Robotics startup Mytra has been quietly operating behind the scenes since its May 2022 founding in a bid to rethink warehouse automation. The outfit brings a solid pedigree to the table, founded by automotive vets from EV firms, including Tesla and Rivian.
Warehouse/fulfillment has been a white-hot category for automation since the pandemic hamstrung the global supply chain in ways that are still being felt. It’s a highly competitive space as well, with big names like Amazon, Locus and Zebra/Fetch making headway and setting the stage for an explosion of interest in the bipedal humanoid form factor.
Even as the world has come back to life post-pandemic, labor shortages remain a major sticking point for the industry, as with so many others. There’s still plenty of room for players to make an impact, however. Estimates suggest that between 5% and 10% of global warehouses are automated in any meaningful sense.
Like so many others, Mytra co-founder and CEO Chris Walti discovered automation’s shortcomings the hard way. He was previously at Tesla, where the hard way tends to be par for the course. Walti spent seven years at the carmaker, first in engineering, then mobile robotics and ultimately as the senior manager/lead for what would become Optimus.
He describes his journey through Tesla as an ongoing cycle of looking for solutions, determining that nothing in market was suitable for its specifics needs, and then going and building the things themselves. That started with autonomous mobile robot (AMR) solutions.
“I was pulled into manufacturing and automation through the Model 3 ramp,” Walti told TechCrunch. “Tesla was struggling to get our automation systems up and running, so we ended up setting up a manual warehouse as a pressure release valve for the manufacturing system. About six months later, they were like, ‘Can you just take over the automation system that’s causing a lot of these challenges?’”
Image Credits: Mytra
Among the industry shortcomings that emerged for Tesla’s specific needs was an inability to find AMRs that could move around payloads as heavy as 3,000 pounds. Those are the sorts of demands one bumps up against when making cars. So the team went to work building their own solutions in-house.
“And then Elon [Musk] was like, ‘We should build a humanoid,” Walti said. “My team was tapped to lead that. I led the internal hiring effort for that team. Everything you saw on AI Day was a product of those efforts.” He added that “at some point, [Optimus] became the number one effort in the company. It ended up not really being a fit for what I ended up wanting to do.”
Walti remains bullish about the long-term impact of humanoid robots across a variety of sectors, but he noted that he “think[s] it’s going to be a while before humanoids are truly moving the needle on a production floor.”
Mytra’s solution shares a lot of common DNA with vertical robotic storage solutions produced by companies like AutoStore. Two of the primary differentiators between the startup and existing solutions, according to Walti, are its ability manage heavy payloads and its dynamism.
“There are literally trillions of different ways that I can move one of these pallets or bookshelves from point A to point B within the system,” he explained. “Which is fundamentally unique. This is the most kinematically free system that has been conceived.”
In spite of maintaining stealth until now, Mytra has already drummed up interest with big names. The startup has a pilot with grocery giant Albertsons, along with “another half-dozen Fortune 50 customers that are in varying stages in the pipeline.”
Mytra also recently closed a $50 million Series B, bringing its total funding up to $78 million. Investors include Greenoaks and Eclipse.
Robotics startup Mytra has been quietly operating behind the scenes since its May 2022 founding in a bid to rethink warehouse automation. The outfit brings a solid pedigree to the table, founded by automotive vets from EV firms, including Tesla and Rivian.
Warehouse/fulfillment has been a white-hot category for automation since the pandemic hamstrung the global supply chain in ways that are still being felt. It’s a highly competitive space as well, with big names like Amazon, Locus and Zebra/Fetch making headway and setting the stage for an explosion of interest in the bipedal humanoid form factor.
Even as the world has come back to life post-pandemic, labor shortages remain a major sticking point for the industry, as with so many others. There’s still plenty of room for players to make an impact, however. Estimates suggest that between 5% and 10% of global warehouses are automated in any meaningful sense.
Like so many others, Mytra co-founder and CEO Chris Walti discovered automation’s shortcomings the hard way. He was previously at Tesla, where the hard way tends to be par for the course. Walti spent seven years at the carmaker, first in engineering, then mobile robotics and ultimately as the senior manager/lead for what would become Optimus.
He describes his journey through Tesla as an ongoing cycle of looking for solutions, determining that nothing in market was suitable for its specifics needs, and then going and building the things themselves. That started with autonomous mobile robot (AMR) solutions.
“I was pulled into manufacturing and automation through the Model 3 ramp,” Walti told TechCrunch. “Tesla was struggling to get our automation systems up and running, so we ended up setting up a manual warehouse as a pressure release valve for the manufacturing system. About six months later, they were like, ‘Can you just take over the automation system that’s causing a lot of these challenges?’”
Image Credits: Mytra
Among the industry shortcomings that emerged for Tesla’s specific needs was an inability to find AMRs that could move around payloads as heavy as 3,000 pounds. Those are the sorts of demands one bumps up against when making cars. So the team went to work building their own solutions in-house.
“And then Elon [Musk] was like, ‘We should build a humanoid,” Walti said. “My team was tapped to lead that. I led the internal hiring effort for that team. Everything you saw on AI Day was a product of those efforts.” He added that “at some point, [Optimus] became the number one effort in the company. It ended up not really being a fit for what I ended up wanting to do.”
Walti remains bullish about the long-term impact of humanoid robots across a variety of sectors, but he noted that he “think[s] it’s going to be a while before humanoids are truly moving the needle on a production floor.”
Mytra’s solution shares a lot of common DNA with vertical robotic storage solutions produced by companies like AutoStore. Two of the primary differentiators between the startup and existing solutions, according to Walti, are its ability manage heavy payloads and its dynamism.
“There are literally trillions of different ways that I can move one of these pallets or bookshelves from point A to point B within the system,” he explained. “Which is fundamentally unique. This is the most kinematically free system that has been conceived.”
