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

GLAM: Global-Local Variation Awareness in Mamba-based World Model

arXiv 25.1 2025 58.8 method

TLDR

GLAM uses Mamba-based modules to perceive global and local state variations, improving world model reasoning for MBRL, outperforming on Atari 100k.

Reasoning

The paper introduces a novel variation-aware world model with dual Mamba modules, showing strong empirical results on Atari 100k. However, it lacks evaluation on more complex or real-world environments, and the claimed improvements may be benchmark-specific.

Read-first score

Read-first score 58.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 35.

Recency 8%
86.7

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

Reproducibility 25%
73

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=github

Topical relevance 42%
50

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=benchmark

Field roles

FrontierReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 297.

Keyword Scores

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

Deep Analysis

Innovations

  • Global-Local variation awareness in Mamba-based world model for model-based reinforcement learning
  • Two parallel reasoning modules: GMamba for global variation patterns and LMamba for local variation between adjacent states
  • Using variation between states to improve reasoning quality and imagination-based training

Methodology

GLAM employs two Mamba-based parallel reasoning modules: GMamba captures global variation patterns across state sequences to predict future state variation, while LMamba focuses on local variation between adjacent states to reason about rewards, termination signals, and visual representations. The model integrates both modules to enhance imagination-based training in model-based reinforcement learning.

Key Results

GLAM outperforms existing methods in normalized human scores on the Atari 100k benchmark.

Resources

PaperCode

Tags