28] Melonee Wise: Yeah 1 to 2 is I mean, I would say that that as we Continue to go into the field. We are at parity or slightly better over the long run then Then in the use cases we're deployed in.

So like, obviously there are some high speed activities that Digit most likely will not be doing anytime soon, but in the use cases that our customers have us targeted at, we are at parity or slightly better.

Ethics of humanoid robots

[00:19:01] Audrow Nash: That's amazing. Oh yeah. And then how do you think of like, while we're on the topic of people doing these jobs, I, so I hosted a space on X and so we talked a bit about this as kind of like about humanoids and like food for thought for this interview.

one of the concerns brought up was kind of the ethical one of having people moved, like basically replacing people in different jobs. And I brought up labor shortages and things like this, but I'd love to hear your perspective on kind of the ethics of robots, especially humanoid robots, and kind of the complexities there.

Yeah.

[00:19:42] Melonee Wise: So let, let's, let's separate some of the concerns. So let's start with just labor in general. So when I first started in this industry, the logistics and manufacturing industry, robotics, kind of focused. In, in the 2014 2016 timeframe, there were about 600, 000 jobs available. So, and in that time, all of the robotics companies that you know and love have been deploying robots.

Into that, that environment

[00:20:19] Audrow Nash: and as fast as they can,

[00:20:20] Melonee Wise: as fast as they can. I mean, I don't know if you've seen the latest numbers from Amazon with their Kiva fleet. It's like 750, 000 robots. They've deployed. Yeah. Remarkable. So, so, since then, the labor, the labor gap has grown to a million unfulfilled jobs.

So, so we have been throwing robots at the problem as fast as we can. And the labor shortage has 000 jobs. So that, that's something to food for thought. however. The, the thing is

[00:21:03] Audrow Nash: 400, 000 and since when did you say that? That's since 2014? Between 2016 and today. That's bonkers. Oh my goodness.

[00:21:10] Melonee Wise: Yeah, just look at the Bureau of Labor Statistics.

[00:21:12] Audrow Nash: I think it's the, is it the, the baby boomers retiring? Is that what, one of the forces on this?

[00:21:19] Melonee Wise: It's the aging labor population. It's a, lack of interest in the jobs. it's a, wage pressure problem. Right? and it's also just, you know, I, I'm one of them and you are one of them. As our parents told us that when we would grow up, yeah, we would grow up, we were going to be special flowers and go to college and do whatever our hopes and dreams were.

I don't know about you, but my, none of my hopes and dreams were to work in a warehouse. but, but

[00:22:01] Audrow Nash: that is, what a way to put it. Yeah. But yes, very true.

[00:22:05] Melonee Wise: That's one aspect of it. But I, I think, you know, on the flip side of it is as time approaches infinity. And robots approach infinity. We eventually are going to have to have a conversation about the nature of certain jobs and whether people do those jobs, right?

Technology is continuing to prove that it can create more jobs. Less than six months ago or nine months ago, at this point, nine months ago, no one had ever heard of a prompt engineer. And now it's like. One of the the more random jobs out there for the kind of Quote unquote artificial intelligence economy that we have So technology is creating jobs.

The problem is is whether the people who are currently doing These other jobs have the skill set, the training, and the capability to move into these new, these new labor positions. And the problem is, is that's a socio political issue in large respect. There is some onus on robotics professionals to develop tools that are easier to use.

and that can be cross trained from someone who's doing a warehousing job, but we've got a long way to go. I mean, most people struggle to use their iPhones and web browsers. And we're talking about like robots are basically iPhones and web browsers on steroids with legs and arms and actually, and so, although, So, although technology is going to continue to create more jobs, we have a problem of helping people transition into those jobs long term.

And eventually, the definition of work is going to change. And I, you know, this is a hard one for me because from a, from a personal perspective, I believe in your universal basic income, and I believe that the way we solve this problem is changing the way we view work. and creating a social safety net and then creating a basis for everyone to be, you know, live.

Yeah. To live. And, and if they want more, they can earn more. Like we create an economy that allows for that. and that, that, that doesn't preclude capitalism. It just, Yeah, it enables a basis for living.

[00:24:48] Audrow Nash: It raises the bottom but keeps, you can still be super ambitious. Yeah. yeah, I know, I feel like that is such a complex issue and I don't, I don't know where to go on it.

Yeah. Because I've heard size for everything. What do you think, so related to all the prompt engineers and stuff, I feel Like, like we all thought all the blue collar jobs were going to be automated and then here comes ChatGPT and then it looks like a lot of the white collar jobs are going to be automated.

how, cause, and then that would be interesting because the people that are doing fairly sophisticated work, it's like, okay, now go learn how to be a plumber and electrician from a lawyer or something like this. there, there was a very funny South Park on this a little bit ago. what do you think about that?

Like, what are your thoughts on this in general, all the AI and knowledge workers and automation in general?

LLMs automation and white collar jobs

[00:25:53] Melonee Wise: Yeah, I, I think that it has come a long way. It's very interesting. I think that there is a class of worker out there today that is going to struggle, specifically in the area of knowledge consolidation.

And so that's where you're seeing a lot of the pain right now. in, in like, you know, you can go to ChatGPT and ask for basically a travel schedule, you know, that challenges travel agents. You can go ask for consolidation of legal law, you know, like legal Opinions for law arguments, right? And that's, in my mind, a lot of what ChatGPT is good at right now is consolidation of knowledge.

And so any job that basically is that activity, yeah, there's a, there's a problem, but what we are seeing is ChatGPT struggles with creative tasks, algorithmic tasks in some ways, ironically enough. Well, complex algorithmic tasks, anything that needs, intuition or intuitive thought, anything that needs specialized knowledge.

