As mentioned last week, it’s just about time for my long-threatened break. Automated will be in Becca’s capable hands for the next two weeks, while I attempt to remember what it is people do all day when they’re not working.
Plans so far: 1. Not traveling. 2. Petting rabbits [pictured]. 3. Reading. 4. Hiking 5. Cleaning. 6. Recording my other podcast. 7. Working on a book proposal about robots (anyone know a good agent?).
The deeper you get into these lists, the more things start wrapping back around to work. Maybe a matinee is in order. Primetime or the Nathan Fielder doc? Two in two weeks seems a tad indulgent.
We still have plenty to wring out of Automated before 2026 wraps up. I have another feature that didn’t quite make it in time for this week’s newsletter that I plan to pub on Dyna. If you open these issues as soon as they hit your inbox, you may still be able to catch my live Humanoid safety webinar with Apptronik and FORT (don’t worry, you can replay it later).
I’ve got a pair of fun panels coming up for SOSV’s upcoming Deep Tech Summit on November 6 (also online): 8:40 a.m. PT Russ Tedrake: Teaching Machines to Learn from Humans and 11:10 a.m. PT Re-Booting Industrial Robotics: Agility Robotics, Inbolt and Standard Bots. Register here.
It looks like I’m going to be back in the Midwest and Silicon Valley one or two more times, as well, but nothing official to promote just yet. In the meantime, if you’re interested in catching a live version of the podcast near you, let me know over on LinkedIn (so far Boston is running away with this one, so let your voices be heard, SF, Pittsburgh, and Austin).
One more small favor, before I go: if you help us hit 5k YouTube subs while I’m out, I’ll probably come back after break. Probably. Meantime, thanks for making Becca's life slightly easier by getting this week's job listings in early.
Orthogenesis fell out of favor before the mid-20th century. The fossil record simply didn’t reflect the notion of evolution working in a single direction, toward some grand end. Popular culture never got the memo, of course. It’s much easy to sleep through the night believing you occupy the true pinnacle of 3.7 billion years or so of evolutionary progress.
A simple X-ray of my L4-L5 lumbar vertebrae, however, is enough to put quick lie to any such notions. At very least, it offers a strong argument that walking upright comes with its own disadvantages. Roboticists inherently grasp such concepts — not just due to their own low back struggles.
Without some grand plan, pre-determined outcome, the concept of “de-evolution,” too, suddenly holds little water outside select Midwestern new wave/art punk bands. Evolution is a function of opportunism. The same can be said for engineering. Biology is an extraordinarily useful guide when it comes to developing hardware embodiments for general purpose AI, but shouldn’t be regarded as the be-all, end-all.
In the world of robotics, the question of how closely to hew to biological precedent is well encapsulated in the end effector. If the human form fact is, indeed, well suited for general purpose tasks, that shouldn’t our most versatile and capable built-in tool — the hand — be more suited above all?
The answer, I’m sorry to report, is that it’s complicated. Those existing complications are further complicated by the needs of capital and the requirements of scaling and reliability. When it comes to successfully deploying an industrial humanoid, you can’t really overstate the degree to which the hand is a unique conversation unto itself.
Boston Dynamics tossed out a lot of numbers during its CES 2026 Atlas unveil, but for me, one really stood out: two. With scaling and cost in mind, the company managed to build the entire system out of two different varieties of in-house actuators. Once you factor in the hands, however, that number is three.
Building a functional, versatile hand from a single variety of actuator is impressive, no doubt. The fact that the hands required their own version independent of the rest of the body is a testament to the uniqueness of the problem. Mobile manipulation for distinct digits needs to be small, dexterous, robust, and powerful. It’s a lot to ask from little actuators. Continue Reading>
Jagdeep Singh believes the industry needs a much larger and more diverse source of training data to build robots that can handle the unpredictability of real work. At Rhoda AI, that starts with internet video. Watch on YouTube >
Andy Zhang (ABB) - What's an industrial giant like ABB to do amid a humanoid hype bubble? Lots. While general purpose nabs headlines, industrial arms and cobots dominate deployment.
Karen Panetta (Tufts University) - From helping create the digital twin to solar-powered race cars, Karen Panetta has lived far more life than can fit into a single podcast. We still give it the old college try.
Ahmad Bahai (Texas Instruments) - Ahmad Bahai believes in the power of the semiconductor. Texas Instruments' CTO is a long-time industry vet with a unique outlook on how the next generation of hardware is fueling the AI revolution.
Google's Google DeepMind team Tuesday unveiled Gemini 4 Argon for Robotics, which, "marks a step change in physical understanding, code generation, and visual-semantic reasoning — unlocking new workflows across the robotics stack and accelerating progress toward physical AGI." Perhaps the most fascinating demo of the bunch finds the team training Apptronik's Apollo 2 robot to play cornhole (bean bag toss, depending on your regional variant) based on a video. The system constructed a digital twin based on the video guide, using the MuJoCo sim, and effectively taught itself the game through trial and error, before it "accurately calculated the physics and whole-body control required for Apptronik’s Apollo 2 humanoid to sink seven consecutive shots in the real world." DeepMind says the system training was accomplished in a manner of hours.
