Physical Object Understanding with a Physically Controllable World Model

arXiv 2026 51.2 method

TLDR

A probabilistic world model trained via autoregressive sequence modeling that learns physical object understanding, 3D manipulation, and physical relationships from videos.

Reasoning

Strengths: novel class of probabilistic world models with efficient training and emergent object understanding, including 3D manipulation and physical relationship computation. Weaknesses: abstract lacks explicit real-world validation or quantitative benchmarks; no mention of comparisons to existing methods.

Read-first score

Read-first score 51.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 40.

Recency 6%

100

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

Citation impact 18%

91.9

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.91905417

Topical relevance 29%

57.1

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 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=none

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

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

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 442.

Keyword Scores

world model

world dynamics prediction

generative world model

video world model

interactive world model

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

New class of probabilistic world models that support estimation of the probability of any visual variable conditioned on any other variables
Training these models efficiently with autoregressive sequence modeling, yielding world models from which rich object understanding emerges
Extracting objects and articulated object subparts by analyzing motion correlations across multiple generated plausible future states
Manipulating discovered objects in 3D using the world model
Computing physical relationships between objects from the world model, enabling applications such as Visual Jenga

Methodology

The paper introduces a probabilistic world model trained via autoregressive sequence modeling. The model estimates probabilities of visual variables (e.g., appearance and dynamics) conditioned on any other variables. It generates multiple plausible future states through sequential inference, and objects and articulated subparts are extracted by analyzing motion correlations across these futures. The model then enables 3D manipulation of discovered objects and computation of physical relationships.

Key Results

The model captures physical laws governing object motion by generating plausible future states. It successfully extracts objects and articulated subparts, demonstrates 3D manipulation, and computes physical relationships for applications like Visual Jenga.

Resources

Paper