[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-google-deepmind--mujoco_mpc":3,"tool-google-deepmind--mujoco_mpc":64},[4,17,27,35,43,56],{"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},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":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"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",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"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",[15,14,13,26,54],{"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":103,"forks":104,"last_commit_at":105,"license":106,"difficulty_score":107,"env_os":108,"env_gpu":109,"env_ram":109,"env_deps":110,"category_tags":121,"github_topics":122,"view_count":10,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":128,"updated_at":129,"faqs":130,"releases":161},1037,"google-deepmind\u002Fmujoco_mpc","mujoco_mpc","Real-time behaviour synthesis with MuJoCo, using Predictive Control","MuJoCo MPC（简称 MJPC）是由 Google DeepMind 打造的一款交互式应用软件与框架，旨在利用 MuJoCo 物理引擎实现实时预测控制。面对复杂机器人任务中行为合成与策略求解的挑战，MJPC 提供了一套高效的解决方案，让用户能够轻松设计并解决各类控制难题。\n\nMJPC 非常适合机器人领域的研究人员、开发者以及控制算法工程师使用。无论是四足机器人运动、双臂协同操作，还是魔方还原与人体动作追踪，MJPC 都能胜任。其技术亮点在于支持多种基于多步射击的规划器：既包含基于导数的 iLQG 和梯度下降法，也提供了无需导数却极具竞争力的预测采样方法。\n\n通过直观的图形用户界面，用户可以实时观察仿真效果并调整参数，极大地降低了实验门槛。MJPC 不仅开源免费，还具备良好的跨平台支持，是探索实时行为合成与模型预测控制的理想平台。","\u003Ch1>\n  \u003Ca href=\"#\">\u003Cimg alt=\"MuJoCo MPC\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_2ab5834b0a6e.png\" width=\"100%\">\u003C\u002Fa>\n\u003C\u002Fh1>\n\n\u003Cp>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Factions\u002Fworkflows\u002Fbuild.yml?query=branch%3Amain\" alt=\"GitHub Actions\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Factions\u002Fworkflow\u002Fstatus\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fbuild.yml?branch=main\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fblob\u002Fmain\u002FLICENSE\" alt=\"License\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flicense\u002Fgoogle-deepmind\u002Fmujoco_mpc\">\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n**MuJoCo MPC (MJPC)** is an interactive application and software framework for\nreal-time predictive control with [MuJoCo](https:\u002F\u002Fmujoco.org\u002F), developed by\nGoogle DeepMind.\n\nMJPC allows the user to easily author and solve complex robotics tasks, and\ncurrently supports multiple shooting-based planners. Derivative-based methods include iLQG and\nGradient Descent, while derivative-free methods include a simple yet very competitive planner\ncalled Predictive Sampling.\n\n- [Overview](#overview)\n- [Graphical User Interface](#graphical-user-interface)\n- [Installation](#installation)\n  - [macOS](#macos)\n  - [Ubuntu](#ubuntu)\n  - [Build Issues](#build-issues)\n- [Predictive Control](#predictive-control)\n- [Contributing](#contributing)\n- [Known Issues](#known-issues)\n- [Citation](#citation)\n- [Acknowledgments](#acknowledgments)\n- [License and Disclaimer](#license-and-disclaimer)\n\n## Overview\n\nTo read the paper describing this software package, please see our\n[preprint](https:\u002F\u002Farxiv.org\u002Fabs\u002F2212.00541).\n\nFor a quick video overview of MJPC, click below.\n\n[![Video](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_7562f0da6031.jpg)](https:\u002F\u002Fdpmd.ai\u002Fmjpc)\n\nFor a longer talk at the MIT Robotics Seminar in December 2022 describing our results, click\nbelow.