用强化学习让智能体通关宝可梦红版前半段，揭示奖励设计漏洞。
Pokemon Red via Reinforcement Learning
- 构建简化环境与DRL训练流程，实现基础通关。
- 在数万步长周期中完成初代城市区域，验证可行性。
- 暴露奖励信号漏洞，适合研究大模型与分层算法。
《宝可梦红》作为经典Game BoyJRPG，为智能体测试带来多重挑战：多任务并行、长达数万步的长期决策、困难探索及庞大策略空间。本文提出一种简化环境与深度强化学习（DRL）训练方法，展示一个基准智能体成功完成游戏前半段至完成水都城（Cerulean City）的目标。实验包含多种消融分析，揭示奖励设计中的脆弱性，发现智能体会利用特定奖励信号进行投机行为。此外，论文讨论现有局限，并指出类似《宝可梦》的游戏对未来大型语言模型代理、层次化训练算法和先进探索方法的研究具有巨大潜力。源代码见：https://github.com/MarcoMeter/neroRL/tree/poke_red
原文摘要 · Abstract (English)
Pokémon Red, a classic Game Boy JRPG, presents significant challenges as a testbed for agents, including multi-tasking, long horizons of tens of thousands of steps, hard exploration, and a vast array of potential policies. We introduce a simplistic environment and a Deep Reinforcement Learning (DRL) training methodology, demonstrating a baseline agent that completes an initial segment of the game up to completing Cerulean City. Our experiments include various ablations that reveal vulnerabilities in reward shaping, where agents exploit specific reward signals. We also discuss limitations and argue that games like Pokémon hold strong potential for future research on Large Language Model agents, hierarchical training algorithms, and advanced exploration methods. Source Code: https://github.com/MarcoMeter/neroRL/tree/poke_red
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。