And so, it's, yeah, it's hollowing out the center, but I don't think that it is getting to the point where it, It's challenging the specialized knowledge sets or anything that requires a human interface or reasoning contextually in this social framework, right? Like a lot of like what lawyers and doctors and other people do is they're making decisions not only based on case law.

But also the social context, the social framework, and the emotions of what is happening at the time. You know, can you imagine, ChatGPT diagnosing you with cancer and just reading it out?

[00:28:00] Audrow Nash: Like, oh my god, right, that little prompt pops up and yeah, it's like, congratulations! Like a sad face emoji.

[00:28:06] Melonee Wise: Yeah, you have cancer.

So I think that there, there is, there is a We're not there yet. It would probably or might get there. It depends on how we evolve as a society and how comfortable we get with these kind of interactions. You can see some of that in some cultures that are more robot facing or more technology adopting. But I think where we're seeing it right now in a lot of the pain is the consolidation of knowledge kind of work group where Yeah, if that's what your job is, yeah, ChatGPT can do it.

It's basically a database searching tool.

[00:28:49] Audrow Nash: It is, and it just formats the information for you. I, I was reflecting on this recently, and the way, the kind of conclusion, or The, I don't know, the, the, what, what I arrived at was that I think it's, it's, there's like knowledge, like how do you do stuff? And there's what do I do?

And it's not very good at the what, like if I'm solving a thorny programming problem, it's not much help if I describe it to it and ask, what can I do? It's super generic advice. But if I, like, I'm having to use a new, large framework at work, and I don't know how to do things in it, but I can be like, I know exactly what I need to do.

What should I do? And it tells me very well. So I'm getting up and running probably like 10 times faster than I would before this. So it's like knowledge is cheap, but like the intuition on how to proceed and understanding all the complexities or nuances of what you're doing, and like common sense checking it.

Seems to be like, there's a ways to go there, in my opinion.

[00:29:57] Melonee Wise: Well, and, and it, it also, I think one of the other things is you have to make sure to check it, right? Because it does storytell, and make up, information. Actually, one of the funnier things that, my co founders and I did was we asked it who founded Fetch Robotics, ChatGPT, and it didn't get it right at all.

And we were wondering about this because two of my co founders are very. Under the radar, they don't have much social media. They don't have much about themselves online and it didn't even know who they were. So, so it's, it's one of these things where it has only the information that it has access to, right?

And so. It is still when, when I say knowledge consolidation, that's what it's doing. It's consolidating knowledge that's available and it has access to.

[00:30:53] Audrow Nash: Yeah, I think that's a big point. I really like that, that way of phrasing it. Knowledge consolidation. Let's see. Going back to Digit, do you, Are there, are you guys attempting to put an LLM or anything on Digit for like having it do, I don't know, or I guess what's your, what do you think about LLMs and their use in robotics, and are you guys interested in trying this at Agility?

[00:31:26] Melonee Wise: Yeah, so Yes, we have an innovation group within Agility, and we have done some pretty interesting demos with LLMs, largely showing how, if you assume that Digit has a set of skills, Like, walk, pick up, place, and you have that interface of skills, then you can start asking Digit from, through a large language model to do arbitrary tasks, as long as they can be broken down into those composite skills.

So, very recently we did a demo day in San Francisco, with a large group of Interested Parties, and one of the things that we showed off was a large language model demo and someone was like, take the box that is the color of Darth Vader's lightsaber and put it on the, podium. Labeled, the same number as the movie number that Darth Vader appeared in for the first time and it did it.

And like, like, there's so much to impact there in so many ways, like, and, but all of that, basically, because it basically Digit was in a, like an environment with some boxes and some podiums that were numbered and labeled and they had colors and other iconography on them. And so, the large language model basically unpacked that all into basic, like an action tree, you know, a behavior tree, and then using the skills that Digit had, it went and executed.

[00:33:22] Audrow Nash: Mm hmm. That's really, it's very interesting and it's so funny that like now you can be like, you can give, you're trying to give your robot instructions and you program it with like a riddle and it has to solve it and this kind of thing. Like you can be like esoteric and whatever. It's so funny. Do you what do you think about its impact like LLMs?

In robotics, do you think, do you think it's going to be like very useful for high level control, or is it kind of a flash in the pan and we can do some neat demos with it? Or is it just like user interface changes where it's, you can talk to it and it's a bit easier to then do action selection from there?

How do you think it sits?

[00:34:06] Melonee Wise: I think we'll initially see it, you know, because one of the things that we are building at Agility is, is kind of this skill framework for Digit. So as we start doing more and more with our customers, probably the first place you'll see us potentially using these tools is for, our customers to describe in natural language what they want.

The, the robot to do from a, from a workflow perspective. It's like, I want you to move this tote from the put wall to the conveyor. If you have a, if the tote has an error, I would like you. To put it in the, hospital, quote unquote, hospital area of the warehouse, but, you know,

[00:34:56] Audrow Nash: So it's super fast templating of actions effectively, like building a behavior tree.

[00:35:02] Melonee Wise: You're, you're taking natural language of like someone who talks in business logic. And making it easy for them to describe that without them having to know like specifically, okay, digit has to walk over here and identify the tote. And then digit has to grab the tote. Like all of that will be basically derived from the natural language description.

[00:35:29] Audrow Nash: That's super cool. Yeah, I do think that kind of thing I could see speeding up robotics adoption. Significantly, because they, because you don't have to have someone who's an expert in the field per se, translate it into like an expert in manufacturing or logistics, go and take all the words, relate them to what it means, then program it in the robot.

