We can’t give ourselves too much credit for good timing. For the past decade or so, I’ve covered robotics for a niche — though quite committed — audience. Now that physical AI is the greatest thing since sliced bread (I’m mostly dusting off this old cliché to point out that sliced bread is also a surprisingly recent invention that will celebrate 100 years on the market in 2028), simply covering the category equates to good timing, more often than not.
One of the best and most annoying parts of doing the Automated podcast is the fact that we’re often sitting on months of backlog. A good problem to have, no doubt. One day, when I take my long-threatened first vacation, our regular release cadence won’t be impacted.
The downside, however, is that things are happening so quickly in this space that even a few weeks can make a conversation seem impossibly dated. That’s one of the reasons you’ll see features based on the podcast appear before the podcasts themselves. But even with the built-in delay, our timing lends false credence to the notion that we have any idea what we’re doing.
This week’s newsletter is a great example. Take the Russ Tedrake interview that dropped two weeks back. The researcher’s long-teased physical AI startup, Walden, finally came out of stealth. Also, this week’s guest, Ayanna Howard (featured above on the cover of her own 2021 kids book, illustrated by Elena Bia), was just named President of Atlanta’s Spelman College.
The lesson here — if there is one — is that covering very smart and capable people means reaping some small benefits from the smart and capable things they’ll inevitably do.
We closed both days of this year’s Humanoid Robot Forum with three-person panels, designed to offer a bigger picture of the industry halfway through 2026. It’s a nice way to balance a series of one-person presentations, engaging in conversation, while taking a pragmatic look at the challenges the form factor currently faces.
The day two panel (which we’ll explore in an upcoming piece) featured Ani Kelkar (McKinsey), Erin McColl (Toyota Research Institute), Elizabeth Samara Rubio (Noble Machines), and Rebecca Yeung (Silicon Foundry, Dexterity) discussing questions of scaling and deployment.
On day one, I spoke with Aadeel Akhtar (PSYONIC), Olivia Norton (Sanctuary AI), and Bren Pierce (KINISI) about humanoids’ other biggest problems: namely dexterity and data collection/training. But before all of that, we had to address the bear in the room. Kinisi, which made its Automate debut by winning 2025’s Startup Challenge, just revealed it had been acquired.
“I sold the company yesterday. To my other company. It's very similar to Elon Musk,” Pierce winked, “you take one company, sell it to your other one.” The CEO cofounded restaurant automation firm Bear Robotics in 2017, before serving as chief robotics officer for the next four years.
In 2023, he founded Kinisi with chief of staff Shannon Needham. Among the U.K. company’s guiding philosophies is a drive to “build robots that actually get work done.” That entails utilizing available technologies, rather than the full-stack from-scratch approach currently driving many of the company's bipedal competitors.
Perhaps the biggest shortcut to deployment was the design to build a bidexterous humanoid torso on top of Bear’s wheeled base and navigation stack.
“For the last six months, I found myself repeating what I was trying to do before at Bear,” he adds. “ ‘Okay, now I need to mass-produce this, I need fleet management, over-the-air updates, navigation, a sales team. And at the same time, I know a company with 300 employees that's already doing all of this.’ ”
The procedure is a cholecystectomy — a standard gallbladder removal. The patient, a zonked-out pig. UCSD Healthcare Network members Drs. Ryan Broderick and Shanglei Liu are present to perform the operation. That last bit, however, is where things get a bit tricky. Both surgeons are present in the room with the porcine patient, but their operations are happening in a technological parallel.
Even this isn’t a remarkable element in and of itself. We tend to forget how far ahead of the adoption curve robotic surgery has been. Intuitive’s da Vinci system cleared FDA approval a little over a quarter century ago. By 2012, it was performing somewhere in the neighborhood of 200,000 procedures annually, with hysterectomies and prostate removals topping the list.
In essence, this surgery doesn’t stray far from that work. Our medical professionals are performing the gallbladder removal remotely (albeit only a few feet away in this instance), using a joystick similar to the one they’ve trained on for past procedures. What makes this particular instance remarkable, however, is the change in form factor.
UCSD Electrical and Computer Engineering professor Michael Yip has successfully completed another key step in his ongoing work to bring humanoids into the operating room. The Unitree G1 systems have returned with modifications, most notably to their end effectors, for better grasping the necessary medical equipment. The software stack has also received a major update since we last spoke at A3’s 2025 Humanoid Robot Forum, allowing for sub-millimeter-scale accuracy.
An efficient sterilization process wasn’t the focus this time around in a controlled environment, leaving the involved parties to wipe the robots down with rubbing alcohol. “With a humanoid robot, unlike a human, you can actually pass it through complete decontamination processes, whether it's gas or aerosol liquid,” Yip explains. “That's what the future will look like in my opinion.”
When you attend as many robotics and physical AI events as I do, presentations have a way of blending into one another. It’s not a lack of good insights, so much as the difficulty of introducing novel views of familiar subjects. After watching his talk on exoskeletons on the stage I hosted at the Davos Tech Summit, I pulled Loris Roveda aside for some follow-up questions.
What jumped out at me was the IDSIA USI-SUPSI associate professor’s argument for supplying exoskeletons with situational awareness. It’s a fairly novel concept for those of us who generally think of these as assistive systems. Industrial exoskeletons, for example, adapt to their wearer’s movements. Once it senses that you’re lifting up a box, it will imbue your legs with some extra oomph.