\n\n[![2022Talk](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_d8b6a2a05b78.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=2xVN-qY78P4)\n\nA more recent, December 2023 talk at the IEEE Technical Committee on Model-Based Optimization\nis available here:\n\n[![2023Talk](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_9c36c9f1bbff.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=J-JO-lgaKtw&t=0s)\n\n### Example tasks\n\nQuadruped task:\n\n[![Quadruped](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_0b37aa5f88c4.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=esLuwaWz4oE)\n\n\nBimanual manipulation:\n\n[![Bimanual](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_ec761cde01bd.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=aCNCKVThKIE)\n\n\nRubik's cube 10-move unscramble:\n\n[![Unscramble](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_8e8d7503ec13.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=ZRRvVWV-Muk)\n\nHumanoid motion-capture tracking:\n\n[![Tracking](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_5be75a0886cc.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=tEBVK-MO1Sw)\n\n## Graphical User Interface\n\nFor a detailed dive of the graphical user interface, see the\n[MJPC GUI](docs\u002FGUI.md) documentation.\n\n## Installation\nMJPC is tested with [Ubuntu 20.04](https:\u002F\u002Freleases.ubuntu.com\u002Ffocal\u002F) and [macOS-12](https:\u002F\u002Fwww.apple.com\u002Fby\u002Fmacos\u002Fmonterey\u002F). In principle, other versions and Windows operating system should work with MJPC, but these are not tested.\n\n### Prerequisites\nOperating system specific dependencies:\n\n#### macOS\nInstall [Xcode](https:\u002F\u002Fdeveloper.apple.com\u002Fxcode\u002F).\n\nInstall `ninja` and `zlib`:\n```sh\nbrew install ninja zlib\n```\n\n#### Ubuntu 20.04\n```sh\nsudo apt-get update && sudo apt-get install cmake libgl1-mesa-dev libxinerama-dev libxcursor-dev libxrandr-dev libxi-dev ninja-build zlib1g-dev clang-12\n```\n\n### Clone MuJoCo MPC\n```sh\ngit clone https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\n```\n\n### Build and Run MJPC GUI application\n1. Change directory:\n```sh\ncd mujoco_mpc\n```\n\n2. Create and change to build directory:\n```sh\nmkdir build\ncd build\n```\n\n3. Configure:\n\n#### macOS-12\n```sh\ncmake .. -DCMAKE_BUILD_TYPE:STRING=Release -G Ninja -DMJPC_BUILD_GRPC_SERVICE:BOOL=ON\n```\n\n#### Ubuntu 20.04\n```sh\ncmake .. -DCMAKE_BUILD_TYPE:STRING=Release -G Ninja -DCMAKE_C_COMPILER:STRING=clang-12 -DCMAKE_CXX_COMPILER:STRING=clang++-12 -DMJPC_BUILD_GRPC_SERVICE:BOOL=ON\n```\n**Note: gRPC is a large dependency and can take 10-20 minutes to initially download.**\n\n4. Build\n```sh\ncmake --build . --config=Release\n```\n\n6. Run GUI application\n```sh\ncd bin\n.\u002Fmjpc\n```\n\n### Build and Run MJPC GUI application using VSCode\nWe recommend using [VSCode](https:\u002F\u002Fcode.visualstudio.com\u002F) and 2 of its\nextensions ([CMake Tools](https:\u002F\u002Fmarketplace.visualstudio.com\u002Fitems?itemName=ms-vscode.cmake-tools)\nand [C\u002FC++](https:\u002F\u002Fmarketplace.visualstudio.com\u002Fitems?itemName=ms-vscode.cpptools))\nto simplify the build process.\n\n1. Open the cloned directory `mujoco_mpc`.