Problem
Research questions and friction points this paper is trying to address.
game generation
coding agents
end-to-end synthesis
sparse queries
underspecified assumptions
GAMEGO: Training Game-Dev Agents with Synthetic Trajectories Anchored in Real-World Assets
🤖 AI Summary
This study addresses the incompleteness and low quality of games generated by coding agents due to sparse user queries. To this end, we propose GameGoCoder, a novel framework that introduces an industry-practice-inspired requirement document transformation mechanism to expand brief game seeds into comprehensive requirement specifications. Furthermore, we construct a dataset and benchmark comprising 50,000 trajectories, and incorporate task-specific dynamic compression techniques to effectively balance constraint preservation with design space exploration. Experimental results demonstrate that GameGoCoder surpasses existing baselines across multiple game development benchmarks, achieving performance comparable to state-of-the-art large language models. All code, datasets, and models associated with this project have been fully open-sourced.
📝 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. 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. 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. 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.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Game Development Agent
Synthetic Trajectories
Dynamic Compression
Product Requirements Document
End-to-End Game Synthesis
🔎 Similar Papers
No similar papers found.
