[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-Unity-Technologies--ml-agents":3,"tool-Unity-Technologies--ml-agents":64},[4,17,27,35,44,52],{"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",140436,2,"2026-04-05T23:32:43",[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":10,"last_commit_at":41,"category_tags":42,"status":16},4292,"Deep-Live-Cam","hacksider\u002FDeep-Live-Cam","Deep-Live-Cam 是一款专注于实时换脸与视频生成的开源工具，用户仅需一张静态照片，即可通过“一键操作”实现摄像头画面的即时变脸或制作深度伪造视频。它有效解决了传统换脸技术流程繁琐、对硬件配置要求极高以及难以实时预览的痛点，让高质量的数字内容创作变得触手可及。\n\n这款工具不仅适合开发者和技术研究人员探索算法边界，更因其极简的操作逻辑（仅需三步：选脸、选摄像头、启动），广泛适用于普通用户、内容创作者、设计师及直播主播。无论是为了动画角色定制、服装展示模特替换，还是制作趣味短视频和直播互动，Deep-Live-Cam 都能提供流畅的支持。\n\n其核心技术亮点在于强大的实时处理能力，支持口型遮罩（Mouth Mask）以保留使用者原始的嘴部动作，确保表情自然精准；同时具备“人脸映射”功能，可同时对画面中的多个主体应用不同面孔。此外，项目内置了严格的内容安全过滤机制，自动拦截涉及裸露、暴力等不当素材，并倡导用户在获得授权及明确标注的前提下合规使用，体现了技术发展与伦理责任的平衡。",88924,"2026-04-06T03:28:53",[13,14,15,43],"视频",{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":23,"last_commit_at":50,"category_tags":51,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":53,"name":54,"github_repo":55,"description_zh":56,"stars":57,"difficulty_score":23,"last_commit_at":58,"category_tags":59,"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",[14,60,43,61,15,62,26,13,63],"数据工具","插件","其他","音频",{"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":80,"owner_url":81,"languages":82,"stars":113,"forks":114,"last_commit_at":115,"license":116,"difficulty_score":117,"env_os":118,"env_gpu":119,"env_ram":120,"env_deps":121,"category_tags":127,"github_topics":128,"view_count":23,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":136,"updated_at":137,"faqs":138,"releases":164},4116,"Unity-Technologies\u002Fml-agents","ml-agents","The Unity Machine Learning Agents Toolkit (ML-Agents) is an open-source project that enables games and simulations to serve as environments for training intelligent agents using deep reinforcement learning and imitation learning.","ml-agents 是 Unity 推出的一款开源工具包，旨在让游戏和仿真场景成为训练智能体的“演练场”。它解决了传统 AI 训练中环境构建困难、成本高昂的痛点，允许开发者直接在熟悉的 2D、3D 或 VR\u002FAR 项目中，利用深度强化学习和模仿学习技术培育出具备自主决策能力的智能角色。\n\n这套工具非常适合游戏开发者、AI 研究人员以及技术爱好者使用。对于开发者而言，它可以用于创建更聪明的非玩家角色（NPC）、实现自动化测试或验证游戏设计；研究人员则能借助其丰富的预设环境和灵活的 Python API，轻松复现前沿算法或开展多智能体协作与对抗研究。\n\nml-agents 的技术亮点在于其强大的兼容性与扩展性。它不仅内置了 PPO、SAC 等多种主流算法，支持从单智能体到复杂的多智能体博弈场景，还允许用户通过行为克隆等技术让 AI 直接模仿人类操作。此外，它支持与 Gym 和 PettingZoo 等标准接口无缝对接，并提供了跨平台的推理引擎，让训练好的模型能高效运行于各类设备之上。无论是想快速上手的新手，还是追求定制化算法的专家，都能在其中找到适合的工作流。","# Unity ML-Agents Toolkit\n\n[![docs badge](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-reference-blue.svg)](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest)\n\n[![license badge](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-Apache--2.0-green.svg)](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease\u002F4.0.0\u002FLICENSE.md)\n\n([latest release](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Freleases\u002Ftag\u002Flatest_release)) ([all releases](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Freleases))\n\n**The Unity Machine Learning Agents Toolkit** (ML-Agents) is an open-source project that enables games and simulations to serve as environments for training intelligent agents. We provide implementations (based on PyTorch) of state-of-the-art algorithms to enable game developers and hobbyists to easily train intelligent agents for 2D, 3D and VR\u002FAR games. Researchers can also use the provided simple-to-use Python API to train Agents using reinforcement learning, imitation learning, neuroevolution, or any other methods. These trained agents can be used for multiple purposes, including controlling NPC behavior (in a variety of settings such as multi-agent and adversarial), automated testing of game builds and evaluating different game design decisions pre-release. The ML-Agents Toolkit is mutually beneficial for both game developers and AI researchers as it provides a central platform where advances in AI can be evaluated on Unity’s rich environments and then made accessible to the wider research and game developer communities.\n\n## Features\n- 17+ [example Unity environments](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FLearning-Environment-Examples.html)\n- Support for multiple environment configurations and training scenarios\n- Flexible Unity SDK that can be integrated into your game or custom Unity scene\n- Support for training single-agent, multi-agent cooperative, and multi-agent competitive scenarios via several Deep Reinforcement Learning algorithms (PPO, SAC, MA-POCA, self-play).\n- Support for learning from demonstrations through two Imitation Learning algorithms (BC and GAIL).\n- Quickly and easily add your own [custom training algorithm](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FPython-Custom-Trainer-Plugin.html) and\u002For components.\n- Easily definable Curriculum Learning scenarios for complex tasks\n- Train robust agents using environment randomization\n- Flexible agent control with On Demand Decision Making\n- Train using multiple concurrent Unity environment instances\n- Utilizes the [Inference Engine](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FInference-Engine.html) to provide native cross-platform support\n- Unity environment [control from Python](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FPython-LLAPI.html)\n- Wrap Unity learning environments as a [gym](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FPython-Gym-API.html) environment\n- Wrap Unity learning environments as a [PettingZoo](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FPython-PettingZoo-API.html) environment\n\n## Releases & Documentation\n\n> **⚠️ Documentation Migration Notice**\n> We have moved to [Unity Package documentation](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest) as the **primary developer documentation** and have **deprecated** the maintenance of [web docs](https:\u002F\u002Funity-technologies.github.io\u002Fml-agents\u002F). Please use the Unity Package documentation for the most up-to-date information.\n\nThe table below shows our latest release, including our `develop` branch which is under active development and may be unstable. A few helpful guidelines:\n\n- The [Versioning page](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FVersioning.html) overviews how we manage our GitHub releases and the versioning process for each of the ML-Agents components.\n- The [Releases page](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Freleases) contains details of the changes between releases.\n- The [Migration page](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FMigrating.html) contains details on how to upgrade from earlier releases of the ML-Agents Toolkit.\n- The `com.unity.ml-agents` package is [verified](https:\u002F\u002Fdocs.unity3d.com\u002F2020.1\u002FDocumentation\u002FManual\u002Fpack-safe.html) for Unity 2020.1 and later. Verified packages releases are numbered 1.0.x.\n\n|      **Version**       |  **Release Date**   |                                  **Source**                                   |                                                 **Documentation**                                                  |                                      **Download**                                      |                  **Python Package**                   |                                   **Unity Package**                                   |\n|:----------------------:|:-------------------:|:-----------------------------------------------------------------------------:|:------------------------------------------------------------------------------------------------------------------:|:--------------------------------------------------------------------------------------:|:-----------------------------------------------------:|:-------------------------------------------------------------------------------------:|\n|     **Release 23**     | **August 28, 2025** | **[source](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Ftree\u002Frelease_23)** |              **[docs](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@4.0\u002Fmanual\u002Findex.html)**               | **[download](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Farchive\u002Frelease_23.zip)** | **[1.1.0](https:\u002F\u002Fpypi.org\u002Fproject\u002Fmlagents\u002F1.1.0\u002F)** |                                       **4.0.0**                                       |\n| **develop (unstable)** |         --          |    [source](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Ftree\u002Fdevelop)     | [docs](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Ftree\u002Fdevelop\u002Fcom.unity.ml-agents\u002FDocumentation~\u002Findex.md)   |    [download](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Farchive\u002Fdevelop.zip)     |                         --                            |                                          --                                           |\n\n\n\nIf you are a researcher interested in a discussion of Unity as an AI platform, see a pre-print of our [reference paper on Unity and the ML-Agents Toolkit](https:\u002F\u002Farxiv.org\u002Fabs\u002F1809.02627).