Add WAM paper: Efficient Sim-to-Real Transfer of World-Action Models from Synthetic Priors

[279object]

Paper

• Title: Efficient Sim-to-Real Transfer of World-Action Models from Synthetic Priors
• Short name: ESTRT
• arXiv ID: 2606.31101v1
• Paper: https://arxiv.org/pdf/2606.31101
• Authors: Zixing Wang, Kausik Sivakumar, Jinghuan Shang, Yafei Hu, Zhaoming Xie, Ran Gong, Xiaohan Zhang, Karl Schmeckpeper
• Published: 2026-06-30
• arXiv categories: cs.RO
• Matched keywords: world-action model, world-action models

README Entry

- **ESTRT**: "Efficient Sim-to-Real Transfer of World-Action Models from Synthetic Priors", arXiv 2026. ![](https://img.shields.io/badge/Unified--DiT-be123c) ![](https://img.shields.io/badge/Explicit-f43f5e)
  [[📄 Paper](https://arxiv.org/pdf/2606.31101)]

Robot Decision

• Decision: accept
• Confidence: high
• Taxonomy: joint_diffusion_unified_explicit
• Taxonomy confidence: high
• README heading: ### Diffusion-based Generation
• Badges: https://img.shields.io/badge/Unified--DiT-be123c, https://img.shields.io/badge/Explicit-f43f5e

Reason

Paper explicitly studies world-action models for sim-to-real robotic manipulation, using action-conditioned video/dynamics prediction for visuomotor control.

Evidence

• world-action model trained from synthetic demonstrations
• predict future visual observations together with robot actions
• zero-shot deployment on real Franka manipulation tasks

Taxonomy Evidence

• build upon Cosmos Policy, a video diffusion model adapted for visuomotor control
• actions, future observations, and value estimates are encoded as latent frames within a unified diffusion process
• trained with ~800 fully synthesized simulation demonstrations per task and zero real-world demonstrations
• deployed directly on a real Franka Research 3

Human Review Checklist

• This paper belongs in the WAM survey
• README section is correct
• Badges are correct
• Short name is correct
• Paper link is correct
• Merge this PR if accepted