[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-FLAIROx--JaxMARL":3,"tool-FLAIROx--JaxMARL":64},[4,17,27,35,48,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,43,44,45,15,46,26,13,47],"数据工具","视频","插件","其他","音频",{"id":49,"name":50,"github_repo":51,"description_zh":52,"stars":53,"difficulty_score":10,"last_commit_at":54,"category_tags":55,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,46],{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},2181,"OpenHands","OpenHands\u002FOpenHands","OpenHands 是一个专注于 AI 驱动开发的开源平台，旨在让智能体（Agent）像人类开发者一样理解、编写和调试代码。它解决了传统编程中重复性劳动多、环境配置复杂以及人机协作效率低等痛点，通过自动化流程显著提升开发速度。\n\n无论是希望提升编码效率的软件工程师、探索智能体技术的研究人员，还是需要快速原型验证的技术团队，都能从中受益。OpenHands 提供了灵活多样的使用方式：既可以通过命令行（CLI）或本地图形界面在个人电脑上轻松上手，体验类似 Devin 的流畅交互；也能利用其强大的 Python SDK 自定义智能体逻辑，甚至在云端大规模部署上千个智能体并行工作。\n\n其核心技术亮点在于模块化的软件智能体 SDK，这不仅构成了平台的引擎，还支持高度可组合的开发模式。此外，OpenHands 在 SWE-bench 基准测试中取得了 77.6% 的优异成绩，证明了其解决真实世界软件工程问题的能力。平台还具备完善的企业级功能，支持与 Slack、Jira 等工具集成，并提供细粒度的权限管理，适合从个人开发者到大型企业的各类用户场景。",70612,"2026-04-05T11:12:22",[26,15,13,45],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":79,"owner_email":79,"owner_twitter":79,"owner_website":79,"owner_url":80,"languages":81,"stars":112,"forks":113,"last_commit_at":114,"license":115,"difficulty_score":10,"env_os":116,"env_gpu":117,"env_ram":116,"env_deps":118,"category_tags":131,"github_topics":132,"view_count":10,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":136,"updated_at":137,"faqs":138,"releases":173},1052,"FLAIROx\u002FJaxMARL","JaxMARL","Multi-Agent Reinforcement Learning with JAX","JaxMARL是一个专注于多智能体强化学习（MARL）的开源工具库，基于JAX框架构建，旨在简化多智能体环境下的算法开发与实验验证。它集成了多种经典MARL环境（如MPE、Overcooked、Multi-Agent Brax等）和主流算法，支持高效训练与评估。针对多智能体系统中环境复杂度高、协作机制难建模等问题，JaxMARL通过提供标准化接口和预置环境，降低开发门槛，同时利用JAX的GPU加速能力提升计算效率。  \n\n该工具特别引入SMAX环境，简化了StarCraft Multi-Agent Challenge的实现，无需依赖游戏引擎即可进行实验。适合研究人员和开发者快速验证MARL算法效果，尤其适用于需要对比不同方法在协作、竞争或通信任务中表现的场景。其丰富的文档和交互式教程（如Colab笔记）进一步降低了上手难度。JaxMARL的模块化设计和灵活的环境扩展性，使其成为多智能体强化学习领域的实用工具。","\u003Ch1 align=\"center\">JaxMARL\u003C\u002Fh1>\n\n\u003Cp align=\"center\">\n       \u003Ca href=\"https:\u002F\u002Fpypi.python.org\u002Fpypi\u002Fjaxmarl\">\n        \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fpyversions\u002Fjaxmarl.svg\" \u002F>\u003C\u002Fa>\n       \u003Ca href=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fjaxmarl\">\n        \u003Cimg src=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fjaxmarl.svg\" \u002F>\u003C\u002Fa>\n       \u003Ca href= \"https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002FLICENSE\">\n        \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-Apache2.0-blue.svg\" \u002F>\u003C\u002Fa>\n       \u003Ca href= \"https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002Fjaxmarl\u002Ftutorials\u002FJaxMARL_Walkthrough.ipynb\">\n        \u003Cimg src=\"https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg\" \u002F>\u003C\u002Fa>\n       \u003Ca href= \"https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.10090\">\n        \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FarXiv-2311.10090-b31b1b.svg\" \u002F>\u003C\u002Fa>\n       \u003Ca href= \"https:\u002F\u002Fjaxmarl.foersterlab.com\u002F\">\n        \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-green\" \u002F>\u003C\u002Fa>\n       \n\u003C\u002Fp>\n\n[**Installation**](#install) | [**Quick Start**](#start) | [**Environments**](#environments) | [**Algorithms**](#algorithms) | [**Citation**](#cite)\n---\n\n\u003Cdiv class=\"collage\">\n    \u003Cdiv class=\"column\" align=\"centre\">\n        \u003Cdiv class=\"row\" align=\"centre\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_d9474378e31f.gif\" alt=\"Overcooked\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_d98398151fb7.png\" alt=\"mabrax\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_b09629a485f6.gif\" alt=\"STORM\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_1dcbdfa0569a.png\" alt=\"hanabi\" width=\"20%\">\n        \u003C\u002Fdiv>\n        \u003Cdiv