Papers
arxiv:2611.24154

Game2Player: Scaling Gameplay Experience for Visual Agents through Environment-Player Co-Development

Published on Nov 18
· Submitted by
Rui Yang
on Nov 19
#2 Paper of the day
Authors:
,
,
,
,
,
,
,
,
,
,
,
,
,

Abstract

Games are a rich testbed for visual agents, demanding perception, long-horizon planning, and precise control. However, collecting human gameplay is costly, and learning from gameplay videos requires inferring missing actions. Existing datasets also cover a limited range of games, often without explicit tasks or verifiable outcomes. To address these gaps, we introduce Game2Player, a framework that co-develops game environments and players to collect diverse and verifiable gameplay experience at scale. In the Game2Player loop, a Builder agent adapts existing web games, exposes their internal state, and defines tasks with state-based progress measures and success checks. A Coding Player and a VLM Player then attempt the tasks and report problems to the Builder for repair, so that play serves as both environment validation and training data collection. With Game2Player, we build 568 verified web games with 3,436 tasks, collect 15.7K trajectories (307 hours of gameplay), and hold out 50 games as the benchmark PlayerBench. To evaluate the collected experience as training data, we train VLM agents with progress-weighted SFT, assigning greater weight to steps that advance task progress. Game2Player-9B improves progress from 14.3% to 31.2% on PlayerBench and from 30.0% to 53.1% on the external GameWorld benchmark. More broadly, our work suggests a path for visual agents to scale their gameplay experience by building, verifying, and learning from existing game worlds.

Community

Paper submitter

Game2Player scales gameplay experience for visual agents by turning existing web games into verifiable environments. A coding player and a visual player test tasks, report problems, and collect training trajectories, yielding 568 games, 3,436 tasks, and 307 hours of gameplay. Training on these trajectories raises a 9B model’s progress from 14.3% to 31.2% on held-out PlayerBench games, with gains that transfer to GameWorld. The environment–player feedback loop improves task verification while producing data for visual agent training.

Sign up or log in to comment

Models citing this paper 2

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2611.24154 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2611.24154 in a Space README.md to link it from this page.

Collections including this paper 1