Agility Robotics CEO on how Digit 5 fixes "islands of automation"

Most humanoid robots doing real work today do it inside a cage.

Even the ones with paying customers have typically worked behind physical safety barriers, fenced off from the people on the warehouse floor. That keeps humans safe. It also keeps the robot doing one job in one spot.

And that, of course, is a problem when the whole pitch for a humanoid is that it can go where people go and do what people do.

So Agility Robotics launched Digit 5 earlier this month as its answer: a humanoid built to work cage-free, near people. The headline numbers are out there already, and we covered them here: 50-pound payload, 10:1 run-to-charge ratio, $300 million in orders, and a $2.5 billion SPAC that will make Agility one of the first publicly traded humanoid companies.

So when I sat down with CEO Peggy Johnson for the podcast, I was less interested in the what than the why. Why the cage matters so much, what three years of commercial deployments taught Agility, and how a company with real customers thinks about a market that's suddenly packed with competitors.

"That barrier really has kept humanoids from scaling," Johnson told me. "So bringing that barrier down is going to allow us to scale throughout these facilities and take a number of different jobs on in different locations."

The reason comes down to how factories and warehouses actually look. Johnson describes them as islands of automation: a conveyor belt here, a put wall there, AMRs rolling product in from the loading dock. Almost none of it moves, and none of it is going anywhere soon.

"Humans are the connectors right now between all of that automation," she says. "That automation, with the exception of AMRs, doesn't move. It's deployed and it's there for another 10-plus years."

Inside a safety cage, a humanoid is just one more island. At GXO, Digit 4 has moved 100,000 totes doing essentially one task: picking them off AMRs and setting them on a conveyor. Outside the cage, though, the same robot can chain jobs together, like palletizing, machine tending, and picking and placing parts.

"We knew if we could come outside of the safety barrier, we would be able to do end-to-end workflows, and that's what we'll accomplish with Digit 5," Johnson says.

That's a different model from most industrial automation, which is built to do one thing very fast, a million times a day. There will always be a place for that, and humanoid robots are not going to take that place. But the flexible, judgment-heavy work in between is where humans have traditionally been plugged in, and it's where Agility thinks humanoids fit.

The demand is there, Johnson says, because the humans are getting harder to find. She expects the labor gap in these facilities to grow about four and a half times over the next several years, driven by what she calls a trifecta: younger workers who want digital jobs, a retiring workforce, and a wave of reshored manufacturing.

A lot of what's new in Digit 5 came from customers rather than engineers. The payload jump from 35 to 50 pounds, for one. And the battery.

"When you're in a lab, you don't think about that, but when you're in front of real customers and they have a labor gap and they say, 'Can one robot take on three shifts?' we stepped up to that demand," she says.

Agility's answer was fast charging rather than swappable batteries, a call Johnson says followed the EV industry. Swappable packs come with baggage in a factory: they "have to be protected and held apart from other parts of the automation," she says. Nine minutes of charging for every 90 of work gets Digit to about 20 hours a day.

Then there's safety, where the rules simply didn't exist yet.

"There were standards for autonomous mobile robots, which are wheeled robots, which really have a lower safety bar to meet," Johnson says. "But when you are as large as a humanoid and you have limbs, very powerful limbs that can pick up boxes like this, 50-pound weights, you need to have safe maneuvering and safe handling."

Agility has been working with ANSI and ISO on a humanoid safety protocol. In the meantime, Digit 5 plays it very safe. Seven pairs of cameras give it stereo vision all the way around. ("We don't have eyes behind our head, but Digit does," Johnson says.) Someone walking by at a distance makes Digit stop and wait. Someone closer triggers what Johnson calls "that extraordinary action to bring itself down and to power down its limbs," a crouch it holds until the person has passed.

In the demo footage, that buffer looked like 10 feet or so. That's a lot of caution, and Johnson is fine with it. The zone should shrink as Digit logs more hours in messy real-world environments and regulators get comfortable, she says.

"We're starting very cautiously, as the EV industry did."

(Johnson reached for the EV industry three times in our conversation: for charging, for safety, and for software upgrades. For a company about to go public, it's not a bad comparison to lean on.)

