Greetings, time travelers, from the final day of the A3 Business Forum. We’re presently packing up our mobile podcasting studio – well, my producer Jana is. I’m sitting here, hoping she doesn’t see me as I write up this week’s newsletter before heading to the airport.
Special thanks to everyone who pushed back their own flights to catch the first live Automated recording. Double extra-special thanks to GM’s Mikell Taylor for making me look like I knew what I was doing up there. It was a great chat. We discussed her current work in automotive, her time at Amazon, launching a startup in 2017, and, of course, the time she took a robot to prom.
That episode should be dropping soon. Also be on the lookout for the episode we recorded on-site (but not on-stage) with the equally delightful Erik Nieves of Plus One Robotics. Thanks everyone for the off-camera conversations this week, including fellow travelers Chris Luecke and Jacob Hall. I took away a bunch that will no doubt feed into future newsletter episodes.
News about the upcoming Humanoid Robot Forum can be found below. I will almost certainly be knee-deep in programming meetings soon. Also, we’ll be digging a bit deeper into the Serve/Diligent a bit more in next week’s edition. Lots more Automated stuff to share with you soon, and a quick reminder to send all podcast guest suggestions to podcast@automate.org. Talk soon.
Skild.AI hasn’t taken the easy road. Abhinav Gupta acknowledges that generalized AI will be a significantly harder nut to crack than bespoke models tailored to different robot embodiments. Ultimately, however, the startup’s cofounder and president doesn’t believe there’s much choice with this particular question.
“Learning a common model across different form factors is a necessity,” he explains. “This is the only way we believe that this problem is going to get solved.”
Gupta has been steeped in questions about robot intelligence for much longer than most of the industry. In addition to research stints with Meta, Google, and the Allen Institute for AI (AI2), he’s served as a professor at Carnegie Mellon’s Robotics Institute for more than 16 years. With papers published on topics like reinforcement learning, grasping, and navigation, he’s well aware that there are still plenty of challenges ahead.
While a one-size-fits-all robot model isn’t exactly industry consensus, a growing number of startups are banking on such a generalized approach. Gupta cites two primary drivers for his own thinking on the subject. The first comes down to the broader problem of data. One thing everyone in the space seems to agree about is that there just isn’t enough of it. Like, not even close.
Various methods have been deployed to generate that data, with a wide range of efficacy. One of the best ways is to let robots interact with the real world. It’s plausible that — one day in the not-so-distant future — we’ll have a sort of critical mass of systems operating in the real, gathering and refining datasets in the process. But is there a way to achieve scale that doesn’t require having systems that are already fully trained on that data?
UK-based Humanoid has been coming in hot with the partnership news. Just last week saw both the announcement of a Siemens pilot and Schaeffler’s order of “hundreds” or units over the next half-decade. It’s a solid opening for 2026 for a company that really hit the ground running around the midway point last year.
There’s one thing you won’t see listed among this recent dispatch of press releases: robot legs. The aforementioned deployments — and, for that matter, all Humanoid industrial partnerships for the foreseeable future — will have none to stand on. While having the option of both legs and wheels has been a major selling point for the firm since day one, you’re likely to see the latter in the vast majority of industry deployments — Humanoid’s primary focus, at least for the next few years.
The breakdown offers an interesting bit of insight into the form factors, as Humanoid is in a somewhat unique position of having a stake in both. AMR proponents have long insisted that wheels are a perfectly acceptable — and sometimes superior — choice for the industrial setting. When industrial humanoids began to populate VC balance sheets, it was like encore time at the ZZ Top concert: all about legs.
If you had asked me, say, six years ago, what was next for industrial robotics, I no doubt would have pointed to mobile manipulation. Though my prediction would have been decidedly more conservative, resembling an AMR fitted with vertically elevated robot arms for bin picking. If I was feeling adventurous in a given moment, it might have more closely resembled Array, the automated order picking machine Locus told us about a month or two back.
Last-mile delivery firm, Serve Robotics, on Tuesday announced plans to acquire Diligent Robotics. Founded in 2017, Austin-based Diligent has grown into one of the better-known robotics startups in the patient-facing healthcare space. The company’s Moxi robot has been deployed at 25 hospitals across the United States, offering assistance to nurses and other health care workers. According to the company, the robot has made 1.25 million deliveries to date. Serve, meanwhile, spun out of Uber in 2021. The company currently has 2,000 robots deployed in U.S. markets that have made north of 100,000 deliveries combined. Existing customers include 7-11 and former owner, Uber. In a release tied to the news, cofounder and CEO, Andrea L. Thomaz stressed the limited resources at hand when Diligent was founded, just under a decade ago. The firm certainly preceded the current robotics and AI funding boom by several years, as well as a global pandemic that drove worldwide nursing shortages. Serve will pay out stock valued at $29 million to acquire the company, with a potential added $5.3 million, should undisclosed milestones be met. The deal is expected to close in Q1 of 2026, pending regulatory approval.
