Surgical WAM: A World-Action Model for Data-Efficient Surgical Robot Learning

2026-08-11

AI Summary

This work addresses the challenge of limited availability of action-labeled demonstration data in surgical robot learning, where tasks demand high precision in contact-rich interactions, long-horizon reasoning, and bimanual coordination. The authors propose Surgical WAM, a generative world-action model that jointly represents endoscopic visual observations and executable action primitives. For the first time, it effectively transfers visual dynamics priors learned from unlabeled surgical videos to closed-loop robotic control. Built upon the Cosmos Policy architecture, the approach integrates unsupervised video pretraining with receding-horizon control, enabling efficient fine-tuning with only sparse action annotations. Evaluated on four simulated surgical tasks, the method improves average success rates from 63.5% to 77.8%, with a notable 20-percentage-point gain on PegTransfer and especially pronounced gains in contact-intensive and bimanual coordination tasks.

Abstract

Learning reliable surgical manipulation policies is bottlenecked by the scarcity of action-labeled demonstrations: teleoperated surgical robot (e.g., dVRK) trajectories with synchronized kinematics are costly to collect, while surgical tasks demand precise contact handling, long-horizon reasoning, and bimanual coordination. Endoscopic video is comparatively inexpensive and abundant relative to synchronized video--kinematics trajectories, and a natural way to exploit it is to learn world models of surgical scenes. However, existing surgical world models use video primarily for simulation or policy evaluation, and rarely translate the learned dynamics into closed-loop control. This gap raises our central question: under a fixed budget of action-labeled demonstrations, does action-free video pretraining improve closed-loop surgical manipulation? To answer it, we introduce the Surgical World-Action Model (Surgical WAM), a unified generative model built on Cosmos Policy that jointly predicts future endoscopic observations and executable surgical robot action chunks. Surgical WAM first learns surgical visual dynamics from action-free video and is then fine-tuned on the fixed action-labeled budget; at deployment, it acts as a closed-loop, receding-horizon controller that executes a short prefix of each predicted action chunk and replans from the resulting observation. On a suite of four simulated surgical manipulation tasks, video pretraining improves the average success rate from 63.5% to 77.8%, including an absolute gain of 20 percentage points on PegTransfer, with the largest improvements on contact-rich and bimanual tasks. These results demonstrate that action-free video provides transferable visual dynamics priors for learning surgical robot control with limited action supervision, positioning data-efficient video pretraining as a practical path toward scaling up surgical robot learning.

Problem

• surgical robot learning
• data efficiency
• action-labeled demonstrations
• world models
• video pretraining

Innovation

• world-action model
• data-efficient learning
• surgical robot control
• video pretraining
• closed-loop planning