AI 自己造游戏来练自己,成绩翻倍
训练 AI 玩游戏,最缺的不是算法,是数据:真人玩太贵,看视频又不知道手该往哪放。这篇干脆让 AI 自己动手——一个 AI 负责把现成的网页游戏改造成带明确任务和进度条的训练场,另一个 AI 负责进去玩、把卡住的地方反馈回去修,玩的过程既是验收也是攒数据。就这么自产自销,造出 568 个游戏、3436 个任务、15.7 万条轨迹,拿这些数据训练的 9B 模型,在测试游戏上进度从 14.3% 提到 31.2%,在外部基准上从 30% 提到 53.1%。它不是你明天能用上的东西,但它指向一条路:AI 的「经验」可以自己造、自己验、自己学,不用再等人类喂。
📄 原文摘要(英文)
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.