Add WAM paper: iMaC: Translating Actions into Motion and Contact Images for Embodied World Models

[279object]

Paper

• Title: iMaC: Translating Actions into Motion and Contact Images for Embodied World Models
• Short name: iMaC
• arXiv ID: 2606.09813v1
• Paper: https://arxiv.org/pdf/2606.09813
• Authors: Zhenyu Wu, Xiuwei Xu, Yukun Zhou, Yifan Li, Qiuping Deng, Xiaofeng Wang, Zheng Zhu, Bingyao Yu, Ziwei Wang, Jiwen Lu, Haibin Yan
• Published: 2026-06-08
• arXiv categories: cs.RO, cs.CV
• Matched keywords: embodied world model, embodied world models

README Entry

- **iMaC**: "iMaC: Translating Actions into Motion and Contact Images for Embodied World Models", arXiv 2026.
  [[📄 Paper](https://arxiv.org/pdf/2606.09813)]

Robot Decision

• Decision: accept
• Confidence: high
• Taxonomy: world_model_for_evaluation
• Taxonomy confidence: high
• README heading: ### World Model for Evaluation
• Badges: None

Reason

The paper explicitly develops an action-conditioned embodied world model for robotic manipulation, predicting future observations conditioned on robot actions for policy evaluation and closed-loop rollout.

Evidence

• embodied world model conditioned on future joint actions
• predicts future observation chunks for robot policy evaluation
• uses URDF/FK motion images and contact images as action controls

Taxonomy Evidence

• embodied world models promise to serve as real-world simulators for robot policy evaluation
• iMaC can evaluate the relative performance of different policy checkpoints
• world-model success estimates strongly positively correlated with real-world policy performance

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