\n\nIf you use Unity or the ML-Agents Toolkit to conduct research, we ask that you cite the following paper as a reference:\n\n```\n@article{juliani2020,\n  title={Unity: A general platform for intelligent agents},\n  author={Juliani, Arthur and Berges, Vincent-Pierre and Teng, Ervin and Cohen, Andrew and Harper, Jonathan and Elion, Chris and Goy, Chris and Gao, Yuan and Henry, Hunter and Mattar, Marwan and Lange, Danny},\n  journal={arXiv preprint arXiv:1809.02627},\n  url={https:\u002F\u002Farxiv.org\u002Fpdf\u002F1809.02627.pdf},\n  year={2020}\n}\n```\n\nAdditionally, if you use the MA-POCA trainer in your research, we ask that you cite the following paper as a reference:\n\n```\n@article{cohen2022,\n  title={On the Use and Misuse of Absorbing States in Multi-agent Reinforcement Learning},\n  author={Cohen, Andrew and Teng, Ervin and Berges, Vincent-Pierre and Dong, Ruo-Ping and Henry, Hunter and Mattar, Marwan and Zook, Alexander and Ganguly, Sujoy},\n  journal={RL in Games Workshop AAAI 2022},\n  url={http:\u002F\u002Faaai-rlg.mlanctot.info\u002Fpapers\u002FAAAI22-RLG_paper_32.pdf},\n  year={2022}\n}\n```\n\n\n## Additional Resources\n\n* [Unity Discussions](https:\u002F\u002Fdiscussions.unity.com\u002Ftag\u002Fml-agents)\n* [ML-Agents tutorials by CodeMonkeyUnity](https:\u002F\u002Fwww.youtube.com\u002Fplaylist?list=PLzDRvYVwl53vehwiN_odYJkPBzcqFw110)\n* [Introduction to ML-Agents by Huggingface](https:\u002F\u002Fhuggingface.co\u002Flearn\u002Fdeep-rl-course\u002Fen\u002Funit5\u002Fintroduction)\n* [Community created ML-Agents projects](https:\u002F\u002Fdiscussions.unity.com\u002Ft\u002Fpost-your-ml-agents-project\u002F816756)\n* [ML-Agents models on Huggingface](https:\u002F\u002Fhuggingface.co\u002Fmodels?library=ml-agents)\n* [Blog posts](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FBlog-posts.html)\n* [Discord](https:\u002F\u002Fdiscord.com\u002Fchannels\u002F489222168727519232\u002F1202574086115557446)\n\n## Community and Feedback\n\nThe ML-Agents Toolkit is an open-source project and we encourage and welcome contributions. If you wish to contribute, be sure to review our [contribution guidelines](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FCONTRIBUTING.html) and [code of conduct](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease\u002F4.0.0\u002FCODE_OF_CONDUCT.md).\n\nFor problems with the installation and setup of the ML-Agents Toolkit, or discussions about how to best setup or train your agents, please create a new thread on the [Unity ML-Agents discussion forum](https:\u002F\u002Fdiscussions.unity.com\u002Ftag\u002Fml-agents). Be sure to include as many details as possible to help others assist you effectively. If you run into any other problems using the ML-Agents Toolkit or have a specific feature request, please [submit a GitHub issue](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues).\n\nPlease tell us which samples you would like to see shipped with the ML-Agents Unity package by replying to [this discussion thread](https:\u002F\u002Fdiscussions.unity.com\u002Ft\u002Fhelp-shape-the-future-of-ml-agents\u002F1661019).\n\n## Privacy\n\nIn order to improve the developer experience for Unity ML-Agents Toolkit, we have added in-editor analytics. Please refer to \"Information that is passively collected by Unity\" in the [Unity Privacy Policy](https:\u002F\u002Funity3d.com\u002Flegal\u002Fprivacy-policy).\n","# Unity 机器学习智能体工具包\n\n[![文档徽章](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-reference-blue.svg)](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest)\n\n[![许可证徽章](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-Apache--2.0-green.svg)](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease\u002F4.0.0\u002FLICENSE.md)\n\n([最新版本](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Freleases\u002Ftag\u002Flatest_release)) ([所有版本](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Freleases))\n\n**Unity 机器学习智能体工具包**（ML-Agents）是一个开源项目，它使游戏和模拟环境能够作为训练智能体的平台。我们基于 PyTorch 提供了当前最先进的算法实现，让游戏开发者和爱好者可以轻松地为 2D、3D 以及 VR\u002FAR 游戏训练智能体。研究人员也可以使用我们提供的简单易用的 Python API，通过强化学习、模仿学习、神经进化或其他方法来训练智能体。这些训练好的智能体可用于多种用途，包括控制 NPC 行为（在多智能体和对抗性等多种场景中）、自动化测试游戏版本，以及在发布前评估不同的游戏设计决策。ML-Agents 工具包对游戏开发者和人工智能研究者都大有裨益，因为它提供了一个中心化平台，可以在 Unity 丰富的环境中评估人工智能领域的进展，并将这些成果开放给更广泛的研究和游戏开发社区。\n\n## 功能特性\n- 17+ 个 [示例 Unity 环境](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FLearning-Environment-Examples.html)\n- 支持多种环境配置和训练场景\n- 灵活的 Unity SDK，可集成到您的游戏或自定义 Unity 场景中\n- 支持通过多种深度强化学习算法（PPO、SAC、MA-POCA、自我对弈）训练单智能体、多智能体协作以及多智能体竞争场景。\n- 支持通过两种模仿学习算法（BC 和 GAIL）从示范中学习。\n- 可快速简便地添加您自己的 [自定义训练算法](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FPython-Custom-Trainer-Plugin.html) 和\u002F或组件。\n- 针对复杂任务可轻松定义课程式学习场景\n- 使用环境随机化训练鲁棒性更强的智能体\n- 通过按需决策实现灵活的智能体控制\n- 可同时使用多个 Unity 环境实例进行训练\n- 利用 [推理引擎](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FInference-Engine.html) 提供原生跨平台支持\n- 可从 Python 控制 Unity 环境 [Python LLAPI](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FPython-LLAPI.html)\n- 可将 Unity 学习环境封装为 [gym](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FPython-Gym-API.html) 环境\n- 可将 Unity 学习环境封装为 [PettingZoo](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FPython-PettingZoo-API.html) 环境\n\n## 发布与文档\n\n> **⚠️ 文档迁移通知**\n> 我们已将 [Unity Package 文档](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest) 作为**主要的开发者文档**，并已**弃用**对 [网页文档](https:\u002F\u002Funity-technologies.github.io\u002Fml-agents\u002F) 的维护。请使用 Unity Package 文档以获取最新信息。\n\n下表展示了我们的最新版本，包括处于积极开发中且可能不稳定的 `develop` 分支。以下是一些有用的指南：\n\n- [版本管理页面](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FVersioning.html) 概述了我们如何管理 GitHub 发布以及 ML-Agents 各组件的版本化进程。\n- [发布页面](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Freleases) 包含了各版本之间的变更详情。\n- [迁移页面](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FMigrating.html) 提供了从早期版本的 ML-Agents Toolkit 升级的详细说明。\n- `com.unity.ml-agents` 包已在 Unity 2020.1 及更高版本上通过[验证](https:\u002F\u002Fdocs.unity3d.com\u002F2020.1\u002FDocumentation\u002FManual\u002Fpack-safe.html)。经过验证的包版本号为 1.0.x。\n\n|      **版本**       |  **发布日期**   |                                  **源代码**                                   |                                                 **文档**                                                  |                                      **下载**                                      |                  **Python 包**                   |                                   **Unity 包**                                   |\n|:----------------------:|:-------------------:|:-----------------------------------------------------------------------------:|:------------------------------------------------------------------------------------------------------------------:|:--------------------------------------------------------------------------------------:|:-----------------------------------------------------:|:-------------------------------------------------------------------------------------:|\n|     **Release 23**     | **2025年8月28日** | **[源代码](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Ftree\u002Frelease_23)** |              **[文档](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@4.0\u002Fmanual\u002Findex.html)**               | **[下载](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Farchive\u002Frelease_23.zip)** | **[1.1.0](https:\u002F\u002Fpypi.org\u002Fproject\u002Fmlagents\u002F1.1.0\u002F)** |                                       **4.0.0**                                       |\n| **develop（不稳定）** |         --          |    [源代码](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Ftree\u002Fdevelop)     | [文档](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Ftree\u002Fdevelop\u002Fcom.unity.ml-agents\u002FDocumentation~\u002Findex.md)   |    [下载](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Farchive\u002Fdevelop.zip)     |                         --                            |                                          --                                           |\n\n\n\n如果您是对此感兴趣的科研人员，并希望探讨 Unity 作为 AI 平台，请参阅我们关于 Unity 和 ML-Agents Toolkit 的[预印本论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F1809.02627)。