class=\"row\" align=\"centre\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_3bd0062c375e.png\" alt=\"coin_game\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_050111482457.gif\" alt=\"MPE\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_3e3947af079d.gif\" alt=\"jaxnav\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_066fbfdec8cf.gif\" alt=\"SMAX\" width=\"20%\">\n        \u003C\u002Fdiv>\n    \u003C\u002Fdiv>\n\u003C\u002Fdiv>\n\n## Multi-Agent Reinforcement Learning in JAX\n\nJaxMARL combines ease-of-use with GPU-enabled efficiency, and supports a wide range of commonly used MARL environments as well as popular baseline algorithms. Our aim is for one library that enables thorough evaluation of MARL methods across a wide range of tasks and against relevant baselines. We also introduce SMAX, a vectorised, simplified version of the popular StarCraft Multi-Agent Challenge, which removes the need to run the StarCraft II game engine. \n\nFor more details, take a look at our [blog post](https:\u002F\u002Fblog.foersterlab.com\u002Fjaxmarl\u002F) or our [Colab notebook](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002Fjaxmarl\u002Ftutorials\u002FJaxMARL_Walkthrough.ipynb), which walks through the basic usage.\n\n\u003Ch2 name=\"environments\" id=\"environments\">Environments 🌍 \u003C\u002Fh2>\n\n| Environment | Reference | README | Summary |\n| --- | --- | --- | --- |\n| 🔴 MPE | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1706.02275) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fmpe) | Communication orientated tasks in a multi-agent particle world\n| 🍲 Overcooked | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1910.05789) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fovercooked) | Fully-cooperative human-AI coordination tasks based on the video game of the same name | \n| 🥘 OvercookedV2 | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.17821) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fovercooked_v2) | Partially observable and stochastic extention of Overcooked. Fully-cooperative. | \n| 🦾 Multi-Agent Brax | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2003.06709) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fmabrax) | Continuous multi-agent robotic control based on Brax, analogous to Multi-Agent MuJoCo |\n| 🎆 Hanabi | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1902.00506) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fhanabi) | Fully-cooperative partially-observable multiplayer card game |\n| 👾 SMAX | Novel | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fsmax) | Simplified cooperative StarCraft micro-management environment |\n| 🧮 STORM: Spatial-Temporal Representations of Matrix Games | [Paper](https:\u002F\u002Fopenreview.net\u002Fforum?id=54F8woU8vhq) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fstorm) | Matrix games represented as grid world scenarios\n| 🧭 JaxNav | [Paper](https:\u002F\u002Fwww.arxiv.org\u002Fabs\u002F2408.15099) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fjaxnav) | 2D geometric navigation for differential drive robots\n| 🪙 Coin Game | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1802.09640) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fcoin_game) | Two-player grid world environment which emulates social dilemmas\n| 💡 Switch Riddle | [Paper](https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper_files\u002Fpaper\u002F2016\u002Fhash\u002Fc7635bfd99248a2cdef8249ef7bfbef4-Abstract.html) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fswitch_riddle) | Simple cooperative communication game included for debugging\n| 🤖 JaxRobotarium | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.06771) | [Source](https:\u002F\u002Fgithub.com\u002FGT-STAR-Lab\u002FJaxRobotarium) | Multi-robot environment with open access sim2real through the [Robotarium](https:\u002F\u002Fwww.robotarium.gatech.edu\u002F)\n \n\u003Ch2 name=\"algorithms\" id=\"algorithms\">Baseline Algorithms 🦉 \u003C\u002Fh2>\n\nWe follow CleanRL's philosophy of providing single file implementations which can be found within the `baselines` directory. We use Hydra to manage our config files, with specifics explained in each algorithm's README. Most files include `wandb` logging code, this is disabled by default but can be enabled within the file's config.