GameWAM: A World Action Model for Video Games

arXiv 2026 45.4 method

TLDR

GameWAM introduces a World-Action Model that jointly generates future visuals and keyboard-mouse actions for closed-loop video game gameplay and GUI control.

Reasoning

Strengths include a novel unified formulation, mode-specific action generation, and long-horizon replanning with fewer native actions. Weaknesses: the abstract lacks detailed benchmark results and only partially reports appendix content, so generalizability claims remain limited.

Read-first score

Read-first score 45.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 53.

Recency 6%

100

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

Topical relevance 29%

75.7

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 18%

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

Reproducibility 18%

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

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

Frontier

Rank sensitivity

Stability: volatile; rank range: 453.

Keyword Scores

world model

world dynamics prediction

generative world model

interactive world model

video world model

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

Introduces GameWAM, to the authors' knowledge the first World-Action Model for native closed-loop gameplay and GUI control.
Jointly generates future visual observations and executable keyboard-mouse trajectories through parallel visual and action generative processes with block-causal conditioning and flow matching.
Constructs synchronized gameplay and GUI trajectories to support joint world-action learning.
Handles heterogeneous native control by predicting a gameplay/GUI mode at each action step and using mode-specific prediction distributions and continuous-action normalization.
Uses block-cycle control for long-horizon interaction, predicting beyond the committed horizon, executing only a short action prefix, and replanning from new observations with within-cycle context and hierarchical cross-cycle history.
Identifies Low-Frequency Action Source Imprinting (LASI), a source-sensitivity failure mode in generative control.

Methodology

GameWAM jointly models future visual observations and executable keyboard-mouse trajectories using parallel visual and action generative processes with block-causal conditioning and flow matching. It is trained on synchronized gameplay and GUI trajectories, and at each action step predicts a gameplay/GUI mode, using mode-specific prediction distributions and continuous-action normalization for heterogeneous native controls. For long-horizon interaction, it employs block-cycle control that predicts beyond the committed horizon, executes a short action prefix, and replans from new observations while preserving temporal continuity via within-cycle context and hierarchical cross-cycle history.

Key Results

GameWAM achieves competitive task success while executing fewer native actions than the compared agents. The authors also uncover Low-Frequency Action Source Imprinting (LASI), in which low-frequency components of the sampled action source systematically steer coarse generated camera motion under fixed conditioning.

Limitations

Exhibits Low-Frequency Action Source Imprinting (LASI), a source-sensitivity failure mode in generative control.

Resources

Paper