\n\n如果您使用 Unity 或 ML-Agents Toolkit 进行研究，我们恳请您引用以下论文作为参考文献：\n\n```\n@article{juliani2020,\n  title={Unity: A general platform for intelligent agents},\n  author={Juliani, Arthur and Berges, Vincent-Pierre and Teng, Ervin and Cohen, Andrew and Harper, Jonathan and Elion, Chris and Goy, Chris and Gao, Yuan and Henry, Hunter and Mattar, Marwan and Lange, Danny},\n  journal={arXiv preprint arXiv:1809.02627},\n  url={https:\u002F\u002Farxiv.org\u002Fpdf\u002F1809.02627.pdf},\n  year={2020}\n}\n```\n\n此外，如果您在研究中使用 MA-POCA 训练器，我们也恳请您引用以下论文作为参考文献：\n\n```\n@article{cohen2022,\n  title={On the Use and Misuse of Absorbing States in Multi-agent Reinforcement Learning},\n  author={Cohen, Andrew and Teng, Ervin and Berges, Vincent-Pierre and Dong, Ruo-Ping and Henry, Hunter and Mattar, Marwan and Zook, Alexander and Ganguly, Sujoy},\n  journal={RL in Games Workshop AAAI 2022},\n  url={http:\u002F\u002Faaai-rlg.mlanctot.info\u002Fpapers\u002FAAAI22-RLG_paper_32.pdf},\n  year={2022}\n}\n```\n\n\n## 其他资源\n\n* [Unity 讨论区](https:\u002F\u002Fdiscussions.unity.com\u002Ftag\u002Fml-agents)\n* [CodeMonkeyUnity 的 ML-Agents 教程](https:\u002F\u002Fwww.youtube.com\u002Fplaylist?list=PLzDRvYVwl53vehwiN_odYJkPBzcqFw110)\n* [Huggingface 关于 ML-Agents 的入门介绍](https:\u002F\u002Fhuggingface.co\u002Flearn\u002Fdeep-rl-course\u002Fen\u002Funit5\u002Fintroduction)\n* [社区创建的 ML-Agents 项目](https:\u002F\u002Fdiscussions.unity.com\u002Ft\u002Fpost-your-ml-agents-project\u002F816756)\n* [Huggingface 上的 ML-Agents 模型](https:\u002F\u002Fhuggingface.co\u002Fmodels?library=ml-agents)\n* [博客文章](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FBlog-posts.html)\n* [Discord 社区](https:\u002F\u002Fdiscord.com\u002Fchannels\u002F489222168727519232\u002F1202574086115557446)\n\n## 社区与反馈\n\nML-Agents Toolkit 是一个开源项目，我们鼓励并欢迎社区贡献。如果您希望参与贡献，请务必阅读我们的[贡献指南](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.ml-agents@latest\u002Findex.html?subfolder=\u002Fmanual\u002FCONTRIBUTING.html)和[行为准则](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease\u002F4.0.0\u002FCODE_OF_CONDUCT.md)。\n\n如您在安装或配置 ML-Agents Toolkit 时遇到问题，或希望讨论如何更好地设置和训练您的智能体，请在 [Unity ML-Agents 讨论区](https:\u002F\u002Fdiscussions.unity.com\u002Ftag\u002Fml-agents) 创建新主题。请尽可能提供详细信息，以便他人更有效地帮助您。如果您在使用 ML-Agents Toolkit 时遇到其他问题，或有特定的功能需求，请[提交 GitHub 问题](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues)。\n\n请通过回复[此讨论帖](https:\u002F\u002Fdiscussions.unity.com\u002Ft\u002Fhelp-shape-the-future-of-ml-agents\u002F1661019)，告诉我们您希望 ML-Agents Unity 包中包含哪些示例。\n\n## 隐私政策\n\n为了提升 Unity ML-Agents Toolkit 的开发者体验，我们在编辑器中加入了分析功能。有关更多信息，请参阅 [Unity 隐私政策](https:\u002F\u002Funity3d.com\u002Flegal\u002Fprivacy-policy) 中的“由 Unity 被动收集的信息”。","# Unity ML-Agents 快速上手指南\n\nUnity ML-Agents Toolkit 是一个开源项目，允许将游戏和模拟环境作为训练智能代理（Agents）的场所。它基于 PyTorch 实现了多种前沿算法（如 PPO、SAC、模仿学习等），帮助开发者轻松为 2D、3D 及 VR\u002FAR 应用训练智能行为。\n\n## 1. 环境准备\n\n在开始之前，请确保您的开发环境满足以下要求：\n\n*   **操作系统**: Windows, macOS 或 Linux\n*   **Unity 编辑器**: 版本 **2020.1** 或更高（推荐使用最新 LTS 版本）\n*   **Python**: 版本 **3.8** 至 **3.10** (推荐 3.8+)\n*   **包管理器**: `pip` (通常随 Python 安装)\n\n> **注意**：本指南假设您已安装好 Unity Hub 和基础 Python 环境。国内用户建议使用清华源或阿里源加速 Python 包下载。\n\n## 2. 安装步骤\n\n### 第一步：安装 Python 依赖\n\n打开终端（Terminal 或 CMD），运行以下命令安装 ML-Agents 的 Python 端库。\n\n**通用安装命令：**\n```bash\npip install mlagents\n```\n\n**国内加速安装（推荐）：**\n```bash\npip install mlagents -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n### 第二步：在 Unity 中安装 Package\n\n1.  打开 Unity 编辑器，创建一个新的 3D 项目或打开现有项目。\n2.  点击菜单栏 **Window > Package Manager**。\n3.  点击左上角的 **+** 号，选择 **Add package by name...**。\n4.  在弹出的对话框中输入包名：\n    ```text\n    com.unity.ml-agents\n    ```\n5.  点击 **Add**。Unity 将自动下载并安装最新的验证版本（当前主要版本为 4.x）。\n\n> **提示**：如果网络较慢导致安装失败，可尝试在 Unity 的 `Packages\u002Fmanifest.json` 文件中手动添加 `\"com.unity.ml-agents\": \"4.0.0\"` (版本号以实际最新为准)，然后重启 Unity。\n\n## 3. 基本使用\n\n以下是训练第一个智能代理的最简流程：\n\n### 步骤 A: 导入示例场景\n为了快速验证，我们直接使用官方提供的示例环境。\n1.  在 Unity 菜单中，点击 **ML-Agents > Import Examples**.\n2.  等待资源导入完成。\n3.  在 Project 窗口导航至 `Packages\u002Fcom.unity.ml-agents\u002FExamples\u002F3DBall\u002FScenes`。\n4.  双击打开 `3DBall.unity` 场景。\n\n### 步骤 B: 配置训练参数\n1.  在场景中选中名为 `LearningBehavior` 的游戏对象（或在 Hierarchy 中找到带有 `Behavior Parameters` 组件的对象）。\n2.  确保 **Behavior Name** 设置为 `3DBall`（这将作为训练时的标识符）。\n3.  无需修改其他代码，默认配置已包含观测空间和行为动作定义。\n\n### 步骤 C: 启动训练\n1.  保持 Unity 场景处于**未运行**状态（不要点击 Play），但需保存场景。\n2.  回到终端，确保位于项目根目录或任意路径。\n3.  运行以下训练命令：\n\n```bash\nmlagents-learn config\u002Fppo\u002F3DBall.yaml --run-id=first_run --env-name=3DBall\n```\n\n*   `--run-id`: 自定义本次训练的标识。\n*   `--env-name`: 必须与 Unity 场景中 `Behavior Parameters` 组件里的 **Behavior Name** 一致。\n*   如果是首次运行且未指定环境路径，脚本会尝试连接正在运行的 Unity 实例。**请先在 Unity 中点击 Play 按钮**，然后再执行上述命令（或者使用 `--env-path` 指定构建好的可执行文件路径进行无头训练）。\n\n**最简单的交互式训练流程（推荐新手）：**\n1.  在 Unity 中点击 **Play** 按钮运行场景。\n2.  在终端执行：\n    ```bash\n    mlagents-learn config\u002Fppo\u002F3DBall.yaml --run-id=first_run\n    ```\n3.  观察终端输出，当看到 `Connected to Unity environment` 时，表示训练已开始。\n4.  训练完成后，模型将保存在 `results\u002Ffirst_run\u002Fmodels\u002F` 目录下。\n\n### 步骤 D: 使用训练好的模型\n训练结束后，将生成的 `.onnx` 模型文件（位于 `results\u002F\u003Crun-id>\u002Fmodels\u002F`）拖拽到 Unity 项目中，替换原示例文件夹中的模型，或赋值给 `Behavior Parameters` 组件的 **Model** 字段，即可让 AI 自主控制平衡球。","某独立游戏团队正在开发一款 3D 平台跳跃游戏，需要为关卡中的敌人设计智能且多变的巡逻与追击行为。\n\n### 没有 ml-agents 时\n- 开发者必须手动编写复杂的有限状态机代码来定义敌人行为，逻辑僵硬且难以覆盖所有玩家操作情况。\n- 调整敌人难度需反复修改硬编码的参数（如反应速度、跳跃时机），每次验证都依赖人工试玩，效率极低。\n- 难以模拟真实玩家的多样化操作习惯，导致测试覆盖率不足，上线后容易发现未被测试到的通关漏洞。\n- 实现多个敌人间的协同围捕或对抗策略几乎不可能，因为传统脚本难以处理高维度的多智能体博弈。\n\n### 使用 ml-agents 后\n- 利用深度强化学习算法（如 PPO），让敌人在 Unity 仿真环境中通过数百万次自我试错自动学会高效的追击与躲避策略。\n- 通过调整奖励函数即可灵活控制敌人风格（激进或保守），无需重写代码，训练出的模型能自适应不同水平的玩家。\n- 借助模仿学习功能，直接导入真人玩家的通关录像作为示范数据，快速训练出行为拟人化的高智能测试代理，自动挖掘关卡漏洞。\n- 原生支持多智能体协作训练，轻松实现敌人小队之间的战术配合，如包抄、诱敌等复杂群体行为。\n\nml-agents 将原本依赖人工经验的静态脚本设计，转化为数据驱动的动态智能进化过程，大幅提升了游戏 AI 的逼真度与开发迭代效率。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FUnity-Technologies_ml-agents_d3d8ad7e.png","Unity-Technologies","Unity Technologies","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FUnity-Technologies_d8ebbff4.png","",null,"http:\u002F\u002Funity.com","https:\u002F\u002Fgithub.com\u002FUnity-Technologies",[83,87,91,95,99,102,106,110],{"name":84,"color":85,"percentage":86},"C#","#178600",54.7,{"name":88,"color":89,"percentage":90},"Python","#3572A5",40.2,{"name":92,"color":93,"percentage":94},"Jupyter Notebook","#DA5B0B",4.5,{"name":96,"color":97,"percentage":98},"ShaderLab","#222c37",0.2,{"name":100,"color":101,"percentage":98},"Shell","#89e051",{"name":103,"color":104,"percentage":105},"Batchfile","#C1F12E",0.1,{"name":107,"color":108,"percentage":109},"Dockerfile","#384d54",0,{"name":111,"color":112,"percentage":109},"C","#555555",19284,4437,"2026-04-05T21:31:33","NOASSERTION",4,"Windows, macOS, Linux","未说明 (训练基于 PyTorch，通常建议使用支持 CUDA 的 NVIDIA GPU，但 README 未指定具体型号或版本；推理引擎支持跨平台原生运行)","未说明",{"notes":122,"python":123,"dependencies":124},"该工具包含两部分核心环境：1. Unity 编辑器 (需版本 2020.1 或更高) 用于运行仿真环境；2. Python 环境用于训练算法。官方文档已迁移至 Unity Package 文档系统。支持将环境封装为 Gym 或 PettingZoo 接口。