\n\n| Algorithm | Reference | README | \n| --- | --- | --- | \n| IPPO | [Paper](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.09533.pdf) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FIPPO) | \n| MAPPO | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2103.01955) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FMAPPO) | \n| IQL | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1312.5602v1) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FQLearning) | \n| VDN | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1706.05296)  | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FQLearning) |\n| QMIX | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F1803.11485) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FQLearning) |\n| TransfQMIX | [Paper](https:\u002F\u002Fwww.southampton.ac.uk\u002F~eg\u002FAAMAS2023\u002Fpdfs\u002Fp1679.pdf) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FQLearning) |\n| SHAQ | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2105.15013) | [Source](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FQLearning) |\n| PQN-VDN | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.04811) | [Source](https:\u002F\u002Fgithub.com\u002Fmttga\u002Fpurejaxql) |\n\n\u003Ch2 name=\"install\" id=\"install\">Installation 🧗 \u003C\u002Fh2>\n\n**Environments** - Before installing, ensure you have the correct [JAX installation](https:\u002F\u002Fgithub.com\u002Fgoogle\u002Fjax#installation) for your hardware accelerator. We have tested up to JAX version 0.4.36. The JaxMARL environments can be installed directly from PyPi:\n\n``` bash\npip install jaxmarl \n```\n\n**Algorithms** - If you would like to also run the algorithms, install the source code as follows:\n\n1. Clone the repository:\n    ``` bash\n    git clone https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL.git && cd JaxMARL\n    ```\n2. Install requirements:\n    ``` bash\n    pip install -e .[algs]\n    export PYTHONPATH=.\u002FJaxMARL:$PYTHONPATH\n    ```\n3. For the fastest start, we recommend using our Dockerfile, the usage of which is outlined below.\n\n**Development** - If you would like to run our test suite, install the additonal dependencies with:\n `pip install -e .[dev]`, after cloning the repository.\n\n\u003Ch2 name=\"start\" id=\"start\">Quick Start 🚀 \u003C\u002Fh2>\n\nWe take inspiration from the [PettingZoo](https:\u002F\u002Fgithub.com\u002FFarama-Foundation\u002FPettingZoo) and [Gymnax](https:\u002F\u002Fgithub.com\u002FRobertTLange\u002Fgymnax) interfaces. You can try out training an agent in our [Colab notebook](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002Fjaxmarl\u002Ftutorials\u002FJaxMARL_Walkthrough.ipynb). Further introduction scripts can be found [here](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Ftutorials).\n\n### Basic JaxMARL API  Usage 🖥️\n\nActions, observations, rewards and done values are passed as dictionaries keyed by agent name, allowing for differing action and observation spaces. The done dictionary contains an additional `\"__all__\"` key, specifying whether the episode has ended. We follow a parallel structure, with each agent passing an action at each timestep. For asynchronous games, such as Hanabi, a dummy action is passed for agents not acting at a given timestep.\n\n```python \nimport jax\nfrom jaxmarl import make\n\nkey = jax.random.PRNGKey(0)\nkey, key_reset, key_act, key_step = jax.random.split(key, 4)\n\n# Initialise environment.\nenv = make('MPE_simple_world_comm_v3')\n\n# Reset the environment.\nobs, state = env.reset(key_reset)\n\n# Sample random actions.