Good morning from backstage at Pittsburgh’s David L. Lawrence Center. Unfortunately, my first Pittsburgh trip in several years is roughly 32 hours long. On the upside, any time I’m not sleeping will be spent here at Robotics & AI Demo Day (some pics here), including a morning panel with Seegrid CEO Chris Baker. The other upside is I expect to be back in the Steel City before the year lets out — perhaps for slightly longer next time out.
As I noted during our chat, this week marks one year for Baker in the chief executive gig. It felt gauche to mention that Automated also turns one, but I did bring plenty of show stickers with me. I’m going to leave any leftovers in my carry-on for my upcoming trip to the Bay in… two days.
Thanks to everyone who has RSVP'd for Tuesday’s Mountain View Happy Hour and live podcast recording with Rodney Brooks. We’re all amazed and grateful how quickly we hit capacity.
We’re adding names to the waitlist and will do our best to accommodate as many people as possible, but it seems like we’ll need to do another one of these soon in my old stomping grounds. The rest of the time I’m in the Bay next week, I’ll either be hosting a stage at A3’s AI conference or recording more in-person episodes of the pod. That time’s all booked up, too, but again, I anticipate being back in SF soon.
When we first pitched the podcast that became Automate, several folks asked us whether we’d be able to find new guests every week. It’s a funny thing to contemplate a year after launch, as we’ve very much had the opposite problem. Interest in robotics and AI is at an all-time high, and there are more voices than we can possibly cover right now (we’re trying our best).
Heck, we’re dropping five new episodes this week alone, and still have a considerable backlog to work through. I suppose these are what people talk about when they talk about “good problems.” This week, we’ve dropped interviews with Amanda Prorok of the University of Cambridge, Dhruv Batra of Yutori, Clara Vu of Magnendo Corp., and Carolina Parada of Google DeepMind (links to all below). We’ll end the five days of Automated celebration on Friday with a Sunday — specifically, CEO Tony Zhou. (Also, thanks Jake and MassRobotics. You're all mensches.)
Here’s a longer list of episodes that have already been recorded (to say nothing of those that are scheduled for recording): Ahmad Bahai (Texas Instruments), Karen Panetta (Tufts), Andie Zhang (ABB), Jagdeep Singh (Rhoda AI), Jan Zizka (Brightpick), Boris Sofman (Bedrock Robotics), Daniele Pucci (Generative Bionics), Inna Lobel (frog), Fady Saad (Cybernetix), Romain Moulin (Exotec), and Paolo Ardoino (Tether).
Plenty cooking with the newsletter, as well. This week, we’ve got features on Tether, Dynamic Creatures, a look at the new Digit, and a Q&A with Agility CEO Peggy Johnson. Plus a roundup of promising new Y Combinator companies from Becca and Liam's look at the wild world of robotic apple picking.
“Even in robotics, I think that there is there is a lot of potential from the financial side,” Paolo Ardoino says. “But I'm super excited, because I started coding when I was eight. I dreamed about this stuff for a long time. Now it's becoming reality. But we need to have our feet on the ground, understanding that we have to distinguish the practicality and the real timelines to put this into production, with the hype of the financial markets.”
Tether’s CEO has been making the interview rounds of late, promoting the USAT stablecoin among various other big money projects. I’ll be the first to admit I’m out of my depth in cryptoland. I never boarded that train, and for all his accomplishments, that subject matter alone doesn’t make the executive a prime candidate for an Automated interview.
Tether does, however, print money in more than one sense. The Stablecoin issuer has begun investing some of its massive war chest (including $10 billion in net profits in 2025 alone). The firm has built a markedly diverse portfolio, ranging from a brain-computer interface to an Argentinian dairy farm. It’s Ardoino and Tether’s interest in robotics that specifically piqued our interest.
While diversification is an overarching theme across its investments, Tether’s two big robotics bets occupy similar space. Both Neura (leading a round listed as “up to” $1.4 billion, currency exchange rates depending) and Generative Bionics (participating in are European firms that have focused significant time and money into building humanoids) received big backing from the digital currency firm. The partnership side is where things begin to diversify. Ardoino says that both companies are aiding in the build-out of two distinct Tether products. For Neura, it’s providing the robots themselves with a kind of currency.
“What we have is a product called WDK (Wallet Development Kit),” the executive explains. “That is a self-custodial wallet framework that can be embedded everywhere, from smartphones to laptops to servers to robot brains. That is really what we are experimenting on with Neura. That technology would allow the robots to hold inside their brain USDT as a stablecoin.”
Earlier this week, Agility unveiled the most comprehensive update to Digit since the biped emerged from an Oregon State lab eight years back [see below]. The system highlights the firm’s pragmatic approach to deployment and scaling, relative to much of the humanoid field. It also reconfirms its head start in terms of real-world use.
Digit 5 has been put through its paces behind the scenes, and is now ready for the spotlight, as Agility maintains plans to go public via a SPAC before year’s end.
The robotics firm has begun disclosing financials ahead of the merger’s close, setting the scene for a level of transparency heretofore unseen from an American humanoid firm. A recent filing revealed that the new robot is the subject of a massive $300 million customer deployment deal, which comprises the lion’s share of commitments.
It also pulled the curtain back on Digit 5’s $8,500-a-month RaaS licensing model, along with "a weighted adoption curve of approximately 800 [units] in 2027, approximately 7,000 in 2030, and up to approximately 25,000 in 2035."
Automated caught up with CEO Peggy Johnson to discuss the new robot and the state of the SPAC.
Safety is a core value-add here. How long do you anticipate it will take before Digit is working in close proximity to people?
