Nvidia’s leather jacket-clad CEO, Jensen Huang, seems to enjoy conducting recent keynotes flanked by an entourage of robots.
Part of this might be the futuristic vibes—a Star Wars-inspired droid is more fun on stage than the utilitarian chips that power it. But lately, these displays seem less about showmanship and more about staking a claim on an increasingly heated space.
Roboticists have been talking about a nearing inflection point. Certain advances in generative AI stand to change the way that general-purpose robots are trained and how they perform tasks. But risks also abound when applying foundation models to real-world scenarios, and truly general-purpose robots are still a ways off.
The feeling is evident in some recent investments from Big Tech companies. OpenAI began hiring new robotics staffers in January (it disbanded its previous team in 2021), and Meta was reported to be forming a humanoid robotics team in February. Google DeepMind demonstrated robots infused with its Gemini LLMs last month.
“The ChatGPT moment for general robotics is just around the corner,” Huang declared at CES this year.
Nvidia rolled out what it claims is the first open foundation model for generalized robot skills and capabilities at its GTC developer conference last month. At CES, it debuted a “world foundation model” called Cosmos, a system for using generated simulation to train robots.
A “crescendo”: Agility Robotics, whose humanoid robot, Digit, appeared at GTC last month, just raised $400 million in new funding, according to a report this week in The Information (the company declined to comment on the round).
Agility’s CTO, Pras Velagapudi, said a host of improvements are converging to make robots more viable: better actuation—the conversion of energy into motion—plus energy storage, on-board compute, “and a huge advancement in AI-based methodologies for controls and for behaviors.”
“What we’ve seen is really this confluence of technological advancement that’s really kind of hit a crescendo right now,” Velagapudi told Tech Brew. “It has all acted as an accelerator for our part of the industry. What we can do now with the software tools that are available on the hardware that we can now put together within the past few years has greatly accelerated the speed at which we’re able to unlock new capabilities and skills.”
Bill Ray, distinguished VP analyst and chief of research at Gartner, said the market research firm expects robotics adoption in industries like light manufacturing to accelerate within the next two to three years.
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
Ray said one challenge right now is the disruption cost caused by reorganizing factories around robots and the headaches if a key robot breaks down. Polyfunctional robots—automata that can perform a range of tasks like a human does—can help to solve this last problem, he said.
“We are on the edge,” Ray said. “The challenges have largely been addressed…The problem is the disruption of deployment…We’re very close to that. And then it’s going to ramp up very, very quickly.”
However, robots that use GenAI to model tasks are still very nascent, Ray said. “We’ve got a couple of academic papers on this topic, and we’ve seen one demonstration from Agility…That’s where we’re at,” he said. “The problem, of course, is the hallucination. You cannot risk the robot just getting it wrong.”
Humanoid debate: Polyfunctional robots are not to be confused with humanoid-shaped robots, which Gartner is decidedly against as a form factor.
“We think the primary purpose of a humanoid robot is to artificially inflate your share price,” Ray said. “The problem with humanoids is that humans are not very well-designed. Our knees bend the wrong way. We have too many teeth. Our eyes are too far away from the things we want to hold.”
Velagapudi said the humanoid designation of its robot, Digit, is not meant to be so much about shape as it is Digit’s ability to act as a human in a workspace.
“Digit is built not to be humanoid in the sense of replicating a human body, but to be human-centric in the sense of being able to interact with spaces that were built for humans, to be able to reach the same things that a human could reach and maneuver around the same areas that are spaced out for what a human would find comfortable,” Velagapudi said.
Edward Mehr, co-founder and CEO of manufacturing-focused robotics firm Machina Labs, said that while humanoid robots can serve purposes and be psychologically exciting, most early versatile robots at work in factories will not be human-shaped. Humanoids will come later in different settings like homes and offices, he said.
“As exciting as humanoid is, in early adoptions, we will see intelligent robots, robots that are as smart as maybe humans and as a craftsman, but then they have a form factor that is useful in an environment like manufacturing,” Mehr said.