Hands, it turns out, are overrated.

"Five fingers generally in these environments are not the most useful tool," she says.

Instead, Digit 5 uses a standard ISO flange at the end of each arm. Agility builds some end effectors itself, based on what it's seen customers need, and can plug into the grippers that 60 years of industrial robotics has already produced. Heavy grippers for 50-pound boxes, small tools for piece parts, three-finger grips where they fit.

I also asked Johnson the religious question: legs or wheels? 

She doesn't see it as a war. Digit works with wheeled robots every day, and Agility's Arc cloud platform coordinates it with conveyors, put walls, and AMRs.

"There are trucks, there are SUVs, there are small cars, there are race cars," she says. "They all do a different thing, but they all operate on the same road."

Legs earn their place in narrow aisles and when lifting 50 pounds up to a high shelf, where wheeled robots, she says, "get a little tippy."

The competition is the other thing I wanted to ask about. We're tracking more than 400 humanoid manufacturers and roughly 800 robots on HumanoidMegaHub right now, with Chinese automakers, American startups, and European companies all piling in. Johnson's case for Agility rests on the unglamorous stuff: 65,000 hours of real-world operation, a fifth-generation robot, and customers who sign contracts.

"It's one thing to dance on a stage," she says. "It's another thing to do real work and actually have someone pay you for it because it's meaningful to them."

Those customers have been Agility's best teachers, she says. Deploying with Amazon, GXO, and Schaeffler meant learning compliance, meeting safety requirements, and plugging into existing systems.

"So it's not enough to throw a piece of metal over the wall and say, 'Here, it can dance.' It has to be able to do useful things."

Agility is also well positioned for the FCC's recent ruling on imports, which has sent some competitors scrambling to restructure supply chains. Asked whether it causes problems, Johnson was blunt.

"No, it doesn't cause us any problems. In fact, we welcome the administration's push for fair competition."

Agility designs in the U.S. and assembles at its RoboFab facility in Salem, Oregon, with a skills-training office in Pittsburgh and a new hardware and physical AI office in Fremont, California. With a CE mark, Digit 5 will deploy in the EU and U.K. as well.

On training, where some competitors are spending enormous sums on data, Agility is taking a blended approach. "We've found the best path forward is a mix," Johnson says. That means its own operational data lake, simulation in NVIDIA's Isaac Sim, and learning from demonstration, where an operator teleoperates Digit through a new task a few times before simulation and real-world data are layered on top.

That's the part Johnson most wanted people to watch. New skills will ship to Digit 5 over time — much like new features arrive on an EV — which means the robot customers get in 2027 won't be the robot they have a year later.

"What Digit 5 can do on day one will be very different than what Digit 5 can do a year later," she says.

Early-access deployments start in the first half of next year. We won't have to wait long to find out.

Much more in the full transcript

John Koetsier: Peggy, it is so good to see you. It's been a long time. We met for the first time at Web Summit, and you were just talking about Digit and all the wonderful things you're doing. And now, of course, your life is insane. You're going public via SPAC, your daughter's getting married, many things are going on, you're launching something new.

John: Tell us about it.

Peggy Johnson: So we are launching Digit 5, and it's our first humanoid that's designed to work in close proximity with humans. That means there's no more physical safety barrier that we've had around the robot up to this point. And that barrier really has kept humanoids from scaling. So bringing that barrier down is going to allow us to scale throughout these facilities and take a number of different jobs on in different locations.

John: You've teased it a little bit before. That's exciting. You have a deck, I think, that shows a little bit about Digit 5 and its capabilities. Do you want to show that now?

Peggy Johnson: Sure, sure. We've been out there for many years. We've been deployed for three years and gathering data from customers, understanding what these environments are all about. We've got 65,000 hours of real-world operation, and through all that, we learned that Digit needed to lift heavier payloads.

So we've increased the payload by about 40 percent. We've gone from 35 pounds to 50 pounds. The other thing we learned is that Digit needs more battery in order to operate for three shifts. That was something—you know, when you're in a lab, you don't think about that, but when you're in front of real customers and they have a labor gap and they say, "Can one robot take on three shifts?" we stepped up to that demand.

