Reproduce “CreativeGame:Toward Mechanic-Aware Creative Game Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hongnan Ma, Han Wang, Shenglin Wang, Tieyue Yin, Yiwei Shi, Yucong Huang, Yingtian Zou, Muning Wen and 1 more
CreativeGame introduces a multi-agent system for iterative HTML5 game generation, leveraging a programmatic CreativeProxyReward and lineage-aware memory to foster mechanic-aware evolution. The system generates playable games, demonstrating how core game mechanics can be reinterpreted and progressively transformed across successive versions.
Key insights
- Explicitly treating game mechanics as planning objects, rather than post-hoc descriptions, enables more interpretable and structurally innovative game evolution.
- Combining multi-agent collaboration with a programmatic reward function and lineage-scoped memory allows for the progressive and observable evolution of complex game systems.
- Integrating runtime validation directly into the iterative generation loop significantly improves the reliability and playability of LLM-generated game code.
- Meaningful creative changes often manifest as reinterpretations of existing mechanics and shifts to explicit causal structures, rather than superficial aesthetic modifications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.