The first half of next year, we'll have early deployments of Digit 5, and then by the end of next year, we'll have general availability. We're able to do that because of the new sensors, the new software, and everything that we've incorporated into the robot. It has 360-degree view, which allows for safe human detection so that we can avoid — or work around — humans that come in close proximity. If they become too close to the robot, Digit 5 will bring itself down to the ground safely, sit quietly until the human passes, and then proceed on its work. It’s quite a feat to be able to come outside of that physical safety barrier that we've been in for the last three years. We learned a lot being in there and have over 65,000 hours of operational deployment that really taught us a lot about what it takes to work in these environments and how it will encounter humans throughout the facilities. We learned quite a bit with Digit 4 deployments.
From ants using pheromone trails to find the shortest path to food to robot teams dividing complex work among specialized members, nature repeatedly shows that cooperation can produce capabilities no individual agent possesses alone.
So many robot demos are still little more than a physical wrapper around a language model. A robot may be able to hold a convincing conversation, but that does not prove it can navigate, manipulate objects, and understand physical space.
A prototype that works 80% of the time may represent only 2% of the work required to ship. Clara explains why edge cases, environmental variability, reliability, and exception handling consume nearly all the effort, and why a task that takes three to five years to commercialize should look like something a robotics graduate student could prototype in three to five weeks.
Why are general-purpose robots an even harder challenge than autonomous vehicles? Roads have lanes, traffic lights, and established rules. Robots operating around people must navigate unstructured environments, adapt to unfamiliar objects, and understand how to “read the room.”
Five generations in, Digit hasn't gained much in the way of actual digits — though modular end effectors could help soon enough. In the meantime, this latest version represents what is likely the biggest update for Agility's biped since it first diverged from Oregon State University's research robot, Cassie. What the system currently lacks in fingers, it makes up for in new safety features, better battery life, and, of course, those brand new knees.
Y Combinator Demo Day was always a big to-do back during my TechCrunch days. I was never that involved with the process, however, as the VC firm didn't seem especially interested in my areas of coverage. Now that I've been gone for a year and a half, however, it's robots, robots, robots. Thankfully, I'm working with Becca again, and she's pulled out her five favorites from this current batch of early-stage startups.
This week, we're talking apple AI. Not the "Hey Siri, play 'You Are the Sunshine of My Life' " variety. Liam's got a piece of lowercase "a" apple picking and the robots that want to get in on the act. Strange to think that in the Hudson Valley where I live, people actually pay money for the privilege of picking apples, as a kind of respite from the city. When it comes to harvesting in bulk, however, farmers are dealing with increasingly tight margins (convincing people to pay you money to perform manual labor only goes so far). Researchers at Cornell believe relief could soon arrive in robot form.
Universal Robots debuted its Gen 7 platform Monday at IMTS in Chicago. The lineup features three new arms — the UR10g-1750, UR17g-1300, and UR18g-950 — featuring new tool flanges outfitted with sensors designed specifically for AI models. The naming convention, meanwhile, isn't just internal code. The first number corresponds to payload in kilograms, while the second states the arm's reach in millimeters. The higher the former, the shorter the latter. The lineup also brings an upgraded control box with 40% more compute in a 30% smaller footprint, along with upgrades to the Polyscope operating system focused on both ISO 10218-1 safety and cybersecurity compliance. The OS also sports APIs for ROS2.
Facebook has three key applications these days: remembering friends' birthdays, learning your neighbors' unhinged political views, and marking the irreversible passage of time. Here's a happy memory, courtesy of friend and former colleague, Veanne: this week marks (Marcs?) 10 years since I was joined on stage by Marc Raibert and Spot. We did a number of firesides over the years, including one occasion when Marc announced Boston Dynamics' plans for a production version of Spot, marking the first time the legendary research firm had ever commercialized a product. (He was also one of our day-one Automated podcast guests.) Video clip in the link below, and a bonus appearance by my beard and pullover sweater in a slightly different hue. It's like those first couple of Simpsons seasons when the artists hadn't quite nailed the character design.
I had the pleasure of chatting with several promising startups at this week's Pittsburgh Robotics' Discovery Day. As expected, gens Z and Alpha were well represented, courtesy of both the city's strong higher ed system, along with all of the great K-12 STEM work from organizations like FIRST. One of the most interesting young companies, however, comes from a recent retiree. After 30 years at Westinghouse, CMU Robotics alum Lyman Petrosky figured it was time to found a startup. As the CEO of ForSight, Petrosky develops wearable technology for the blind, like the Pathfinder system pictured above. The system embeds proximity sensing developed for autonomous driving into a pair of gloves that alert the wearer of their surroundings. The founder tells me ForSight is in the early stages of scaling, with plans to open the product up to small batch pre-orders later this fall. Prices will run around $300 for those who want to use their smartphones for the glove's imaging and compute. You'll also be able to buy a standalone system for somewhere between $ 1,500 and $ 1,700.
I'm concerned we haven't shoved enough long-form video content down your throat this week, so here's another recently uploaded panel from this year's Humanoid Robot Forum. We've got McKinsey's Ani Kelkar, TRI's Erin McColl, Noble Machines' Elizabeth Samara Rubio, and Silicon Foundry's Rebecca Yeung discussing why deployment is so dang tough.
LG's spinning off AI. The Korean electronics giant is transforming its Edge Vision Analytics (EVA) division into SpectraBrain. The new company, which is expected to be fully realized before end of year, will be led by former LG Electronics exec Cho Bong-soo. The startup will commercialize technology that utilizes AI to scan CCTV footage for potential industrial hazards. LG has been dogfooding the tech internally, including its use with mobile patrolling robots.
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.
-3.png?upscale=true&width=1120&upscale=true&name=Untitled%20design%20(2)-3.png)
.avif)