So you need someone who's good at both or can communicate with teams that do both. And so now it's like, ah, you just tell the robot and it does it and it translates it. That's very interesting. Okay, so that's kind of the short term. Do you think like, eventually it's going to be a bunch of robots that are all using LLMs for high level control, and they'll just have like action sets that they can do, and you'll get the high level commands?

from the LLMs and they'll just do it? Or what do you imagine?

[00:36:29] Melonee Wise: I'm, I'm guessing that what you'll have is like a natural language description that you can say, and then you basically evolve it until it's what you want, especially in manufacturing and logistics. And then you say freeze and then run forever.

because People want repeatable, deterministic labor from their robots. They, they don't want every time you give it an instruction, it runs the LLM every time and is like, well, today I decided I'm gonna take a tour around the warehouse before I do this.

[00:37:03] Audrow Nash: I'll move this box to over here every single time it has to decide it.

This kind of thing, that'd be pretty silly. Okay.

[00:37:12] Melonee Wise: But you can consider, you could believe that that would be a normal thing in like a home robot, right? And so it's really a domain specific behavior and way of looking at the problem. It's just in our domain, we care about stability, determinism, repeatability, you know, throughput, reliability, all of those things.

Robot factory + suppliers in North America

[00:37:36] Audrow Nash: Yeah, for sure. Let's see. So going back to Digit, um, one thing that I saw that was very interesting on the website is you guys have recently opened like a robot factory, like a big humanoid robot factory. Will you tell me a bit about that?

[00:37:56] Melonee Wise: Yeah. So we've, we haven't opened it yet. We've kicked off construction of it.

You know, it's, it's opening in the spring. and that will Be, you know, it's designed to produce long term, up to 10, 000 robots in anticipation of, of the, the customer contracts that we currently have. Mm hmm.

[00:38:23] Audrow Nash: That's awesome. And so what parts of the making the robot will be done there? Will it be a lot of assembly or will it be Okay, where do you get the parts from otherwise?

Like the, the motors, the sensors, where, where are they all coming from? I guess it's a diverse

[00:38:43] Melonee Wise: Yeah, it's a diverse set of suppliers and fabricators, um, in largely North America. Wow. Hell

[00:38:52] Audrow Nash: yeah. Is that, I guess that's, that's probably by choice to have it mostly in North America?

[00:39:00] Melonee Wise: You know, we're making sure that we are choosing suppliers and vendors that are not in, I guess, conflicted geopolitical regions.

We're trying our best to do that. and we're trying to make sure that we have a good, diversity of suppliers so that we're not single sourced. but a lot of the North American focus is, is because our initial market is North America. And so, um,

[00:39:35] Audrow Nash: Very interesting. Do you, can you tell, I mean, we talked about it in, I think our last interview, about just.

There's actually a good amount of manufacturing here, one of the, here in the US or in North America. one of the things that, another space, so I was talking to Rgs and he was saying that the US is really good for, and if I, if I am mistaken or anything, I apologize in advance. But what I understood was the US is very good at, very specialized.

Like high precision manufacturing, but when you get more to volume, it's tricky. cause it doesn't seem that there's that much of that, but what's your perspective on this?

[00:40:23] Melonee Wise: Yeah. I mean, I think that that is relatively true. except for in automotive, I think if you, if you look at it, it really comes down to the cost of, of bringing the product back into the United States and in what volume and in what kind of import tariffs and things like that that you might encounter.

It for very large high value goods for smaller electronics, high value electronics, you know, things in the one 5, 000 range. One of the things that we basically run into in the United States is just the labor pool. Like we're already talking about logistics and manufacturing, having a million open jobs right now in the United States, right?

Can you imagine producing the iPhone in the United States? Like, I saw, like, an estimate that you would have to basically build, like, a whole town with millions of people to support the iPhone production, in the United States. And so Like one of the big reasons that actually precludes us from building certain things, especially at high volumes in the United States, is we just don't have the infrastructure and the people to do it.

which is, is different in other parts of the world. They actually have, like, if you look at Shenzhen in, in China, Like, they produce so much there, and you could like, walk down the street and go to like 14 injection molders in like, the, like the span of like 5 city blocks, which is something you would never find in the United States, potentially.

and so there's just this contrast of, of, suppliers. The density of them and the sheer amount of labor that's available to support some of that manufacturing.

[00:42:20] Audrow Nash: Yeah, where, so where are most of your Manufacturers in North America, like where in the U S are they in Mexico? Maybe some in Canada, like where, where are you seeing most of them?

[00:42:35] Melonee Wise: I can't speak for all of that because that's not my like, job at, at agility, but we do use, several large North American machine shops and other vendors for making our, our machine parts and things like that. And they're, They're located in the United States.

Reestablishing a trade class in the US

[00:42:56] Audrow Nash: Gotcha. do you like, one thing that has been interesting is the US I think is investing pretty significantly in like, reshoring, a lot of manufacturing.

So I think there's like. I don't know, trillions of dollars or billions of dollars, like some huge amount of money in flight. And that infrastructure is going to take, I don't know, 10 years or something to spin up. But how do you imagine this changing in, say, 10 years, 15 years, if you have opinions on kind of what's, what's been going on?

[00:43:31] Melonee Wise: I think that in order to achieve that, we have to reestablish our trade class as a country. and I think that's, that's something that a lot of people are working on. But, I think that there are not strong incentives right now. for people to go become tradesmen, to become skilled electricians or skilled mold makers, welders, welders.

Yeah. Yeah. And so it kind of creates this vacuum. It, even if we wanted to reshore it all, could we reshore it all? Because we have tended towards. over the last 20 to 50 years, you know, towards a higher and higher, higher education population, that is trending away from trade based jobs. which leaves a gap for trying to reshore manufacturing in the United States.

