ReactiveGWM: Steering NPC in Reactive Game World Models

arXiv 2026 62 method

TLDR

ReactiveGWM decouples player and NPC controls in game world models, enabling zero-shot strategy transfer for steerable NPC interactions.

Reasoning

The paper introduces a novel decoupling of player and NPC dynamics in game world models, achieving zero-shot transfer across games. Strengths include explicit interaction modeling and lightweight adaptation, but evaluation is limited to two fighting games, raising questions about generalizability.

Read-first score

Read-first score 62, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 54.

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,evaluation,metric

Citation impact 18%

79.3

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.79304674

Topical relevance 29%

77.1

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=none

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: 435.

Keyword Scores

world model

interactive world model

world simulator

generative world model

world dynamics prediction

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Explicit decoupling of player controls from NPC behaviors in game world models
Lightweight additive bias injection for player actions into diffusion backbone
Cross-attention modules for grounding high-level NPC responses (Offense, Control, Defense)
Game-agnostic representation of interactive logic enabling zero-shot strategy transfer
Plug-and-play modules that work with off-the-shelf, unannotated world models without retraining

Methodology

ReactiveGWM uses a diffusion backbone where player actions are injected via a lightweight additive bias, while NPC behaviors are grounded through cross-attention modules that learn a game-agnostic representation of interactive logic. The model is trained and evaluated on two Street Fighter games, with baselines and metrics focusing on player controllability and NPC strategy adherence.

Key Results

ReactiveGWM maintains fine-grain player controllability while achieving robust, prompt-aligned NPC strategy adherence, enabling steerable NPC interactions without domain-specific retraining.

Limitations

Evaluation is limited to two Street Fighter games, so generalization to other game genres or interaction types is unverified.
The paper does not explicitly discuss failure cases or potential degradation in zero-shot transfer scenarios.

Resources

Paper