\nkey_act = jax.random.split(key_act, env.num_agents)\nactions = {agent: env.action_space(agent).sample(key_act[i]) for i, agent in enumerate(env.agents)}\n\n# Perform the step transition.\nobs, state, reward, done, infos = env.step(key_step, state, actions)\n```\n\n### Dockerfile 🐋\nTo help get experiments up and running we include a [Dockerfile](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002FDockerfile) and its corresponding [Makefile](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002FMakefile). With Docker and the [Nvidia Container Toolkit](https:\u002F\u002Fdocs.nvidia.com\u002Fdatacenter\u002Fcloud-native\u002Fcontainer-toolkit\u002Flatest\u002Findex.html) installed, the container can be built with:\n```\nmake build\n```\nThe built container can then be run:\n```\nmake run\n```\n\n## Contributing 🔨\nPlease contribute! Please take a look at our [contributing guide](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002FCONTRIBUTING.md) for how to add an environment\u002Falgorithm or submit a bug report. If you're looking for a project, we also have a few suggestions listed under the roadmap :) \n\n\u003Ch2 name=\"cite\" id=\"cite\">Citing JaxMARL 📜 \u003C\u002Fh2>\nIf you use JaxMARL in your work, please cite us as follows:\n\n``` bibtex\n@inproceedings{\n    flair2024jaxmarl,\n    title={JaxMARL: Multi-Agent RL Environments and Algorithms in JAX},\n    author={Alexander Rutherford and Benjamin Ellis and Matteo Gallici and Jonathan Cook and Andrei Lupu and Gar{\\dh}ar Ingvarsson and Timon Willi and Ravi Hammond and Akbir Khan and Christian Schroeder de Witt and Alexandra Souly and Saptarashmi Bandyopadhyay and Mikayel Samvelyan and Minqi Jiang and Robert Tjarko Lange and Shimon Whiteson and Bruno Lacerda and Nick Hawes and Tim Rockt{\\\"a}schel and Chris Lu and Jakob Nicolaus Foerster},\n    booktitle={The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track},\n    year={2024},\n}\n```\n\n## See Also 🙌\nThere are a number of other libraries which inspired this work, we encourage you to take a look!\n\nJAX-native algorithms:\n- [Mava](https:\u002F\u002Fgithub.com\u002Finstadeepai\u002FMava): JAX implementations of popular MARL algorithms.\n- [PureJaxRL](https:\u002F\u002Fgithub.com\u002Fluchris429\u002Fpurejaxrl): JAX implementation of PPO, and demonstration of end-to-end JAX-based RL training.\n\nJAX-native environments:\n- [Gymnax](https:\u002F\u002Fgithub.com\u002FRobertTLange\u002Fgymnax): Implementations of classic RL tasks including classic control, bsuite and MinAtar.\n- [Jumanji](https:\u002F\u002Fgithub.com\u002Finstadeepai\u002Fjumanji): A diverse set of environments ranging from simple games to NP-hard combinatorial problems.\n- [Pgx](https:\u002F\u002Fgithub.com\u002Fsotetsuk\u002Fpgx): JAX implementations of classic board games, such as Chess, Go and Shogi.\n- [Brax](https:\u002F\u002Fgithub.com\u002Fgoogle\u002Fbrax): A fully differentiable physics engine written in JAX, features continuous control tasks.\n- [XLand-MiniGrid](https:\u002F\u002Fgithub.com\u002Fcorl-team\u002Fxland-minigrid): Meta-RL gridworld environments inspired by XLand and MiniGrid.\n- [Craftax](https:\u002F\u002Fgithub.com\u002FMichaelTMatthews\u002FCraftax): (Crafter + NetHack) in JAX.\n","\u003Ch1 align=\"center\">JaxMARL\u003C\u002Fh1>\n\n\u003Cp align=\"center\">\n       \u003Ca href=\"https:\u002F\u002Fpypi.python.org\u002Fpypi\u002Fjaxmarl\">\n        \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fpyversions\u002Fjaxmarl.svg\" \u002F>\u003C\u002Fa>\n       \u003Ca href=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fjaxmarl\">\n        \u003Cimg src=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fjaxmarl.svg\" \u002F>\u003C\u002Fa>\n       \u003Ca href= \"https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002FLICENSE\">\n        \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-Apache2.0-blue.svg\" \u002F>\u003C\u002Fa>\n       \u003Ca href= \"https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002Fjaxmarl\u002Ftutorials\u002FJaxMARL_Walkthrough.ipynb\">\n        \u003Cimg src=\"https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg\" \u002F>\u003C\u002Fa>\n       \u003Ca href= \"https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.10090\">\n        \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FarXiv-2311.10090-b31b1b.svg\" \u002F>\u003C\u002Fa>\n       \u003Ca href= \"https:\u002F\u002Fjaxmarl.foersterlab.com\u002F\">\n        \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-green\" \u002F>\u003C\u002Fa>\n       \n\u003C\u002Fp>\n\n[**安装**](#install) | [**快速开始**](#start) | [**环境**](#environments) | [**算法**](#algorithms) | [**引用**](#cite)\n---\n\n\u003Cdiv class=\"collage\">\n    \u003Cdiv class=\"column\" align=\"centre\">\n        \u003Cdiv class=\"row\" align=\"centre\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_d9474378e31f.gif\" alt=\"Overcooked\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_d98398151fb7.png\" alt=\"mabrax\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_b09629a485f6.gif\" alt=\"STORM\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_1dcbdfa0569a.png\" alt=\"hanabi\" width=\"20%\">\n        \u003C\u002Fdiv>\n        \u003Cdiv class=\"row\" align=\"centre\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_3bd0062c375e.png\" alt=\"coin_game\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_050111482457.gif\" alt=\"MPE\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_3e3947af079d.gif\" alt=\"jaxnav\" width=\"20%\">\n            \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_readme_066fbfdec8cf.gif\" alt=\"SMAX\" width=\"20%\">\n        \u003C\u002Fdiv>\n    \u003C\u002Fdiv>\n\u003C\u002Fdiv>\n\n## 基于JAX的多智能体强化学习库\n\nJaxMARL（基于JAX的多智能体强化学习库）结合了易用性与GPU加速的高效性能，支持多种常用的MARL环境及主流基线算法。我们的目标是提供一个统一库，可在广泛任务中对MARL方法进行系统评估。我们还引入了SMAX环境，这是经典StarCraft多智能体挑战的向量化简化版本，无需运行StarCraft II游戏引擎。\n\n更多细节请参阅我们的[博客文章](https:\u002F\u002Fblog.foersterlab.com\u002Fjaxmarl\u002F)或[Colab教程](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002Fjaxmarl\u002Ftutorials\u002FJaxMARL_Walkthrough.ipynb)。\n\n\u003Ch2 name=\"environments\" id=\"environments\">环境 🌍 \u003C\u002Fh2>\n\n| 环境 | 参考文献 | 源码 | 简介 |\n| --- | --- | --- | --- |\n| 🔴 MPE | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1706.02275) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fmpe) | 多智能体粒子世界的通信导向任务 |\n| 🍲 Overcooked | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1910.05789) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fovercooked) | 基于同名游戏的全协作人机协作任务 | \n| 🥘 OvercookedV2 | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.17821) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fovercooked_v2) | Overcooked的局部可观测与随机扩展，全协作模式 | \n| 🦾 Multi-Agent Brax | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2003.06709) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fmabrax) | 基于Brax的连续多智能体机器人控制（类比Multi-Agent MuJoCo） |\n| 🎆 Hanabi | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1902.00506) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fhanabi) | 全协作局部可观测多人卡牌游戏 |\n| 👾 SMAX | 原创实现 | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fsmax) | 简化的协作型星际争霸微观管理环境 |\n| 🧮 STORM: 矩阵游戏时空表征 | [论文](https:\u002F\u002Fopenreview.net\u002Fforum?id=54F8woU8vhq) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fstorm) | 以网格世界形式呈现的矩阵游戏 |\n| 🧭 JaxNav | [论文](https:\u002F\u002Fwww.arxiv.org\u002Fabs\u002F2408.15099) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fjaxnav) | 差分驱动机器人的二维几何导航 |\n| 🪙 Coin Game | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1802.09640) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fcoin_game) | 模拟社会困境的双人网格世界环境 |\n| 💡 Switch Riddle | [论文](https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper_files\u002Fpaper\u002F2016\u002Fhash\u002Fc7635bfd99248a2cdef8249ef7bfbef4-Abstract.html) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Fenvironments\u002Fswitch_riddle) | 用于调试的简单协作通信游戏 |\n| 🤖 JaxRobotarium | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.06771) | [源码](https:\u002F\u002Fgithub.com\u002FGT-STAR-Lab\u002FJaxRobotarium) | 支持开放访问sim2real的多机器人环境（基于Robotarium平台） |\n\n\u003Ch2 name=\"algorithms\" id=\"algorithms\">基线算法 🦉 \u003C\u002Fh2>\n\n我们遵循CleanRL的单文件实现理念，所有算法实现位于`baselines`目录。