GAMEGO: Training Game-Dev Agents with Synthetic Trajectories Anchored in Real-World Assets

查看原文 / View Original 编辑 / Edit

Computer Science > Artificial Intelligence arXiv:2610.06910 (cs) [Submitted on 2 Oct 2026] Title:GAMEGO: Training Game-Dev Agents with Synthetic Trajectories Anchored in Real-World Assets Authors:Haoyue Yang, Jingyao Li, Zhengfan Wu, Jing Liu, Xuanle Zhao, Kang Liu

计算机科学 > 人工智能 arXiv:2610.06910 (cs) [提交于 2026 年 10 月 2 日] 标题：GAMEGO：利用锚定真实世界资产的合成轨迹训练游戏开发智能体作者：Haoyue Yang, Jingyao Li, Zhengfan Wu, Jing Liu, Xuanle Zhao, Kang Liu

Abstract: Recent advances in Large Language Models (LLMs) have demonstrated remarkable capabilities in web front-end execution, with browser-based game generation emerging as a particularly prominent frontier. While previous efforts frequently rely on complex multi-turn workflows or focus on static game evaluation benchmarks, this work targets direct end-to-end real-world game synthesis driven by coding agents.

摘要：大型语言模型（LLMs）的最新进展在 Web 前端执行方面展现了卓越的能力，其中基于浏览器的游戏生成已成为一个尤为突出的前沿领域。尽管以往的研究往往依赖于复杂的多轮工作流或侧重于静态游戏评估基准，但本研究旨在实现由编码智能体驱动的直接端到端真实世界游戏合成。

However, generating complex games directly from sparse user queries often forces coding agents to make underspecified assumptions, yielding incomplete mechanics, disconnected gameplay flows, and limited visual aesthetics. To resolve this issue, this paper presents GameGo, a scalable framework that systematically transforms brief game seeds into comprehensive Product Requirements Documents grounded in industry game-development practices.

然而，直接从稀疏的用户查询中生成复杂游戏往往会迫使编码智能体做出不充分的假设，导致机制不完整、游戏流程脱节以及视觉美感有限。为了解决这一问题，本文提出了 GameGo，这是一个可扩展的框架，能够将简短的游戏种子系统地转化为基于行业游戏开发实践的综合产品需求文档。

To retain core gameplay constraints without restricting design exploration, GameGo uses task-specific dynamic compression to maximize information density while preserving instruction following. Based on this pipeline, GameGoData is constructed with 55,060 development trajectories across 2D, 2.5D, and 3D games, alongside GameGoBench, a benchmark comprising 124 diverse game queries.

为了在不限制设计探索的情况下保留核心游戏约束，GameGo 使用了特定任务的动态压缩技术，在保持指令遵循的同时最大化信息密度。基于此流水线，构建了包含 55,060 条涵盖 2D、2.5D 和 3D 游戏开发轨迹的 GameGoData，以及包含 124 个多样化游戏查询的基准测试集 GameGoBench。

Training GameGoCoder on GameGoData yields a model that outperforms matched baselines and is comparable to frontier models across gamedev benchmarks. All code, datasets, and models will be made publicly available.

在 GameGoData 上训练的 GameGoCoder 模型在各项游戏开发基准测试中表现优于匹配的基准模型，并可与前沿模型相媲美。所有代码、数据集和模型都将公开提供。