WorldMind: Decoupled Game World Model for State-Aware NPC Behavior

arXiv 2026 45.2 method

TLDR

Introduces WorldMind, a decoupled game world model with four layers for state-aware NPC behavior, plus BOSS-140K dataset of gameplay videos with internal states.

Reasoning

The paper clearly decomposes world modeling into understanding, decision, control, and generation layers, and introduces a large dataset with paired internal states. However, the abstract is truncated before quantitative results, so empirical strengths and limitations are not fully visible.

Read-first score

Read-first score 45.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 44.

Recency 6%

100

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

Methodology quality 18%

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

Topical relevance 29%

62.9

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 18%

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

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 477.

Keyword Scores

world model

interactive world model

generative world model

video world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

First decoupled framework for state-aware NPC behavior in game world models, separating interactive world modeling into four layers: Understanding, Decision, Control, and Generation.
BOSS-140K dataset: gameplay videos paired with rich internal game states, collected via an automated agent.

Methodology

WorldMind decouples NPC behavior into an Understanding Layer that constructs a compact state from frames, a Decision Layer that plans actions over the state, a Control Layer that translates actions into temporal conditions, and a Generation Layer that synthesizes visual outcomes, forming a closed interaction loop. The BOSS-140K dataset and an automated collection agent are introduced for training and evaluation.

Key Results

WorldMind achieves reliable compact state reconstruction and mechanics-grounded planning, and is preferred over baselines in ~70% of pairwise comparisons for more tactically appropriate and coherent NPC behavior.

Resources

Paper