When I look back on these wild past few weeks, I will fondly remember the apricots. There were those eaten by the armful — comfort eating during an emergency trip back home to the Bay Area. Some, too, were no doubt driven by knowledge I’d likely not come close to their kind until the next time I find myself in proximity to California’s unparalleled Central Coast.
I’m pleased to report, however, as I write this from a café on the south bank of Zurich’s Suhl river ahead of my hosting duties at the first-ever Davos Tech Summit, that the Swiss’s apricot game gives the Golden State a run for its money. I’m excited to report back what wonders France’s apricot game holds while I’m in town next week for Machina. My visit to Chicago was, sadly, a wholly apricot-free endeavor.
There were, on the other hand, plenty of robots.
More than 50,000 human beings registered for A3’s 10th annual Automate conference. This time, there were plenty of humanoids amongst the McCormick Center throngs, as well, thanks in large part to the presence of the NVIDIA-sponsored Humanoid Robot Pavilion, coupled with the similarly named Forum (HRF), which was folded into the larger tradeshow after two years as a standalone event.
A3’s puts the combined number of humanoid and quadrupeds (a large net, admittedly) at more than 50. While the figure is a relative drop in the bucket in the overall context of on-site industrial robots — 1,230 exhibitors registered for the event — it represents a sharp uptick from last year’s show, where humanoids were much discussed and mostly absent.
The dynamic has shifted significantly in 2026. The humanoids are increasing, and so, too, has the skepticism. It’s a shift that came into sharp focus in — of all places — last September’s HRF. Increased mainstream media attention, mind-boggling fundraising numbers, and… aggressive production projections from prominent figures have put companies’ claims under the microscope.
Skepticism is distinct from cynicism, of course, and having played a key role in programming these last two HRFs, I can tell you that pragmatism is foundational to our discussions. I moderated the closing panels for both days, focusing on the biggest challenges to the form factor, deployment, adoption, generalization, training, and dexterity.
In February, Apptronik announced a $590 million Series A extension, bringing the full round to nearly $1 billion. The massive funding news arrived just as the humanoid maker had leased around 60,000 square feet of additional space across the street from its Austin, Texas headquarters.
In a call this week, CEO Jeff Cardenas confirmed that the size has ballooned to roughly 90,000 square feet, which the company has christened “Robot Park.” The space is being devoted to an increasingly prominent trend in the world of physical AI and general-purpose robotics.
It’s a kind of brute-force approach to model training, aiming at a flywheel many proponents believe can only be sufficiently kickstarted through real-world data collection. The space, which Cardenas says formerly served as a Dell server plant, replaces a far smaller area of Apptronik’s offices that was used for model training.
Robot Park already functions as the home to Apollo 2, the latest version of the company’s humanoid robot. In our discussion, Cardenas referred to the system as a “prototype” and “data collection platform,” much like its eponymous predecessor. The labels are meant to distinguish the robot from a production system designed for commercial deployment, rather than an attempt to downplay the advances the system presents.
“It's a significant upgrade over Apollo 1,” says Cardenas. “A next generation of actuation is in Apollo 2. It has a bigger battery, better at achieving that four-hour runtime, depending on what it's doing. The battery was safety rated this time so that we could get out to customer sites. New sensors on the robot, so new camera placement for data collection. We've tested out a variety of different end effectors and hands on Apollo 2 overall.”
The road to generalization is paved with smaller, simpler tasks. The current generation of humanoid robots, prized for their flexibility, will make their way into factories and warehouses one job at a time. Tote moving is likely the example you’ve most often seen touted — transferring bins from one location to the other.
Is it something most humans can do? Sure. Is it something most humans want to be spending their precious time on Earth doing? Probably not so much. In a conversation shortly after the wrap of our Humanoid Robot Forum, Agility Robotics CTO Pras Velagapudi describes this as the sort of in-between pieces of automation.
They’re not particularly stimulating for the humans tasked with staffing them, but robotics hasn’t quite found a way to bridge the pieces.
“There's a lot of utility already in having robots that fill these blue automation roles where they're between inconvenient parts of the process,” says Velagapudi.
