When I recently connected with Owen Kent over email to discuss his role as co-founder and chief marketing officer at ATDev and working with Meta and artificial intelligence, he distilled his work into an eminently cogent argument.
“My phone cost $800. My wheelchair cost $40,000. The phone got smarter every year. The wheelchair didn’t,” he told me. “That isn’t a technology gap. It’s a design philosophy gap, and we started ATDev to close it.”
For ATDev, who describes its mission as “[creating] innovative, assistive technology that empowers people to live independently, confidently, and without barriers,” building assistive technologies is done with the disability community instead of for them. Indeed, Kent himself is of the community, as he’s coped with spinal muscular atrophy most of his life. He’s also a user of a power wheelchair, and runs his business via voice and lip joystick. A native Coloradan, Kent studied film at UC Berkeley and met his roommate and future ATDev co-founder there in Todd Roberts.
One of ATDev’s two products is called RAMMP, or the Robotic Assistive Mobility and Manipulation Platform. The open-source platform is made possible by a $41 million award from ARPA-H, or the Advanced Research Projects Agency for Health, a subsidiary of the Department of Health and Human Services. ATDev describes RAMMP on its website as aiming to “empower people with mobility issues to live more independently by combining advanced robotics, artificial intelligence, and smart mobility” while adding the program “is designed to enable greater autonomy, helping users perform daily tasks safely and independently.” Kent explained RAMMP is overseen by the University of Pittsburgh from Dr. Rory Cooper’s Human Engineering Research Laboratories, alongside Carnegie Mellon, Cornell, Northeastern, Purdue, Kinova Robotics and LUCI Mobility. ATDev serves as the industry partner, with Kent saying the company’s job is to turn research into practicum by “[taking] the research and [making] it work in real devices, in real homes.”
In late July, Meta published a blog post in which the company shares how its AI models are assisting the aforementioned University of Pittsburgh to “transform assistive robotics.” The piece is a rather technical read, but the crux is simple: AI can help engender independence amongst users. Meta cites a Department of Transportation report showing more than 100,000 wheelchair-related injuries are treated in U.S. emergency rooms each year, often due to trips and falls. “As a result, the need for smarter, safer, and more adaptable technology is urgent,” Meta wrote.
When asked how the collaboration with Meta came together, Kent explained it happened “the right way round” through research. An electric wheelchair, he went on to say, “has to be able to see” things like a doorway, a curb, a person, etc. He added the perception team at Pitt built that aspect of RAMMP using Meta’s open-source vision models in DINOv3 and SAM, which inspired Meta’s to write and publish its aforementioned blog post.
So how does RAMMP work?
“Think platform, not product,” Kent said. “RAMMP has a lighter mobility base—the target is roughly a third of the weight of today’s 450-pound complex-rehab chairs—a seat that moves with you, sensors all the way around that build a picture of the room and understand what’s in it, and a Kinova robotic arm for reaching and grabbing. The software ties it together, and the core design choice is what we call a spectrum of autonomy. You decide, task by task, how much the chair does. Drive it yourself, or hand it more of the work. The person stays in charge."
For Meta’s part, Kent said the reason the collaboration works lies in openness. RAMMP is an open-source project whilst Meta releases its models openly to the public. The net effect, he told me, is university labs (such as those at Pitt) have the ability to “put state-of-the-art vision on a wheelchair without a licensing wall.” Better still, the models run locally on the wheelchair itself and require no active network connection. “For a device people sit in for 16 hours a day, that matters,” Kent said. “Your chair can’t stop seeing because you drove into an elevator.”
I’ve long been a strong stumper of AI’s potential to be an assistive technology to do good for people with disabilities, from writing tools to coding to research and captioning and much more. For his part, Kent feels similarly, telling me he believes AI-as-accessibility is “undervalued” overall industry-wide. The primary focus, he said, is chasing humanoid robots that someday will walk into your kitchen and cook dinner or something. But there already are millions of people sitting in robots right now by way of their electric wheelchairs—and it clearly deserves AI goodness too.
"It’s a robot,” Kent said. “It’s just an underpowered, under-sensored, unintelligent one. The most useful robot in the world may be the one that’s already in the room, and it’s getting a fraction of the attention.”
He added: “The industry measures AI productivity in percentages. For disabled people, it isn’t a percentage. It’s the difference between doing the work and not doing it.”
When asked how RAMMP has changed his life, Kent noted his everyday chair remains a typical power chair. RAMMP is in the middle of a five-year program. He did, however, share an anecdote of an April Phase 1 demo in which the RAMMP chair found the accessibility button and opened a door on its own. Then the chair picked up a cup and got coffee before bringing the cup up for the user to drink. Then the chair went down a curb and back up while staying level. At the end, the user, whom Kent identified as DJ, said “one more drink, please.” The chair obliged and gave DJ one more sip.
“That second sip is the part I keep thinking about,” Kent said. “Anyone can stage a robot doing something once. Doing it again because the person asked is what reliability looks like, and reliability is the whole game. A cup of coffee sounds small. It isn’t small when getting one has always meant waiting for someone else. I watched that end, on a wheelchair, in a real room.”
As to feedback on RAMMP, Kent said it has been strong. It speaks volumes that Meta has written about the project, as well as that ARPA-H director Dr. Alicia Jackson has endorsed the RAMMP program publicly. More than titanic tech companies or government officials, Kent believes the most salient feedback comes from actual wheelchair users. In the spring, ATDev surveyed 175 people who use power chairs and found 87% of respondents said a RAMMP-equipped chair “would be useful to them.” Another 87% reported they would use a RAMMP chair were it available today.
Do people really want this robotic wheelchair technology, I asked.
“The better question is why nobody has built it,” Kent said. “People have wanted it for decades. Most of us just stopped expecting it.”
Disability, Kent told me, has always been “an engine of innovation.” It just rarely gets due credit.
Looking towards the future, Kent hopes RAMMP can eventually be “the next curb cut” whereby an individual grad student can iterate on or a small company can build upon thanks to the program’s open-source nature. That’s how a category of product, stagnant for four decades, “finally moves,” according to Kent. Moreover, he hopes young people with spinal muscular atrophy, or SMA, to grow up with more possibilities than he had—“not because the world got more sympathetic [but] because the tools got better.”
“I’m not building for a market I studied. I’m building for a life I’ve lived,” Kent said.
