Welcome, early adopter, to the first issue of Automated. You just gained a lifetime’s bragging rights for every room you enter. You also hold in your hand something I’ve been waiting to launch since my first day on the job. In fact, the Automated newsletter and its sister podcast of the same name were my two big pitches when I first met with most of the A3 management team back in March.
Naming is always the hardest bit, but something I felt strongly about from the outset was maintaining a unified brand. Automated the newsletter and Automated the podcast are symbiotic entities, in that the interviews from the latter will inform many feature pieces in the former.
The podcast gives us a direct line into automation’s leading voices. Just look at the first four episodes: Marc Raibert, Deepu Talla, Melonee Wise, and Rodney Brooks (early peek here, full interview coming next week!). We have another dozen or so already recorded, and I’m pacing back and forth in my office to stop myself from just listing them off here (I’m pretty sure producer Jana would fly to New York to murder me).
Each issue of the newsletter, meanwhile, will open on a feature, often informed by some element of this conversation. It’s a chance to pull on a thread and explore a larger trend relevant to robotics, AI, or automation. It could be technical, research, cultural, political, economic, or even historical. Turns out there are a lot of weeks in a year, so I’ll need to write a lot of these, some informed by the podcast, some by events, some by the week’s news.
Automated the newsletter also marks the return of my job postings. I suppose that never really died, but can existing only on LinkedIn really be called living? As I’ve said many times, connecting people with work has long been one of the most rewarding parts of this job. I've worked in publishing for most of my adult life and know first-hand how brutal job hunting almost always is.
We’ve made some tweaks to the section, which will hopefully improve the discovery process. The biggest is the fact that the section is now [deep breath] weekly. This, in turn, means, that not every company is necessarily going to make it in every week. Instead, we’ll be cycling through to keep things fresh.
That note brings me to the other big change. We’ll also be highlighting companies at the top of the section, by pulling out a few select jobs and including additional information, like location.
Lastly, and arguably most importantly, is the news piece. Without it, this would just be a letter. Here’s where we round up the big and most interesting bits from the week, including new developments, research, macro trends, and everything else I think you should know from the worlds of robotics, AI, and automation.
I had an hour-long call with Deepu Talla toward the end of last month. It was the second time I’d spoken with the NVIDIA exec for any length of time. We’d met a few years back at the chip giant’s South Bay headquarters, as he walked me through the Jetson ecosystem. The August conversation was cleaned up a touch and edited into the second episode of this newsletter’s eponymous podcast.
We got a bit more into our respective backgrounds — it was a podcast, I couldn’t help myself. Discussing his work history, I was struck by a parallel between our careers I’d not previously noticed. Before we were robot guys, we were phone guys. I reviewed them and Talla helped build them.
His transition was a quick one from the sound of it. NVIDIA hired him for its Tegra team in February 2013. By April, he was the company’s vice president and general manager Robotics and Edge AI. Talla joined the team as founder Jensen Huang was in the throes of an early pivot from mobile into robotics.
“We have a saying inside NVIDIA, ‘if you’re not at least 10 years before a market or a technology takes shape, you’re probably late,’” Talla told me about the then-risky move. “I think it was perfect timing for us to jump into physical AI and autonomous robotics applications.” Continue Reading >
Ali Agha describes it as FieldAI’s “aha moment.” Still, it would take another two years before the company nailed down the architecture for its FFMs (field foundational model). The CEO is quick to distance the term from the similarly initialed LLMs (large language models), which much of the competition is attempting to backward engineer into automation.
“It’s a very physics-first mentality,” he says of FieldAI’s models. “We don’t take a gigantic neural network and pass a lot of data and hope for success as we’ve done in ChatGPT. We don’t rely on one pure end-to-end model to pass raw data from one side — like LiDAR and raw images — and get the motor velocity, actions, and torque, on the other end.”
The approach has garnered interest from investors and customers, alike. On the former, FieldAI came out of semi-stealth just under a month ago by announcing it had raised $405 million over a quiet couple of rounds. Notable supporters include Bezos Expeditions, BHP Ventures, Canaan Partners, Emerson Collective, Intel Capital, Khosla Ventures, NVentures (NVIDIA), the Gates Frontier, and Samsung.
Being able to meaningfully reference embodied AI in the first sentence of your pitch deck is something akin to a VC blank check these days, but the OC-based startup has something even more important when it comes to sustained funding: deployments.
“These are massive customers,” Agha says, maintaining client confidentiality. “These are among the top three or five construction companies in Japan and the U.S., each of which can have hundreds of sites and deploy thousands of robots – potentially. Similarly, we’re with the largest energy providers in the world.”
We capped off last week by hosting Intrinsic CSO and the most affable guy in machine learning, Torsten Kroeger, on Automate Live. We covered a lot of ground, but the bit that stuck with me into the weekend was about a kind of bifurcation in embodied AI between specialized and generalized models for robotics systems. Alphabet/Intrinsic generally sides with the former, but there's a lot of fascinating discussion happening in labs across the world. "Right now we have two camps in the robotics community," Kroeger told me. "One thinks they’ll have one model, that works for any embodiment, any task, and all they need to do to throw more data at the model, and at some point it's going to work. And that might very well be right. There's a different part of the community that, thinks they're going to have multiple models that are either embodiment-specific, function-specific, or task specific."