“They're connecting pieces of automation that don't fit together well. This happens all over manufacturing logistics right now. Someone needs to unload an AMR, someone needs to put totes into a shelf. And it's exactly the sort of use cases the Digits are really good for, when it's, you don't want to modify the infrastructure a lot. They're strenuous, repetitive types of applications.”
Velagapudi’s HRF presentation was sandwiched between two big pieces of news for the Oregon-based humanoid firm. First NVIDIA announced that it had tapped the Oregon-based firm as the first partner for Halos for Robots — of its autonomous driving platform focused on automation safety. After the talk, Agility announced a bigger bombshell — it was set to become the first U.S.-based humanoid firm to go public.
The move, says, Velagapudi, reflects the form factor’s, specifically Digit’s, commercial readiness.
Rick Faulk (Locus Robotics) - When Amazon bought Kiva in 2012, the robotics startup's former customers were in a bind. Quiet Logistics did something about it, creating a behemoth in the process.
Aya Durbin (Boston Dynamics) - After helping deploy industrial robots at 6 River System, Aya Durbin is looking to do the same with Boston Dynamics' Atlas Humanoid.
Andrew Barry (Generalist AI) - Generalist AI's cofounder and CTO, Andrew Barry, discusses physical AI's difficulties and jaw-dropping breakthroughs.
Waabi this week announced that it has achieved cross-platform zero-shot generalization between two different vehicles. The company’s Driver platform was trained on a Peterbilt 579 and transferred to Volvo VNL Autonomous, without requiring updates to sim and real-world data, engineering, or fine-tuning, per the company. This is all in spite of the trucks’ different builds, controls, and sensor arrays. “This is a defining moment for physical AI,” says founder and CEO Raquel Urtasun. “This capability has the potential to transform far more than transportation. It is the foundation for a new generation of intelligent machines that can adapt, scale, and operate across the physical world, creating possibilities and opportunities we can scarcely imagine today.
In addition to raising the $11 million that landed his startup in Becca’s roundup this week, Jay Li’s robot hand startup settled an ongoing lawsuit with his former bosses at Tesla. The suit alleged that Proception had benefited from trade secrets gained during Li’s time as technical lead for the Optimus project. “I think it’s kind of like a resilience test, or pressure test,” Li said in a recent interview with my old bosses. “People say that what doesn’t kill you makes you stronger, right?”
Startups like Panthalassa are looking to the ocean for inspiration. It is an inspiring place to look, to be fair. Get yourself to the nearest coastline, stare for long, unbroken stretches, and just wait for the startup ideas to start pouring in. For Panthalassa, it’s wave energy that fuels their work. The company is looking to build floating AI data centers powered by the motion of the ocean. That, it turns out, is a hard problem. Really hard. Unsolved, according to some. Matthew chats with some key researchers about how far the community has gotten in wave energy — and how much is still left to go.
Japanese exoskeleton maker Cyberdyne is looking to Silicon Valley for the latest in physical AI innovation. The company teamed with Pegasus Tech Ventures. To launch a ¥10 billion ($62 million) corporate venture capital fund designed to back Bay Area startups. The deal also requires that the company receiving the funds partner with Cyberdyne, giving the established company access to the nascent technology. “Cyberdyne has been one of the corporations looking for the next generation of technology,” Pegasus CEO Anis Uzzaman told A3. “This was the right time to take a step forward to help them integrate their technologies with what is going on in the U.S. top AI engines. They are asking us to bridge the gap here.”
Researchers at MIT are showcasing a new chip that greatly reduces the compute and memory requirements for 3D mapping. Among the other use cases is little pocket-sized robots, which can now navigate complex environments, all while requiring roughly the same power as an LED. Liam’s got the story.
Sure, we've devoted roughly half of this newsletter to mobile manipulation, but what if mobile pedipulation is actually where it's at? Researchers at the University of Michigan gave this little biped the ability to move objects around with its wheels. It's not exactly what I call bi-dexterous mobile pedipulation — in fact, the research paper refers to the actions as "scooting" and "lateral sliding." The research adds that the system was able to grab a "one kg object from under a desk and slid[e] a four kg object over a distance of 0.228 m via scooting."
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=1120&upscale=true&name=Russ%20Tedrake%20(1).png)
