The authors built P3-WAM on a pretrained video diffusion transformer, representing actions as compact tokens encoding articulated hand poses and camera trajectories.
Unified world action model jointly learns perception, prediction, and planning
P3-WAM integrates 3D hand and camera reconstruction, action-conditioned video prediction, and goal-directed action-future planning in a single video diffusion transformer.
Top University
Zirui Wu · Chen Guo · Tianjian Jiang · Zeren Jiang · Zeming Li · Lingchen Yang · +8 more
HKUST (GZ) · Primatix · ETH Zurich · Visual Geometry Group, University of Oxford · HKUST
Research Digest··3 min read
The authors introduce P3-WAM, a unified world action model that trains 3D interaction reconstruction, action-conditioned world prediction, and joint action-future planning in one architecture.
Why this paper
From HKUST (GZ) and 5 others
In one line
A single model can jointly reconstruct 3D hand and camera motion, predict action-conditioned video futures, and plan goal-directed manipulation trajectories, with unified training improving all three.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§