That sets the bar pretty high for Boston Dynamics AI in terms of expectations.
Supported by the South Korean carmaker Hyundai, the nonprofit research center is on a mission to push the frontiers of robotics and artificial intelligence. Since the institute was announced in August 2022, Raibert has hired 150 people and is on the way to hiring 200 more.
Raibert said the goal of the institute “is to work on the really important and hard problems, not the incremental problems, like making the reliability of a robot a few percent better, or making it cost less.” His ambitions for the institute run along the lines of the legendary New Jersey research center that gave birth to the transistor, the laser, and the solar cell.
“We want to be the Bell Labs of robotics and artificial intelligence,” he said, “a place really making the future happen, with a lot of talented people working together.”
Despite the name, the institute is separate from Boston Dynamics, the Waltham company that Raibert started in 1992 and Hyundai acquired in 2020. That company is now largely focused on selling merch-moving robots to warehouses. Raibert is still chairman but handed over the chief executive role in 2020.
On a recent visit to the institute’s offices, a free lunch was laid out in the kitchen, and clusters of employees sat chatting. Nearby were some of the gangly-looking aluminum walking robots that Raibert built as an MIT and Carnegie Mellon professor, dating back to the early 1980s.
Raibert was wearing one of his trademark Hawaiian shirts, a charcoal fleece jacket, and a pair of well-worn jeans as he led a tour of the institute, which works with a coterie of different robots — not just those made by Boston Dynamics. But in one area, there’s a bright yellow Boston Dynamics ‘Spot” robot with a long arm that can snap a picture of a visitor, pick up a dry-erase marker, and then draw its own version of the visage. In another open workshop area, employees have purchased a few kid’s bicycles, and are trying to program robots to be able to repair them.
The institute is focusing on several broad areas of research, Raibert said, one of which he calls “watch, understand, do.” Robots should be able to watch a person or another robot performing a task, understand what is happening, and then do the task itself. That eliminates the need to program robots for each new task.
A second area is making robots more dexterous. People have tried to get robots to pick up and handle items for 50 years, Raibert said, “and still they’re nowhere near approaching how human beings or animals can manipulate things.”
A third is what he calls “physical AI” — combining new artificial intelligence software with robots to help them learn about their environments, move in more agile ways, and respond to changes happening around them.
Raibert built robots at Boston Dynamics that could dance the mashed potato, open doors, and do backflips. What is new for him here are AI models that can give added levels of intelligence and awareness to machines — helping them better understand the where they are and what they’re handling, rather than just trying to avoid bumping into things, as most of today’s bots do.
These new AI models would help the robot not only use its camera to identify a red apple sitting on a counter but also understand how to pick one up and put it into a bowl without damaging it.
“If you’re going to handle one,” Raibert says, “you might want to know what it weighs, what is the friction on the surface, and how much pressure would crush it. You’re connecting the thing — an apple — to all those characteristics.”
The new technology developed at the institute will benefit Hyundai, which has divisions working on self-driving cars, “ultra-mobility vehicles” that can travel through terrain that would stop today’s vehicles, and new kinds of electric aircraft. But Raibert wants to be clear the institute is not a product development shop for its South Korean sugar daddy.
“We’re rabidly noncommercial here,” he said. “We’re trying to focus on long-term research.”
Raibert has spent much of the past year focused on recruiting for the institute. Among the people he has hired are chief technology officer Al Rizzi, who came over from Boston Dynamics; Lael Odhner, a cofounder of the Charlestown robotics startup RightHand Robotics; and Kate Darling, who joined from the MIT Media Lab to focus on the ethics and societal impact of AI and robotics.
“He’s hiring unbelievable talent, and they’re growing like crazy,” said Tom Ryden, executive director of Mass Robotics, a Boston nonprofit that supports the sector. “I think it’s going to be great for the ecosystem.”
The institute’s Cambridge office has 11 visiting professors who spend time there working on projects. It also funds research at 10 different universities, including MIT’s Computer Science and Artificial Intelligence Lab, just a few blocks from its office.
The institute also is establishing a European branch, which will open in Switzerland this year and be led by Marco Hutter, a professor at the public university ETH Zurich. Several jobs are posted for roles developing new AI software and some require applicants to have published at least one research paper in a top-tier scientific journal.
Boston Dynamics built an international reputation in part through its series of humorous, impressive, and sometimes alarming YouTube videos. But one thing the Boston Dynamics AI Institute hasn’t gotten around to yet is setting up a YouTube channel.
“We don’t have anything yet that’s completely worthy of a video,” Raibert says. “But I hope soon.”
Scott Kirsner can be reached at [email protected]. Follow him @ScottKirsner.
Eugen Boglaru is an AI aficionado covering the fascinating and rapidly advancing field of Artificial Intelligence. From machine learning breakthroughs to ethical considerations, Eugen provides readers with a deep dive into the world of AI, demystifying complex concepts and exploring the transformative impact of intelligent technologies.