使用Hydra管理配置文件，各算法README包含具体说明。大部分文件包含`wandb`日志记录代码（默认关闭，可通过配置启用）。\n\n| 算法 | 参考文献 | 源码 | \n| --- | --- | --- | \n| IPPO | [论文](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.09533.pdf) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FIPPO) | \n| MAPPO | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2103.01955) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FMAPPO) | \n| IQL | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1312.5602v1) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FQLearning) | \n| VDN | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1706.05296)  | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FQLearning) |\n| QMIX | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1803.11485) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FQLearning) |\n| TransfQMIX | [论文](https:\u002F\u002Fwww.southampton.ac.uk\u002F~eg\u002FAAMAS2023\u002Fpdfs\u002Fp1679.pdf) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FQLearning) |\n| SHAQ | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2105.15013) | [源码](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fbaselines\u002FQLearning) |\n| PQN-VDN | [论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.04811) | [源码](https:\u002F\u002Fgithub.com\u002Fmttga\u002Fpurejaxql) |\n\n\u003Ch2 name=\"install\" id=\"install\">安装 🧗 \u003C\u002Fh2>\n\n**环境安装** - 安装前请确保已根据硬件加速器配置正确的[JAX环境](https:\u002F\u002Fgithub.com\u002Fgoogle\u002Fjax#installation)（已测试至JAX 0.4.36版本）。可通过PyPi直接安装JaxMARL环境：\n\n``` bash\npip install jaxmarl \n```\n\n**算法运行** - 如需运行算法，请按以下步骤安装源码：\n\n1. 克隆仓库：\n    ``` bash\n    git clone https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL.git && cd JaxMARL\n    ```\n2. 安装依赖：\n    ``` bash\n    pip install -e .[algs]\n    export PYTHONPATH=.\u002FJaxMARL:$PYTHONPATH\n    ```\n3. 推荐使用我们的Dockerfile快速启动（具体用法见下文）\n\n**开发模式** - 如需运行测试套件，克隆仓库后通过以下命令安装开发依赖：\n`pip install -e .[dev]`\n\n\u003Ch2 name=\"start\" id=\"start\">快速入门 🚀 \u003C\u002Fh2>\n\n我们借鉴了[PettingZoo（多智能体强化学习环境库）](https:\u002F\u002Fgithub.com\u002FFarama-Foundation\u002FPettingZoo)和[Gymnax（纯JAX环境库）](https:\u002F\u002Fgithub.com\u002FRobertTLange\u002Fgymnax)的接口设计。可通过[Colab教程](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002Fjaxmarl\u002Ftutorials\u002FJaxMARL_Walkthrough.ipynb)体验智能体训练，更多示例脚本见[教程目录](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Ftree\u002Fmain\u002Fjaxmarl\u002Ftutorials)。\n\n### 基础JaxMARL API 使用示例 🖥️\n\n动作、观测、奖励和终止信号均以智能体名称为键的字典形式传递，支持不同动作\u002F观测空间。终止字典包含额外的`\"__all__\"`键表示全局终止状态。我们采用并行结构，每个时间步所有智能体同步执行动作。对于异步游戏（如Hanabi），非行动智能体将接收空动作。\n\n```python \nimport jax\nfrom jaxmarl import make\n\nkey = jax.random.PRNGKey(0)\nkey, key_reset, key_act, key_step = jax.random.split(key, 4)\n\n# 初始化环境\nenv = make('MPE_simple_world_comm_v3')\n```\n\n# 重置环境。\nobs, state = env.reset(key_reset)\n\n# 采样随机动作。\nkey_act = jax.random.split(key_act, env.num_agents)\nactions = {agent: env.action_space(agent).sample(key_act[i]) for i, agent in enumerate(env.agents)}\n\n# 执行步进转换。\nobs, state, reward, done, infos = env.step(key_step, state, actions)\n```\n\n### Dockerfile（容器配置文件） 🐋\n为帮助实验快速部署，我们提供了[Dockerfile](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002FDockerfile)及其配套的[Makefile](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002FMakefile)。在安装Docker和[Nvidia容器工具包](https:\u002F\u002Fdocs.nvidia.com\u002Fdatacenter\u002Fcloud-native\u002Fcontainer-toolkit\u002Flatest\u002Findex.html)后，可通过以下命令构建容器：\n```\nmake build\n```\n构建完成后可通过以下命令运行容器：\n```\nmake run\n```\n\n## 贡献指南 🔨\n欢迎贡献！请查看我们的[贡献指南](https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fblob\u002Fmain\u002FCONTRIBUTING.md)了解如何添加环境\u002F算法或提交缺陷报告。