A few years back, the A3 team asked me to moderate Automate's first humanoid panel. It was a great conversation, but still felt like an odd fit. Sure, people were talking about the form factor at the show, but you'd really have to go hunting to find anything on the floor that fit the bill, beyond a couple of Pepper robots well past their prime. Upon joining the association last March, one of my first tasks was helping the conference team program that year's Humanoid Robot Forum. The lineup was great, the show went well, and the event marked an evolution in the way some of us discuss the space. Many of the speakers and attendees found themselves ready to re-examine their own views on the subject. How necessary are legs, really? What does real scalability look like? Are we underestimating how difficult mobile manipulation truly is?
This week at Forum, the people who pay me money to write this newsletter announced that the larger event is absorbing the former, as Automate hits the Windy City June 22-25. NVIDIA —- which has obviously taken a special interest in the category courtesy of its Jetson developer platform — is sponsoring the annex. It will be comprised of two key parts. There's the Humanoid Robot Pavilion, a dedicated section of the show floor, where you can engage in prolonged staring contests with the latest models. That bit is free to everyone who purchases a ticket to Automate and uses the code words: "I listen to the Automated podcast every week" at the door. I'm told that last bit isn't strictly necessary, but I figure better safe than sorry. HRF, the programmed conference bit, will remain a separate (but on-site) paid event set for Tuesday and Wednesday afternoon.
While the end product of human evolution is a bit of a mixed bag from a design perspective, folks I've spoken with tend to thing we got a good deal on the hand front. Hey, there's a reason we primates love lording our opposable thumb over most of the animal kingdom (word to the wise, do not try this with a Koala). There are plenty of robotics folks who are of the opinion that humanoids, too, would be better off with something that looked as close to human as possible. After all, one of the form factor's biggest selling points is its ability to operate well in human-centric environments. Certainly this spills over into much of the tooling, totes, and other objects with which these systems are expected to regularly interact. I have, however, always had a soft spot for the superhuman argument. However Icarian it is to believe we can one-up millions of years of evolution with a bit of elbow grease and a metric tonne of venture capital, why not give it a shot. Sure, two arms are great for carrying around totes, but that doesn't mean a third wouldn't offer a key advantage in some situations.
This research from EPFL and MIT suggests that our hands' lack of symmetry can be a disadvantage for some jobs. Again, we love our thumbs (imagine hitchhiking or rating a movie without them — oh wait, you can't, it's impossible). When combined with the index finger, it makes a great grasper (again, never turn your back on a Koala). In fact, the biggest downside of the whole deal is that —- generally speaking — we only have one per hand. The robot hands featured in this Nature paper, on the other hand, have graspers for days. The researchers note, "We designed a robotic hand with a uniform, symmetrical structure. Each finger can bend bi-directionally, increasing dexterity." Oh, and it detaches from the robot's wrist and uses the fingers for "loco-manipulation," to help it get across the room.
Airbus is piloting a bunch of UBTech humanoids to help build planes. Not to be actual airplane pilots. Not yet, at least.
Project Stardust is an extremely worthy (and personal) new project from past newsletter subject, Katherine Zealand.
These robots bloom when exposed to sunlight or people. Bonus points for the wildly uncanny (and this is coming from a guy who just wrote about a robot hand that walks across the floor at will) phrase, " 'living-like’ architecture for functional and creative applications."
Eric Danziger (Invisible AI) - Invisible AI's CEO discusses why making real robots that really work in the real world is real, real hard.
Vincent Vanhoucke (Waymo) - When it comes to AI at Alphabet, Vincent Vanhoucke has seen it all, from research scientist at Google Brain to head of Google Robotics to his current role at Waymo.
Rohit Sharma (True Ventures)- The investor discusses how the firm determines early stage successes and charts his family’s journey from India farm to Palo Alto boardroom.
The Association for Advancing Automation (A3) is North America’s largest automation trade association representing more than 1,400 organizations involved in robotics, artificial intelligence, machine vision & imaging, motion control & motors, and related automation technologies.
.png?upscale=true&width=420&upscale=true&name=Eric_Square_Automated%20(1).png)
