Add WAM paper: WAM-TTT: Steering World-Action Models by Watching Human Play at Test Time

[279object]

Paper

• Title: WAM-TTT: Steering World-Action Models by Watching Human Play at Test Time
• Short name: WAM-TTT
• arXiv ID: 2607.06988v1
• Paper: https://arxiv.org/pdf/2607.06988
• Authors: Yusen Feng, Bingchen Han, Jiangran Lyu, Kai Liu, Yixin Zheng, Yuxuan Wan, Weiheng Liu, Sun Han, Ruiqin Li, Yulong Zhang, Fangfu Liu, Xuesong Shi, Libin Liu, Yizhou Wang, Zhizheng Zhang, He Wang
• Published: 2026-07-08
• arXiv categories: cs.RO, cs.AI
• Matched keywords: WAM, world action model, world action models, world-action model, world-action models

README Entry

- **WAM-TTT**: "WAM-TTT: Steering World-Action Models by Watching Human Play at Test Time", arXiv 2026. ![](https://img.shields.io/badge/Multi--Dit-9f1239) ![](https://img.shields.io/badge/Cross--Attention-f97316)
  [[📄 Paper](https://arxiv.org/pdf/2607.06988)]

Robot Decision

• Decision: accept
• Confidence: high
• Taxonomy: joint_diffusion_multistream_cross_attention
• Taxonomy confidence: medium
• README heading: ### Diffusion-based Generation
• Badges: https://img.shields.io/badge/Multi--Dit-9f1239, https://img.shields.io/badge/Cross--Attention-f97316

Reason

Paper explicitly proposes test-time steering/adaptation of world-action models for robot manipulation control using human videos and video prediction.

Evidence

• "test-time training framework for steering world action models"
• WAM jointly models visual dynamics and actions for robot control
• experiments on real-robot manipulation tasks

Taxonomy Evidence

• build WAM-TTT on top of LDA, a pretrained world-action model
• diffusion transformer block contains video expert and action expert
• two experts communicate through joint attention
• unlabeled human videos adapted through self-supervised video prediction

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