Earlier this week, someone uttered the words “conference season,” and my brain left my body for a minute. It was a nice — though brief — vacation. Anyway, it’s more or less back where it’s supposed to be, because I’ve got a newsletter to write. I also have several conferences to plan for — starting with Automate in a couple of weeks.
It’s my newsletter, and I’m the managing editor, so I’m kicking off this week by plugging the panels I’m moderating there and at the attached Humanoid Robot Forum. Come, watch. I’ll have stickers.
Tues 6.23
11:45 AM - 12:45 PM (CDT) - Automate Show Theater - Booth 2884
Navigating Your Digital Transformation Feat: Annemarie Große Frie (Siemens AG), Zain Gulamali (Microsoft), Ricky Watts (Intel)
2:00 PM - 2:30 PM (CDT) - Grand Ballroom S100
Developing Safety Standards for Humanoid Robots Feat: Carole Franklin (A3), Kevin Reese (Agility Robotics), Federico Vicentini (Boston Dynamics)
4:00 PM - 5:00 PM (CDT) - Grand Ballroom S100
Hands, Manipulation, and the State of Physical AI Feat: Aadeel Akhtar (Psyonic), Bren Pierce (Kinisi), James Wells (Sanctuary AI)
Weds 6.24
3:00 PM - 4:00 PM (CDT) Grand Ballroom S100
Scaling, Delivering, and Deploying: Humanoids in the Real World Feat: Ani Kelkar, (McKinsey), Erin McColl (Toyota Research Institute), Elizabeth Samara Rubio (Noble Machines), Rebecca Yeung (Strategic Advisor)
Thurs 6.25
10:30 AM - 11:15 AM (CDT) Automate Show Theater - Booth 2884
Automated Podcast Live Feat Ali Agha (FieldAI)
And then I’m flying home for long enough to do a load of laundry before flying out to Europe for the inaugural Davos Tech Summit, where I’ll be hosting the tech stage on July 2-3. Full agenda here.
Meantime, we’ve got some great newsletter interviews lined up this week. Texas Instruments CTO Ahmad Bahai kicks us off with a conversation about futureproofing semiconductor production, before we check in with Andromeda CEO Grace Brown about her move to San Francisco. Over on the pod, Daniel Rausch, Amazon’s vice president of Alexa and Echo, joins us to talk about the company’s smarter smart assistant.
Our contributors have some great stuff this week. Liam spoke with one of the researchers behind the 20-legged Argus robot, and has an interesting piece on a project to improve first-responder drone navigation. Rebecca chatted up one of the cofounders of robot forklift firm Cavalla and has our weekly investment roundup, with quotes from some of the top players.
In publishing, it’s known as “killing your darlings.” It’s been alternately ascribed to the usual suspects for quotations of uncertain origin — Faulkner, Wilde, Chekov, Eudora Welty.
It’s universal advice about avoiding myopia that applies across disciplines. Texas Instruments CTO, Ahmad Bahai, puts it well toward the end of our upcoming podcast interview, noting, “The worst thing [that] can happen for a researcher is to fall so much in love with their own ideas, they ignore the reality of what’s going on around them, and say, 'I've worked on this for five years. I have to make it work.' ”
While the sentiment is transferable, the consequences can be starkly different. In the case of the sorts of technologies Bahai and I discuss, it’s the question of paying attention to one’s surroundings. It’s about reading the research, following the market, and appreciating the sorts of breakthroughs that shatter orthodoxies. Sunk cost is a powerful influence on a personal level, and anyone who has operated in a corporate environment can tell you that even the most progressive-seeming businesses are not especially nimble.
Here’s another overtired, but largely accurate cliché for you: hardware is hard. Maintaining a successful technology company requires the ability to read the writing on the wall. This requires a much longer horizon when dealing with atoms.
“Even when you wake up to the fact that AI is going to be big, a new device takes five, six years to develop. It's not like tomorrow,” says Bahai. “If you just wait for the next market to start something, then you're always behind the curve and not having like opportunity to capture these growth opportunities. That's why I think we need to look at these trends and these exponentials from a foundational technology viewpoint and then react to the markets.”
The last time we spoke, Grace Brown was exhausted. You wouldn’t know it from the clips. Managing a 12-hour time difference and running off fumes from a fresh $23 million AUD ($16M USD) raise, Andromeda’s founder brought her A game to one of the Automated podcast’s inaugural episodes.
“I just remember being so exhausted,” she admits with a laugh. Brown is still navigating the endless parade of challenges facing an early-stage robotics founder, though a move halfway around the world has brought us significantly closer together in time zones. Andromeda’s September Series A fueled Brown’s movement from Melbourne to San Francisco, effectively taking much of the company’s operations along in the process.
“We've got a team of just under 10 people [in California], and will be hiring more throughout the year,” says Brown. “We're thinking a lot about how to do manufacturing in this next phase. There are a lot of questions there that are being discussed actively. My head of machine learning has moved over. We have engineers who kind of like rotate through a lot. My COO is here in San Francisco. My VP of product is here in San Francisco.”
