⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

Dexterous World Models

arXiv 25.12 2025 66.4 method

TLDR

A video diffusion framework that models dexterous human actions inducing dynamic changes in static 3D scenes for interactive digital twins.

Reasoning

Strengths: novel integration of dexterous hand motion with scene-conditioned video diffusion, hybrid dataset combining synthetic and real videos. Weaknesses: limited to egocentric hand actions, no evaluation of generalization or long-term dynamics.

Read-first score

Read-first score 66.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 49.

Recency 8%
86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%
70

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 25%
70

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=dataset,experiment

Reproducibility 25%
50

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,dataset,github

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 91.

Keyword Scores

interactive world model
9
video world model
9
world model
8
generative world model
8
world dynamics prediction
8
world simulator
6
model-based reinforcement learning world model
1

Deep Analysis

Innovations

  • Scene-action-conditioned video diffusion framework for modeling dexterous human actions in static 3D scenes
  • Conditioning on static scene renderings and egocentric hand mesh renderings to ensure spatial consistency and encode geometry/motion cues
  • Hybrid interaction video dataset combining synthetic egocentric interactions with fixed-camera real-world videos for joint locomotion and manipulation learning

Methodology

DWM is a video diffusion model conditioned on (1) static 3D scene renderings following a specified camera trajectory and (2) egocentric hand mesh renderings that encode both geometry and motion cues. Training uses a hybrid dataset: synthetic egocentric interactions provide fully aligned supervision for joint locomotion and manipulation, while fixed-camera real-world videos contribute diverse and realistic object dynamics. The model generates temporally coherent videos of plausible human-scene interactions.

Key Results

DWM generates realistic and physically plausible interactions such as grasping, opening, and moving objects while maintaining camera and scene consistency, demonstrating a first step toward video diffusion-based interactive digital twins.

Limitations

  • As a first step, the framework may have limited generalization and temporal consistency; no explicit limitations are stated in the abstract.

Resources

PaperProject

Tags