UltraWorld: Learning Interactive Ultrasound World Models from Untracked Clinical Videos with Acoustic Sampling Map
We present UltraWorld, a self-distillation recipe that transfers priors from clinical ultrasound videos into interactive world models without real action annotations.
ProofPaper ↗
Key points
- World models can enable autonomous ultrasound scanning by predicting the outcomes of probe motions from local observations.
- Reliable action following further requires modeling ultrasound's cross-sectional sampling geometry.
- Anatomical masks sampled along programmable trajectories through 3D anatomy provide spatial guidance for synthesizing action--video pairs.
- To further improve action following, we introduce the Acoustic Sampling Map (AsMap), which represents probe poses and imaging settings as pixel-wise 3D sampling positions, beam directions, and depths.
Sources (1)
- [1]UltraWorld: Learning Interactive Ultrasound World Models from Untracked Clinical Videos with Acoustic Sampling MaparXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 10:02 AM
We present UltraWorld, a self-distillation recipe that transfers priors from clinical ultrasound videos into interactive world models without real action annotations.
World models can enable autonomous ultrasound scanning by predicting the outcomes of probe motions from local observations.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 7, 2026ΔWAM: Distilling Action Tangent Fields into World Action Models
- Oct 7, 2026MIMESIS: Learning User Simulators as Training Environments for Interactive Agents
- Oct 6, 2026Self-Retrospection Distillation: Turning Post-hoc Experiences into Prior Foresight
- Oct 5, 2026From Traces to Agentic Worlds: Agentic Language World Models for Interactive Environment Simulation
- Sep 29, 2026LongTake: Learning to Sustain Dynamics in Long-Horizon Video Generation
- Jun 9, 2026Powering the future of robotics in Europe