\n2. Configure the project with CMake (a pop-up should appear in VSCode)\n3. Set compiler to `clang-12`.\n4. Build and run the `mjpc` target in \"release\" mode (VSCode defaults to\n   \"debug\"). This will open and run the graphical user interface.\n\n### Build Issues\nIf you encounter build issues, please see the\n[Github Actions configuration](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fblob\u002Fmain\u002F.github\u002Fworkflows\u002Fbuild.yml).\nThis provides the exact setup we use for building MJPC for testing with Ubuntu 20.04 and macOS-12.\n\n# Python API\nWe provide a simple Python API for MJPC. This API is still experimental and expects some more experience from its users. For example, the correct usage requires that the model (defined in Python) and the MJPC task (i.e., the residual and transition functions defined in C++) are compatible with each other. Currently, the Python API does not provide any particular error handling for verifying this compatibility and may be difficult to debug without more in-depth knowledge about MuJoCo and MJPC.\n\n## Installation\n\n### Prerequisites\n1. Build MJPC (see instructions above).\n\n2. Python 3.10\n\n3. (Optionally) Create a conda environment with **Python 3.10**:\n```sh\nconda create -n mjpc python=3.10\nconda activate mjpc\n```\n\n4. Install MuJoCo\n```sh\npip install mujoco\n```\n\n### Install API\nNext, change to the python directory:\n```sh\ncd python\n```\n\nInstall the Python module:\n```sh\npython setup.py install\n```\n\nTest that installation was successful:\n```sh\npython \"mujoco_mpc\u002Fagent_test.py\"\n```\n\nExample scripts are found in `python\u002Fmujoco_mpc\u002Fdemos`. For example from `python\u002F`:\n```sh\npython mujoco_mpc\u002Fdemos\u002Fagent\u002Fcartpole_gui.py\n```\nwill run the MJPC GUI application using MuJoCo's passive viewer via Python.\n\n### Python API Installation Issues\nIf your installation fails or is terminated prematurely, we recommend deleting the MJPC build directory and starting from scratch as the build will likely be corrupted. Additionally, delete the files generated during the installation process from the `python\u002F` directory.\n\n## Predictive Control\n\nSee the [Predictive Control](docs\u002FOVERVIEW.md) documentation for more\ninformation.\n\n## Contributing\n\nSee the [Contributing](docs\u002FCONTRIBUTING.md) documentation for more information.\n\n## Known Issues\n\nMJPC is not production-quality software, it is a **research prototype**. There\nare likely to be missing features and outright bugs. If you find any, please\nreport them in the [issue tracker](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fissues).\nBelow we list some known issues, including items that we are actively working\non.\n\n- We have not tested MJPC on Windows, but there should be no issues in\n  principle.\n- Task specification, in particular the setting of norms and their parameters in\n  XML, is a bit clunky. We are still iterating on the design.\n- The Gradient Descent search step is proportional to the scale of the cost\n  function and requires per-task tuning in order to work well. This is not a bug\n  but a property of vanilla gradient descent. It might be possible to ameliorate\n  this with some sort of gradient normalisation, but we have not investigated\n  this thoroughly.