We found a battery that operates in—it's called a run-to-charge ratio—and it's 10 to 1. So you can imagine 90 minutes running, nine minutes of charging. Very fast, quick charging. That means Digit can operate 20 out of 24 hours in a—

John: Nice. Nice.

Peggy Johnson: Yeah.

John: Swappable battery. That's just, go charge it 10 minutes. There I go. I'm done.

Peggy Johnson: It is. And we didn't do swappable. We kind of followed the EV industry in that they biased more toward a fast charge rather than swappable. There are a lot of other elements around swappable batteries. They have to be protected and held apart from other parts of the automation in a factory. That just brings in a lot of additional elements. So having a fast-charge battery really solves a problem for these facility managers.

John: Love that look. If we can go back a slide right there where Digit is on its heels. I can't do that personally. I don't have flexible enough ankles. That gets really low. Wow. Impressive.

Peggy Johnson: It does. First, Digit is built to operate in areas that humans operate in, but can also do a little bit more. It can get down into the squat position, it can lift high overhead, those 50-pound weights that I was talking about, and it can do it over and over and over again. And it doesn't have a knee problem. It doesn't have a back problem like humans. This is difficult work for humans.

So humans applaud this kind of robot coming in and taking the tasks off their hands that are very repetitive, very dirty, sometimes dangerous because of the repetitive motion on knees and backs. And that's something that has caused this labor gap.

Over the next several years, that labor gap is going to grow about four and a half times because of a few things, really. There are three things going on. Younger people want digital jobs. Largely, many of them are seeking that first, so not many of them are stepping into these jobs. There's a retiring workforce that's happening, then we have a lot of reshoring of manufacturing. So it's a trifecta of pressure for these facilities looking for humans, and that's what's caused this large and actually growing labor gap.

So to have Digit 5s being able to step in to operate in different parts of the factory and support the humans, it's really an "and" that allows the facility managers to get the job done.

John: So two things that people are always telling me when I go visit a robot factory or a place where robots are working. One, they say there's no regulatory environment yet for bipedal robots. There's just one for wheeled robots. The second thing is no robot can work together with humans in the same space.

They have to have the cages. Are you breaking both those boundaries today?

Peggy Johnson: We are. In fact, we have been helping—we are involved in the standards operation. Well, here's Digit now outside of the safety barriers. You can see Digit able to pick up parts with very high flexibility, place them, palletize, depalletize, and go back and recharge when it needs to.

John: Mm-hmm.

Peggy Johnson: And now you see a human here coming by. Digit detects that. It has seven pairs of cameras over it, so it actually has very good 360-degree vision to keep humans safe when it comes into close proximity.

John: You say seven pairs? So it has binocular vision wherever it's looking?

Peggy Johnson: Wherever it's looking, which actually makes it very safe because we don't have eyes behind our head, but Digit does. So we have cameras watching all the time, monitoring, detecting when humans come in.

And you're right, there was no standard for this. There were standards for autonomous mobile robots, which are wheeled robots, which really have a lower safety bar to meet. But when you are as large as a humanoid and you have limbs, very powerful limbs that can pick up boxes like this, 50-pound weights, you need to have safe maneuvering and safe handling.

We've been working with the ANSI and ISO standards to incorporate a humanoid safety protocol. And so Digit will deploy that in Digit 5 when we deploy our safe human detection at the first part of next year.

John: So that's pretty exciting. That sounds, first part of next year, I guess that sounds like first quarter or something like that.

Peggy Johnson: Yeah, the first half we'll be doing early-access deployments with our existing customers. And then the first half of the next year we'll be doing wide deployments across a number of customers.

John: In the video you're showing there, Digit was walking with a reasonable speed, right? I mean, it's not running around and it's not going crazy, but it was walking with a reasonable speed. It looked like a reasonable, normal human gait. It wasn't significantly slow like we've seen some humanoids.

Is that correct?

Peggy Johnson: That is correct. So you don't want it to be super fast because it is operating around humans, but what you need to do is meet the metrics that these facilities need: throughput metrics, accuracy, and uptime. Here it's detecting a human, and then proceeding once the human has safely passed.