[00:44:36] Audrow Nash: Yeah. It's such an interesting thing. So like I talked with. on the Sense Think Act Podcast, I talked with like AMT, Association for Manufacturing Technology and a few other manufacturing organizations and some of the things that stuck out to me, it's like, Actually, community college and, like community college is a wonderful way to get a trades education and you can get like, so you train for six months or something fairly short and you immediately get a high paying job for that.

Like not CEO high paying, but like enough to raise a family, probably a couple hundred thousand or more.

[00:45:15] Melonee Wise: Yeah, because, because there's so few trades people, right? And they're trying more and more to get the word out. I know AMT is, I mean, they're really trying. A3 is, they're trying to pull more people into being, you know, robotic operators, for example.

That's a very high paying job, but just getting young people interested in it, them seeing the value in it, and seeing it as a career path that is meaningful.

[00:45:46] Audrow Nash: Do you think, so a thing that I have been aware of is there's, and I don't, this doesn't, definitely doesn't speak for everyone, but I've been seeing a lot of people thinking universities are not a terribly good deal for this kind of thing.

So it's like some people are saying things like I won't have my kids go to college and whatever because maybe trades and this kind of thing could be a good Alternative, and a lot of, a lot of people just end up getting a lot of debt. But, do you think that kind of society's going to start having, like, will we move back into a better balance with trades versus everyone goes to a four year college with not as many exceptions and this kind of thing?

[00:46:32] Melonee Wise: I don't know. I doubt it, if I were to guess because The United States, we have a very manifest destiny approach to life, right? What does it mean? Of, well, if you have a desire, you should go get it, right? And so we raise Our children and our cultural bias is towards achieving what we want, living our dreams.

Passion, passion, passion. Yeah. And until we start reframing, um, some of these things as something that you can be passionate I mean, engineering, for example, has had a really bad, bad rap for a long time, right? Right? You go and you look at TV and every engineer is homely and weird and doesn't have a girlfriend and, you know, sits at home and He's on their computer all the time, and everyone's been told that they have to be the smartest person in the world.

And what have we had for, you know, the last 20 years? A dearth of engineers. And so, you know, we've been trying to change it as a community, trying to help people understand engineering isn't just about being smart, it's about being creative, it's about problem solving. You're not going to be Sheldon. Maybe you want to be, but, you know, you don't have to be Sheldon if you don't want to be.

And, and that is something that like, societally, we have to change about any of these roles that we want people to go into. Whether it's engineering, or a trades job, or, or something like that, but if, if young people don't believe that they're either qualified for the role, in the case of engineering, or it's, something to aspire to, In the case of some trades, you know, how do we convince people that they should aspire to it?

And so a lot of it is reframing what, what people should be aspiring to and raising our children to believe that it's okay to do these things. But if you spend all your time telling your kids the best thing, change the world.

[00:48:56] Audrow Nash: Yeah, these kinds of things.

[00:48:58] Melonee Wise: Yeah, or go to university, then they're going to feel like they're failing if they don't.

[00:49:04] Audrow Nash: Yeah, it's such an interesting problem in a sense. What do you think would be, do you have any ideas what a good solution is? It's reframe it, but how would you reframe it? and I know clearly we've been struggling with this as a community, but what do you think?

[00:49:21] Melonee Wise: I think that's a hard one because I think it, it, It really comes down to what people value, right?

Like, do they value stability? Do they value, you know, wealth? Do they value career growth? and. Those are stories you have to tell, but I also think that one of the reasons people have historically not wanted to adopt trade jobs because they have limited career growth at some point and limited wage growth, for example.

And this comes back to some of the question before, which is, you know, universal basic income. What does it mean to be successful? You know, how do you create, how do you go and reach beyond and, and get more? and we don't have great stories there.

[00:50:25] Audrow Nash: Yeah, so coming up with some sort of good story so that people can, you can just chill and you can do a job, and it can be a meaningful one. That's say a trade, or if you want to go really big and you like. Like it's going to be like, I mean, you with Fetch, for example, I imagine that was, that was a hard path to choose and it was a great, like you did very well and everything is great because of it.

But it was also probably very difficult, I imagine.

[00:50:53] Melonee Wise: And it was high risk, right? Yeah. High risk. But I, I think the other thing that, that has changed. Very much in the last 50 years or 60 years is the disappearance of the pension has also had a very big impact I think on trade adoption.

[00:51:12] Audrow Nash: I don't, so the pension, pensions are like you, the retirement accounts that your company contributes.

[00:51:19] Melonee Wise: No, no, so pensions were The last salary that you made in the last year of work, you got 80 percent of it for the rest of your life. Oh wow. Oh wow. Yeah.

[00:51:35] Audrow Nash: So why did they disappear? They're expensive. I guess that makes sense.

[00:51:41] Melonee Wise: Yeah. Go look up like what happened with GM. Like they had a very large pension program and it almost bankrupted the company.

[00:51:48] Audrow Nash: So it's kind of like, I don't know, you hear about like. The fall of different countries or something. And they eventually are like way over leveraged financially or something to their citizens because of like generous retirement plans, like this kind of happened with pensions. And so we stepped back with it, but it removes some incentive to go do trade jobs that had good pensions.

Interesting.

[00:52:15] Melonee Wise: I mean, there's a couple of jobs still left in the world that still get pensions. Teachers tend to get pensions. Government jobs get pensions, but they used to be that tradespeople got pensions. huh.

[00:52:30] Audrow Nash: Would the, it's the pension, the pension is paid by the employer, correct? So yeah, that does sound expensive for this kind of thing.