In spite of maintaining stealth until now, Mytra has already drummed up interest with big names. The startup has a pilot with grocery giant Albertsons, along with “another half-dozen Fortune 50 customers that are in varying stages in the pipeline.”
Mytra also recently closed a $50 million Series B, bringing its total funding up to $78 million. Investors include Greenoaks and Eclipse.
The savvy marketers at Boston Dynamics produced two major robotics news cycles last week. The larger of the two was, naturally, the electric Atlas announcement. As I write this, the sub-40 second video is steadily approaching five million views. A day prior, the company tugged at the community’s heart strings when it announced that the original hydraulic Atlas was being put out to pasture, a decade after its introduction.
The accompanying video was a celebration of the older Atlas’ journey from DARPA research project to an impressively nimble bipedal ’bot. A minute in, however, the tone shifts. Ultimately, “Farewell to Atlas” is as much a celebration as it is a blooper reel. It’s a welcome reminder that for every time the robot sticks the landing on video there are dozens of slips, falls and sputters.
Image Credits: Boston Dynamics
I’ve long championed this sort of transparency. It’s the sort of thing I would like to see more from the robotics world. Simply showcasing the highlight reel does a disservice to the effort that went into getting those shots. In many cases, we’re talking years of trial and error spent getting robots to look good on camera. When you only share the positive outcomes, you’re setting unrealistic expectations. Bipedal robots fall over. In that respect, at least, they’re just like us. As Agility put it recently, “Everyone falls sometimes, it’s how we get back up that defines us.” I would take that a step further, adding that learning how to fall well is equally important.
The company’s newly appointed CTO, Pras Velagapudi, recently told me that seeing robots fall on the job at this stage is actually a good thing. “When a robot is actually out in the world doing real things, unexpected things are going to happen,” he notes. “You’re going to see some falls, but that’s part of learning to run a really long time in real-world environments. It’s expected, and it’s a sign that you’re not staging things.”
A quick scan of Harvard’s rules for falling without injury reflects what we intuitively understand about falling as humans:
Protect your headUse your weight to direct your fallBend your kneesAvoid taking other people with you
As for robots, this IEEE Spectrum piece from last year is a great place to start.
“We’re not afraid of a fall—we’re not treating the robots like they’re going to break all the time,” Boston Dynamics CTO Aaron Saunders told the publication last year. “Our robot falls a lot, and one of the things we decided a long time ago [is] that we needed to build robots that can fall without breaking. If you can go through that cycle of pushing your robot to failure, studying the failure, and fixing it, you can make progress to where it’s not falling. But if you build a machine or a control system or a culture around never falling, then you’ll never learn what you need to learn to make your robot not fall. We celebrate falls, even the falls that break the robot.”
Image Credits: Boston Dynamics
The subject of falling also came up when I spoke with Boston Dynamics CEO Robert Playter ahead of the electric Atlas’ launch. Notably, the short video begins with the robot in a prone position. The way the robot’s legs arc around is quite novel, allowing the system to stand up from a completely flat position. At first glance, it almost feels as though the company is showing off, using the flashy move simply as a method to showcase the extremely robust custom-built actuators.
“There will be very practical uses for that,” Playter told me. “Robots are going to fall. You’d better be able to get up from prone.” He adds that the ability to get up from a prone position may also be useful for charging purposes.
Much of Boston Dynamics’ learnings around falling came from Spot. While there’s generally more stability in the quadrupedal form factor (as evidenced from decades trying and failing to kick the robots over in videos), there are simply way more hours of Spot robots working in real-world conditions.
Image Credits: Agility Robotics
“Spot’s walking something like 70,000 kms a year on factory floors, doing about 100,000 inspections per month,” adds Playter. “They do fall, eventually. You have to be able to get back up. Hopefully you get your fall rate down — we have. I think we’re falling once every 100-200 kms. The fall rate has really gotten small, but it does happen.”
Playter adds that the company has a long history of being “rough” on its robots. “They fall, and they’ve got to be able to survive. Fingers can’t fall off.”
Watching the above Atlas outtakes, it’s hard not to project a bit of human empathy onto the ’bot. It really does appear to fall like a human, drawing its extremities as close to its body as possible, to protect them from further injury.
With a 99% success rate over about 20 hours of live demos, Digit still took a couple of falls at ProMat.
We have no proof, but we think our sales team orchestrated it so they could talk about Digits quick-change limbs and durability. #ConspiracyTheories pic.twitter.com/aqC5rhvBTj
— Agility Robotics (@agilityrobotics) April 6, 2023
When Agility added arms to Digit, back in 2019, it discussed the role they play in falling. “For us, arms are simultaneously a tool for moving through the world — think getting up after a fall, waving your arms for balance, or pushing open a door — while also being useful for manipulating or carrying objects,” co-founder Jonathan Hurst noted at the time.
I spoke a bit to Agility about the topic at Modex earlier this year. Video of a Digit robot falling over on a convention floor a year prior had made the social media rounds. “With a 99% success rate over about 20 hours of live demos, Digit still took a couple of falls at ProMat,” Agility noted at the time. “We have no proof, but we think our sales team orchestrated it so they could talk about Digits quick-change limbs and durability.”
As with the Atlas video, the company told me that something akin to a fetal position is useful in terms of protecting the robot’s legs and arms.
The company has been using reinforcement learning to help fallen robots right themselves. Agility shut off Digit’s obstacle avoidance for the above video to force a fall. In the video, the robot uses its arms to mitigate the fall as much as possible. It then utilizes its reinforcement learnings to return to a familiar position from which it is capable of standing again with a robotic pushup.
One of humanoid robots’ main selling points is their ability to slot into existing workflows — these factories and warehouses are known as “brownfield,” meaning they weren’t custom built for automation. In many existing cases of factory automation, errors mean the system effectively shuts down until a human intervenes.
"We'll be singing / when we're winning." 🔊 pic.twitter.com/51DYD1Avvg
— Brian Heater (@bheater) August 17, 2021
“Rescuing a humanoid robot is not going to be trivial,” says Playter, noting that these systems are heavy and can be difficult to manually right. “How are you going to do that if it can’t get itself off the ground?”