John: So that's interesting. In that case, it just stood there and waited till the human had passed. In the other case, it actually went down to sort of a low crouch. That's just by design, that when a human is near, I'm going to stop moving, let the human do what it does, and I'm going to continue when the human is gone.

Is that correct?

Peggy Johnson: Yeah, and the difference between those two videos, in the first one the female worker was much closer to the humanoid.

John: Mm-hmm.

Peggy Johnson: Digit 5 takes that extraordinary action to bring itself down and to power down its limbs. In the second video, the gentleman was walking outside of a barrier, or I would say a radius, that would cause any damage to him.

So he stops, proceeds on his way there.

John: Essentially, you're operating with an excess of caution there, right? I mean, you're essentially going like, "We're not even going to come close to having a problem because we're staying..." That was like 10 feet away or something like that. In other cases, you're staying significantly far away, even going down.

I assume that as things prove out, as things test out, as you get more familiarity and confidence and regulators get that as well, then you can be a little less, "I'll stop whenever a human comes within 10 feet," type of thing.

Peggy Johnson: Exactly. And we'll do this hand in hand with regulators and as AI continues to improve. And as you know, AI improves with more data.

John: Mm-hmm.

Peggy Johnson: The more hours we can log in these real-world environments where unexpected things happen, the better that our safety abilities will be able to improve. And so it's really a mix of all of those things that over time will allow it to take on different maneuvers in these facilities. But we're starting very cautiously, as the EV industry did. I think we're definitely following the lead there.

John: Swappable grippers I saw on there, so you can choose different hands depending on what it needs to do. That's effectively—what kind of hands do you have available?

Peggy Johnson: Well, it's interesting. They're end effectors, actually, in robotics.

John: Yes.

Peggy Johnson: There are a number of ways that you can look at this. Five fingers generally in these environments are not the most useful tool.

John: Mm-hmm.

Peggy Johnson: You can see here Digit has on some pretty heavy-looking end effectors. That's Digit looking to pick up a 50-pound box and move it across a floor. There are other times when Digit needs to pick up a very tiny tool, maybe piece parts, and stack them into a bin for shipment to a customer. At other times it may need a three-finger grip.

And so we have to cover the gamut of what is needed in these facilities, and we decided to make it a standard ISO flange on the end so that we could take advantage of not only grippers that we were developing, but also grippers that are in the industry already.

There's been 60 years of development of grippers. You see a lot of these single-arm robots doing—

John: Mm-hmm.

Peggy Johnson: We can take advantage of that ecosystem by creating an ISO flange that's standard. So we've got a few that we're building ourselves because we've seen how customers are using the device, and then we'll take advantage of the others that are out there.

John: Peggy, maybe contextualize a little bit where we are in humanoids generally, like the ecosystem, where we are in this moment in history. You're one of two companies that very shortly will be publicly traded and offering humanoids globally.

You'll be getting, I think it's somewhere around $600 million cash injection as a result of this. That is a nice little war chest to continue building Digit 5, Digit 6, and all the other things that you need to do. Where do you see us at this moment in history? Where are we going?

Peggy Johnson: Okay, so at this moment in history, we are starting with tasks that Digit can perform very well, very capably, and there's a real need, there's a real demand because facilities can't find humans to do that.

I will say the company, before I arrived, had looked at a number of entry points for humanoids. There are reasons that they couldn't fully get into certain areas because there wasn't an ecosystem around it. Here, with the fact that there are grippers out there, there are a number of use cases that Digit can do very well now in this logistics manufacturing space.

Most companies that ship something are moving product around, material handling, so there was a demand, and we could fill that demand with Digit. That's where it starts. That's the entry point. Over time, and now with access to AI, we are able to continue to add skills to Digit's abilities, and we do that through real deployments.

So when we started, when we were in the safety barriers initially, we were doing one task over and over again. Now with Digit 5 and our ability to come outside that safety barrier, that opens up the skills that we can do across the facility. So, for instance, at GXO, we've already moved 100,000 bins.