具体 Python 包版本 (如 mlagents) 需参考 PyPI 发布的最新版本 (当前示例为 1.1.0)。","未说明 (依赖 PyTorch 和 Python API，通常需 Python 3.8+)",[125,126],"PyTorch","Unity Engine (2020.1+)",[13,62],[129,130,131,132,133,134,135],"reinforcement-learning","unity3d","deep-learning","unity","deep-reinforcement-learning","neural-networks","machine-learning","2026-03-27T02:49:30.150509","2026-04-06T11:56:24.213192",[139,144,149,154,159],{"id":140,"question_zh":141,"answer_zh":142,"source_url":143},18763,"为什么在使用 Internal Brain 类型时 Unity 编辑器会崩溃？","该问题通常与旧版本的 TensorFlowSharp 插件有关。项目已迁移至 Barracuda 引擎，不再依赖 TensorFlowSharp。如果您遇到此问题，请确保升级到最新版本的 ml-agents，并使用 Barracuda 进行模型推理，旧的 Internal Brain 崩溃问题在新架构下已不再相关。","https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues\u002F918",{"id":145,"question_zh":146,"answer_zh":147,"source_url":148},18764,"是否支持 TensorFlow 1.8 及更高版本？如何将高版本 TensorFlow 模型转换为 Unity 可用格式？","早期版本对 TensorFlow 1.8+ 支持有限，但项目切换到 Barracuda 后已改善兼容性。如果您需要将 TensorFlow 模型（如 1.13.1）转换为 .nn 格式，请使用 Barracuda 提供的转换工具。相关修复已合并到开发分支并在 v0.10 及以后版本中发布。","https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues\u002F1389",{"id":150,"question_zh":151,"answer_zh":152,"source_url":153},18765,"编译时出现错误：'The type or namespace name TensorFlow could not be found'，如何解决？","此错误通常是因为 .NET 框架版本不匹配。TensorFlowSharp 是基于 .NET 4.6.1 构建的，而您的项目可能 targeting .NET 4.6。解决方案是将 Unity 项目的 Scripting Runtime Version 升级为 .NET 4.x Equivalent（Unity 2018.x 默认支持 .NET 4.7），或者确保安装了正确版本的 TFSharpPlugin 并定义了 ENABLE_TENSORFLOW 符号。","https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues\u002F552",{"id":155,"question_zh":156,"answer_zh":157,"source_url":158},18766,"训练或运行时报错 'KeyError: [BrainName]' 是什么原因？","这通常是因为配置文件中的行为名称（Behavior Name）与实际文件名或字典键不匹配。例如，代码中查找的键是 'ChameleonLearning'（无空格），但配置中写的是 'Chameleon Learning'（有空格）。请检查您的配置文件（yaml）和场景中的 Brain 设置，确保行为名称与文件名完全一致（包括空格和大小写）。","https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues\u002F2427",{"id":160,"question_zh":161,"answer_zh":162,"source_url":163},18767,"遇到 'Fatal Error: Could not allocate memory' 内存分配错误怎么办？","此类内存错误有时与特定的重置逻辑或对象创建时机有关。有用户反馈通过调整代码逻辑，避免在 FixedUpdate 之外频繁创建\u002F销毁游戏对象，或者优化 Agent 的重置（Reset）流程解决了问题。建议检查 Agent 的 Done() 调用频率以及场景对象的实例化逻辑，确保没有内存泄漏或瞬时大量分配。","https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues\u002F3024",[165,170,175,180,185,190,195,200,205,210,215,220,225,230,235,240,245,250,255,260],{"id":166,"version":167,"summary_zh":168,"released_at":169},109259,"release_23_tag","# [4.0.0] - 2025-08-28\r\n## 主要变更\r\n**com.unity.ml-agents** \r\n升级至推理引擎 2.2.1 (#6212)\r\n最低支持的 Unity 版本更新为 6000.0。(#6207)\r\n将扩展包 com.unity.ml-agents.extensions 合并到主包 com.unity.ml-agents 中。(#6227)\r\n## 次要变更\r\n**com.unity.ml-agents** \r\n从包中移除了损坏的示例 (#6230)\r\n迁移至 Unity 包文档，作为主要的开发者文档。(#6232)\r\n**ml-agents \u002F ml-agents-envs** \r\n将 grpcio 版本提升至 >=1.11.0,\u003C=1.53.2 (#6208)","2025-09-02T16:19:50",{"id":171,"version":172,"summary_zh":173,"released_at":174},109260,"release_22","# [3.0.0] - 2024-09-02\r\n## 主要变更\r\ncom.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\r\n升级至 Sentis 2.0.0 (#6137)\r\n升级至 Sentis 1.3.0-pre.3 (#6070)\r\n升级至 Sentis 1.3.0-exp.2 (#6013)\r\n将支持的最低 Unity 版本更新为 2023.2。(#6071)\r\nml-agents \u002F ml-agents-envs\r\n升级至 PyTorch 2.1.1。(#6013)\r\n## 次要变更\r\ncom.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\r\n新增 no-graphics-monitor 选项。(#6014)\r\nml-agents \u002F ml-agents-envs\r\n更新 Installation.md (#6004)\r\n更新 Using-Virtual-Environment.md (#6033)\r\n## 错误修复\r\ncom.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\r\n修复升级后 CI 测试失败的问题 (#6141)\r\n修复 Google Protocol Buffers 的程序集引用缺失问题 (#6099)\r\n修复 ModelRunner 中 Tensor 的 Dispose 方法缺失问题 (#6028)\r\n修复 3DBall 示例包，移除 Barracuda 依赖 (#6030)\r\nml-agents \u002F ml-agents-envs\r\n修复 migrating.md 中示例代码缩进问题 (#5840)\r\n修复持续集成测试问题 (#6079)\r\n修复点赞格式错误问题 (#6078)\r\n将 numpy 版本提升至 >=1.23.5,\u003C1.24.0 (#6082)\r\n将 ONNX 版本提升至 1.15.0 (#6062)\r\n将 Protocol Buffers 版本提升至 >=3.6,\u003C21 (#6062)","2024-10-05T14:04:56",{"id":176,"version":177,"summary_zh":178,"released_at":179},109261,"release_21","## [3.0.0-exp.1] - 2023-10-09\r\n### 主要变更\r\n#### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\r\n- 将 ML-Agents 升级至 Sentis 1.2.0-exp.2，并弃用 Barracuda。(#5979)\r\n- 最低支持的 Unity 版本更新为 2022.3。(#5950)\r\n- 添加了批处理射线检测传感器选项。(#5950)\r\n\r\n#### ml-agents \u002F ml-agents-envs\r\n- 更新至 PyTorch 1.13.1 (#5982)\r\n- 弃用对 Python 3.8.x 和 3.9.x 的支持 (#5981)\r\n\r\n### 次要变更\r\n#### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\r\n- 向 DecisionRequester 添加了 DecisionStep 参数 (#5940)\r\n  - 这将允许在使用多智能体时错开执行时机，从而提高运行稳定性。\r\n\r\n#### ml-agents \u002F ml-agents-envs\r\n- 增加了超时 CLI 和 YAML 配置文件支持，用于指定环境超时时间。(#5991)\r\n- 新增训练配置功能，可在整个训练过程中均匀分布检查点。(#5842)\r\n- 更新了训练区域复制器，添加了仅在构建模式下才复制训练区域的条件。(#5842)\r\n\r\n### 错误修复\r\n#### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\r\n- 在启用 .NET Standard 2.1 的情况下使用 IAsyncEnumerable\u003CT> 时出现编译器错误 (#5951)\r\n#### ml-agents \u002F ml-agents-envs","2023-10-09T19:14:39",{"id":181,"version":182,"summary_zh":183,"released_at":184},109262,"release_20","# 软件包版本\n\n注意：为了获得最佳体验，强烈建议您使用来自同一发行版的软件包组合。\n\n| 软件包 | 版本 |\n|----------|---------|\n| com.unity.ml-agents (C#) | v2.3.0-exp.3 |\n| com.unity.ml-agents.extensions (C#) | v0.6.1-preview |\n| ml-agents (Python) | v0.30.0 |\n| ml-agents-envs (Python) | v0.30.0|\n| gym-unity (Python) | v0.30.0 |\n| Communicator (C#\u002FPython) |v1.5.0 |\n\n# 发行说明\n### 主要变更\n\n#### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\n- 最低支持的 Unity 版本已更新至 2021.3。(#)\n\n#### ml-agents \u002F ml-agents-envs\n- 使用 ML-Agents 自定义训练器插件将您的训练器添加到软件包中。(#)\n  - ML-Agents 自定义训练器插件是一个可扩展的插件系统，用于基于高级训练器 API 定义新的训练器，更多信息请参阅[此处](..\u002Fdocs\u002FPython-Custom-Trainer-Plugin.md)。\n- 对核心模块进行了重构，使 ML-Agents 的内部类更加通用，适用于各种强化学习算法。(#)\n- ML-Agents 支持的最低 Python 版本已更改为 3.8.13。(#)\n- PyTorch 的最低支持版本已更改为 1.8.0。(#)\n- 为 PPO 添加了共享批评家的可配置性。(#)\n- 我们已将 `UnityToGymWrapper` 和 `PettingZoo` API 移至 `ml-agents-envs` 软件包。未来所有这些环境都将由 `ml-agents-envs` 软件包进行版本管理 (#)\n\n### 次要变更\n#### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\n- 为 RayPerceptionSensor 添加了一个开关，允许光线按从左到右的顺序排列。(#)\n  - 当前的交错顺序仍然是默认设置，但将被弃用。\n- 增加了对启用或禁用附加在相机传感器上的摄像机对象的支持，以提高性能。(#)\n\n#### ml-agents \u002F ml-agents-envs\n- 将遮蔽 torch 的路径重命名为 “mlagents\u002Ftrainers\u002Ftorch_entities”，并更新相应的导入语句 (#)\n\n\n### 错误修复\n#### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\n#### ml-agents \u002F ml-agents-envs\n\n","2022-11-29T21:24:33",{"id":186,"version":187,"summary_zh":188,"released_at":189},109263,"release_19","# 软件包版本\n\n注意：为了获得最佳体验，强烈建议您同时使用来自同一发行版的软件包。\n\n| 软件包 | 版本 |\n|----------|---------|\n| com.unity.ml-agents (C#) | v2.2.1-exp.1 |\n| com.unity.ml-agents.extensions (C#) | v0.6.1-preview |\n| ml-agents (Python) | v0.28.0 |\n| ml-agents-envs (Python) | v0.28.0|\n| gym-unity (Python) | v0.28.0 |\n| Communicator (C#\u002FPython) |v1.5.0 |\n\n# 发行说明\n## 主要变更\n\n### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\n- 最低支持的 Unity 版本已更新为 2020.3。(#5673)\n- 添加了一项新功能，可在运行时动态复制训练区域。(#5568)\n- 将 Barracuda 更新至 2.3.1-preview (#5591)\n- 将 Input System 更新至 1.3.0 (#5661)\n\n### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n\n## 次要变更\n\n### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\n- 增加了从任意检查点而非仅最新检查点初始化行为的能力 (#5525)\n- 新增获取堆叠观测值只读视图的功能 (#5523)\n\n### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n- 将 gym-unity 中的 gym 版本设置为 gym 0.20.0 发布版 (#5540)\n- 增加了对 `beta`、`epsilon` 和 `学习率` 使用独立调度的支持（仅影响 PPO 和 POCA）。(#5538)\n- 默认行为已更改为重启崩溃的 Unity 环境，而非退出程序。(#5553)\n  - 可通过 3 个新的 YAML 选项配置重启频率和生命周期限制：\n    1. env_params.max_lifetime_restarts (--max-lifetime-restarts) [默认=10]\n    2. env_params.restarts_rate_limit_n (--restarts-rate-limit-n) [默认=1]\n    3. env_params.restarts_rate_limit_period_s (--restarts-rate-limit-period-s) [默认=60]\n- 现在支持在训练和推理过程中进行确定性动作选择 (#5619)\n  - 新增了一个 `--deterministic` CLI 标志，用于在策略中确定性地选择概率最高的动作。同样也可以通过在运行选项配置的 `network_settings` 下添加 `deterministic: true` 来实现。（#5597）\n  - 现在会序列化额外的张量，以支持 ONNX 中的确定性动作选择。（#5593）\n  - 支持在编辑器中进行具有确定性动作选择的推理。（#5599）\n- 向 LL-API 添加了最小化的分析数据收集功能 (#5511)\n- 更新了 GridWorld 示例的 Colab 笔记本，使用 DQN 展示 Python API 的用法以及如何导出为 ONNX 格式。（#5643）\n\n## 错误修复\n### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\n- 将 gRPC 原生库更新为适用于 arm64 和 x86_64 的通用版本。此更改应使 ml-agents 能够在 Mac M1 上使用。（#5283，#5519）\n- 修复了一个错误，即 ml-agents 代码无法在不支持分析功能的平台上编译（PS4\u002F5、XBoxOne）。