If these systems are truly going to ensure uninterrupted automation, they’ll need to fall well and get right back up again.
“Every time Digit falls, we learn something new,” adds Velagapudi. “When it comes to bipedal robotics, falling is a wonderful teacher.”
You could easily walk the entire Automate floor without spotting a single humanoid. There was a grand total of three, by my count — or, rather, three units of the same nonworking prototype. Neura was showing off its long-promised 4NE-1 robot, amid more traditional form factors. There was a little photo setup where you could snap a selfie with the bot, and that was about it.
Notably absent at the annual Association for Advancing Automation (A3) show was an Agility booth. The Oregon company made a big showing at last year’s event, with a small army of Digits moving bins from a tote wall to a conveyer belt a few feet away. It wasn’t a complex demo, but the mere sight of those bipedal robots working in tandem was still a showstopper.
Agility chief product officer Melonee Wise told me that the company had opted to sit this one out, as it currently has all the orders it can manage. And that’s really what these trade shows are about: manufacturers and logistics companies shopping around for the next technological leg up to remain competitive.
How large a role humanoids will play in that ecosystem is, perhaps, the biggest question on everyone’s mind at the moment. Amid the biggest robotics hype cycle I’ve witnessed firsthand, many are left scratching their heads. After all, the notion of a “general purpose” humanoid robot flies in the face of decades’ worth of orthodoxy. The notion of the everything robot has been a fixture of science fiction for the better part of a century, but the reality has been one of single-purpose systems designed to do one job well.
Image Credits: Brian Heater
While there wasn’t much of a physical presence, the subject of humanoids loomed large at the event. As such, A3 asked me to moderate a panel on the subject. I admit I initially balked at the idea of an hourlong panel. After all, the ones we do at Disrupt tend to run 20 to 25 minutes. By the end of the conversation, however, it was clear we easily could have filled another hour.
That was due, in part, to the fact that the panel was — as one LinkedIn commenter put it — “stacked.” Along with Wise, I was joined by Boston Dynamics CTO Aaron Saunders, Apptronik CEO Jeff Cardenas and Neura CEO David Reger. I kicked the panel off by asking the audience how many in attendance would consider themselves skeptical about the humanoid form factor. Roughly three-quarters of the people present raised their hands, which is more or less what I’d anticipate at this stage in the process.
As for A3, I would say it has entered the cautiously optimistic phase. In addition to hosting a panel on the subject at Automate, the organization is holding a Humanoid Robot Forum in Memphis this October. The move echoes the 2019 launch of A3’s Autonomous Mobile Robot (AMR) Forum, which presaged the explosive growth in warehouse robotics during the pandemic.
Investors are less measured in their optimism.
Image Credits: Figure
“A year after we laid our initial expectations for global humanoid robot [total addressable market] of $6bn, we raise our 2035 TAM forecast to $38bn resulting from a 4-fold increase in our shipments estimate to 1.4mn units with a much faster path to profitability on a 40% reduction in bill of materials,” Goldman Sachs researcher Jacqueline Du wrote in a report published in February. “We believe our revised shipment estimate would cover 10%-15% of hazardous, dangerous and auto manufacturing roles.”
There are, however, plenty of reasons to be skeptical. Hype cycles are hard to navigate when you’re in the middle of them. The amount of money currently changing hands (see: Figure’s most recent raise of $675 million) gives one pause in the wake of various startup collapses across other fields. It also comes during a time when robotics investments have slowed after a few white-hot years.
One of the biggest risks at this stage is the overpromise. Every piece of new technology runs this risk, but something like a humanoid robot is a lightning rod for this stuff. Much like how eVTOL proponents see the technology as finally delivering on the promise of flying cars, the concept of personal robot servant looks within reach.
The fact that these robots look like us leads many to believe they can — or soon will be able to — do the same things as us. Elon Musk’s promise of a robot that works in the Tesla factory all day and then comes home to make you dinner added fuel to that fire. Tempering expectations isn’t really Musk’s thing, you know? Others, meanwhile, have tossed around the notion of a general intelligence for humanoid robots — a thing that is a ways off (“five to 10 years” is a time frame I often hear bandied about).
Image Credits: Apptronik/Mercedes
“I think we need to be careful about the hype cycles, because we ultimately need to deliver the promise and potential,” Cardenas said. “We’ve been through this before, with the DARPA Robotics challenge, where there’s a lot of excitement going into it, and we crashed into reality coming out of that.”
One source of disconnect is the question of what these systems can deliver today. The answer is murky, partly because of the nature of partnership announcements. Agility announced it was working with Amazon, Apptronik with Mercedes, Figure with BMW and Sanctuary AI with Magna. But every partnership so far needs to be taken for what it is: a pilot. The precise number of robots deployed in any specific partnership is never disclosed, and the figure is often single digits. It makes perfect sense: These are all operating factories/warehouses. It would be wildly disruptive to just slot in a new technology at scale and hope for the best.
Pilots are important for this reason, but they should not be mistaken for market fit. As of this writing, Agility is the only one of the bunch that has confirmed with TechCrunch that it’s ready for the next step. On the discussion panel, Wise confirmed that Agility will be announcing specifics in June. Cardenas, meanwhile, stated that the company plans to heavily pilot in the “back half” of 2024, with plans to move beyond early next year.
Neura and Boston Dynamics are simply too early stage for the conversation. Neura promised to show off some demos at some point in July, moving 4NE-1 beyond what has up until now been a series of rendered videos, coupled with the nonfunctioning units shown at Automate.
As for when we’ll see more of the electric Atlas beyond a 30-second video, Saunders says, “[the video] is just meant to be an early peek. We’re planning on getting into the pilot and some of the more pragmatic pieces next year. So far, we’re focused mainly on building up the focus and technology. There are a lot of hard problems left to solve in the manipulation and the AI spaces. Our team is working on it right now, and I think as those features get more robust, we’ll have more to show off.”