Prior to the move, Andromeda made strong inroads in Melbourne’s senior living community with the early version of its Abi robot. All told, the company has “dozens” of the colorful humanoid deployed in assisted living facilities, the majority of which are in Australia, where the company maintains a support staff. Abi is also being manufactured Down Under, though that’s one of many aspects the startup is reassessing as it looks to scale production.
Brown and Andromeda’s move to the U.S. is about more than just capital and access to Silicon Valley talent (though both strongly factor in). Living facilities in the States also represent a huge opportunity for Abi.
For years, voice assistants were built around rules, scripted responses, and carefully designed commands. But with the rise of large language models and generative AI, Amazon had to rethink what Alexa could be and how people might use it. Watch on YouTube>
Daniela Rus (MIT CSAIL)- This episode has it all: physical AI, sperm whale alphabets, and some cutting edge insight into autonomy. It's also recorded in front of a live audience at MassRobotics. You're not going to want to miss it.
When I first started touring university labs, I mostly encountered robots designed for different spaces. Over the years, it’s been common to purchase systems like industrial arms, or Rethink Baxter robots. Just the other month, I caught a glimpse of my first wild Pepper in some time. Manufacturers have mostly written off the R&D category as a money loser. That’s not to say the category has been wholly devoid of admirable attempts. Willow Garage famously gave us PR2, which I imagine continues to darken many a laboratory door (though the system’s true legacy may well be ROS’s continuing ubiquity). Over the last few years, another company has quietly begun dominating research spaces. It’s fair, certainly, to suggest that Unitree’s current commercially available systems aren’t cut out for blue-collar work, but their low cost makes them ideal candidates for research facilities. It’s the kind of thing you can buy for a lab without having to name an entire wing of the building after a PayPal cofounder.
Those are a few of the reasons this bit of NVIDIA news is a no-brainer. This week at Computex in Taipei, the chip giant announced a new reference humanoid designed specifically for research purposes. The system is built on top of Unitree’s H2 Plus system, powered by NVIDIA’s Isaac GR00T software and sporting a pair of Sharpa’s five-fingered hand. Ai2, ETH Zurich, Stanford Robotics Center, and UC San Diego’s Advanced Robotics and Controls Laboratory are launch partners for the design.
The notion of research robotics as a kind of financial dead end is no doubt evolving in the current climate. Even when the playing field was significantly smaller, providing hardware and software solutions to research facilities has always been an excellent way to get roboticists invested in your systems at the outset of their careers. We’ve seen a number of open source projects pop up in recent years with limited success. Fauna’s first robot, Sprout, was also positioned as a research system, though the future of that product is unclear post-Amazon acquisition. A representative from RoboStore reached out to tell me that it will be selling the H2 Plus humanoid in the U.S.
That’s a direct quote from Boyuan Chen, the lead researcher Liam interviewed for this piece about the wild new Argus robot. And, indeed, Mr. Chen, I agree. Of course, the quote doesn’t end there. Here’s the full thing: “Twenty legs sound like a lot until you look at the design: every leg is identical, same actuator, same camera, same cable-driven drum. There is one part to mass-produce and one part to maintain. That is mechanically simpler, not harder, than designing four highly specialized limbs.” Argus is the product of the deceptively simple notion of moving beyond humanoids and quadrupeds to find the “optimal” robot body. The study uses “dynamic isotropy,” which prioritizes symmetry above all, allowing the system a broad freedom of movement. In the case of the Argus, that means omnidirectional movement, climbing trees, and scaling walls. Twenty legs sounds like a lot, until a robot straight out of The Prisoner starts chasing you up a tree.
You know what they say about the weather in San Francisco: wait 20 seconds, and there will be another Waymo to take you away from it. The last time I was in my beloved city by the Bay, it contained Waymultitudes. I was out in front of an event at the San Francisco-Marin Food Bank, playing count the self-driving cars, only to lose track in the high double digits. The good news is, next time I’m back home, there should be more variety, which is to say, a different kind of Waymo car cruising the hills alongside the familiar, white self-driving Jaguar.
The Google-owned autonomous vehicle division this week announced that it has officially added Ojai to its fleet, after finishing employee-only pilots with the machine learning minivan — dogfooding drives, if you will. Sharing a name with a Southern California valley that Wikipedia tells me derived its name from the indigenous Ventureño Chumash word for “moon,” the cars are manufactured by five-year-old Chinese startup Zeekr. They’re also the first to showcase sixth-generation Waymo Driver technology. “This deep understanding of real-world requirements is why the Waymo Driver utilizes a custom, multi-modal sensing suite where high-resolution cameras, advanced imaging radar, and lidar work as a unified system,” Waymo writes. “Using these diverse inputs, the Waymo Driver can confidently navigate the ‘long tail’ of one-in-a-million events we regularly encounter when driving millions of miles a week, leaving nothing to the imagination of a single lens.