And also it's interesting because like teachers, like, I don't know, I have family who, are teachers or, have been teachers and they're not paid very fairly or not very much. So it's like you give them a pension, but like you're only paying them, I don't know, a third of what they probably should be paid at least.

Yeah, you get the thing and you get 80 percent of that, but it's still super, super low. what an interesting thing. I wonder, um, so. Back to robots with Digit. So one thing that struck me from looking at Digit, in some of the videos, it doesn't really have hands. It has like flippers for where I, where we have all the fingers that it can use to pull in and I guess grab things and then it probably pinches.

I don't, I don't remember seeing thumbs. No. tell me a bit about that.

Digit’s hands

[00:53:45] Melonee Wise: So, that's one minute. Instantiation of Digit's hands.

[00:53:54] Audrow Nash: Yeah, very modular, I suppose.

[00:53:57] Melonee Wise: Yeah. You know, when you look at where we're going with the end effectors of Digit, Digit, um, as we evolve the design, we'll have kind of, interchange point, like all.

Robots where we will be able to change out the end effectors of the robot based on the task It's it's what industrial robots have been doing for a long time We've tried to focus on having relatively simplistic grippers to start or I guess end effectors I should say that solve A large swath of problems, with the simplest design.

And that's one, approach that we've taken for doing tote manipulation. And it's fairly robust for some of the, the, the tasks that we've been focused on, but we are right now going to add a different type of gripper to digits repertoire, for handling totes. so it's, you know, We're trying to create MVP products, right?

And so we, we are not trying to solve. And swallow all of the complexity at once. Digit has the ability to have other end effectors. We will make other end effectors for Digit. But our priority is not to make high dexterity hands. Because honestly, like, I haven't seen a problem yet. That, that we need high dexterity hands for digit for the set of use cases that we're tackling right now.

[00:55:44] Audrow Nash: Gotcha. So, it's just unnecessary complexity and you can get most of the way there with just very simple like pinch in grippers kind of thing. Yeah. What do you think, so To me, what it seems is, and this kind of goes back even before you were involved with agility, it seems like it's like the min, as you said, an MVP, it's a minimum product that, can do something useful and can find market fit and this kind of thing.

what do you think of? There's been several companies entering the humanoid space and they seem to be making like full fledged humanoids with not, to my knowledge, they don't have too much of an application in mind for them and maybe they do, but what are your thoughts on some of the other humanoid?

Robot companies or humanoid initiatives.

Thoughts on humanoids

[00:56:43] Melonee Wise: Yeah, so if you look at 'em, some of them are very impressive and have been around for a long time. Like, look at Boston Dynamics.

[00:56:49] Audrow Nash: Mm-Hmm. . Oh yeah. Yeah. I mean, super impressive. some of them are some back folks and stuff, just bonkers.

[00:56:57] Melonee Wise: Yeah, and well, and, and remember they, they started from a very different place, right?

They started a long time ago as the Petman and they were testing hazmat suits, right? Yeah. And they're hydraulic based. And so they have different challenges. Superpower. Yeah. But I think that when you look at a lot of the stuff that sprung up recently, a lot of it is very startup heavy. they're still even just trying to figure out why they're building it.

many of them haven't declared a market or, or shown my impression too, or shown working robots. I mean, let's be honest, there's a lot of videos, not a lot of reality, and, and I'm not trying to criticize, I'm just saying that this is, you know, I went through this in the early days of the AMR market.

There was like five companies who were actually building hardware, and everyone else was showing really cool videos of hardware that they were going to build. Right. and, and we're, we're in that stage right now with mobile manipulation robots. and the thing that I find interesting though, is, is some of the, the players that are getting into the game is recently are kind of funny.

They're just throwing a lot of money at the problem and just hiring any engineer they can. And they're like, you roboticists, you guys are taking too long. It's like, what the hell? As a community, I've been working on these problems for a very long time. And like, magically, like thinking that you can throw a ton of engineers at it and like get good results is I don't know.

It's a mythical man month, basically, and We'll see. But I, I think that there are some really interesting competitors out there. I, you know, and I'm excited about what they're building because, you know, there's plenty of room. I mean, look at what happened in the AMR market, you know. Many companies, yeah.

Yeah, there's Locus, Auto, Mirror, Fetch. You know, we all did very, very well,

[00:59:24] Audrow Nash: So you're thinking of it like AMRs, I suppose, seeing all these humanoid companies. I was, I was feeling like it was a bit like autonomous vehicles with, all the, like the, the feeling that I'm getting with the investor interest and kind of the hype cycles around it.

It feels a lot like the early days of autonomous cars in like 2014 or something when people were like We're going to have self driving cars in four years or something like that.

[00:59:57] Melonee Wise: They haven't put enough money in for you to believe it's like autonomous cars yet. It's more like AMRs. Gotcha. Yeah, like the total amount of money in mobile manipulation right now in humanoids, quote unquote, is closer to the AMR space still.

Like the total dollars put in from venture.

[01:00:16] Audrow Nash: How much would that be, out of curiosity, if you have a good guess? Is it like tens of billions or is it?

[01:00:23] Melonee Wise: It's under five to ten billion. That's like. That's AMRs and then over 5 to 10 billion is automotive.

[01:00:33] Audrow Nash: Over 5 to 10 billion. How much, how much more over? Is it like 50 billion or is it?

[01:00:37] Melonee Wise: I don't know. I don't, I, I, I'm guessing it's in that, in that frame range because like just look at the, what wasn't there a, like an autonomous car company recently that got some massive, insane amount of money. I don't know. It was like, what was it? Well, was it, I thought it was 80, no, 8 billion or something insane like that?

