Hungarian Team Helps Build a Surgical-Robot ImageNet With Open Data

Researchers at Hungary’s Óbuda University are working with international partners to tackle one of medical robotics’ biggest bottlenecks: the shortage of high-quality, large-scale data. Their vehicle is Open-H-Embodiment, an open-source dataset intended to give surgical robots a kind of collective memory and speed progress toward autonomous surgery.

The comparison to ImageNet is deliberate. Just as that dataset reshaped computer vision, Open-H-Embodiment is positioned as foundational infrastructure for a new generation of surgical AI, providing reusable data to train and benchmark models rather than relying on scattered, small-scale experiments.

Open-H-Embodiment is not a single algorithm but a shared record of surgical motion, video and context that can be reused across experiments. By pooling data from many robots and settings, it seeks to reduce the fragmentation that has slowed the field.

The project’s scale marks a departure from earlier efforts. Past datasets often captured only a few hours from a single robot; Open-H-Embodiment offers roughly 780 hours of synchronized video and kinematic data.

That resource draws on nearly 50 institutions worldwide and spans 20 robot platforms, including Intuitive Surgical’s widely known da Vinci Surgical System. The contributor list includes Hungarian researchers from Óbuda University’s Antal Bejczy Center for Intelligent Robotics—Eszter Lukács, Tamás Haidegger and Kristóf Takács—underscoring the team’s role in the international effort.

The full dataset and its documentation are now available in a GitHub repository, according to the project team. Such diversity matters because surgical robots differ in kinematics, sensors and control interfaces; a dataset spanning multiple platforms can help models learn transferable representations instead of overfitting to one machine or one laboratory’s workflow.