Meta's open computer-vision models are powering a new assistive robotic wheelchair platform

Meta published a blog post on July 27, 2026 describing how its open-source computer vision models, DINOv3, the Segment Anything Model, and a lightweight detector called RF-DETR, are powering the perception system behind RAMMP, the Robotic Assistive Mobility and Manipulation Platform being developed by the University of Pittsburgh's Human Engineering Research Laboratories. The project, backed by up to $41.5 million from the Advanced Research Projects Agency for Health, targets a real gap: more than 5.5 million Americans use wheelchairs, and over 100,000 wheelchair-related injuries are treated in U.S. emergency rooms each year, many from falls and collisions the platform is designed to help prevent. The engineering challenge is running capable perception models on battery-powered, size-constrained hardware that has to work reliably in the outside world, not a lab, which the team addresses through smaller memory footprints, lower-precision inference, and efficient batching so the system can still detect doors, cups, curbs, and other hazards in real time. Sivashankar Sivakanthan, chief of staff to RAMMP, said that for assistive robotics “performance is not measured by benchmark accuracy alone, but by whether a system can operate reliably in the unpredictability of everyday life.” Pittsburgh's HERL supplies the biomedical engineering and user research, while commercial partner ATDev focuses on turning the research into something that can actually be manufactured and deployed, with a broader consortium that includes Kinova Robotics, Carnegie Mellon, Cornell, Northeastern, and Purdue. ARPA-H director Alicia Jackson framed the goal plainly: “Millions of Americans with limited mobility deserve technology that meets them where they are, not technology that asks them to adapt.”