[01:01:09] Audrow Nash: That's quite a lot. I wonder which company that was.

[01:01:15] Melonee Wise: Yeah, I don't know, let me see

[01:01:19] Audrow Nash: if I can Look if you like. Yeah, we have time. Yeah. One of the wonderful things about these long forum Things doesn't really matter. and I'll cut it out if we wait too long.

[01:01:30] Melonee Wise: Yeah, I'm trying to find it.

[01:01:37] Audrow Nash: so we couldn't find the company that had the high valuation, but, so thinking more about humanoids. What do you imagine is a timeline for them? so you're going to, you guys are working on them in a logistics and manufacturing space. what do you kind of imagine the progression? Like when will I see one in a grocery store kind of thing or like, like when will I see them in my day to day life or will it be, will they be relegated mostly to like manufacturing and logistics for a while?

Okay, so I mean, but there's a bunch of, it seems like there's a lot of excitement where they're like, in two years, there's going to be one in your home.

Timeline for humanoids

[01:02:28] Melonee Wise: Yeah. Okay, Audrow, let's reflect. Yes. In, in 2004, 2004, they said that everyone will have an autonomous car in 10 years. Oh, man. Oh, man. So, now, in 2004, if you derated that and you said autonomous cars would be part of your everyday life, do you feel that's true today?

Because I don't. Okay. So now, today, we're starting with, with, you know, humanoid bipedal mobile manipulation robots, right? I don't think that that'll be part of your daily life for another 20 or 30 years. I think we'll spend the next 10 years in industrial, light industrial environments. There's a lot of safety work that has to be done to get them out of the warehouse and into your house.

[01:03:41] Audrow Nash: And just, I feel like I have to ask, This doesn't get super accelerated because of LLMs, or what are you thinking there?

[01:03:53] Melonee Wise: no, because it doesn't make the hardware any cheaper. It doesn't make the controls any easier. It makes the programming of them a little bit easier, but it doesn't make some of the more fundamental problems.

Like do LLMs make autonomous cars go faster? So why would they make bipedal navigating robots go any faster either?

[01:04:19] Audrow Nash: Yeah, now, thought experiment to just, what if, so if OpenAI reveals that they have made super intelligent something or another and it's like Einstein for every single field all at once, how does that work?

Like, does, we still have all these hard problems and the timeline kind of stays unchanged or what do you think?

[01:04:45] Melonee Wise: I think the timeline still changes. It remains unchanged. And maybe that's naive of me, but like, I mean, technology has been constantly progressing. You know, and although ChatGPT is super interesting and showing very interesting progress, how much has it fundamentally changed your day to day in the last 12 months?

[01:05:12] Audrow Nash: It's a better Google to me is what it's been.

[01:05:15] Melonee Wise: It's not, I mean, yes, people are very excited about it. It's very powerful in very specific ways, but it's not the, You know, singularity moment that, that everyone, it's nowhere near that, right? It's, it's like you said, it's a fancier Google right now. and I, I think though that like, you know, some of the thoughts that Bill Gates put, Bill Gates put out about like personalized agents is kind of interesting.

but there's a lot of other things that we have to, to deal with. Or I'm like, what if, what if you had your own ChatGPT?

What if you had your own ChatGPT?

[01:06:04] Audrow Nash: Hmm, like they had a history and this kind of thing.

[01:06:07] Melonee Wise: That you know, that was like it took all of your data, all of it. It has everything about you, your medical records, everything, everything you ever wrote, everything you ever did, all of it.

And was like your own personal agent and could be your the business version of you and your home version of you and did all of these things and could manage the complexities of your life. Super cool, right? Like that's probably the next thing that might happen with this kind of technology. It would be super cool.

Yeah. But there's so many problems we have to figure out in order for Someone like me and potentially you to even want to give it all of our data And then what it what do we do when people want to advertise to us with that data? And how do we set our own boundaries? And how do we deal with the complex social interactions that come out of that?

Like, okay, so you and I have personal assistants. That are these agents, right? Yeah. And you, you say, you say to your agent, hey, see if Melonee wants to have dinner. Okay. And my agent is like, well, Melonee's already got plans with friends that we are both friends with. Oh yeah, yeah. But you weren't invited, right?

And so do you want my agent to go to your agent and be like, well, you know, and so how do you even keep secrets between agents? How do you define that? How do we like, and so I think it's a super interesting space, but I think there's a lot of. Social fallout that we haven't thought about. And it's the same with all of HRI, right?

Like there's all this contextualized social interaction and it's not just, you know, in quotes, social interaction. It's, it's highly contextualized to our, you know, region, our, our cultural backgrounds, things like that. And. I don't know if we're ready or we're like, we're at that point where the technology has that sophistication yet.

[01:08:14] Audrow Nash: that's a good point. Yeah, I guess. Yeah, there's a whole bunch of things that need to be worked out and you really have to see how people react. So this kind of thing can't really be rushed, I would imagine.

[01:08:25] Melonee Wise: Well, I mean, they'll always be the people at the forefront and they'll be learning all the painful lessons.

[01:08:31] Audrow Nash: Yep. Yeah, that distribution of like the early adopters and this kind of thing. What do you, so one thing that's been interesting to me, there's again, from a recent space, we had someone come on, that's like just about to graduate with a degree in computer science where they, feel like it's a hard.

Like, I think the feeling is because of things like chat GBT, that there's no reason to go into something like computer science because you're just going to get in a thing and be automated right away for this kind of thing. I think that's the feeling. I wanted to hear your thoughts on it. Like, do you think it's a good time to be a roboticist or,

Best time to be in robotics or worst time?

