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

Object-Centric World Models for Causality-Aware Reinforcement Learning

AAAI 26 2026 53.6 method

TLDR

STICA uses object-centric Transformers as world models with causality-aware policy networks for sample-efficient RL on object-rich benchmarks.

Reasoning

The paper introduces a novel integration of object-centric representations and causality-aware decision-making, showing strong empirical results on benchmarks. However, it lacks real-world experiments and does not address scalability or limitations of the approach.

Read-first score

Read-first score 53.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 34.

Recency 8%
100

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

Methodology quality 25%
70

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

Topical relevance 42%
48.6

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 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

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 399.

Keyword Scores

world model
10
model-based reinforcement learning world model
9
world dynamics prediction
8
world simulator
3
generative world model
2
interactive world model
1
video world model
1

Deep Analysis

Innovations

  • Object-centric Transformers as the world model
  • Causality-aware policy and value networks
  • Token-level dynamics and interaction prediction
  • Causality-guided decision-making via attention layers

Methodology

STICA represents each observation as a set of object-centric tokens, along with tokens for the agent action and the resulting reward. The world model, based on object-centric Transformers, predicts token-level dynamics and interactions. The policy and value networks estimate token-level cause-effect relations and incorporate them into attention layers to guide decision-making.

Key Results

On object-rich benchmarks, STICA consistently outperforms state-of-the-art agents in both sample efficiency and final performance.

Resources

Paper

Tags