We can do that very well. What's next is palletizing, things like machine tending, picking and placing parts, and that is where we say it unlocks these end-to-end workflows. So a single robot can do a number of different things, which is actually quite different from most automation, which is very fixed. It's very focused on doing one thing very well.

We knew if we could come outside of the safety barrier, we would be able to do end-to-end workflows, and that's what we'll accomplish with Digit 5.

John: That's super interesting, and it's always interesting to think about the factory of the future. What's that look like, right? There's always going to be a place for super high-speed, do one thing, do it a million times a day, high-repetition automation, right? And then there's always a need for something a little bit flexible.

It's changing, it's adapting, or there are decisions to be made, and it's a little bit complex, and that's where we've plugged in humans, right? And now we'll be able to plug in humanoids.

Peggy Johnson: Right. And many of these factories, the depreciation cycle is very long on this automation. So you might imagine a factory well into the future being able to be built fully automated. But today, with different types of automation—we call it islands of automation—there's a conveyor belt here, a put wall there, there's an autonomous mobile robot that might roll up with product on it. Humans are the connectors right now between all of that automation.

That automation, with the exception of AMRs, doesn't move. It's deployed and it's there for another 10-plus years. So until that time when factories are fully automated, humanoids can step in and pick up that labor gap that exists now.

John: I have to ask you a religious question, Peggy. People who like wheels really like wheels. People who like legs really like legs. Talk about legs versus wheeled platforms.

Peggy Johnson: Sure. And, by the way, we talk to many wheeled robots because we work with them. So they'll come shooting up from the loading dock with product on their flat surface. Digit then can pick that product off and set it on a conveyor belt. That's what it does at GXO, for instance.

It just picks off the AMR, sets it on the conveyor belt over and over again. And I think of it the same way there are different types of vehicles. There are trucks, there are SUVs, there are small cars, there are race cars. They all do a different thing, but they all operate on the same road. And this is where, for us, it's not really a religious war.

We built Digit and Digit 5 to operate in these very tight, narrow spaces, as you see in the picture here. Oftentimes other types of robotics can't fit down narrow aisles. They also get a little tippy when you pick up 50 pounds and you reach overhead to put it on a high shelf. They can tip over.

So it's really looking for the right tool for each task, and we do interoperate quite well, as I said, with the AMRs and really all types of automation, and we talk to them as well. So we have some software in the background called ARC, our cloud platform, that helps us communicate with other devices, whether that's conveyor belts or robots or put walls in the facility.

John: The level of competition in this space is just getting off the charts. I'm tracking 400-plus manufacturers. I'm tracking 800 or so robots. Not all those are full humanoids. And everybody, it seems, wants to be in this space. You get automotive manufacturers in China all jumping in, and you've got the Americans, you've got the European companies.

When you look at the competitive landscape, what gives you confidence that what you're bringing to the market and what you have to offer is going to be successful in the years ahead?

Peggy Johnson: Yeah. So we've been doing this for a while, as I said, and we've posted a number of points on the board, if you will. We've got the 65,000 hours of real-world operation. We are on our fifth-generation robot that's built for being cooperatively safe out in these environments. That's quite a bit different from the competition that we see out there right now.

We have real proof points, real customers doing real work in real facilities, and it's actually given our customers the confidence to back Digit 5 with $300 million of multi-year—

John: That's not small.

Peggy Johnson: —contracts over a couple of years. It's not small. So it's one thing to dance on a stage. It's another thing to do real work and actually have someone pay you for it because it's meaningful to them.

They can't find humans to step in, and so there's a real need, and Digit can fill that need.

John: To look at the partner list that's just on screen right now, right? Amazon—Amazon might have the most robots of any company on the planet right now. I think that the last number I heard, I'm probably way out of date, like a million, is just insane, right? Schaeffler, they're in a lot of conversations because they're building components, they're helping actuators, other things like that, working with companies like Humanoid in the UK and others. Talk about your partner list here and what that means to you.

Peggy Johnson: Right. So we've learned a lot from them. As I said, a lab environment's one thing, and dancing on a stage is one thing. To get out there in these environments and have to meet their high bars, you have to understand the regulatory environment. You have to understand compliance. You have to be safe. We've talked about that.