Figure Tuesday confirmed February reports that its latest fundraising round valued the humanoid maker at $39 billion. The South Bay firm says it has more than $1 billion in “committed capital” for its Series C raise. I used to think the company’s name was a reference to the human figure – but I'm increasingly of the mind that it’s about that 11-figure valuation. After dissolving a partnership with OpenAI (which is reportedly working on its own humanoid hardware), Figure moved much of its embodied AI work in house under the Helix banner. Some of the funding will go to that work, and much will be devoted to scaling up robot manufacturing.
AWS confirmed that it has sunset RoboMaker, seven years after launching the Gazebo-based simulation service. The enterprise giant notes in a statement, “When AWS decides to retire a service or feature, it is typically because its capabilities are better addressed by newer AWS solutions or offerings from our AWS Partner Network partners that better meet customer needs.” In keeping with that, AWS is offering guidance for users looking to transition to its Batch service. The Robot Report, which was first to spot the news, suggests RoboMaker’s fortunes were ultimately too dependent on its ties to a single company, Roomba-maker, iRobot.
ANSI and A3 last week published the first major revision to the R15.06 robot industrial safety standard in nearly 15 years. One of the more interesting tidbits I learned while researching the topic is a lot of folks in that world never particularly loved the term "cobot." In fact, it’s on the outs in this version of text. Todd Dickey, chairman of subcommittee behind the U.S. version of the paper, told me, “In actuality there is no such thing as a ‘cobot,’ rather there are robots that utilize collaborative technologies.” The paper instead refers to “collaborative applications,” and not "coapps," a truly horrible portmanteaux I just made up.
Dyna Robotics is the latest embodied AI firm to line up some of Silicon Valley’s deepest pockets. The Bay Area-based firm this week announced a $120 million Series A, a mere six months after raising a $23.5 million seed. I’ve launched a podcast and a newsletter in that same timeframe. We are not the same. Investors include Salesforce Ventures, NVentures (NVIDIA), the Amazon Industrial Innovation Fund, Samsung Next, and LG Technology Ventures. The company claims to have released a foundation model with a success rate north of 99%, with robots, “running sixteen hours a day at hotels, restaurants, laundromats, and gyms.” I’ll be chatting with the founders soon, so look for something a bit more substantial in next week's newsletter.
You may have seen Grace Brown in our recently released Automated podcast trailer. We spoke a couple of months back after her appearance at the Humanoid Summit in June. A lot impressed me about the Melbourne-based startup, including the amount of progress they’d already made in delivering their 3D printed humanoid robot, Abi, to Australian assisted living facilities. Investors have seen a lot to like in Andromeda, as well, it seems, as the company just announced a $23 million AU ($15 million USD) Series A, featuring Forerunner Ventures, Rethink Impact, and Main Sequence Ventures. Much of the skepticism around these kinds of systems centers around fears that they might further alienate people in care facilites. As Brown noted in our interview, “If Abi wasn’t there, [the resident] would be in his room, by himself, with the same four walls, not interacting with anyone, most likely not doing anything.”
RoboBall is back, and it keeps getting bigger. More than 20 years after NASA’s then-director of the Robotics and Automation Design Lab (RAD Lab), Robert Ambrose, scrapped his idea for a big robot ball, he’s picked up where he left off, as a professor at Texas A&M. Not only has RoboBall returned – it’s already spawned another sequel. There’s the two-foot-diameter RoboBall II and the six-foot-diameter RoboBall III. The second is mostly designed for testing and data collection purposes, while the team hopes a version of the third may some day carry payloads for real-life missions.
Google’s notoriously secret X Labs have opened up the miniblinds and let a bit more sunshine through in recent years, including a new podcast from “Captain of Moonshots,” Astro Teller. Over on its blog, the division recently spotlighted Materra, a project hoping to tackle the great floating garbage patch that is humankind’s plastic problem. The stuff is everywhere, including our brains, which may explain why we’ve had so much trouble efficiently recycling it. Aided by robotics and AI, Materra is taking a very Google approach to the subject, by collecting as much data as possible. “They started with their own household garbage, using spectrometers and analytical chemistry to reveal the molecular composition of each package,” Google notes. “They then trained machine learning algorithms to properly identify the molecular makeup of every piece of plastic packaging.”
Marc Raibert (RAI Institute) - The Boston Dynamics founder talks physical AI, general purpose robotics, and guitar tube amps.
Deepu Talla (NVIDIA)- Just over a decade ago, NVIDIA got out of mobile processing and into robotics. The move is paying off in the form of the chipmaker’s Jetson platform.
Melonee Wise (Fetch Founder)- When we recorded this, Wise was Agility's CPO. Who can say where the Robot Ninja will strike next?
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, and motors and related automation technologies.
