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

Learning a Particle Dynamics Model with Real-world Videos

arXiv 2026 51.3 method

TLDR

Learns particle dynamics from real-world videos using Gaussian splatting and rendering supervision, overcoming sim-to-real gap.

Reasoning

Strengths include a novel framework for training on unlabeled real videos and a new dataset of 500 videos. Weaknesses are the focus on particle dynamics rather than a full world model, and potential limitations in complex scenes.

Read-first score

Read-first score 51.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 31.

Recency 6%
100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%
75.6

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.75574928

Methodology quality 18%
70

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

Topical relevance 29%
44.3

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

Reproducibility 18%
38

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

Citation velocity 12%
0

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 226.

Keyword Scores

world dynamics prediction
9
world model
7
video world model
5
world simulator
4
generative world model
3
interactive world model
2
model-based reinforcement learning world model
1

Deep Analysis

Innovations

  • Training neural object dynamics models directly from unlabeled real-world videos, overcoming the sim-to-real gap
  • Particle-based dynamics model compatible with Gaussian splatting, operating on dense particles with scales and rotations
  • Training via rendering supervision without requiring particle-level labeled states
  • Introduction of a real-world dataset of about 500 videos capturing diverse object interactions

Methodology

The paper proposes a particle-based dynamics model that operates on dense particles derived from Gaussians (with scales and rotations) and predicts their position and rotation changes over time. The model is trained via rendering supervision from unlabeled real-world videos, eliminating the need for particle-level state labels. A dataset of about 500 videos is used for training and evaluation.

Key Results

The model is trained on a real-world dataset of about 500 videos, enabling learning of object dynamics from unlabeled video data. No quantitative results are reported in the abstract.

Tags

particle dynamicsneural dynamicsreal-world videosphysics simulationworld modelssim-to-realCV