[01:09:15] Melonee Wise: I think it's a great time to be a roboticist.

This is the, probably the next 50 years are going to be like the heyday of robotics. Like of, of mobile robotics, mobile, mobile manipulation robotics. I think from 1960 to 2000 was probably the heyday of industrial robotic arms. 2000 to 2015 was probably the heyday of collaborative robotic arms. you know, 2014 till now and going forward is the heyday of autonomous mobile robots.

You know, I, I think that. And I think that, that now is the time. I mean, I think that Willow Garage really kicks something off and we're like, we're in it. So yeah, you should definitely become a roboticist. It's our time.

Advice for new roboticists

[01:10:11] Audrow Nash: That's my feeling too. I have a feeling that it is probably the best time so far to be a roboticist.

What do you, so for someone who's feeling lost, especially with all the advancing technology. What advice do you have for them? Like, how, how do they, how do they get a foothold in this world? And I don't know, how do they do well?

[01:10:34] Melonee Wise: Learn to, yeah, learn to program, learn ROS.

[01:10:39] Audrow Nash: Hell yeah. The, the thing that I was thinking, it's, It's almost like everyone gets a bunch of junior programmers beneath them, for working. And it's like with JGPT, you get an assistant for this kind of thing. It's like everyone is kind of their own CEO, and you have a bunch of people to delegate to that are all AIs, is how I've been thinking of it a bit.

It's not so much. Like you still have a lot of autonomy in how you choose to move forward and you don't have to do a lot of the grunt work, from these things. And like, I mean, I'm, I'm using Angular, which is a complex web framework made by Google in work for Intrinsic. And, I am up and running super fast because of ChatGPT.

It's like, that's a, that's a real power there.

[01:11:27] Melonee Wise: Yeah. And I think the other thing is, is. I don't know what your early career space was like as an engineer, but there, there was this time early in my career where I was always afraid to ask a question, you know, and like, go bother someone who knew more than me.

And now you don't have to, you can bother ChatGPT, you know, like, so your ability to ask questions and to fail. You know, or, or ask dumb questions is totally enabled by having this knowledge consolidator.

[01:12:10] Audrow Nash: Yeah, I think that's a good point. I do think there's an interesting case there where it's actually not very good at answering questions about the bleeding edge of things.

And so It's very good at answering everything that's kind of in the like very well known space, but I wonder, I feel like that's something to be a little wary of, where if you're like, I don't know, how do I do this super hard thing and it gives you a super generic answer, like talking to you or someone who's like actually done very hard things is going to be so much more informative than just asking these same questions to ChatGPT.

[01:12:48] Melonee Wise: Sure, sure, but that's when you're in the mid cycle of your career. I was speaking like early stages of your career You just got out of university and you're struggling with something that you know You feel like would be almost wasting your mentor's time, right?

[01:13:06] Audrow Nash: Yes, that is a very good point. Yeah, you could just bug ChatGPT.

Yeah, what an interesting thing. So we are Coming to the end of the time one of the things that I wanted to ask about And we've just been talking about other things, but, so. Agility was doing Amazon trials, with their robots. Yeah. How did that go? And tell me a bit about that. It's great.

Amazon trials

[01:13:32] Melonee Wise: We, you know, there were a whole bunch of, news media articles put out.

There's a bunch of videos. It's going really well. so we've been working with Amazon for. Quite a while. in, in some different applications, you saw one of the applications in some of the videos that were highlighted, as part of their demo day. And now we're moving on to some other phases of the project, which are really exciting.

and we're continuing to work with them on, on these projects and excited to start deploying more robots with them.

[01:14:13] Audrow Nash: Yeah, that's so cool. Are they, how does it, I suppose you have other customers too, but they're one of the big ones and they're one of the, like, I mean, you said they have 750, 000 robots, if I remember correctly, Kiwa robots, which is bonkers.

So, I mean, there's clearly lots of potential to scale. I think that's really awesome, but you're, you're also working with other companies too. Oh, yeah. And all very same, similar use case at the moment, which is picking up those totes. Okay. Super cool. I'm glad that's going well. It's so cool. I, I really, it's exciting to see a humanoid robot that's doing a really.

Practical job in a sense and making a lot of sense for the ROI of these like justifying itself with a good return on investment for the companies that are investing in it because I think that was my big skepticism for the space was I think it'd be hard to get a good ROI for a lot of the. More complex ones, at least initially.

[01:15:23] Melonee Wise: Yeah, but customers wouldn't be working with us if they didn't believe that there was a return on investment.

Agility Robotics in the next 2-5 years

[01:15:30] Audrow Nash: Hell yeah. Let's see. So what, what do you think is the future? Like tell me the next two, five years. For agility, where are you guys headed?

[01:15:42] Melonee Wise: Yeah. So over the next couple of years, we're going to be very focused on expanding our, let's call it skill set.

So, you know, as I was telling you, we look at Digit as a platform that has composable skills. and as we start working with more and more customers, we're going to be expanding the set of skills that Digit has, so that would be in the areas of. You know, tote manipulation, but also tote stacking, destacking, tote, tote wrangling, those types of things.

But then also moving into other types of containers, like corrugate boxes, palletization, depalletization. And so it's just Looking at the, the space of, of activities in the warehouse and slowly branching out across a swath of similar activities within the warehouse. So there's a lot of processes that require taking some kind of container, whether it's a tote or a box, um, from a shelf or to a shelf.

From a conveyor or to a conveyor, from a cart or to a cart. and so now you've got all these skills. It's like, okay. And then now, if you know how to take something to and from a conveyor and to, and from a shelf and to, and from a cart, now you can go from a shelf to a conveyor, from a conveyor to a cart.

