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

Multimodal Dreaming: A Global Workspace Approach to World Model-Based Reinforcement Learning

arXiv 25.2 2025 57.9 method

TLDR

Combining Global Workspace theory with world models improves sample efficiency and robustness to missing modalities in RL.

Reasoning

The paper presents a novel integration of Global Workspace theory with world models (GW-Dreamer), showing improved training efficiency and robustness compared to baselines. Strengths include clear methodology and emergent robustness; weaknesses include lack of real-world validation and limited discussion of limitations.

Read-first score

Read-first score 57.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 39.

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=baseline,evaluation,metric,result

Topical relevance 42%
55.7

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: 240.

Keyword Scores

world model
10
model-based reinforcement learning world model
9
world dynamics prediction
8
generative world model
6
world simulator
3
interactive world model
2
video world model
1

Deep Analysis

Innovations

  • Integration of Global Workspace theory with world model-based reinforcement learning
  • Dreaming (mental simulation) performed within the Global Workspace latent space
  • Emergent robustness to missing observation modalities (images or simulation attributes)

Methodology

The paper proposes GW-Dreamer, an RL system that combines a Global Workspace (GW) with a world model. It is compared against standard PPO and the original Dreamer algorithm. The training and evaluation setup involves measuring environment steps and robustness to modality absence, though specific environments, hyperparameters, and metrics are not detailed in the abstract.

Key Results

GW-Dreamer requires fewer environment steps for training compared to baselines, and it exhibits strong robustness when one observation modality (images or simulation attributes) is absent, unlike the comparison algorithms.

Tags