You also have to plug into their system. So it's not enough to throw a piece of metal over the wall and say, "Here, it can dance." It has to be able to do useful things. And Digit listened, we learned, and we deployed, and that feedback loop from these customers has been invaluable, to tell you the truth.

It's really helped us meet, I would say, a commercial bar that's very different than just a safety bar inside of a lab. And that's put us ahead of the pack right now, and we intend to stay there.

John: Now, the landscape shifted probably a month, two months ago when the FCC came down with their ruling on importing and everything like that. And there have been a number of different ways that companies are responding to that. And some, of course, since it's only two months, it's based on who they are, right?

Some companies are tremendously vertically integrated. Others are trying to source everything inside the United States. How does the FCC decision impact you, and does it cause any problems?

Peggy Johnson: No, it doesn't cause us any problems. In fact, we welcome the administration's push for fair competition. But for us from the start, we've been designed and we assemble in the U.S.

John: Mm-hmm.

Peggy Johnson: Unlike many of the others. We have something called RoboFab where we assemble all of our units.

It's based in Salem, Oregon, because our founders were from Oregon State University, so they located the headquarters there. We also have an office in Pittsburgh that does a lot of our skills training of the robot, and we just opened an office in Fremont that does our hardware and physical AI.

So between the three of them, we are very much focused on continuing to grow our roadmap over the next many, many years and to add more skills over time. Our supply chain has been significantly focused on the U.S., so we feel good about our ability to stay ahead and to stay out of any shortages, if you will.

So we've been comfortable with that since the start. Very U.S.-focused, U.S.-designed, U.S.-assembled, and U.S.-deployed. By the way, we will deploy with Digit 5 into the EU and Great Britain because we'll have the CE mark that's needed for those markets.

John: Nice. I love those locations. Oregon has had a longtime big presence from Intel. You've got some smart companies there doing amazing things. Pittsburgh, obviously, massive engineering and manufacturing capability there, right? And you can't get much better than California for AI as well, so that makes a ton of sense.

What have I not asked you? What have I missed? What do you want to share that I've totally forgotten about?

Peggy Johnson: Wow. We've covered a lot of things. I will say, keep an eye on Digit 5 because we now are able, with our simulation and the data resources that we have, we're training Digit 5 very, very quickly. So we'll have new skills coming out. You can think of it a little bit like EV vehicles that get an upgrade and all of a sudden they can do something new.

We're very, very similar. So as we come outside of that safety barrier, we will be adding more skills, more capabilities. Those end effectors give us a lot of flexibility. So what Digit 5 can do on day one will be very different than what Digit 5 can do a year later.

John: That's amazing. And it's really interesting. It's literally, it seems like every—I'm going to say week, but it almost seems like every day—there's a new company that's coming out with amazing stories on training robots and doing it very, very effectively. Skild AI came out and, you know, $100 million revenue rate just yesterday, right? And there are others as well.

And you wonder, I mean, obviously Figure committed to spending $3.5 billion that they don't have on index and training data and crunching all that. So it's interesting to see if there's an advantage in just using smarter and cheaper ways of training than forcing it. I don't know. What are your thoughts there?

Peggy Johnson: There are different schools of thought. We've found the best path forward is a mix. So we do—we have, first of all, a very large data lake of operational data. All of those hours that we've gathered for all of the past many years are a great base for training our physical AI.

On top of that, we have access to a number of industry tools where we can simulate data, like Isaac Sim from NVIDIA. Then we also have something called learning from demonstration, where you can teleoperate the robot to do something it's never done before.

So if you want it to pick cereal boxes off a shelf, you would first, kind of through a joystick-type action, operate the robot, have it walk over, pick up the cereal box, set it down where you want. You can do that a few times. You add simulation data and then real-world data on top of that, and that tends to be the fastest way to teach Digit new skills. And so that's the path that we're on.

John: I want to thank you for taking this time. I really appreciate it, and right before the wedding as well. So—

Peggy Johnson: Yes.

John: Great to see you and hear you, and thank you for taking the time.

Peggy Johnson: Thank you, John. Thanks for having me.