如果您正在寻找项目方向，路线图下也列出了一些开发建议 :)\n\n\u003Ch2 name=\"cite\" id=\"cite\">引用JaxMARL 📜 \u003C\u002Fh2>\n若您在研究中使用了JaxMARL，请通过以下方式引用：\n\n``` bibtex\n@inproceedings{\n    flair2024jaxmarl,\n    title={JaxMARL: Multi-Agent RL Environments and Algorithms in JAX},\n    author={Alexander Rutherford and Benjamin Ellis and Matteo Gallici and Jonathan Cook and Andrei Lupu and Gar{\\dh}ar Ingvarsson and Timon Willi and Ravi Hammond and Akbir Khan and Christian Schroeder de Witt and Alexandra Souly and Saptarashmi Bandyopadhyay and Mikayel Samvelyan and Minqi Jiang and Robert Tjarko Lange and Shimon Whiteson and Bruno Lacerda and Nick Hawes and Tim Rockt{\\\"a}schel and Chris Lu and Jakob Nicolaus Foerster},\n    booktitle={The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track},\n    year={2024},\n}\n```\n\n## 相关项目 🙌\n本项目受到以下库的启发，推荐您查看这些优秀作品：\n\n原生JAX算法：\n- [Mava](https:\u002F\u002Fgithub.com\u002Finstadeepai\u002FMava)：JAX实现的主流MARL算法\n- [PureJaxRL](https:\u002F\u002Fgithub.com\u002Fluchris429\u002Fpurejaxrl)：PPO的JAX实现，展示端到端JAX强化学习训练\n\n原生JAX环境：\n- [Gymnax](https:\u002F\u002Fgithub.com\u002FRobertTLange\u002Fgymnax)：包含经典控制、bsuite和MinAtar任务的经典RL实现\n- [Jumanji](https:\u002F\u002Fgithub.com\u002Finstadeepai\u002Fjumanji)：涵盖简单游戏到NP难组合问题的多样化环境\n- [Pgx](https:\u002F\u002Fgithub.com\u002Fsotetsuk\u002Fpgx)：JAX实现的国际象棋、围棋、将棋等经典棋盘游戏\n- [Brax](https:\u002F\u002Fgithub.com\u002Fgoogle\u002Fbrax)：用JAX编写的全微分物理引擎，支持连续控制任务\n- [XLand-MiniGrid](https:\u002F\u002Fgithub.com\u002Fcorl-team\u002Fxland-minigrid)：受XLand和MiniGrid启发的元强化学习网格世界\n- [Craftax](https:\u002F\u002Fgithub.com\u002FMichaelTMatthews\u002FCraftax)：JAX实现的Crafter+NetHack游戏","# JaxMARL 快速上手指南\n\n---\n\n## 环境准备\n### 系统要求\n- Python ≥ 3.9\n- JAX ≥ 0.4.36（需根据硬件加速器安装对应版本）\n- 推荐使用 Linux 系统（支持 GPU 加速）\n\n### 前置依赖\n```bash\n# 安装 JAX（推荐使用清华镜像加速）\npip install --upgrade pip\npip install jax --index-url https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n\n# 安装基础工具（如使用 Docker 可跳过）\nsudo apt-get update && sudo apt-get install -y git\n```\n\n---\n\n## 安装步骤\n### 仅安装环境（推荐国内用户）\n```bash\npip install jaxmarl -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n### 安装完整版（含算法）\n```bash\n# 克隆仓库\ngit clone https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL.git && cd JaxMARL\n\n# 安装依赖（国内用户可加 -i 参数指定镜像源）\npip install -e .[algs]\nexport PYTHONPATH=.\u002FJaxMARL:$PYTHONPATH\n```\n\n### 使用 Docker（推荐快速启动）\n```bash\n# 构建镜像（需提前安装 Docker 和 Nvidia Container Toolkit）\nmake build\n\n# 运行容器\nmake run\n```\n\n---\n\n## 基本使用\n### 最简环境测试\n```python\nimport jax\nfrom jaxmarl import make\n\nkey = jax.random.PRNGKey(0)\nkey, key_reset, key_act, key_step = jax.random.split(key, 4)\n\n# 初始化环境\nenv = make('MPE_simple_world_comm_v3')  # 使用 PettingZoo 风格接口\n\n# 重置环境\nobs, state = env.reset(key_reset)\n\n# 采样随机动作\nkey_act = jax.random.split(key_act, env.num_agents)\nactions = {agent: env.action_space(agent).sample(key_act[i]) \n           for i, agent in enumerate(env.agents)}\n\n# 执行一步\nobs, state, reward, done, infos = env.step(key_step, state, actions)\n```\n\n### 运行教程示例\n```bash\n# 本地运行 Colab 笔记本（需安装 jupyter）\npip install jupyter\njupyter notebook jaxmarl\u002Ftutorials\u002FJaxMARL_Walkthrough.ipynb\n```\n\n---\n\n> 📌 提示：  \n> 1. 国内用户安装依赖时建议添加 `-i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple` 使用清华镜像加速  \n> 2. Docker 用户可通过 `make run` 直接进入预配置环境  \n> 3. 环境列表详见官方文档：https:\u002F\u002Fjaxmarl.foersterlab.com\u002F","一家游戏开发公司正在研发一款需要多个AI代理协作完成任务的策略类手游，团队需要训练智能体在动态环境中与玩家协同完成资源收集与分配。\n\n### 没有 JaxMARL 时\n- 需要从头实现复杂的多智能体强化学习算法（如QMIX、MAPPO），耗费大量研发周期\n- 环境配置依赖StarCraft II等重型游戏引擎，部署和调试过程繁琐\n- 训练效率受限于CPU计算能力，单次完整训练需要12小时以上\n- 论文复现困难，不同算法的超参数调整需要反复验证\n- 多人协作场景中的通信机制需要自行设计底层实现\n\n### 使用 JaxMARL 后\n- 直接调用内置的QMIX、IPPO等算法模块，核心训练代码缩减至50行以内\n- 通过`jaxmarl.make()`一键启动SMAX环境，无需安装StarCraft II客户端\n- 利用JAX的XLA编译加速，在相同硬件条件下训练时间缩短至2.5小时\n- 通过配置YAML文件即可复现论文中的超参数组合，支持一键切换算法对比\n- 使用预置的通信协议模块，快速实现智能体间的策略协调\n\nJaxMARL显著降低了多智能体强化学习的研究与应用门槛，通过模块化设计和JAX生态优势，使开发者能够专注于算法创新而非基础设施搭建。