Image Credits: Brian Heater
Boston Dynamics isn’t starting from scratch, of course. After more than a decade of Atlas, the company has as much humanoid expertise as any, while the launches of Spot and Stretch have taught the firm plenty about commercializing products after decades of research.
So, why did it take so long to see the company’s swing at the commercial humanoid category? “We wanted to make sure that we understood where the value is placed,” Saunders said. “It’s really easy to make demo videos and show cool things, but it takes a long time to find ROI [return on investment] cases that justify the human form.”
Neura has easily the most diverse portfolio of the companies present onstage. In fact, one gets the sense that whenever the company is finally ready to launch a humanoid in earnest, it will be just another form factor in the company’s portfolio, rather than the driving force. Meanwhile, when the electric Atlas eventually launches, it will be Boston Dynamics’ third commercially available product.
As Digit is Agility’s only offering at the moment, the company is wholly committed to the bipedal humanoid form factor. For its part, Apptronik splits the difference. The Austin-based firm has been taking a best-tool-for-the-job approach to the form factor. If, for example, legs aren’t needed for a specific environment, the company can mount the upper half of its robot onto a wheeled base.
Image Credits: Tesla
“I think at the end of the day, it’s about solving problems,” Cardenas said. “There are places where you don’t need a bipedal robot. My view is that bipedal form factors will win the day, but the question is how do you actually get them out there?”
Not every terrain requires legs. Earlier this week, Diligent Robotics co-founder and CEO Andrea Thomaz told me that part of the reason her company targeted healthcare first is the prevalence of ADA (Americans with Disabilities Act) compliant structures. Anywhere a wheelchair can go, a wheeled robot should be able to follow. Because of that, the startup didn’t have to commit to the very difficult problem of building legs.
Legs have benefits beyond the ability to handle things like stairs, however. Reach is an important one. Legged robots have an easier time reaching lower shelves, as they can bend at the legs and the waist. You could, theoretically, add a very large arm to the top of an AMR, but doing so introduces all kinds of new problems like balance.
Safety is something that has thus far been under-addressed in conversations around the form factor. One of humanoid robots’ key selling points is their ability to slot into existing workflows alongside other robotic or human co-workers.
But robots like these are big, heavy and made of metal, therefore making them a potential hazard to human workers. The subject has been top of mind for Wise, in particular, who says further standards are needed to ensure that these robots can operate safely alongside people.
For my part, I’ve been advocating for a more standardized approach to robot demos. Videos of humanoids, in particular, have obscured what these robots can and can’t do today. I would love to see disclosures around playback speed, editing, the use of teleop and other tricks of the trade that can be used to deceive (intentionally or not) viewers.
“It’s very hard to distinguish what is and isn’t progress,” Wise said, referring to some recent videos of Tesla’s Optimus robot. “I think one thing that we, as a community, can do better is being more transparent about the methodologies that we’re using. It’s fueling more power for the hype cycle. I think the other problem that we have is, if we look at what’s going on with any humanoid robot in this space, safety is not clear. There isn’t an e-stop on Optimus. There isn’t an e-stop on many of our robots.”
Today Figure confirmed long-standing rumors that it’s been raising more money than God. The Bay Area-based robotics firm announced a $675 million Series B round that values the startup at $2.6 billion post-money.
The lineup of investors is equally impressive. It includes Microsoft, OpenAI Startup Fund, Nvidia, Amazon Industrial Innovation Fund, Jeff Bezos (through Bezos Expeditions), Parkway Venture Capital, Intel Capital, Align Ventures and ARK Invest. It’s a mind-boggling sum of money for what remains a still-young startup, with an 80-person headcount. That last bit will almost certainly change with this round.
Figure already had a lot to work with. Founder Brett Adcock, a serial entrepreneur, bootstrapped the company, putting in an initial $100 million to get it started. Last May, it added $70 million in the form of a Series A. I used to think “Figure” was a reference to the robot’s humanoid design and perhaps an homage to a startup that’s figuring things out. Now it seems it’s might also be a reference to the astronomical funding figure it’s raised thus far.
When Figure launched in 2022, it put out an ambitious goal of creating a walking bipedal robot within a year. The company told TechCrunch that it hit that date. We didn’t see walking video at the time, but it has since surfaced.
Image Credits: Figure
The startup is very much a product of its time: humanoid robots are having a moment. Examples have been showcased by Tesla (though I’d temper your expectations somewhat on that), Apptronik and 1X, among others. Amazon recently began a small pilot with Agility’s Digit robot, which seems to have found its groove supplementing human labor in brownfield warehouses and fulfillment centers.
Most of these efforts — including Figure’s — are working toward that same goal of building robots for industry. Upfront costs are just one reason it makes a lot more sense to focus on the workplace before the home. It’s also one of many reasons it’s important to properly calibrate your expectations of what a system like this can — and can’t — do.
Some companies (namely Tesla again) have perhaps set unrealistic expectations about the current state of the art. I’m speaking primarily about artificial general intelligence, which many roboticists believe is about five years out — though that could well prove optimistic.
“General purpose” gets tossed around a lot when discussing these robots. In essence, it refers to systems that can quickly pick up a variety of tasks the way humans do. Traditional robotics systems are single purpose, meaning they do one thing really well a number of times. Multipurpose systems are certainly out there, and APIs like the kind provided by Boston Dynamics for Spot will go some way toward expanding that functionality.
The eventual goal of generalized AI is, in fact, a big driver for the humanoid form factor. Robots built for a single function are difficult to adapt, while, in theory, a robot built to think like us can do anything we can.
When I visited Figure’s HQ last year, the company had recently built a demo area in the center of its office.
The space’s primary use was showcasing the robot for potential clients and investors. Tellingly, it was set up to resemble a warehouse or factory.
