Stanford-Princeton Team Launches MedOS AI-Robotics System to Assist Clinicians and Reduce Medical Errors

The Stanford-Princeton AI Coscientist Team announced the launch of MedOS, the first AI-XR-cobot system designed to actively assist clinicians inside real clinical environments. Created by an interdisciplinary team led by Drs. Le Cong, Mengdi Wang, and Zhenan Bao with clinical collaborators Drs. Rebecca Rojansky and Christina Curtis, MedOS combines smart glasses, robotic arms, and multi-agent AI to form a real-time co-pilot for doctors and nurses with the mission to reduce medical errors, accelerate precision care, and support overburdened clinical teams. Physician burnout has reached crisis levels, with over 60% of doctors in the United States reporting symptoms according to recent studies, and MedOS (https://ai4med.stanford.edu) is designed to alleviate this not by replacing clinicians but by reducing cognitive overload, catching errors, and extending precision through intelligent automation and robotic assistance.

Built on years of innovation from the team's previous breakthrough, the LabOS (https://ai4lab.stanford.edu), MedOS bridges digital diagnostics with physical action. From operating rooms to bedside diagnostics, the system perceives the world in 3D, reasons through medical scenarios, and acts in coordination with doctors, nurses, and care teams after testing in surgical simulations, hospital workflows, and live precision diagnostics. MedOS introduces a "World Model for Medicine" that combines perception, intervention, and simulation into a continuous feedback loop, using smart glasses and robotic arms to understand complex clinical scenes, plan procedures, and execute them in close collaboration with clinicians. The platform has shown early promise in tasks such as laparoscopic assistance, anatomical mapping, and treatment planning.

MedOS is modular by design, built to adapt across clinical settings and specialties. In surgical simulations, it has demonstrated the ability to interpret real-time video from smart glasses, identify anatomical structures, and assist with robotic tool alignment, functioning as a true clinical co-pilot. This tight integration of perception, planning, and action sets MedOS apart as an active collaborator rather than just a passive assistant in high-stakes procedures. Breakthrough capabilities include a multi-agent AI architecture that mirrors clinical reasoning logic, synthesizes evidence, and manages procedures in real time, with MedOS achieving 97% accuracy on MedQA (USMLE) and 94% on GPQA, beating frontier AI models like Gemini-3 Pro, GPT-5.2 Thinking, and Claude 4.5 Opus.

The system also leverages MedSuperVision, the largest open-source medical video dataset featuring more than 85,000 minutes of surgical footage from 1,882 clinical experts. It has demonstrated success in helping nurses and medical students reach physician-level performance and reducing human error in fatigue-prone environments, with registered nurses improving from 49% to 77% with MedOS assistance and medical students from 72% to 91%. Case studies include uncovering immune side effects of the GLP-1 agonist Semaglutide (Wegovy) from the FDA database and identifying prognostic implications of driver gene co-mutations on cancer patients' survival. MedOS is launching with support from NVIDIA, AI4Science, and Nebius and has been deployed in early pilots, with clinical collaborators now able to request early access.

Dr. Le Cong, leader of the Stanford-Princeton AI Coscientist Team and Associate Professor at Stanford University, emphasized that the goal is not to replace doctors but to amplify their intelligence, extend their abilities, and reduce risks posed by fatigue, oversight, or complexity, describing MedOS as the beginning of a new era of AI as a true clinical partner. Dr. Mengdi Wang, co-leader of the collaboration, noted that MedOS reflects a convergence of multi-agent reasoning, human-centered robotics, and XR interfaces aimed at creating a collaborative loop that helps clinicians manage complexity in real time. MedOS will be showcased at a Stanford event in early March followed by a public unveiling at the NVIDIA GTC conference in March 2026, with GTC session information available at https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81748/. Additional information can be found on the project page at https://medos-ai.github.io/ and the official site at https://ai4medos.com/, with the research paper available at https://medos-ai.github.io/paper.