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FFLAIROx_JaxMARL_d9474378.gif","FLAIROx","Foerster Lab for AI Research","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FFLAIROx_e185e9f2.png","",null,"https:\u002F\u002Fgithub.com\u002FFLAIROx",[82,86,90,94,98,102,105,109],{"name":83,"color":84,"percentage":85},"Python","#3572A5",69.8,{"name":87,"color":88,"percentage":89},"HTML","#e34c26",21.9,{"name":91,"color":92,"percentage":93},"Jupyter Notebook","#DA5B0B",7.1,{"name":95,"color":96,"percentage":97},"JavaScript","#f1e05a",1,{"name":99,"color":100,"percentage":101},"CSS","#663399",0.1,{"name":103,"color":104,"percentage":101},"Dockerfile","#384d54",{"name":106,"color":107,"percentage":108},"Makefile","#427819",0,{"name":110,"color":111,"percentage":108},"Shell","#89e051",780,143,"2026-04-01T17:52:26","Apache-2.0","未说明","需要 NVIDIA GPU 和 CUDA 支持（需根据 JAX 安装要求配置）",{"notes":119,"python":120,"dependencies":121},"需预先安装与硬件匹配的 JAX 版本；推荐使用 Docker 部署；部分算法需额外依赖如 Brax 和 PettingZoo；使用 Hydra 管理配置文件","3.8+（根据 JAX 常规需求推断）",[122,123,124,125,126,127,128,129,130],"jax","flax","optax","gymnax","chex","wandb","hydra-core","brax","pettingzoo",[15],[133,134,135],"marl","multiagent-reinforcement-learning","reinforcement-learning","2026-03-27T02:49:30.150509","2026-04-06T07:13:21.360924",[139,144,149,154,159,164,169],{"id":140,"question_zh":141,"answer_zh":142,"source_url":143},4687,"如何解决Jax版本冲突导致CUDA不可用的问题？","将`jax`版本限制调整为`jax>=0.4.17`以保持兼容性，同时建议在Dockerfile中指定具体版本。维护者已通过PR#97调整版本范围，避免与新版本库冲突。","https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fissues\u002F83",{"id":145,"question_zh":146,"answer_zh":147,"source_url":148},4688,"QMIX算法中RNN的隐藏状态重置逻辑是否正确？","Q-Learning脚本在更新步骤开始时重置环境，确保回放缓冲区轨迹从第一步开始。通过在环境重置后立即处理新观测值，避免使用旧episode的观测值。具体实现参考PR#127。","https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fissues\u002F119",{"id":150,"question_zh":151,"answer_zh":152,"source_url":153},4689,"如何为STORM环境实现IPPO算法？","STORM环境的奖励是个人奖励而非团队奖励。维护者提供了IPPO实现的Colab笔记本，使用CNN提取特征，可通过[此链接](https:\u002F\u002Fcolab.research.google.com\u002Fdrive\u002F1coWWBMYQUukZvrdVurqaAGcCCeGsZQao)获取入门代码。","https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fissues\u002F74",{"id":155,"question_zh":156,"answer_zh":157,"source_url":158},4690,"为什么Jax版本被锁定为0.4.25？","原版本锁定是为了避免JAX API变动导致的兼容性问题。维护者已通过PR#123取消严格版本限制，允许使用更高版本，同时建议监控API变更。","https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fissues\u002F116",{"id":160,"question_zh":161,"answer_zh":162,"source_url":163},4691,"如何获取连续动作版本环境（如SMAX\u002FMPE）的性能基准？","维护者提供了连续动作IPPO实现的参考代码：https:\u002F\u002Fgithub.com\u002Famacrutherford\u002Fsampling-for-learnability\u002Fblob\u002Fmain\u002Fsfl\u002Ftrain\u002Fjaxnav_sfl.py。提交PR时建议包含与离散动作的对比曲线。","https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fissues\u002F128",{"id":165,"question_zh":166,"answer_zh":167,"source_url":168},4692,"运行教程文件时出现Flax配置错误如何解决？","错误由Flax版本不兼容导致，需升级Flax至最新版本。问题已通过PR#123修复，具体更新命令参考仓库的安装指南。","https:\u002F\u002Fgithub.com\u002FFLAIROx\u002FJaxMARL\u002Fissues\u002F122",{"id":170,"question_zh":171,"answer_zh":172,"source_url":163},4693,"如何实现连续动作空间的MAPPO算法？","可基于现有离散动作MAPPO代码修改，维护者建议参考IPPO连续动作实现（链接同上），并欢迎提交PR进行代码审查。",[174,179,184,189,194,199,203],{"id":175,"version":176,"summary_zh":177,"released_at":178},104182,"v0.1.0","Correcting several environment interfaces to match the base interface hence moving to 0.1.0. See Changelog for exact details","2025-06-01T14:47:50",{"id":180,"version":181,"summary_zh":182,"released_at":183},104183,"v0.0.7","updated jax version","2024-12-19T09:44:34",{"id":185,"version":186,"summary_zh":187,"released_at":188},104184,"v0.0.5","Overcooked & Hanabi updates","2024-07-15T15:42:51",{"id":190,"version":191,"summary_zh":192,"released_at":193},104185,"v0.0.4","Added auto reset to specific state in environment base class and added JaxNav!","2024-06-15T14:02:27",{"id":195,"version":196,"summary_zh":197,"released_at":198},104186,"v0.0.3","fixes to hanabi, smax conic obs option","2024-04-02T14:10:27",{"id":200,"version":201,"summary_zh":79,"released_at":202},104187,"v0.0.2","2023-11-16T15:12:59",{"id":204,"version":205,"summary_zh":79,"released_at":206},104188,"v0.0.1","2023-11-16T14:29:19"]