PyoSignal Logo
PyoSignal
Back to Research

OpenGame: Open Agentic Coding for Games

Paper ID: 2604.18394 β€’ 49 Upvotes
Agent Game Development Code Generation LLM Evaluation Safety
OpenGame: Open Agentic Coding for Games

πŸ“ 핡심 μš”μ•½

OpenGame은 κ²Œμž„ κ°œλ°œμ„ μœ„ν•œ μ—μ΄μ „νŠΈ ν”„λ ˆμž„μ›Œν¬λ‘œ, κ²Œμž„ μ—”μ§„ λ§ˆμŠ€ν„°λ§μ— νŠΉν™”λœ μ½”λ“œ LLMκ³Ό μžλ™ 디버깅 μ‹œμŠ€ν…œμ„ 톡해 λ³΅μž‘ν•œ κ²Œμž„ κ°œλ°œμ„ μ§€μ›ν•˜λ©°, κ°œλ°œμžκ°€ μ‹€μ œ κ²Œμž„ κ°œλ°œμ— LLM을 ν™œμš©ν•  수 μžˆλ„λ‘ λ•λŠ”λ‹€.

πŸ“– 상세 λ‚΄μš©

κΈ°μ‘΄ LLM은 λ³΅μž‘ν•œ κ²Œμž„ κ°œλ°œμ—μ„œ 파일 κ°„ 뢈일치, 씬 μ—°κ²° 였λ₯˜, 논리적 λͺ¨μˆœ λ“±μ˜ 문제둜 어렀움을 κ²ͺμŠ΅λ‹ˆλ‹€. μ΄λŸ¬ν•œ 문제λ₯Ό ν•΄κ²°ν•˜κΈ° μœ„ν•΄, λ³Έ λ…Όλ¬Έμ—μ„œλŠ” μ›Ή κ²Œμž„ κ°œλ°œμ„ μœ„ν•œ μ˜€ν”ˆ μ†ŒμŠ€ μ—μ΄μ „νŠΈ ν”„λ ˆμž„μ›Œν¬μΈ OpenGame을 μ œμ‹œν•©λ‹ˆλ‹€. OpenGame은 ν”„λ‘œμ νŠΈ ꡬ쑰λ₯Ό ν•™μŠ΅ν•˜λŠ” Template Skillκ³Ό κ²€μ¦λœ μˆ˜μ • 사항을 κΈ°λ‘ν•˜λŠ” Debug Skill을 톡해 μ•ˆμ •μ μΈ μ•„ν‚€ν…μ²˜λ₯Ό κ΅¬μΆ•ν•˜κ³  톡합 였λ₯˜λ₯Ό μ²΄κ³„μ μœΌλ‘œ μˆ˜μ •ν•©λ‹ˆλ‹€. λ˜ν•œ, κ²Œμž„ μ—”μ§„ λ§ˆμŠ€ν„°λ§μ„ μœ„ν•΄ GameCoder-27BλΌλŠ” μ½”λ“œ LLM을 κ°œλ°œν•˜κ³ , Build Health, Visual Usability, Intent Alignmentλ₯Ό ν‰κ°€ν•˜λŠ” OpenGame-Benchλ₯Ό λ„μž…ν–ˆμŠ΅λ‹ˆλ‹€. λ‹€μ–‘ν•œ κ²Œμž„ ν”„λ‘¬ν”„νŠΈμ—μ„œ OpenGame은 μƒˆλ‘œμš΄ SOTAλ₯Ό λ‹¬μ„±ν–ˆμŠ΅λ‹ˆλ‹€.

πŸ”‘ μ£Όμš” λ‚΄μš© (Key Points)

  • κ²Œμž„ κ°œλ°œμ— νŠΉν™”λœ μ—μ΄μ „νŠΈ ν”„λ ˆμž„μ›Œν¬ OpenGame μ œμ‹œ
  • ν”„λ‘œμ νŠΈ ꡬ쑰 ν•™μŠ΅ 및 μžλ™ 디버깅 μ‹œμŠ€ν…œ (Template Skill, Debug Skill) λ„μž…
  • κ²Œμž„ μ—”μ§„ λ§ˆμŠ€ν„°λ§μ„ μœ„ν•œ μ½”λ“œ LLM (GameCoder-27B) 개발 및 평가 νŒŒμ΄ν”„λΌμΈ (OpenGame-Bench) ꡬ좕

πŸ’‘ 싀무적 κ°€μΉ˜ (Relevance)

OpenGame은 κ²Œμž„ κ°œλ°œμžκ°€ LLM을 ν™œμš©ν•˜μ—¬ κ²Œμž„ 개발 ν”„λ‘œμ„ΈμŠ€λ₯Ό μžλ™ν™”ν•˜κ³  νš¨μœ¨μ„±μ„ 높일 수 μžˆλ„λ‘ μ§€μ›ν•˜λ©°, 특히 λ³΅μž‘ν•œ κ²Œμž„ 둜직 및 λ‹€μ–‘ν•œ 에셋 톡합 문제λ₯Ό ν•΄κ²°ν•˜λŠ” 데 도움을 쀄 수 μžˆλ‹€.

βœ… μΆ”μ²œ μ•‘μ…˜ (Actionable Items)

  • OpenGame ν”„λ ˆμž„μ›Œν¬λ₯Ό ν™œμš©ν•˜μ—¬ κ°„λ‹¨ν•œ μ›Ή κ²Œμž„ 개발 μ‹€ν—˜ μ§„ν–‰
  • GameCoder-27B LLM을 ν™œμš©ν•˜μ—¬ κ²Œμž„ 둜직 생성 및 μˆ˜μ • μ‹€ν—˜ μ§„ν–‰
  • OpenGame-Benchλ₯Ό ν™œμš©ν•˜μ—¬ κ°œλ°œν•œ κ²Œμž„μ˜ λΉŒλ“œ μƒνƒœ, μ‚¬μš©μ„±, μ˜λ„ μΌμΉ˜λ„ 평가 μ‹€ν—˜ μ§„ν–‰