About the Tutorial
An inflection point for surgical robot learning
For the first time, the field has open, cross-embodiment data at scale, attempts at generalist policy architectures that transfer from kitchen-table manipulation to the operating room, and peer-reviewed demonstrations of autonomous soft-tissue surgical steps. Yet the path from a learned policy to a system a surgeon trusts in the OR remains poorly charted.
This tutorial is explicitly instructional. We pair perspective talks from industry, academia, and practicing surgery with a guided hands-on session on the Open-H-Embodiment dataset — so attendees leave able to load the data, run a pretrained policy, and reason about how to improve the current state of the art.
Intended audience: robot-learning researchers, graduate students, and engineers entering surgical robotics, plus practitioners in surgical data science seeking a shared dataset and a current map of open problems. We assume comfort with deep learning and Python, but no surgical background.
Core Questions
Five decisions that gate progress
Each is revisited from the industry, academic, and clinical perspective during the talks, and made real through the hands-on session.
What data should we collect?
Which sensing streams and platforms yield data that actually transfers, and where the largest collection gaps remain.
What data should we annotate?
Where scarce expert-annotation budget is best spent, and how to combine sparse labels with foundation-model pretraining.
Which task should we automate?
Choosing subtasks that are clinically meaningful, technically tractable, and safely evaluable.
Bridging industry, academia & surgery
Surfacing what is lost in translation across the loop, and what each community needs from the others.
Training the next medical roboticist
What a newcomer should learn first, and what infrastructure lowers the barrier to entry.
Invited Speakers
Industry, academia, and clinical practice
Four speakers span the perspectives below. Names and affiliations will be announced here as they are confirmed.
Clinical perspective
Clinical reality — what autonomy must solve in the OR. Grounding every challenge in clinical need: what data and which tasks matter, what annotations are meaningful, and what it takes to earn surgeon trust.
Academic perspective
Perception and learning for autonomy. Surgical perception, motion planning, and vision-language models for surgery — annotation strategy, benchmarking, and the infrastructure newcomers need.
Industry perspective
What ships, and the data behind it. Surgical activity and skill recognition from the seat of a company that deploys models against real surgical video at scale, and how production constraints shape data collection.
Academic frontier
End-to-end soft-tissue autonomy. How far learned policies can go on real anatomy — anchoring the "which task to automate" and "what data to collect" questions with concrete results.
Format
A half-day that never sits still
Four 25-minute talks supply the perspectives; a 60-minute hands-on block on a shared Open-H Colab notebook lets attendees run the methods themselves; a closing panel turns the speakers toward each other. Pacing changes every 20–30 minutes.
Organizers