\n\n## Citation\n\nIf you use MJPC in your work, please cite our accompanying [preprint](https:\u002F\u002Farxiv.org\u002Fabs\u002F2212.00541):\n\n```bibtex\n@article{howell2022,\n  title={{Predictive Sampling: Real-time Behaviour Synthesis with MuJoCo}},\n  author={Howell, Taylor and Gileadi, Nimrod and Tunyasuvunakool, Saran and Zakka, Kevin and Erez, Tom and Tassa, Yuval},\n  archivePrefix={arXiv},\n  eprint={2212.00541},\n  primaryClass={cs.RO},\n  url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2212.00541},\n  doi={10.48550\u002FarXiv.2212.00541},\n  year={2022},\n  month={dec}\n}\n```\n\n## Acknowledgments\n\nThe main effort required to make this repository publicly available was\nundertaken by [Taylor Howell](https:\u002F\u002Fthowell.github.io\u002F) and the Google\nDeepMind Robotics Simulation team.\n\n## License and Disclaimer\n\nAll other content is Copyright 2022 DeepMind Technologies Limited and licensed\nunder the Apache License, Version 2.0. A copy of this license is provided in the\ntop-level LICENSE file in this repository. You can also obtain it from\nhttps:\u002F\u002Fwww.apache.org\u002Flicenses\u002FLICENSE-2.0.\n\nThis is not an officially supported Google product.\n","\u003Ch1>\n  \u003Ca href=\"#\">\u003Cimg alt=\"MuJoCo MPC\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_2ab5834b0a6e.png\" width=\"100%\">\u003C\u002Fa>\n\u003C\u002Fh1>\n\n\u003Cp>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Factions\u002Fworkflows\u002Fbuild.yml?query=branch%3Amain\" alt=\"GitHub Actions\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Factions\u002Fworkflow\u002Fstatus\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fbuild.yml?branch=main\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fblob\u002Fmain\u002FLICENSE\" alt=\"License\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flicense\u002Fgoogle-deepmind\u002Fmujoco_mpc\">\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n**MuJoCo MPC (MJPC)** 是一个交互式应用程序和软件框架，用于通过 [MuJoCo](https:\u002F\u002Fmujoco.org\u002F) 进行实时预测控制（Real-time Predictive Control），由 Google DeepMind 开发。\n\nMJPC 允许用户轻松编写和解决复杂的机器人任务，目前支持多种基于多重射击法（Multiple Shooting）的规划器。基于导数的方法包括 iLQG（迭代线性二次高斯控制）和梯度下降（Gradient Descent），而无导数方法包括一个简单但极具竞争力的规划器，称为预测采样（Predictive Sampling）。\n\n- [概述](#overview)\n- [图形用户界面](#graphical-user-interface)\n- [安装](#installation)\n  - [macOS](#macos)\n  - [Ubuntu](#ubuntu)\n  - [构建问题](#build-issues)\n- [预测控制](#predictive-control)\n- [贡献](#contributing)\n- [已知问题](#known-issues)\n- [引用](#citation)\n- [致谢](#acknowledgments)\n- [许可证和免责声明](#license-and-disclaimer)\n\n## 概述\n\n要阅读描述此软件包的论文，请参阅我们的 [预印本](https:\u002F\u002Farxiv.org\u002Fabs\u002F2212.00541)。\n\n如需快速了解 MJPC 的视频概述，请点击下方。\n\n[![Video](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_7562f0da6031.jpg)](https:\u002F\u002Fdpmd.ai\u002Fmjpc)\n\n如需观看 2022 年 12 月在 MIT 机器人研讨会上描述我们成果的更长演讲，请点击下方。\n\n[![2022Talk](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_d8b6a2a05b78.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=2xVN-qY78P4)\n\n2023 年 12 月在 IEEE 基于模型的优化技术委员会上的最新演讲可在此处观看：\n\n[![2023Talk](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_9c36c9f1bbff.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=J-JO-lgaKtw&t=0s)\n\n### 示例任务\n\n四足机器人任务：\n\n[![Quadruped](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_0b37aa5f88c4.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=esLuwaWz4oE)\n\n\n双手操作：\n\n[![Bimanual](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_ec761cde01bd.