Right. And so we look at it as, Building up of a composable skill space that then eventually can be deployed into different applications. And eventually, Digit has all the skills to form the basis for an app store for labor. And then you start looking at the, the workflows and the tasks that Digit can do.

And, you know, building out the next thing based on the skill set that Digit already has. And so as you start gathering all these skills. It's like any person. The more skills you can do, the more jobs you can do.

[01:17:55] Audrow Nash: I like that. Yeah. It's a, it's a cool thing. You keep building the capabilities that keeps opening up applications that keeps letting you grow your market.

And then it's just, it's like a nice flywheel in a sense. And you said app store at some point like that. Yeah. Yeah. It's, it's so interesting that app stores are, like it's kind of, I guess maybe where

[01:18:15] Melonee Wise: we'll call it a skill store. Fine.

[01:18:17] Audrow Nash: Skill store. Yeah. Yeah. Skillstore. But it's just, it's so interesting because I, in a lot of these interviews, I talk with them and then it's like the long term vision is to get to something like a, an app store or a skill store for this kind of thing.

And it makes a lot of sense because then you have the diversity of application that, like it becomes generally useful. Do you, do you think that, Humanoid robots are going to become like, will they be like the silly metaphor, but will they be like the spreadsheet of the computer age where it's like, you have something flexible enough that it justifies things, places getting one.

And then from that, like you get it, it does its core application and now a bunch of people are buying it, but then you can also add other. Programs on it, that also provide some value like spreadsheets, justifying computers.

Humanoids and spreadsheets? Is this the tipping point?

[01:19:16] Melonee Wise: I think so. I think, I think the thing is, is in the industrial landscape, it's harder to do that because.

The primary motivator for a return on investment is, is like the primary work task. but I think when you look at like, if you look at like, retail applications or storefront applications, you know, Forgoing the fact that Pepper had limited utility because it had limited utility, if Pepper was a fully capable humanoid robot, like, you could believe Then yeah, then you would have kind of Pepper's primary activity, but there was probably a whole bunch of other things that Pepper could have done where LLMs, ironically enough, would be perfect.

Yeah, for sure. Like imagine walking into a retail store and saying, Hey Pepper, can you help me find a pink blouse? Or, you know, then that becomes very interesting in that space. But I think when the return on investment In the industrial application is very. You know, task oriented as opposed to like a retail or grocery space or a, hardware store, for example, you know, like if you ever tried to find someone to help you find the screw you're looking for at Home Depot or Lowe's.

It's, it's like searching for a needle in a haystack.

[01:20:51] Audrow Nash: And then they're walking somewhere and now you're following them. Yeah, this kind of thing. Yeah, for sure.

[01:20:56] Melonee Wise: And so, but imagine if, if the utility of those types of robots is, is more of what we're talking about, where maybe the, the original application that you maybe have is, restocking for the robot, but when it's not restocking, an individual can walk up to it and ask it to find a screw.

[01:21:16] Audrow Nash: Yeah. Or even while it's restocking, just like the people that are working there. Okay, very interesting. do you, let's see, like, I guess, wrapping up, what are you excited about in robotics now, in general?

What are you excited about in robotics now?

[01:21:34] Melonee Wise: Hmm, what am I excited about in robotics? I don't know. I, I think I'm, I think I'm most excited maybe about, the, the growing interest in making robots usable.

and, and I, I, I think that that's, that's something that's still going to take us a long time as a community, but I'm, I'm very, excited by the progress we're making there. I, I somewhat wish there was like. an academic version of an, industrial conference that was more like an academic conference where like companies could go and just present their HRI work.

I think it would be very interesting because one of the things that you, when you look out into the. Or their usability work for robotics is there aren't a lot of places for us to talk about it as a community and, and a lot of the research has limited data sets that are limited to, to like university students or whatever users they could scrounge up on a Sunday.

as opposed to some of the companies that have, you know, thousands of hours of interactions with hundreds of people at a time, kind of data sets. and, and I'd really, I. I wish that we had more of a community and an opportunity to talk about and a venue for talking about kind of how do we advance usability for robotics.

[01:23:23] Audrow Nash: Mm hmm. Why do you think someone's not doing that? Or is it a new idea? Because I think it seems like a great idea.

[01:23:32] Melonee Wise: probably some of it is proprietary work. Like, I will admit that I had a very strong interest in it, but Fetch never showed any of its UI ever. It like, like, there's very few videos of it online.

Wrapping up

[01:23:47] Audrow Nash: Yeah, gotcha. Everyone's holding their cards close to their chest for that kind of thing. Yeah. Okay. Well, uh, do you have any links or contact info you'd like to share with our watchers and listeners?

[01:24:02] Melonee Wise: I don't know. I'm on Twitter, Twitter or X, whatever it's called these days. And, at Melonee Wise, I mean, all my handles are Melonee Wise.

[01:24:13] Audrow Nash: Okay. Hell yeah. And I'll put a link to Agility in the episode. Hell yeah. Okay. Well, it's been great talking to you. And hearing your opinion on a lot of things. it's an awesome perspective and I really value it.

[01:24:26] Melonee Wise: Awesome. it's nice seeing you and hopefully we'll grab a beer sometime.

[01:24:31] Audrow Nash: Hope so. All right.

See ya.

That's it. I, for one, had my opinions changed on humanoids from this interview, but what did you think? Do you agree with Melonee that we're not going to see humanoids outside of manufacturing and logistics for 10 years or so? What other low hanging fruit might humanoids be used for? If you're not already, consider subscribing to never miss an interview and I'll see you next time.