ActWorld: From Explorable to Interactive World Model via Action-Aware Memory

arXiv 2026 63.4 method

TLDR

ActWorld extends navigation-centric world models to support object interaction via action-aware memory and a new dataset.

Reasoning

Strengths: addresses the navigation-interaction gap with a novel hierarchical memory and a 100K interaction video dataset. Weaknesses: abstract lacks detailed experimental results and scalability discussion.

Read-first score

Read-first score 63.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

92.8

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.92808755

Methodology quality 18%

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

Topical relevance 29%

64.3

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 18%

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

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 403.

Keyword Scores

interactive world model

world model

video world model

world dynamics prediction

world simulator

generative world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Extending navigation-centric world models to support mid-rollout object interaction within a chunk-autoregressive framework
Construction of a 100K interaction video dataset with per-chunk captions via chain-of-thought reasoning
Hierarchical action-aware memory design that routes history compression by interaction importance, complemented by a persistent memory bank for event-update and object-identity tokens

Methodology

ActWorld uses a chunk-autoregressive framework to generate interactive world models. It addresses the data bottleneck by constructing a 100K interaction video dataset annotated with per-chunk captions via chain-of-thought reasoning. To overcome the memory bottleneck, it introduces a hierarchical action-aware memory that compresses history based on interaction importance and maintains a persistent memory bank for event-update and object-identity tokens across long rollouts.

Key Results

ActWorld supports both flexible navigation and rich object interaction within a single model, substantially improving interaction fidelity over navigation-only baselines without sacrificing viewpoint control.

Resources

Paper