（#5628）\n\n### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n- 修复了一个错误，即在训练过程中批评家网络未被归一化。（#5595）\n- 修复了课程学习因 i 导致崩溃的错误。","2022-01-14T20:55:53",{"id":191,"version":192,"summary_zh":193,"released_at":194},109264,"release_18","# 包版本\n\n注意：為獲得最佳體驗，強烈建議您統一使用來自同一個發佈版本的包。\n\n| 包名 | 版本 |\n|----------|---------|\n| com.unity.ml-agents (C#) | v2.1.0-exp.1 |\n| com.unity.ml-agents.extensions (C#) | v0.5.0-preview |\n| ml-agents (Python) | v0.27.0 |\n| ml-agents-envs (Python) | v0.27.0|\n| gym-unity (Python) | v0.27.0 |\n| Communicator (C#\u002FPython) |v1.5.0 |\n\n# 發佈說明\n## 次要變更\n### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\n - 將 Barracuda 更新至 2.0.0-pre.3。(#5385)\n - 修復了在沒有 `Agent` 的情況下添加行為參數時出現的 `NullReferenceException`。(#5382)\n - 在編輯器中為 `VectorSensorComponent` 新增了堆疊選項。(#5376)\n### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n - 將 cattrs 依賴庫版本鎖定為 1.6。(#5397)\n - 為圖像輸入非常小的環境新增了一個全連接視覺編碼器。(#5351)\n - 在倉庫中新增了演示 Python API 使用方法的 Colab 筆記本。(#5399)\n## Bug 修復\n### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\n - `RigidBodySensorComponent` 現在會在以無法生成有用觀測值的方式使用時顯示警告。(#5387)\n - 更新了文檔，增加註釋說明 `GridSensor` 不適用於 2D 環境。(#5396)\n - 修復了在劇集結束時收集最後一次觀測之前傳感器無法正確重置的問題。(#5375)\n### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n - 修正了連續動作版 SAC 目標熵計算錯誤的問題。(#5372)\n - 修復了 TensorBoard 中直方圖統計信息無法正確報告的問題。(#5410)\n - 修復了導入使用 ResNet 編碼器的模型時出現的錯誤。(#5358)\n","2021-06-09T22:01:42",{"id":196,"version":197,"summary_zh":198,"released_at":199},109265,"release_17","# ML-Agents 17 版本\n\n# 包版本\n\n注意：为获得最佳体验，强烈建议您同时使用同一发行版中的各个包。\n\n| 包名 | 版本 |\n|----------|---------|\n| com.unity.ml-agents (C#) | v2.0.0 |\n| com.unity.ml-agents.extensions (C#) | v0.4.0-preview |\n| ml-agents (Python) | v0.26.0 |\n| ml-agents-envs (Python) | v0.26.0|\n| gym-unity (Python) | v0.26.0 |\n| Communicator (C#\u002FPython) |v1.5.0 |\n\n\n### 破坏性变更\n#### 最低版本支持\n- 最低支持的 Unity 版本已更新至 2019.4。(#5166)\n#### C# API 变更\n- 进行了多项破坏性接口更改。更多详情请参阅[迁移指南](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_17_docs\u002Fdocs\u002FMigrating.md)。\n- 一些先前标记为 `Obsolete` 的方法已被移除。如果您正在使用这些方法，则需要将其替换为受支持的替代方法。(#5024)\n- 在 `IDiscreteActionMask` 中禁用离散动作的接口已更改。`WriteMask(int branch, IEnumerable\u003Cint> actionIndices)` 已被 `SetActionEnabled(int branch, int actionIndex, bool isEnabled)` 所取代。(#5060)\n- `IActuator` 现在实现了 `IHeuristicProvider` 接口。(#5110)\n- `ISensor.GetObservationShape()` 已被移除，并新增了 `GetObservationSpec()`。`ITypedSensor` 和 `IDimensionPropertiesSensor` 接口也被移除。(#5127)\n- `ISensor.GetCompressionType()` 已被移除，并新增了 `GetCompressionSpec()`。`ISparseChannelSensor` 接口也被移除。(#5164)\n- 抽象方法 `SensorComponent.GetObservationShape()` 已不再被调用，因此已被移除。(#5172)\n- `SensorComponent.CreateSensor()` 已被 `SensorComponent.CreateSensors()` 所取代，后者返回一个 `ISensor[]` 数组。(#5181)\n- 默认的 `InferenceDevice` 现为 `InferenceDevice.Default`，其等价于 `InferenceDevice.Burst`。如果您依赖于之前的行为，可以显式地将 Agent 的 `InferenceDevice` 设置为 `InferenceDevice.CPU`。(#5175)\n#### 模型格式变更\n- 使用 ML-Agents 1.x 版本训练的模型，如果是在循环神经网络上训练的，在推理时将不再适用。(#5254)\n- `.onnx` 模型的输入名称已更改。所有输入占位符现在都使用前缀 `obs_`，从而取消了视觉观察和向量观察之间的区分。此外，LSTM 的输入和输出也发生了变化。使用此版本创建的模型无法与先前版本的包兼容。(#5080, #5236)\n- `.onnx` 模型的离散动作输出现在包含离散动作的实际值，而非 logits。使用此版本创建的模型也无法与先前版本的包兼容。(#5080)\n\n### 从 com.unity.ml-agents.extensions 移至 com.unity.ml-agents 的功能\n#### Match3\n- Match-3 集成工具已从 `com.unity.ml-agents.extensions` 移至 `com.unity.ml-agents`。(#5259)\n- `Match3Sensor` 已","2021-04-27T15:53:47",{"id":201,"version":202,"summary_zh":203,"released_at":204},109266,"release_16","# ML-Agents 16 版本\n\n# 包版本\n\n注意：为获得最佳体验，强烈建议您同时使用来自同一版本的包。\n\n| 包名 | 版本 |\n|----------|---------|\n| com.unity.ml-agents (C#) | v1.9.1 |\n| com.unity.ml-agents.extensions (C#) | v0.3.1-preview |\n| ml-agents (Python) | v0.25.1 |\n| ml-agents-envs (Python) | v0.25.1|\n| gym-unity (Python) | v0.25.1 |\n| Communicator (C#\u002FPython) |v1.5.0 |\n\n### 主要变更\n#### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n- `--resume` 标志现在支持在奖励提供者增加或网络架构发生变化时恢复实验，以及加载部分模型。更多详情请参阅 [此处](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_16_docs\u002Fdocs\u002FTraining-ML-Agents.md#loading-an-existing-model)。(#5213)\n\n### 错误修复\n#### com.unity.ml-agents (C#)\n- 修复了使用演示记录器时出现的错误警告。(#5216)\n\n#### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n- 修复了使用 LSTM 时导致方差增大的问题。同时还修复了在 POCA 和 `sequence_length` \u003C `time_horizon` 的情况下使用 LSTM 时的问题。(#5206)\n- 修复了一个 bug：即使启用了 `save_replay_buffer`，SAC 的重放缓冲区在运行结束时也不会被保存。(#5205)\n- ELO 现在能够在从检查点加载时正确恢复。(#5202)\n- 在 Python API 中，修复了调用 `set_action_single_agent` 时，`validate_action` 期望的维度不正确的问题。(#5208)\n- 在 `GymToUnityWrapper` 中，如果环境已经结束但仍调用 `step()`，将抛出适当的警告。(#5204)\n- 修复了一个问题：使用其中一个 `gym` 封装器会覆盖用户设置的日志级别。(#5201)\n\n","2021-04-14T00:26:21",{"id":206,"version":207,"summary_zh":208,"released_at":209},109267,"release_15","# ML-Agents 15 版本\n\n# 包版本\n\n注意：为了获得最佳体验，强烈建议您将来自同一版本的包一起使用。\n\n| 包名 | 版本 |\n|----------|---------|\n| com.unity.ml-agents (C#) | v1.9.0 |\n| com.unity.ml-agents.extensions (C#) | v0.3.0-preview |\n| ml-agents (Python) | v0.25.0 |\n| ml-agents-envs (Python) | v0.25.0|\n| gym-unity (Python) | v0.25.0 |\n| Communicator (C#\u002FPython) |v1.5.0 |\n\n### 主要变更\n#### com.unity.ml-agents (C#)\n- 新增了 `BufferSensor` 和 `BufferSensorComponent` 类（[文档](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Fmain\u002Fdocs\u002FLearning-Environment-Design-Agents.md#variable-length-observations)）。它们允许智能体观察数量可变的实体。示例请参见 [Sorter 环境](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_15_docs\u002Fdocs\u002FLearning-Environment-Examples.md#sorter)。（#4909）\n- 新增了 `SimpleMultiAgentGroup` 类和 `IMultiAgentGroup` 接口（[文档](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_15_docs\u002Fdocs\u002FLearning-Environment-Design-Agents.md#groups-for-cooperative-scenarios)）。这些功能允许智能体以组为单位获得奖励并结束回合。示例请参见 [Cooperative Push Block](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_15_docs\u002Fdocs\u002FLearning-Environment-Examples.md#cooperative-push-block)、[Dungeon Escape](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_15_docs\u002Fdocs\u002FLearning-Environment-Examples.md#dungeon-escape) 和 [Soccer](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_15_docs\u002Fdocs\u002FLearning-Environment-Examples.md#soccer-twos) 环境。（#4923）\n#### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n- 新增了 MA-POCA 训练器。这是一个新的训练器，使智能体能够学习如何在群体中协作。在实例化 `SimpleMultiAgentGroup` 后，请将配置 YAML 中的训练器设置为 `poca` 以使用此功能。（#5005）\n\n### 次要变更\n#### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\n- 将 com.unity.barracuda 更新至 1.3.2-preview。（#5084）\n- 在 `com.unity.ml-agents` 示例中添加了 3D Ball 场景。（#5077）\n#### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n- 已弃用 RewardSignals 的 `encoding_size` 设置。请改用 `network_settings`。（#4982）\n- 传感器名称现在会传递到 `ObservationSpec.name`。（#5036）\n\n### 错误修复\n#### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n- 修复了一个导致 GAIL 在智能体可通过自我牺牲结束回合的环境中失败的问题。（#4971）\n- 使当形状不同的观测值被发送到训练器时显示的错误信息更加清晰。（#5030）\n- 修复了一个阻止课程计划在自对弈模式下递增的问题。（#5098）\n","2021-03-17T21:50:50",{"id":211,"version":212,"summary_zh":213,"released_at":214},109268,"release_14","# ML-Agents 14 版本发布\n\n# 包版本\n\n注意：为获得最佳体验，强烈建议您同时使用来自同一发行版的包。\n\n| 包名 | 版本 |\n|----------|---------|\n| com.unity.ml-agents (C#) | v1.8.1 |\n| com.unity.ml-agents.extensions (C#) | v0.2.0-preview |\n| ml-agents (Python) | v0.24.1 |\n| ml-agents-envs (Python) | v0.24.01|\n| gym-unity (Python) | v0.24.1 |\n| Communicator (C#\u002FPython) |v1.4.0 |\n\n\n### 小幅变更\n#### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\n- 更新了 `cattrs` 的版本依赖，允许在 Python 3.8 或更高版本上使用 `>=1.1.0`。(#4821)\n\n### 错误修复\n#### com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\n- 修复了一个问题：在同帧内排队 InputEvent 时，会覆盖前一个事件的数据。\n\n\n","2021-03-09T03:51:04",{"id":216,"version":217,"summary_zh":218,"released_at":219},109269,"release_13","# ML-Agents Release 13\r\n\r\n# Package Versions\r\n\r\nNOTE: It is strongly recommended that you use packages from the same release together for the best experience.\r\n\r\n| Package | Version |\r\n|----------|---------|\r\n| com.unity.ml-agents (C#) | v1.8.0 |\r\n| com.unity.ml-agents.extensions (C#) | v0.1.0-preview |\r\n| ml-agents (Python) | v0.24.0 |\r\n| ml-agents-envs (Python) | v0.24.0 |\r\n| gym-unity (Python) | v0.24.0 |\r\n| Communicator (C#\u002FPython) |v1.4.0 |\r\n\r\n# Major Features and Improvements\r\n## com.unity.ml-agents \u002F com.unity.ml-agents.extensions\r\n- Add an InputActuatorComponent to allow the generation of Agent action spaces from an InputActionAsset.