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

R-WoM: Retrieval-augmented World Model For Computer-use Agents

arXiv 25.10 2025 59.7 method, application

TLDR

Proposes R-WoM, a retrieval-augmented world model that improves LLM-based simulation for computer-use agents by grounding predictions in external tutorials.

Reasoning

Strengths include systematic probing of LLM world model capabilities and a novel retrieval-augmented approach that shows significant improvements on real-world benchmarks. Weaknesses are the narrow focus on digital environments and reliance on tutorial retrieval, which may limit generalization.

Read-first score

Read-first score 59.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 42.

Recency 8%
86.7

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

Methodology quality 25%
80

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

Topical relevance 42%
60

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: 180.

Keyword Scores

world model
10
world dynamics prediction
9
world simulator
7
generative world model
6
interactive world model
5
model-based reinforcement learning world model
5
video world model
0

Deep Analysis

Innovations

  • Systematic probing of LLMs' core world modeling capabilities (future state prediction and reward estimation) through three tasks: next-state identification, full-procedure planning alignment, and milestone transition recognition.
  • Proposal of Retrieval-augmented World Model (R-WoM) that grounds LLM simulations by incorporating factual, up-to-date knowledge retrieved from external tutorials.

Methodology

The study first probes LLMs' world modeling abilities via three tasks: next-state identification, full-procedure planning alignment, and milestone transition recognition. It then proposes R-WoM, which augments LLM simulations with retrieved external tutorial knowledge. Experiments are conducted on subsets of OSWorld and Webarena, comparing against baselines.

Key Results

R-WoM achieves relative improvements of up to 23.4% and 16.3% on subsets of OSWorld and Webarena, respectively, with particular advantage in longer-horizon simulations.

Limitations

  • LLMs' tendency toward hallucination and reliance on static training knowledge.
  • Compounding errors that inhibit long-horizon simulations.
  • Performance rapidly degrades in full-procedure planning, highlighting limitations in reliably modeling environment dynamics over long horizons.

Resources

Paper

Tags