Most people believe that warehouse work is the first step to broader adoption and is perhaps the eventual arrival of a home robot. After all, corporations will happily invest a good chunk of money into a product they believe will save them money in the long run. Also, it’s much easier to fill a day’s work with one or two extremely repetitive tasks. Consumers will almost certainly demand something indistinguishable from generalization before paying the equivalent of a new car to buy one.
It’s worth noting that today’s news also finds Figure signing a partnership with generative AI pioneer OpenAI.
The goal of the deal is to “develop next generation AI models for humanoid robots,” according to Figure. The near-term application for Large Language Models will be the ability to create more natural methods of communication between robot and their human colleagues. “The collaboration aims to help accelerate Figure’s commercial timeline by enhancing the capabilities of humanoid robots to process and reason from language,” the company notes.
Natural language allows people to give the systems commands and gives humans a better understanding of what the robot is doing (hence the ability to “reason” in language). These are, after all, much more complex systems than a human-piloted forklift, for example. If they’re going to operate autonomously, you’re going to need a more direct method of communication — especially on a busy warehouse or factory floor. Language process allows for human assistance in correcting mistakes.
Image Credits: Figure
“We’ve always planned to come back to robotics and we see a path with Figure to explore what humanoid robots can achieve when powered by highly capable multimodal models,” OpenAI VP, Peter Welinder, said in a statement. “We’re blown away by Figure’s progress to date and we look forward to working together to open up new possibilities for how robots can help in everyday life.”
Another thing that makes the deal interesting is OpenAI’s investment in direct competitor, 1X. One wonders whether such a deal is OpenAI rethinking its investments, or if this is simply the company playing the field. My guess at the moment is the latter. If you’re in OpenAI’s position, you might as well work with as many promising companies as you can, and Figure has certainly demonstrated some real progress in the eight months since it took its first steps.
Take the video below, posted a little over a week ago. Figure says the robot’s operations are roughly 16.7% the speed of a human doing the same task. That is, it’s very slow and methodical — deliberate, even. That much is clear from the video. And it’s always good to see a robot operating at actual speed in a demo video, no matter how well produced it happens to be. People have told me in hushed tones that some folks try to pass off sped up videos without disclosing as much. It’s the kind of thing that feeds into consumers’ already unrealistic expectations of what robots can do.
Microsoft’s investment finds Figure utilizing Azure for storage, training and ‘AI infrastructure.
“We are excited to collaborate with Figure and work towards accelerating AI breakthroughs,” said Microsoft Corporate VP, Jon Tinter in a statement. “Through our work together, Figure will have access to Microsoft‘s AI infrastructure and services to support the deployment of humanoid robots to assist people with real world applications.”
Somewhat interestingly, Figure was not included in Bill Gates’ recent list of exciting robotics startups, though two other humanoid companies (Agility and Apptronik) were.
The Amazon Innovation Fund’s participation in this round is also particularly notable, as it can often serve as a pipeline to real-world deployment in fulfillment centers — take Agility as a key example.
The autonomous part is important as well, given the propensity to pass off tele-op for autonomy. One of the reasons autonomy is so difficult in cases like this is all the variations you can’t account for. While warehouses tend to be fairly structured environments, any number of things can occur in the real world that will knock a task off-kilter. And the less structured these tasks become, the larger the potential for error. A lot of questions remain, including how many takes it took to get this right. One thing this absolutely has going for it is the fact that the action is captured in one continuous shot, meaning the company didn’t cobble together a series of actions through creative editing.
Image Credits: Figure
Mechatronics are easier to judge in a short video than AI and autonomy, and from that perspective, the Figure 01 robot appears quite dexterous. In fact, if you look at the angle and positioning of the arms, you’ll notice that it’s performing the carry in a manner that would be quite uncomfortable for most people. It’s important to note that just because the robot looks like a person doesn’t mean that it has to behave exactly like one. My educated guess is that the positioning of the tote has to do with the robot’s center of gravity and perhaps the fact that it appears to be extremely top heavy.
Figure says the money will go toward accelerating its go-to-market. The company has already signed a deal with BMW for robotics deployment.
Agility’s Digit wasn’t the only humanoid holding court at Modex in Atlanta this week. On the opposite end of the Georgia World Congress Center, Reflex Robotics, a younger and smaller startup, was drawing its own crowd. The booth was centered on a single demo running throughout the week. Passersby requested something from the Reflex robot, and it spring into action, grabbing the item off the shelf (it didn’t hurt that the company was giving out free food and beverage).
From the look of it, most showgoers were impressed by the system’s speed and accuracy. The hardware is an in-house design, featuring a “torso” mounted to a base that allows the arms and sensors to dynamically move up and down. It makes for a surprisingly dexterous robot that can access shelves at a variety of heights, while maneuvering tight spaces. The system has a wheeled base, which is perfectly effective for navigating these kinds of layouts.
Image Credits: Brian Heater
Something that needs be mentioned is the fact that the system is primarily tele-operated. It’s an example of a human in the loop system, whereby it can be controlled remotely. The only real constraints there are potential latency issues if the operator is moved too far. Co-founder and CEO Ritesh Ragavender compares the interface to a video game. He adds that the robot is “approaching” human-level efficiency.
The team is moving toward a system that requires fewer and fewer people. As the robot grows more autonomous and capable, the human element will shift from control to supervisor. Humans stick around as a kind of fail-safe, should the robot get into trouble or encounter something entirely new. For now, however, the human-to-robot ratio is 1:1. Ragavender says he believes the company will achieve a 1:2 ratio in warehouses and 3PLs (third-party logistics) by early next year.
Image Credits: Brian Heater
The robot putting in the hours at Modex is a second-generation system, which is currently being selectively piloted with potential customers, including some “big names,” per Ragavender. He adds that the current timeline involves having 10 to 20 Reflex robots in the world, followed by “hundreds” next year.
