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

DreamSAC: Learning Hamiltonian World Models via Symmetry Exploration

arXiv 26.3 2026 56.3 method

TLDR

DreamSAC uses symmetry exploration and Hamiltonian world models to improve extrapolative generalization in 3D physics simulations.

Reasoning

The paper introduces a novel unsupervised exploration strategy and a Hamiltonian-based world model with contrastive learning to capture physical invariances, outperforming baselines on extrapolation tasks. However, it is limited to simulated 3D physics environments and lacks real-world validation or discussion of broader applicability.

Read-first score

Read-first score 56.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 47.

Recency 8%
100

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

Topical relevance 42%
67.1

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 25%
50

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

Reproducibility 25%
30

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 514.

Keyword Scores

world model
10
world dynamics prediction
8
generative world model
7
model-based reinforcement learning world model
7
interactive world model
6
world simulator
5
video world model
4

Deep Analysis

Innovations

  • Symmetry Exploration: an unsupervised exploration strategy where an agent is intrinsically motivated by a Hamiltonian-based curiosity bonus to actively probe and challenge its understanding of conservation laws, thereby collecting physically informative data.
  • Hamiltonian-based world model that learns from collected data using a novel self-supervised contrastive objective to identify the invariant physical state from raw, view-dependent pixel observations.

Methodology

DreamSAC uses a Hamiltonian-based world model trained on data collected via Symmetry Exploration, an unsupervised exploration strategy with a Hamiltonian-based curiosity bonus. The model employs a self-supervised contrastive objective to learn invariant physical states from raw pixel observations. It is evaluated against state-of-the-art baselines in 3D physics simulations on extrapolation tasks.

Key Results

DreamSAC significantly outperforms state-of-the-art baselines in 3D physics simulations on tasks requiring extrapolation.

Resources

Paper

Tags