Unified World Models: Memory-Augmented Planning and Foresight for Visual Navigation

arXiv 25.10 2025 72.6 method, application

TLDR

UniWM integrates visual foresight and planning in a unified memory-augmented world model for visual navigation, achieving up to 30% improvement on benchmarks.

Reasoning

The paper's strength lies in its unified architecture combining memory-augmented world modeling with planning, demonstrated through strong empirical results across multiple benchmarks and zero-shot generalization. Weaknesses include a narrow focus on visual navigation and lack of discussion on failure cases or limitations.

Read-first score

Read-first score 72.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 51.

Methodology quality 25%

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

Recency 8%

86.7

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

Topical relevance 42%

72.9

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%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 43.

Keyword Scores

world model

world dynamics prediction

generative world model

video world model

interactive world model

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

Unified world model (UniWM) integrating egocentric visual foresight and planning within a single multimodal autoregressive backbone
Hierarchical memory mechanism fusing short-term perceptual cues with longer-term trajectory context
Explicit grounding of action selection in visually imagined outcomes

Methodology

UniWM is a unified, memory-augmented world model that uses a multimodal autoregressive backbone to integrate visual foresight and planning. It employs a hierarchical memory mechanism to combine short-term perceptual cues with long-term trajectory context for stable reasoning over extended horizons. The model is evaluated on four benchmarks (Go Stanford, ReCon, SCAND, HuRoN) and the 1X Humanoid Dataset, with zero-shot testing on TartanDrive.

Key Results

UniWM improves navigation success rates by up to 30% against strong baselines, substantially reduces trajectory errors, generalizes zero-shot to the unseen TartanDrive dataset, and scales naturally to high-dimensional humanoid control.

Resources

Paper