“I think the fact that you have a system that’s getting smarter the more it works, coupled with remote operators and a really low-cost robot is what allows us to scale this and make it work,” says Ragavender. “We’re ready to ship today.” The system’s current price point is “below $50,000.” Scaling manufacturing should be able to bring that number down further.
Image Credits: Reflex Robotics
The New York–based startup is still extremely lean, with just five people. The founders all come from MIT, in addition to companies like Boston Dynamics and Tesla. Reflex has thus far raised a seed round, led by Khosla Ventures. Ragavender says the startup will likely pursue a Series A later this year.
Atlas lies motionless in a prone position atop interlocking gym mats. The only soundtrack is the whirring of an electric motor. It’s not quiet, exactly, but it’s nothing compared to the hydraulic jerks of its ancestors.
As the camera pans around the robot’s back, its legs bend at the knees. It’s a natural movement, at first, before crossing into an uncanny realm, like something out of a Sam Raimi movie. The robot, which appeared to be lying on its back, has effectively switched positions with this clever bit of leg rotation.
As Atlas fully stands, it does so with its back to the camera. Now the head spins around 180 degrees, before the torso follows suit. It stands for a moment, offering the camera its first clear view of its head — a ring light forming the perimeter of a perfectly round screen. Once again, the torso follows the head’s 180, as Atlas walks away from the camera and out of frame.
A day after retiring the hydraulic version of its humanoid robot, Boston Dynamics just announced that — like Bob Dylan before it — Atlas just went electric.
The pace is fast, the steps still a bit jerky — though significantly more fluid than many of the new commercial humanoids to which we’ve been introduced over the past couple of years. If anything, the gait brings to mind the brash confidence of Spot, Atlas’ cousin whose branch on the evolutionary tree forked off from the humanoid a few generations ago.
All-new Atlas
The new version of the robot is virtually unrecognizable. Gone is the top-heavy torso, the bowed legs and plated armor. There are no exposed cables anywhere to be found on the svelte new mechanical skeleton. The company, which has warded off reactionary complaints of robopocalypse for decades, has opted for a kinder, gentler design than both the original Atlas and more contemporary robots like the Figure 01 and Tesla Optimus.
The new robot’s aesthetic more closely matches that of Agility’s Digit and Apptronik’s Apollo. There’s a softer, more cartoonish design to the traffic-light-headed robot. It’s the “All New Atlas,” according to the video. Boston Dynamics has bucked its own trend by maintaining the research name for a product it will be positioning toward commercialization. SpotMini became Spot. Handle became Stretch. For now, however, Atlas is still Atlas.
“We might revisit this when we really get ready to build and deliver in quantity,” Boston Dynamics CEO Robert Playter tells TechCrunch. “But I think for now, maintaining the branding is worthwhile.”
The executive’s statement betrays the still early stages of the project. Boston Dynamics’ current timeline has the electric Atlas beginning pilot testing at Hyundai facilities earlier next year, with full production a few years further down the road.
“We’re going to be doing experiments with Hyundai on-site, beginning next year,” says Playter. “We already have equipment from Hyundai on-site. We’ve been working on this for a while. To make this successful, you have to have a lot more than just cool tech. You really have to understand that use case, you’ve got to have sufficient productivity to make investment in a robot worthwhile.”
Doing a 180
Image Credits: Boston Dynamics
What’s most striking about the 40-second “All New Atlas” teaser is the robot’s movements. They’re a reminder that building a humanoid robot doesn’t require making the robot as human as possible. As an investor pointed out to me years back, billions of years of evolution hasn’t made us humans perfect machines. If we are going to create machines in our own images, why not build ones that can do things we can’t?
“We built a set of custom, high-powered and very flexible actuators at most joints,” says Playter. “That’s a huge range of motion. That really packs the power of an elite athlete into this tiny package, and we’ve used that package all over the robot.”
One thing worth keeping in mind while watching the footage is that Boston Dynamics has made its name across decades of viral videos. Recent additions to the canon are just as likely to showcase a ’bot’s dance moves as they are anything genuinely useful in an industrial setting. For that reason, it’s difficult to decouple what the company has deemed real functionality and what is just a bit of showing off.
Starting in the prone position, for instance, is an opportunity to showcase that cool reverse crab leg trick — but it’s practical, as well. As Boston Dynamics was more than happy to showcase in the hydraulic Atlas’ farewell video, falling down is part of the job — and so, too, is getting up. The truth of the matter is that most of the current crop of industrial robots require human intervention when they fail. A robot that can simply dust itself off and get back to work, on the other hand, is a big win for productivity.
The system’s ability to turn on a dime also lends considerably to its productivity potential. It brings to mind Agility’s Digit demos (the company is notably the only one of its ilk demoing systems at this scale), wherein a robot walks to a shelf, turns around, walks to the conveyer belt, turns around and walks back. Multiply that job by hundreds — or even thousands — of times a day, and you begin to see the value of shaving off precious seconds.
“It’s going to be capable of a set of motions that people aren’t,” explains Playter. “There will be very practical uses for that.”
Significantly reducing the robot’s turn radius is also important in tight spaces. Remember, these machines are meant to be brownfield solutions — that is, they’re designed to be plugged into existing workflows in existing spaces. Increased maneuverability could ultimately mean the difference between working in a setting and having to redo the layout.
Head and hands
Image Credits: Boston Dynamics
The hands aren’t brand new for the video, having previously made appearances on the hydraulic model. They do, however, also represent the company’s decision to not hew entirely to human design as a guiding force. Here, the difference is as simple as opting for three fingers, instead of four on the end effectors.
“There’s so much complexity in a hand,” says Playter. “When you’re banging up against the world with actuators, you have to be prepared for reliability and robustness. So, we designed these with fewer than five fingers to try to control their complexity. We’re continuing to explore generations of those. We want compliant grasping, adapting to a variety of shapes with rich sensing on board, so you understand when you’re in contact.
Internally, the most contentious aspect of the design may well be the head. The big, round display has shades of a cosmetic mirror.