Roveda points to an inherent flaw in this design: the amount of time it takes for the sensors to read the motion, and estimate the nature and amount of constructive force required for the wearer to best carry out a task.
“Until now, [exoskeletons] were purely reactive,” he explains. “For example, we have force torque sensors. We apply a force and the robot will follow. This is purely reactive, so we start the motion and the exoskeleton moves after that with the delay. The sensors can be very good, but it will always be reactive.”
Roveda’s pitch involves a system that is intelligent — and fast — enough to offer aid at the outset of such motions. I suggest the word “predictive” to define the approach. The associate professor prefers “proactive."
Dr. Howard’s career has spanned NASA robotics, field robotics, healthcare robotics, assistive technology, AI ethics, and engineering leadership. Across all of that work, one theme has remained constant: technology should help improve the human condition. Watch on YouTube>
Yoel Fink (MIT)- Materials engineer and longtime MIT professor, Yoel Fink, discusses what it means to live a life based on first principles.
Russ Tedrake (Researcher/Stealth Startup)- After a decade serving as TRI's SVP of Robotics Research, the long time MIT professor is launching a new physical AI startup.
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.
We’ll have a more fleshed-out take on this news next week, including a chat with chief product officer Dave Johnson. Meantime, another big physical AI player just emerged from stealth — though everyone in the industry saw this one coming. Russ Tedrake’s startup, Walden, is now officially official, after several months of teases. The Cambridge, Massachusetts firm is hitting the ground running, with a $300 million investment round led by Deviation Capital and Tedrake’s former employer, Toyota. In fact, Walden is being treated as a TRI (Toyota Research Institute) spinout, as of this January. Other key players were involved, including NVIDIA, Boeing, AE Ventures, and Samsung Ventures, which participated in the round that values the brand-new company at $1.1 billion.
The company is built on the back of Tedrake’s pioneering work around Diffusion Policy and Large Behavior Models (LBMs). “Core advances in physical AI, and all of the excitement and attention surrounding it, has made disruptive change possible,” Tedrake says in a release. “But providing real value to customers and building a robust and scalable business requires a deep understanding and respect for how manufacturing is done today. The best way to make fast and positive progress is by working closely together with the real experts.” Seems Walden is going fairly full-stack here, as well. Per Deviation investor, Colin Beirne, “The Walden team is generational, bringing together a rare, interdisciplinary group of experts spanning foundational robotics research, large-scale production hardware, and operational and business leadership.
We had questions about that recent, massive Tether-led funding round (plus humanoids, physical AI training, etc.), so we carved out some time with David Reger at Automate the other week. Becca and I chatted with Neura’s CEO, and she’s repurposed some of the key takeaways in a feature format. Spoiler alert, but I felt obligated to highlight the quote that closes the piece out: “We’ve shown [at] CES already a dancing robot. We brought it into the club. It was cool. But this is not the purpose of this company. The purpose of the company is making humans dance and walk around while robots actually go and work.” 💃
Boston Dynamics Tuesday confirmed it's been testing Spot in a last-mile delivery scenario. The setup is only just modified from the standard robot, with a shelving unit on the back, where people (or, presumably, robots like, say, Stretch) can place multiple payloads. The legged robot then navigates to the door, walks up the porch stairs, and gives a kind of bow, as the packages slide off. Amazon has reportedly been testing quadrupeds for this manner of scenario, a fact seemingly cemented with its recent purchase of legged/wheeled quadruped producer, RIVR. “So much of logistics is already automated, but we believe that the final frontier of logistics automation is that last 50 feet,” Spot SV, Marco da Silva, says in a release. “We’ve done things to automate inside of the hub, inside of sort facilities, but now we’re trying to bridge the gap to the final mile.”
This week, Liam explores how telepresence robots could help connect older adults while strengthening their physical and mental well-being. The piece derives from recent research out of the University of Nottingham, which also explores how such systems might be used as a springboard to integrate more robotics and other age-tech into these communities.
We’ve hit the point in the Waymo hype cycle where expansion into another four major American cities barely registers as standalone news — unless, of course, you frequently find yourself in Denver, Las Vegas, San Diego, or Tampa. The Google spinoff continues its creep across the contiguous United States with plans to launch fully autonomous ride-hailing in those markets “soon.” Those spots will also be graced by the rollout of the new human-supervised (for now) Hyundai IONIQ 5.
Recent A3 Division I recruit, Aaron Prather, gets on the boards with this feature exploring the parallels between grocery store robots like Simbe’s Tally and their manufacturing/logistics forefathers. “Viewed individually, [robot adoption by supermarket chains doesn’t] appear transformative,” he notes in a kind of thesis statement. “Viewed collectively, they suggest grocery retail has reached the point where automation is shifting from isolated pilots to enterprise strategy.”
Affordable robot manipulators — we love to see it. This is the perfectly named GEM (Good Enough Manipulator), a project from University College London PhD candidate Joe Clinton. Per the promotional page (which also offers downloadable files and instructions), the arm’s build of materials comes in under $500. The system is 3D printable, has five degrees of freedom, can lift payloads up to 1.2 kilograms (~2.7 pounds), and is compatible with Hugging Face’s LeRobot repository (it also utilizes Hugging Face-owned Pollen Robotics’ PincOpen for a gripper).
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.