\r\n  Projects wanting to use this feature will need to add the\r\n  [Input System Package](https:\u002F\u002Fdocs.unity3d.com\u002FPackages\u002Fcom.unity.inputsystem@1.1\u002Fmanual\u002Findex.html) at version 1.1.0-preview.3 or later. (#4881)\r\n## ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\r\n- TensorFlow trainers have been removed, please use the Torch trainers instead. (#4707)\r\n- A plugin system for `mlagents-learn` has been added. You can now define custom\r\n  `StatsWriter` implementations and register them to be called during training.\r\n  More types of plugins will be added in the future. (#4788)\r\n\r\n# Minor Changes\r\n## com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\r\n- The `ActionSpec` constructor is now public. Previously, it was not possible to create an\r\n  ActionSpec with both continuous and discrete actions from code. (#4896)\r\n- `StatAggregationMethod.Sum` can now be passed to `StatsRecorder.Add()`. This\r\n  will result in the values being summed (instead of averaged) when written to\r\n  TensorBoard. Thanks to @brccabral for the contribution! (#4816)\r\n- The upper limit for the time scale (by setting the `--time-scale` parameter in mlagents-learn) was\r\n  removed when training with a player. The Editor still requires it to be clamped to 100. (#4867)\r\n- Added the IHeuristicProvider interface to allow IActuators as well as Agent implement the Heuristic function to generate actions.\r\n  Updated the Basic example and the Match3 Example to use Actuators.\r\n  Changed the namespace and file names of classes in com.unity.ml-agents.extensions. (#4849)\r\n- Added `VectorSensor.AddObservation(IList\u003Cfloat>)`. `VectorSensor.AddObservation(IEnumerable\u003Cfloat>)`\r\n  is deprecated. The `IList` version is recommended, as it does not generate any\r\n  additional memory allocations. (#4887)\r\n- Added `ObservationWriter.AddList()` and deprecated `ObservationWriter.AddRange()`.\r\n  `AddList()` is recommended, as it does not generate any additional memory allocations. (#4887)\r\n- The Barracuda dependency was upgraded to 1.3.0. (#4898)\r\n- Added `ActuatorComponent.CreateActuators`, and deprecated `ActuatorComponent.CreateActuator`.  The\r\n  default implementation will wrap `ActuatorComponent.CreateActuator` in an array and return that. (#4899)\r\n- `InferenceDevice.Burst` was added, indicating that Agent's model will be run using Barracuda's Burst backend.\r\n  This is the default for new Agents, but existing ones that use `InferenceDevice.CPU` should update to\r\n  `InferenceDevice.Burst`. (#4925)\r\n\r\n## ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\r\n- Tensorboard now logs the Environment Reward as both a scalar and a histogram. (#4878)\r\n- Added a `--torch-device` commandline option to `mlagents-learn`, which sets the default\r\n  [`torch.device`](https:\u002F\u002Fpytorch.org\u002Fdocs\u002Fstable\u002Ftensor_attributes.html#torch.torch.device) used for training. (#4888)\r\n- The `--cpu` commandline option had no effect and was removed. Use `--torch-device=cpu` to force CPU training. (#4888)\r\n- The `mlagents_env` API has changed, `BehaviorSpec` now has a `observation_specs` property containing a list of `ObservationSpec`. For more information on `ObservationSpec` see [here](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Fmaster\u002Fdocs\u002FPython-API.md#behaviorspec). (#4763, #4825)\r\n\r\n# Bug Fixes\r\n## com.unity.ml-agents (C#)\r\n- Fix a compile warning about using an obsolete enum in `GrpcExtensions.cs`. (#4812)\r\n- CameraSensor now logs an error if the GraphicsDevice is null. (#4880)\r\n- Removed unnecessary memory allocations in `ActuatorManager.UpdateActionArray()` (#4877)\r\n- Removed unnecessary memory allocations in `SensorShapeValidator.ValidateSensors()` (#4879)\r\n- Removed unnecessary memory allocations in `SideChannelManager.GetSideChannelMessage()` (#4886)\r\n- Removed several memory allocations that happened during inference. On a test scene, this\r\n  reduced the amount of memory allocated by approximately 25%. (#4887)\r\n- Removed several memory allocations that happened during inference with discrete actions. (#4922)\r\n- Properly catch permission errors when writing timer files. (#4921)\r\n- Unexpected exceptions during training initialization and shutdown are now logged. If you see\r\n  \"noisy\" logs, please let us know! (#4930, #4935)\r\n\r\n## ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\r\n- Fixed a bug that would cause an exception when `RunOptions` was dese","2021-02-24T22:19:52",{"id":221,"version":222,"summary_zh":223,"released_at":224},109270,"release_12","# ML-Agents Release 12\r\n\r\n# Package Versions\r\n\r\nNOTE: It is strongly recommended that you use packages from the same release together for the best experience.\r\n\r\n| Package | Version |\r\n|----------|---------|\r\n| com.unity.ml-agents (C#) | v1.7.2 |\r\n| ml-agents (Python) | v0.23.0 |\r\n| ml-agents-envs (Python) | v0.23.0 |\r\n| gym-unity (Python) | v0.23.0 |\r\n| Communicator (C#\u002FPython) |v1.3.0 |\r\n\r\n# Bug Fixes\r\n## com.unity.ml-agents (C#)\r\n- Add analytics package dependency to the package manifest. (#4794)\r\n## ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\r\n- Fixed the docker build process. (#4791)","2020-12-23T01:34:35",{"id":226,"version":227,"summary_zh":228,"released_at":229},109271,"release_11","# ML-Agents Release 11\r\n\r\n# Package Versions\r\n\r\nNOTE: It is strongly recommended that you use packages from the same release together for the best experience.\r\n\r\n| Package | Version |\r\n|----------|---------|\r\n| com.unity.ml-agents (C#) | v1.7.0 |\r\n| ml-agents (Python) | v0.23.0 |\r\n| ml-agents-envs (Python) | v0.23.0 |\r\n| gym-unity (Python) | v0.23.0 |\r\n| Communicator (C#\u002FPython) |v1.3.0 |\r\n\r\n# Major Features and Improvements\r\n## com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\r\n- An individual agent can now take both continuous and discrete actions. You can specify both continuous and discrete action sizes in Behavior Parameters. (#4702, #4718)\r\n\r\n## ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\r\n - PyTorch trainers now support training agents with both continuous and discrete action spaces. (#4702)\r\n\r\n# Bug Fixes and Minor Changes\r\n## com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\r\n - In order to improve the developer experience for Unity ML-Agents Toolkit, we have added in-editor analytics. Please refer to \"Information that is passively collected by Unity\" in the [Unity Privacy Policy](https:\u002F\u002Funity3d.com\u002Flegal\u002Fprivacy-policy). (#4677)\r\n- The FoodCollector example environment now uses continuous actions for moving and discrete actions for shooting. (#4746)\r\n- Removed noisy warnings about API minor version mismatches in both the C# and python code. (#4688)\r\n\r\n## ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\r\n - ActionSpec._validate_action() now enforces that UnityEnvironment.set_action_for_agent() receives a 1D np.array.\r\n","2020-12-21T21:36:25",{"id":231,"version":232,"summary_zh":233,"released_at":234},109272,"release_10","# ML-Agents Release 10\r\n\r\n# Package Versions\r\n\r\nNOTE: It is strongly recommended that you use packages from the same release together for the best experience.\r\n\r\n| Package | Version |\r\n|----------|---------|\r\n| com.unity.ml-agents (C#) | v1.6.0 |\r\n| ml-agents (Python) | v0.22.0 |\r\n| ml-agents-envs (Python) | v0.22.0 |\r\n| gym-unity (Python) | v0.22.0 |\r\n| Communicator (C#\u002FPython) |v1.2.0 |\r\n\r\n# Major Features and Improvements\r\n## New Demo Environment\r\n - The [Match3 environment](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_10_docs\u002Fcom.unity.ml-agents.extensions\u002FDocumentation~\u002FMatch3.md) was added to the Project, it uses the new utilities added in com.unity.ml-agents.extensions.