GTC SJ 2026: Next Generation Intelligent Surgical Robots

In this talk

1. 1:02

   From today’s robots to data-centric systems

   Mati Azizzian surveys surgical robots by access method, physical design, and autonomy level. Most commercial systems are teleoperated, and their proprietary connections make upgrades and interoperability difficult. He proposes a shared Ethernet backbone with distributed edge compute, where data can be accessed across system components.

2. 10:20

   Lower-latency processing and development

   Azizzian describes requirements including synchronized data, redundant paths for safety-critical signals, and secure access. In a sensor-to-action example, he contrasts about 50 milliseconds on current systems with a 1.2-millisecond input-to-output benchmark using a Holoscan Sensor Bridge and GPU pipeline. He also describes using the same robot compute across simulation and physical deployment.

3. 20:31

   Open data and surgical AI models

   Shawn Hoover says the Open Embodiment coalition grew to 35 partners and released a dataset covering eight robotic embodiments, including nearly 500 hours of clinical data donated by CMR Surgical. He shows a suturing model pretrained on that data and post-trained with a small amount of Johns Hopkins SutureBot data, while noting that many attempts failed. He also presents a world-model surgical simulator and a video tokenizer released for research.

4. 30:54

   Agentic systems for hospital workflows

   Madi describes a hospital automation framework that links operating-room observations with software agents and mobile robots. In a demonstration, the system identifies upcoming instrument needs, checks whether items are available, and can send a robot to retrieve and prepare a tray from sterile processing. Madi says the demo still needs substantial productization.

5. 35:14

   Regulatory outlook and audience question

   The closing remarks envision data-centric, interoperable systems and discuss possible regulatory pathways for foundational models and software-defined medical devices. In response to a question about functional safety, a speaker says Sensor Bridge is configurable and open source, with watermarking and CRC checks under consideration for IEC 61508 certification; broader safety remains specific to each medical device.