CreativeGame: Mechanic-Aware Iterative Game Generation with Multi-Agent System
A recent research publication presents CreativeGame, a multi-agent framework designed for the iterative creation of HTML5 games, which overcomes the challenges associated with game development reliant on large language models. This system addresses problems like fragile runtime behavior from one-time generation, insufficient experience accumulation across versions, and inconsistent creativity assessments. CreativeGame features four essential elements: a proxy reward system based on programmatic signals instead of solely LLM evaluations, lineage-scoped memory for experience retention across iterations, integrated runtime validation within repair and reward mechanisms, and mechanic-guided planning that treats game mechanics as distinct entities for planning, tracking, preservation, and assessment. The paper can be found on arXiv with the identifier 2604.19926.
Key facts
- CreativeGame is a multi-agent system for iterative HTML5 game generation.
- It addresses brittle runtime behavior from single-shot generation.
- It uses a proxy reward centered on programmatic signals.
- Lineage-scoped memory enables cross-version experience accumulation.
- Runtime validation is integrated into repair and reward processes.
- Mechanic-guided planning treats mechanics as explicit objects.
- The paper is published on arXiv with identifier 2604.19926.
- The system aims to improve iterative creative improvement in game generation.
Entities
Institutions
- arXiv