The Departed. Great film — though, honestly, probably doesn’t crack my Scorsese top 10 (also, working with really strong source material there). My (second) biggest takeaway is that the quickest way to master the Boston accent is by saying “microprocessesahs” over and over again. I’m convinced this is also an unspoken secret to the city’s longstanding position as a bleeding edge technology hub, along with other incidentals like colleges, and the like. Startups and big tech firms have planted roots in the greater Boston area, assuring continued growth of that bustling ecosystem, even in the wake of some high profile struggles, like iRobot.
This week, DeepMind’s VP of Embodied Systems, Robotics, Aaron Saunders, announced Google’s plans to build a new physical AI lab facility in Cambridge. Saunders previously spent a couple of decades at nearby Boston Dynamics, making working his way up to the CTO role, before becoming one of a handful of high-profile exits last year. “Since joining a year ago, we have been developing a new hardware team that builds on Google DeepMind’s pioneering leadership in AI and robotics,” he wrote on social media Tuesday. “Today, I’m excited to share that we're creating a first-class lab to house this team and take robotics to the next level.” The company promises more info on the project “later in 2027.”
Interesting piece from Liam this week based on research out of Germany's Chemnitz University of Technology. The studies aim to determine how pedestrians reflexively react to autonomous vs. human-driven vehicles. The gulf between the two isn't as large as you might assume at first glance. It's a somewhat surprising finding, given how novel this technology remains, particularly outside of select markets like San Francisco, where the sight of self-driving vehicles has become commonplace in recent years.
Here’s one of the more pragmatic statements I’ve heard on the topic of teleop vs. full-autonomy of late: “You don't want full autonomy in these spaces. You just want something that works, that works reliably, and that you can supervise and make sure you can hold it liable. There’s a question of liability if it was autonomous. Imagine your autonomous humanoid turns the wrong valve because it misidentifies it one time. What would that look like? You absolutely cannot have that.” That comes from Minerva Humanoids CEO Sandor Felber in this piece from Becca. It gets to the heart of the subject I’ve been mulling over recently: is full autonomy always the right end game?
When I’m leaning towards “no,” the answer almost always has to do with safety. Certainly that’s the case with Minerva, whose humanoids are designed to work in harm’s way, so human beings don’t have to. That includes inspection for power plants and off-shore oil rigs, along with potential explosive device response. “We have to have a human in the loop for now, and we see this as kind of a transition,” Felber adds. “We have our own data collection flywheel, and we are rolling out more and more autonomous features to make it more convenient. But we're also trying to not change too much and make it fully autonomous ahead of time because, frankly, like that's more counterproductive than productive.”
Hugo is the first robot from Boston Dynamics off-shoot, Dynamic Creatures. We spoke with the company's CEO, Marc Theermann for the Automated with Brian Heater podcast a few weeks back (the episode itself is coming soon), following its September launch video. That video features a brief, silent cameo from a pink poodle named Danielle. Hugo, on the other hand, offers a far more realized look at the company's work developing robots for the hospitality industry, including theme parks, a la the pioneering work that's been done over at Disney Research. Hugo features, "43 actuators, animatronic character design, and our work in expressive movement and human-robot interaction. He is designed to move freely through guest environments as a character you don’t just watch, but meet."
“The truth is that in an industrial environment its a little bit easier to figure out, because you can put the robots behind a light curtain,” Theermann told me last month. “The human walks through the light curtain, the robot first slows down, and then totally shuts down. But thats not where we want to go. We want to create robots that you can hug. Thats our North star. So keeping the humans away from the robots is not going to be our answer.”
New England grocery automation giant Symbotic this week announced that it has acquired Cellumation. The German firm’s flagship product is Celluveyor, a conveyor system with “intelligent, omnidirectional movement.” The deal gives Symbotic a stronger foothold in Europe. Cellumation, meanwhile, will maintain its branding and existing customerbase. Per Symbotic CEO Rick Cohen, “Cellumation’s technology is a game-changer. By bringing the celluveyor into the Symbotic portfolio, we are accelerating our ability to help customers achieve higher performance, better space utilization, and lower system costs while continuing to push the boundaries of physical AI across the supply chain.”
"Mr. Field" was my father's name. Please, call me "President Mind." Rivian CEO RJ Scaringe continues making waves with Mind Robotics, the physical AI firm he founded last year. We still don't know a ton about the company, only that Rivian has (unsurprisingly) taken a special interest in their technology, along with investors. Mind has raised more than $1 billion in under a year, officially achieving Mega-Mind status in Silicon Valley. An undisclosed portion of that funding will go toward paying newly-minted president Doug Field whose EV CV includes Tesla, Apple, and, most recently, Ford.
The Association for Advancing Automation (A3) is North America’s largest automation trade association representing more than 1,500 organizations involved in robotics, artificial intelligence, machine vision & imaging, motion control & motors, and related automation technologies.