To hear Victory Boyd discuss it, all roads led back to forklifts. Cavalla’s cofounder tells Rebecca that the startup looked for other problems to address in warehouse automation, but ultimately kept coming back to everyone’s favorite bit of non-standardized industrial equipment. “We really tried to find something new, but it all came back to autonomous forklifts,” Boyd said. “This was very surprising for us because when we were first looking, it looked like it was solved. We saw so many things in the market when we looked up autonomous forklifts. It was very surprising to continuously hear that none of them work. So it was basically a journey of finding out why they don't work.” Build a better forklift and the world will beat a path to your door — or at the very least, get you $250,000 in funding from a Thiel Fellowship.
Liam’s got some new research out of MIT and the University of Pennsylvania aimed at making first-responder drones faster and more efficient while navigating natural disasters. The system is open-source, meaning it's far more accessible and open to user contribution. The system is currently targeted at managing uncertain environments like the wake of a major earthquake, but future applications could apply to more commercial endeavors like last-mile delivery and wind turbine inspection.
Data collection has become one of the more fascinating frontiers of the physical AI conversation of late. For one thing, there’s no precise consensus on the relative efficacy of different sources. There are those who will tell you that simulation has become a far more effective tool than many suggest, while others suggest that video can do a majority of the heavy lifting. Ultimately, however, nothing compares to real-world data collection. As for the best way to go about collecting — that’s another hot topic. We’re seeing massive robot warehouses and companies paying users to collect their own hard work.
Someone at Shift had the bright idea of offering free apartment cleaning from a human wearing the 2026 version of Homer’s cowboy hat. Rather than blowing the lid off expired deli meat, however, Shift’s system offers something immensely valuable for robot training: first-person video of humans performing useful tasks in the wildly unstructured environment that is the home. “That data is valuable enough for us to offer cleaning services free of charge for a limited time,” the company writes. “So, for a limited time, we cover the cost of professional cleaners. In exchange, they record first-person video while they work. You get a spotless apartment. We get training data. Everyone wins.”
The “free” cleaning is currently open to NYC residents. Shift is a product of MicroAgi, a Munich-based physical AI firm founded in late 2025 that also has a large presence in Zurich. Here’s what Shift’s site has to say on the privacy front, “Yes, your privacy is fully protected. Names, faces, or other personal information is automatically anonymized, with any sensitive details blurred before it's ever used. The headset is designed to capture a first-person view focused on the cleaner's hands and the task being performed. We blur all personally identifiable information from screens and ID cards, to pieces of paper and cell phones to help protect both you and your home.”
Speaking of clean, Postmates by way of Uber Eats spinout, Serve Robotics is getting into the fluff and fold biz, through a partnership with NoScrubs, a “fast-growing on-demand laundry service” that may or may not be able to get love from me. This is Serve’s first last-mile offering beyond food, and “an early step toward broader expansion into additional verticals, including dry cleaning, retail, pharmacy, grocery, each of which shares the same last-mile economics that have made sidewalk robots viable for food." Serve currently manages around 2,000 delivery robots in the U.S. Roughly one-quarter of those are in Los Angeles, where NoScrubs will be hanging on the passenger side of your best friend’s ride, trying to holler at you.
I’m not sure I agree with this sentiment from this week’s Luma announcement: “The default solution is to simply scale teleoperation to capture data across every possible task and combination of tasks — a physically and economically impossible approach.” Still, as noted in the Shift piece, the sheer scale of data required to scale “general purpose” robots remains a massive, unsolved challenge. I can certainly get on board with this bit, “physical AI and robotics are in the pre-AI-scale era.” And here’s an even more intriguing bit, “Generalization is not unique to text” (you can tell because they italicized it). The AI startup says its generative AI work in mediums like video, images, and 3D has offered insight into how multimodal AI models can be used to extract key information around perception, reasoning, and physics.
“[T]his lab will be an open science effort,” the company writes. “We will use our scale and expertise to build the substrate and make it available so that any dedicated group of people can use it, modify it, and assemble it into systems of productive work. To accelerate progress, we will collaborate with our peer labs and academia on core research, build evaluations to measure progress and assess safety, and partner with industry leaders on chips, hardware, and physical agent systems.”
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.
.jpg?upscale=true&width=1116&upscale=true&name=cavalla%20800%20(1).jpg)
.jpg?upscale=true&width=1116&upscale=true&name=mit%20routes%20800%20(1).jpg)
![noscrubs_04942copy_720[24] - Edited](https://a3content.automate.org/hs-fs/hubfs/noscrubs_04942copy_720%5B24%5D%20-%20Edited.jpg?upscale=true&width=1116&upscale=true&name=noscrubs_04942copy_720%5B24%5D%20-%20Edited.jpg)
