Surgical WAM: A World-Action Model for Data-Efficient Surgical Robot Learning
Action-free surgical video pretraining lifted simulated robot-task success from 63.5% to 77.8%.
The paper introduces Surgical WAM, a generative model that learns surgical visual dynamics from endoscopic video, then fine-tunes on a fixed set of action-labeled robot demonstrations. At deployment, it predicts short action chunks and replans after observing the result. The authors test it on four simulated surgical manipulation tasks, with the biggest gains on contact-heavy and bimanual work, including a 20-point gain on PegTransfer. ArXiv · AI/CL/LG's note
The paper introduces Surgical WAM, a generative model that learns surgical visual dynamics from endoscopic video, then fine-tunes on a fixed set of action-labeled robot demonstrations. At deployment, it predicts short action chunks and replans after observing the result. The authors test it on four simulated surgical manipulation tasks, with the biggest gains on contact-heavy and bimanual work, including a 20-point gain on PegTransfer. ArXiv · AI/CL/LG's note
score 5