“It was one of the design elements we fretted over quite a bit,” says Playter. “Everybody else had a sort of humanoid shape. I wanted it to be different. We want it to be friendly and open. It provides a palette for a display. Of course, there are sensors buried in there, but also the shape is really intended to indicate some friendliness. That will be important for interacting with these things in the future.”
An Atlas for Christmas
Image Credits: Boston Dynamics
The landscape has changed dramatically in the decade since the hydraulic Atlas’ introduction. Electric Atlas has a fair bit of company, in the form of humanoid robots from Figure, Apptronik, Tesla and 1X, among others.
“For us, there’s obviously been a huge influx of interest. I think that influx has been motivated by three events. Boston Dynamics got acquired [by Hyundai] for nearly a billion dollars. That sort of woke everybody up like, ‘whoa, there’s an exit path.’ Tesla expressing interest in manufacturing sort of validated things we’ve been doing for a long time. And then, the emergence of AI as a tool to help deal with generality is making all of this feasible. We’ve been patient to announce, because we wanted to do enough research to understand that we can solve manipulation problems and be confident in a new generation of machine.
In spite of Boston Dynamics’ big head start in humanoids, Playter says the company got the new robot’s first build together around Christmas 2023. Before that, it was working through many of the more complex problems in simulation.
This week, it seems, the company is finally ready to begin showing off what the robot can do — or at least the early stages of what it’s planning with the system.
General intelligence
One thing you can definitely say about Elon Musk is the guy makes big promises. In the earliest public-facing days of Optimus, when the Tesla ’bot appeared to be little more than a spandex-clad human, the executive spoke of a system that could do it all. Your Optimus could work all day in the factory, do your grocery shopping and then cook you dinner. That’s the dream, right?
The truth of the matter is, of course, one built around baby steps. Robotics firms may already be discussing “general-purpose humanoids,” but their systems are scaling one task at a time. For most, that means moving payloads from point A to B. Truly utilizing the form factor, however, will require a more generalized intelligence.
It appears the app store model might present the clearest path there. Developer access has, after all, been a big part of growing out Spot’s feature set. Playter, however, says Boston Dynamics won’t be taking that approach with Atlas.
“We are definitely going to target an application ourself and not build a platform,” he says. Our experience is that the way to go fast is for us to focus on an application and go solve problems — and not assume someone else is going to solve it for us. I do think AI is an essential piece here. To support the generality of tasks is going to take and will be reinforced with AI techniques.”
The company recently opened access to Spot’s reinforcement learning algorithm for developers. That work will be foundational to Atlas’ growing skillset.
Outside the box
To be successful, Playter explains, humanoids have to move beyond the boxes.
“I think you can do that with so many other robots,” he says. “Humanoids need to be able to support a huge generality of tasks. You’ve got two hands. You want to be able to pick up complex, heavy geometric shapes that a simple box picker could not pick up — and you’ve got to do hundreds of thousands of those. I think the single-task robot is a thing of the past. Stretch is one of the last applications where you can have a robot just moving around boxes and make it work.”
If not boxes, what will the new Atlas be tasked with on the Hyundai show floor? The answer can be found in a video posted by the company back in February, which saw the hydraulic version of the robot interacting with car struts — the Hyundai components to which Playter alluded to earlier.
“Our long history in dynamic mobility means we’re strong and we know how to accommodate a heavy payload and still maintain tremendous mobility,” he says. “I think that’s going to be a differentiator for us, being able to pick up heavy, complex, massive things. That strut in the video probably weighs 25 pounds. Picking up wheels — we’ll launch a video later as part of this whole effort showing a little bit more of the manipulation tasks with real-world objects we’ve been doing with Atlas. I’m confident we know how to do that part, and I haven’t seen others doing that yet.”
Sanctuary AI often isn’t mentioned in the same breath as humanoid robotics firms like Boston Dynamics, Agility, Figure and 1X, but the Canadian company has been operating in the space for some time. In fact, a new robot introduced on Thursday is actually the seventh-generation of its Phoenix line.
While a recent iteration introduced legs into the line, Sanctuary is most concerned with what’s happening from the waist up. Tellingly, new videos of the latest robot are focused on the system’s torso. The Canadian firm is highlighting the system’s human-like movements while sorting product, as well as the speed with which it can learn such tasks.
Much of the humanoid coverage up to this point has (understandably) revolved around mechatronics — specifically how these robots look as they navigate their way through the world. Boston Dynamics’ recent video is a perfect example of how much can be communicated in a few short seconds.
Ultimately, however, robotic intelligence will be as — or perhaps more — important to these systems, going forward. While “general intelligence” is one of those big, abstract terms, the consensus among many roboticists is that we’re still five-10 years out from robots that are capable of learning new tasks in a manner similar to their human co-workers.
That does not mean, however, that today’s systems can’t learn tasks quickly. Sanctuary, for instance, suggests that the new Phoenix is capable of automating new tasks in less than 24 hours.
“With generation seven, we have a system that we believe is the most closely analogous to a person of any available,” co-founder and CEO Geordie Rose says. “We see this as not only the cornerstone of general-purpose AI robotics but a critical step on the path to artificial general intelligence, and we’re thrilled to be leading the charge on it.”
Image Credits: Sanctuary AI
Now the amount of time and consistency almost certainly varies from task to task. Those featured in the demo are relatively simple — separate objects of different colors — in what looks to be a structured and controlled environment. But Sanctuary absolutely deserves credit for already having deployed earlier systems. The company also recently announced a deal that will bring its systems to Magna auto manufacturing facilities.
For the most part, the seventh-generation robot — introduced 12 months after its predecessor — brings further refinements to the line, including increased up time, an improved range of motion, lighter weight and a lower-cost bill of materials.