\r\n\r\n## ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\r\n - PyTorch trainers are now the default. See the [installation docs](https:\u002F\u002Fhttps:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_10_docs\u002Fdocs\u002FInstallation.md) for more information on installing PyTorch. For the time being, TensorFlow is still available; you can use the TensorFlow backend by adding `--tensorflow` to the CLI, or adding `framework: tensorflow` in the configuration YAML. (#4517)\r\n\r\n\r\n# Bug Fixes and Minor Changes\r\n## com.unity.ml-agents \u002F com.unity.ml-agents.extensions (C#)\r\n - The Barracuda dependency was upgraded to 1.1.2 (#4571)\r\n - Utilities were added to com.unity.ml-agents.extensions to make it easier to integrate with match-3 games. See the [readme](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_10_docs\u002Fcom.unity.ml-agents.extensions\u002FDocumentation~\u002FMatch3.md) for more details. (#4515)\r\n - `Agent.CollectObservations()` and `Agent.EndEpisode()` will now throw an exception if they are called recursively (for example, if they call `Agent.EndEpisode()`). Previously, this would result in an infinite loop and cause the editor to hang. (#4573)\r\n\r\n## ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\r\n - The `action_probs` node is no longer listed as an output in TensorFlow models (#4613).\r\n\r\n\r\n","2020-11-20T01:57:45",{"id":236,"version":237,"summary_zh":238,"released_at":239},109273,"release_9","# ML-Agents Release 9\r\n\r\n# Package Versions\r\n\r\nNOTE: It is strongly recommended that you use packages from the same release together for the best experience.\r\n\r\n| Package | Version |\r\n|----------|---------|\r\n| com.unity.ml-agents (C#) | v1.5.0 |\r\n| ml-agents (Python) | v0.21.1 |\r\n| ml-agents-envs (Python) | v0.21.1 |\r\n| gym-unity (Python) | v0.21.1 |\r\n| Communicator (C#\u002FPython) |v1.2.0 |\r\n\r\n\r\n# Bug Fixes and Minor Changes\r\n\r\n## ml-agents (Python)\r\n- Fixed an issue where runs could not be resumed when using TensorFlow and Ghost Training. (#4593)\r\n- Capped cattrs version to 1.0.x. (#4613)\r\n- Capped PyTorch version to 1.6.x. (#4617)\r\n\r\n","2020-11-03T22:47:31",{"id":241,"version":242,"summary_zh":243,"released_at":244},109274,"release_8","# ML-Agents Release 8\r\n\r\n# Package Versions\r\n\r\nNOTE: It is strongly recommended that you use packages from the same release together for the best experience.\r\n\r\n| Package | Version |\r\n|----------|---------|\r\n| com.unity.ml-agents (C#) | v1.5.0 |\r\n| ml-agents (Python) | v0.21.0 |\r\n| ml-agents-envs (Python) | v0.21.0 |\r\n| gym-unity (Python) | v0.21.0 |\r\n| Communicator (C#\u002FPython) |v1.2.0 |\r\n\r\n# Major Features and Improvements\r\n## com.unity.ml-agents (C#)\r\n- Stacking for compressed observations is now supported. An additional setting option `Observation Stacks` is added to the sensor components that support compressed observations. A new class `ISparseChannelSensor` with an additional method `GetCompressedChannelMapping()`is added to generate a mapping of the channels in compressed data to the actual channel after decompression for the python side to decompress correctly. (#4476)\r\n\r\n## ml-agents (Python)\r\n- Added the Random Network Distillation (RND) intrinsic reward signal to the Pytorch trainers. To use RND, add a `rnd` section to the `reward_signals` section of your yaml configuration file. [More information here](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Fmaster\u002Fdocs\u002FTraining-Configuration-File.md#rnd-intrinsic-reward). (#4473)\r\n- The Communication API was changed to 1.2.0 to indicate support for stacked compressed observation. A new entry `compressed_channel_mapping` is added to the proto to handle decompression correctly. Newer versions of the package that wish to make use of this will also need a compatible version of the Python trainers. (#4476)\r\n\r\n\r\n# Bug Fixes and Minor Changes\r\n## com.unity.ml-agents (C#)\r\n- Fixed a bug where accessing the Academy outside of play mode would cause the Academy to get stepped multiple times when in play mode. (#4532)\r\n\r\n## ml-agents (Python)\r\n- In the `VisualFoodCollector` scene, a vector flag representing the frozen state of the agent is added to the input observations in addition to the original first-person camera frame. The scene is able to train with the provided default config file. (#4511)\r\n- Added a new visual scene `Visual3DBall` in the 3DBall example. (#4513)\r\n- Added conversion to string for sampler classes to increase the verbosity of the curriculum lesson changes. The lesson updates would now output the sampler stats in addition to the lesson and parameter name to the console.  (#4484)\r\n- Localized documentation in Russian is added. (#4529)\r\n\r\n\r\n# Acknowledgements\r\nThank you [@SergeyMatrosov](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fpulls?q=is%3Apr+author%3ASergeyMatrosov) for your contributions to this release.\r\n","2020-10-14T23:10:01",{"id":246,"version":247,"summary_zh":248,"released_at":249},109275,"release_7","# ML-Agents Release 7\r\n\r\n# Package Versions\r\n\r\nNOTE: It is strongly recommended that you use packages from the same release together for the best experience.\r\n\r\n| Package | Version |\r\n|----------|---------|\r\n| com.unity.ml-agents (C#) |    v1.4.0 |\r\n| ml-agents (Python) | v0.20.0 |\r\n| ml-agents-envs (Python) | v0.20.0 |\r\n| gym-unity (Python) | v0.20.0 |\r\n| Communicator (C#\u002FPython) |v1.1.0 |\r\n\r\n# Major Features and Improvements\r\n## com.unity.ml-agents (C#)\r\n- The `IActuator` interface and `ActuatorComponent` abstract class were added. These are analogous to `ISensor` and `SensorComponent`, but for applying actions for an Agent. They allow you to control the action space more programmatically than defining the actions in the Agent's Behavior Parameters. See [BasicActuatorComponent.cs](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_7_docs\u002FProject\u002FAssets\u002FML-Agents\u002FExamples\u002FBasic\u002FScripts\u002FBasicActuatorComponent.cs) for an example of how to use them. (#4297, #4315)\r\n## ml-agents (Python)\r\n- Experimental PyTorch support has been added. Use `--torch` when running mlagents-learn, or add `framework: pytorch` to your trainer configuration (under the behavior name) to enable it. Note that PyTorch 1.6.0 or greater should be installed to use this feature; see see [the PyTorch website](https:\u002F\u002Fpytorch.org\u002F) for installation instructions and [the relevant ML-Agents docs](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_7_docs\u002Fdocs\u002FTraining-ML-Agents.md#using-pytorch-experimental) for usage. (#4335)\r\n\r\n\r\n# Breaking Changes\r\n## ml-agents (Python)\r\nThe minimum supported version of TensorFlow was increased to 1.14.0. (#4411)\r\n\r\n\r\n# Known Issues\r\n## ml-agents (Python)\r\n - Soft-Actor Critic (SAC) runs considerably slower when using the PyTorch backend than when using TensorFlow. \r\n\r\n# Bug Fixes and Minor Changes\r\n## com.unity.ml-agents (C#)\r\n- Updated Barracuda to 1.1.1-preview (#4482)\r\n- Enabled C# formatting using `dotnet-format`. (#4362)\r\n- `GridSensor` was added to the `com.unity.ml-agents.extensions` package. Thank you to [Jaden Travnik](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues?q=is%3Apr+author%Jtravnik) from Eidos Montreal for the contribution! (#4399)\r\n- Added `Agent.EpisodeInterrupted()`, which can be used to reset the agent when it has reached a user-determined maximum number of steps. This behaves similarly to `Agent.EndEpsiode()` but has a slightly different effect on training (#4453).