WorldPack: Compressed Memory Improves Spatial Consistency in Video World Modeling

arXiv 25.12 2025 58.3 method

TLDR

WorldPack uses compressed memory (trajectory packing and retrieval) to improve spatial consistency in long-term video world modeling, outperforming SOTA on Minecraft.

Reasoning

The paper introduces a novel compressed memory mechanism for video world models, addressing long-term spatial consistency with efficient context handling. Its strength lies in the clear methodology and strong empirical results on the LoopNav benchmark. However, the evaluation is limited to a single simulated environment (Minecraft), and real-world applicability is not demonstrated.

Read-first score

Read-first score 58.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 48.

Recency 8%

86.7

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

Topical relevance 42%

68.6

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

Methodology quality 25%

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

Reproducibility 25%

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: 355.

Keyword Scores

world model

video world model

generative world model

interactive world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

Deep Analysis

Innovations

Compressed memory mechanism for video world modeling combining trajectory packing and memory retrieval
Trajectory packing to achieve high context efficiency in long-term generation
Memory retrieval to maintain spatial consistency and support long-term spatial reasoning during rollouts

Methodology

WorldPack employs a compressed memory consisting of trajectory packing for high context efficiency and memory retrieval for consistency in rollouts. The model is evaluated on the LoopNav benchmark in Minecraft, comparing against state-of-the-art video world models.

Key Results

WorldPack notably outperforms strong state-of-the-art models on the LoopNav benchmark, demonstrating improved spatial consistency, fidelity, and quality in long-term generation despite using shorter context lengths.

Resources

Paper