The savvy marketers at Boston Dynamics produced two major robotics news cycles last week. The larger of the two was, naturally, the electric Atlas announcement. As I write this, the sub-40 second video is steadily approaching five million views. A day prior, the company tugged at the community’s heart strings when it announced that the original hydraulic Atlas was being put out to pasture, a decade after its introduction.
The accompanying video was a celebration of the older Atlas’ journey from DARPA research project to an impressively nimble bipedal ’bot. A minute in, however, the tone shifts. Ultimately, “Farewell to Atlas” is as much a celebration as it is a blooper reel. It’s a welcome reminder that for every time the robot sticks the landing on video there are dozens of slips, falls and sputters.
Image Credits: Boston Dynamics
I’ve long championed this sort of transparency. It’s the sort of thing I would like to see more from the robotics world. Simply showcasing the highlight reel does a disservice to the effort that went into getting those shots. In many cases, we’re talking years of trial and error spent getting robots to look good on camera. When you only share the positive outcomes, you’re setting unrealistic expectations. Bipedal robots fall over. In that respect, at least, they’re just like us. As Agility put it recently, “Everyone falls sometimes, it’s how we get back up that defines us.” I would take that a step further, adding that learning how to fall well is equally important.
The company’s newly appointed CTO, Pras Velagapudi, recently told me that seeing robots fall on the job at this stage is actually a good thing. “When a robot is actually out in the world doing real things, unexpected things are going to happen,” he notes. “You’re going to see some falls, but that’s part of learning to run a really long time in real-world environments. It’s expected, and it’s a sign that you’re not staging things.”
A quick scan of Harvard’s rules for falling without injury reflects what we intuitively understand about falling as humans:
Protect your headUse your weight to direct your fallBend your kneesAvoid taking other people with you
As for robots, this IEEE Spectrum piece from last year is a great place to start.
“We’re not afraid of a fall—we’re not treating the robots like they’re going to break all the time,” Boston Dynamics CTO Aaron Saunders told the publication last year. “Our robot falls a lot, and one of the things we decided a long time ago [is] that we needed to build robots that can fall without breaking. If you can go through that cycle of pushing your robot to failure, studying the failure, and fixing it, you can make progress to where it’s not falling. But if you build a machine or a control system or a culture around never falling, then you’ll never learn what you need to learn to make your robot not fall. We celebrate falls, even the falls that break the robot.”
Image Credits: Boston Dynamics
The subject of falling also came up when I spoke with Boston Dynamics CEO Robert Playter ahead of the electric Atlas’ launch. Notably, the short video begins with the robot in a prone position. The way the robot’s legs arc around is quite novel, allowing the system to stand up from a completely flat position. At first glance, it almost feels as though the company is showing off, using the flashy move simply as a method to showcase the extremely robust custom-built actuators.
“There will be very practical uses for that,” Playter told me. “Robots are going to fall. You’d better be able to get up from prone.” He adds that the ability to get up from a prone position may also be useful for charging purposes.
Much of Boston Dynamics’ learnings around falling came from Spot. While there’s generally more stability in the quadrupedal form factor (as evidenced from decades trying and failing to kick the robots over in videos), there are simply way more hours of Spot robots working in real-world conditions.
Image Credits: Agility Robotics
“Spot’s walking something like 70,000 kms a year on factory floors, doing about 100,000 inspections per month,” adds Playter. “They do fall, eventually. You have to be able to get back up. Hopefully you get your fall rate down — we have. I think we’re falling once every 100-200 kms. The fall rate has really gotten small, but it does happen.”
Playter adds that the company has a long history of being “rough” on its robots. “They fall, and they’ve got to be able to survive. Fingers can’t fall off.”
Watching the above Atlas outtakes, it’s hard not to project a bit of human empathy onto the ’bot. It really does appear to fall like a human, drawing its extremities as close to its body as possible, to protect them from further injury.
With a 99% success rate over about 20 hours of live demos, Digit still took a couple of falls at ProMat.
We have no proof, but we think our sales team orchestrated it so they could talk about Digits quick-change limbs and durability. #ConspiracyTheories pic.twitter.com/aqC5rhvBTj
— Agility Robotics (@agilityrobotics) April 6, 2023
When Agility added arms to Digit, back in 2019, it discussed the role they play in falling. “For us, arms are simultaneously a tool for moving through the world — think getting up after a fall, waving your arms for balance, or pushing open a door — while also being useful for manipulating or carrying objects,” co-founder Jonathan Hurst noted at the time.
I spoke a bit to Agility about the topic at Modex earlier this year. Video of a Digit robot falling over on a convention floor a year prior had made the social media rounds. “With a 99% success rate over about 20 hours of live demos, Digit still took a couple of falls at ProMat,” Agility noted at the time. “We have no proof, but we think our sales team orchestrated it so they could talk about Digits quick-change limbs and durability.”
As with the Atlas video, the company told me that something akin to a fetal position is useful in terms of protecting the robot’s legs and arms.
The company has been using reinforcement learning to help fallen robots right themselves. Agility shut off Digit’s obstacle avoidance for the above video to force a fall. In the video, the robot uses its arms to mitigate the fall as much as possible. It then utilizes its reinforcement learnings to return to a familiar position from which it is capable of standing again with a robotic pushup.
One of humanoid robots’ main selling points is their ability to slot into existing workflows — these factories and warehouses are known as “brownfield,” meaning they weren’t custom built for automation. In many existing cases of factory automation, errors mean the system effectively shuts down until a human intervenes.
"We'll be singing / when we're winning." 🔊 pic.twitter.com/51DYD1Avvg
— Brian Heater (@bheater) August 17, 2021
“Rescuing a humanoid robot is not going to be trivial,” says Playter, noting that these systems are heavy and can be difficult to manually right. “How are you going to do that if it can’t get itself off the ground?”
If these systems are truly going to ensure uninterrupted automation, they’ll need to fall well and get right back up again.
“Every time Digit falls, we learn something new,” adds Velagapudi. “When it comes to bipedal robotics, falling is a wonderful teacher.”