\r\n- Previously, `com.unity.ml-agents` was not declaring [built-in packages](https:\u002F\u002Fdocs.unity3d.com\u002F2020.2\u002FDocumentation\u002FManual\u002Fpack-build.html) as dependencies in its `package.json`. The relevant dependencies are now listed. (#4384)\r\n- Fixed the sample code in the custom SideChannel example. (#4466)\r\n\r\n## ml-agents (Python)\r\n- Compressed visual observations with >3 channels are now supported. In `ISensor.GetCompressedObservation()`, this can be done by writing 3 channels at a time to a PNG and concatenating the resulting bytes. (#4399)\r\n- The Communication API was changed to 1.1.0 to indicate support for concatenated PNGs (see above). Newer versions of the package that wish to make use of this will also need a compatible version of the trainer. (#4462)\r\n- A CNN (`vis_encode_type: match3`) for smaller grids, e.g. board games, has been added. (#4434)\r\n- You can now again specify a default configuration for your behaviors. Specify `default_settings` in your trainer configuration to do so. (#4448)\r\n- A bug in the observation normalizer that would cause rewards to decrease when using `--resume` was fixed. (#4463)\r\n\r\n\r\n# Acknowledgements\r\nThank you [@NeonMika ](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues?q=is%3Apr+author%3ANeonMika), [@armando-fandango](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues?q=is%3Apr+author%3Aarmando-fandango), [@Sebastian-Schuchmann ](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues?q=is%3Apr+author%3ASebastian-Schuchmann ), and everyone at Unity for their contributions to this release.\r\n","2020-09-21T17:21:03",{"id":251,"version":252,"summary_zh":253,"released_at":254},109276,"release_6","# ML-Agents Release 6\r\n\r\n# Package Versions\r\n\r\nNOTE: It is strongly recommended that you use packages from the same release together for the best experience.\r\n\r\n| Package | Version |\r\n|----------|---------|\r\n| com.unity.ml-agents (C#) |\tv1.3.0 |\r\n| ml-agents (Python) | v0.19.0 |\r\n| ml-agents-envs (Python) | v0.19.0 |\r\n| gym-unity (Python) | v0.19.0 |\r\n| Communicator (C#\u002FPython) |v1.0.0 |\r\n\r\n# Breaking Changes\r\n## ml-agents (Python)\r\n*   The minimum supported Python version for ml-agents-envs was changed to 3.6.1. (#4244)\r\n\r\n# Bug Fixes and Minor Changes\r\n## com.unity.ml-agents (C#) \r\n*   Academy.EnvironmentStep() will now throw an exception if it is called recursively (for example, by an Agent's CollectObservations method). Previously, this would result in an infinite loop and cause the editor to hang. (#1551)\r\n## ml-agents (Python)\r\n*   The supported versions of `numpy` ml-agents-envs was changed to not allow 1.19.0 or later. This was done to reflect a similar change in TensorFlow's requirements. (#4274)\r\n*   Model checkpoints are now also saved as .nn files during training. (#4127)\r\n*   Model checkpoint info is saved in TrainingStatus.json after training is concluded (#4127)\r\n*   The CSV statistics writer was removed (#4300).\r\n*   StatsSideChannel now stores multiple values per key. This means that multiple calls to `StatsRecorder.Add()` with the same key in the same step will no longer overwrite each other. (#4236)\r\n## ml-agents-envs (Python)\r\n*   The interaction between EnvManager and TrainerController was changed; EnvManager.advance() was split into two stages, and TrainerController now uses the results from the first stage to handle new behavior names. This change speeds up Python training by approximately 5-10%. (#4259)\r\n\r\n# Known Issues\r\n## com.unity.ml-agents (C#) \r\n*   On macOS 10.15, if the ML-Agents package is installed from a website download, you may receive a “file cannot be opened issue” when running scenes that use ML-Agents. Workarounds include installing the package using the Unity Package Manager, or following the instructions [here](https:\u002F\u002Fsupport.apple.com\u002Fen-us\u002FHT202491). (specifically, the section titled “How to open an app that hasn’t been notarized or is from an unidentified developer”). \r\n\r\n\r\n# Acknowledgements\r\nThank you [ChristianCoenen](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues?q=is%3Apr+author%3AChristianCoenen), [niskander](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fissues?q=is%3Apr+author%3Aniskander), and everyone at Unity for their contributions to this release.\r\n","2020-08-17T17:41:17",{"id":256,"version":257,"summary_zh":258,"released_at":259},109277,"release_5","# Package Versions #\r\nAs part of ML-Agents Release 5, we will be versioning the different packages that make up the release.\r\n\r\n**NOTE**: It is strongly recommended that you use packages from the same release together for the best experience.\r\n\r\n| Package | Version |\r\n|-------------|----------|\r\n| com.unity.ml-agents (C#) | v1.2.0 |\r\n| ml-agents (Python) | v0.18.1 |\r\n| ml-agents-envs (Python) | v0.18.1 |\r\n| gym-unity (Python) | v0.18.1 |\r\n| Communicator (C#\u002FPython) | v1.0.0 |\r\n\r\n### Bug Fixes\r\n#### ml-agents \u002F ml-agents-envs \u002F gym-unity (Python)\r\n- Summary writer no longer crashes if Hyperparameters could not be written  (#4265)\r\n- Reduce numpy version in ml-agents-envs setup (#4274)\r\n- Fix curriculum lesson incrementing past final lesson (#4279)\r\n","2020-07-31T17:15:58",{"id":261,"version":262,"summary_zh":263,"released_at":264},109278,"release_4","# Package Versions #\r\n**NOTE**: It is strongly recommended that you use packages from the same release together for the best experience.\r\n| Package | Version |\r\n|-------------|----------|\r\n| com.unity.ml-agents (C#) | v1.2.0 |\r\n| ml-agents (Python) | v0.18.0 |\r\n| ml-agents-envs (Python) | v0.18.0 |\r\n| gym-unity (Python) | v0.18.0 |\r\n| Communicator (C#\u002FPython) | v1.0.0 |\r\n\r\n# Major Features and Improvements\r\n## ml-agents (Python)\r\n- The Parameter Randomization feature has been refactored and merged with the Curriculum. This enables sampling of new parameters per episode to improve robustness. The `resampling-interval` parameter has been removed and the config structure updated. More information [here](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_4_docs\u002Fdocs\u002FTraining-ML-Agents.md). (#4065)\r\n- Curriculum has been refactored and is now specified at the level of the parameter, not the behavior. It is also now possible to specify a sampler in the lesson of a Curriculum. More information [here](https:\u002F\u002Fgithub.com\u002FUnity-Technologies\u002Fml-agents\u002Fblob\u002Frelease_4_docs\u002Fdocs\u002FTraining-ML-Agents.md).(#4160)\r\n\r\n# Breaking Changes\r\n## ml-agents (Python)\r\n- The training configuration yaml file format has changed. In order to use curriculum or environment randomization, you must update your configuration file with the new section `environment_parameters`. We provide an update script to help this migration. Run `python -m mlagents.trainers.upgrade_config -h` with the latest version to see the script usage.\r\n# Bug Fixes and Minor Changes\r\n## com.unity.ml-agents (C#)\r\n- `SideChannelsManager` was renamed to `SideChannelManager`. The old name is still supported, but deprecated. (#4137)\r\n- `RayPerceptionSensor.Perceive()` now additionally stores the GameObject that was hit by the ray. (#4111)\r\n- Fixed an issue where `RayPerceptionSensor` would raise an exception when the list of tags was empty, or a tag in the list was invalid (unknown, null, or empty string). (#4155)\r\n- The Barracuda dependency was upgraded to 1.0.1 (#4188)\r\n## ml-agents-envs (Python)\r\n- Added new Google Colab notebooks to show how to use `UnityEnvironment`. (#4117)\r\n## ml-agents (Python)\r\n- Fixed an error when setting `initialize_from` in the trainer configuration YAML to null. (#4175)\r\n- Fixed a rare crash in StatsReporter when using threaded trainers (#4201)\r\n## Example Environments (C#)\r\n- Fixed issue with FoodCollector, Soccer, and WallJump when playing with keyboard. (#4147, #4174)\r\n# Known Issues\r\n## com.unity.ml-agents (C#)\r\n- On macOS 10.15, if the ML-Agents package is installed from a website download, you may receive a “file cannot be opened issue” when running scenes that use ML-Agents. Workarounds include installing the package using the Unity Package Manager, or following the instructions [here](https:\u002F\u002Fsupport.apple.com\u002Fen-us\u002FHT202491). \r\n# Acknowledgements \r\nThank you @StefanoCecere, @yongjun823, @poehlerflorian, @ChristianCoenen, @textarcana, @furkan-celik and everyone at Unity for their contributions to this release. \r\n","2020-07-15T22:47:47"]