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=aCNCKVThKIE)\n\n\n魔方 10 步还原：\n\n[![Unscramble](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_8e8d7503ec13.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=ZRRvVWV-Muk)\n\n人形机器人动作捕捉跟踪：\n\n[![Tracking](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_readme_5be75a0886cc.jpg)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=tEBVK-MO1Sw)\n\n## 图形用户界面\n\n有关图形用户界面（Graphical User Interface）的详细介绍，请参阅 [MJPC GUI](docs\u002FGUI.md) 文档。\n\n## 安装\nMJPC 已在 [Ubuntu 20.04](https:\u002F\u002Freleases.ubuntu.com\u002Ffocal\u002F) 和 [macOS-12](https:\u002F\u002Fwww.apple.com\u002Fby\u002Fmacos\u002Fmonterey\u002F) 上经过测试。原则上，其他版本和 Windows 操作系统也应能与 MJPC 配合使用，但未经过测试。\n\n### 前置条件\n操作系统特定的依赖项（Dependencies）：\n\n#### macOS\n安装 [Xcode](https:\u002F\u002Fdeveloper.apple.com\u002Fxcode\u002F)。\n\n安装 `ninja` 和 `zlib`：\n```sh\nbrew install ninja zlib\n```\n\n#### Ubuntu 20.04\n```sh\nsudo apt-get update && sudo apt-get install cmake libgl1-mesa-dev libxinerama-dev libxcursor-dev libxrandr-dev libxi-dev ninja-build zlib1g-dev clang-12\n```\n\n### 克隆 MuJoCo MPC\n```sh\ngit clone https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\n```\n\n### 构建并运行 MJPC GUI 应用程序\n1. 更改目录：\n```sh\ncd mujoco_mpc\n```\n\n2. 创建并进入构建目录：\n```sh\nmkdir build\ncd build\n```\n\n3. 配置：\n\n#### macOS-12\n```sh\ncmake .. -DCMAKE_BUILD_TYPE:STRING=Release -G Ninja -DMJPC_BUILD_GRPC_SERVICE:BOOL=ON\n```\n\n#### Ubuntu 20.04\n```sh\ncmake .. -DCMAKE_BUILD_TYPE:STRING=Release -G Ninja -DCMAKE_C_COMPILER:STRING=clang-12 -DCMAKE_CXX_COMPILER:STRING=clang++-12 -DMJPC_BUILD_GRPC_SERVICE:BOOL=ON\n```\n**注意：gRPC 是一个大型依赖项，初始下载可能需要 10-20 分钟。**\n\n4. 构建\n```sh\ncmake --build . --config=Release\n```\n\n6. 运行 GUI 应用程序\n```sh\ncd bin\n.\u002Fmjpc\n```\n\n### 使用 VSCode 构建并运行 MJPC GUI 应用程序\n我们推荐使用 [VSCode](https:\u002F\u002Fcode.visualstudio.com\u002F) 及其 2 个扩展（[CMake Tools](https:\u002F\u002Fmarketplace.visualstudio.com\u002Fitems?itemName=ms-vscode.cmake-tools) 和 [C\u002FC++](https:\u002F\u002Fmarketplace.visualstudio.com\u002Fitems?itemName=ms-vscode.cpptools)）来简化构建过程。\n\n1. 打开克隆的目录 `mujoco_mpc`。\n2. 使用 CMake 配置项目（VSCode 中应会出现弹出窗口）\n3. 将编译器设置为 `clang-12`。\n4. 以 \"release\"（发布）模式构建并运行 `mjpc` 目标（VSCode 默认为 \"debug\"（调试））。这将打开并运行图形用户界面。\n\n### 构建问题\n如果遇到构建问题，请参阅 [Github Actions 配置](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fblob\u002Fmain\u002F.github\u002Fworkflows\u002Fbuild.yml)。这提供了我们用于在 Ubuntu 20.04 和 macOS-12 上构建 MJPC 进行测试的确切设置。\n\n# Python API\n我们为 MJPC 提供了一个简单的 Python API。此 API 仍处于实验阶段，需要用户具备一定的经验。例如，正确的用法要求（在 Python 中定义的）模型与 MJPC 任务（即在 C++ 中定义的残差和转移函数（residual and transition functions））相互兼容。目前，Python API 不提供任何特定的错误处理来验证这种兼容性，如果没有关于 MuJoCo 和 MJPC 的更深入知识，可能难以调试。\n\n## 安装\n\n### 前置条件\n1. 构建 MJPC（参见上述说明）。\n\n2. Python 3.10\n\n3. （可选）创建一个带有 **Python 3.10** 的 conda 环境：\n```sh\nconda create -n mjpc python=3.10\nconda activate mjpc\n```\n\n4. 安装 MuJoCo\n```sh\npip install mujoco\n```\n\n### 安装 API\n接下来，切换到 python 目录：\n```sh\ncd python\n```\n\n安装 Python 模块：\n```sh\npython setup.py install\n```\n\n测试安装是否成功：\n```sh\npython \"mujoco_mpc\u002Fagent_test.py\"\n```\n\n示例脚本位于 `python\u002Fmujoco_mpc\u002Fdemos` 中。例如从 `python\u002F` 目录：\n```sh\npython mujoco_mpc\u002Fdemos\u002Fagent\u002Fcartpole_gui.py\n```\n将通过 Python 使用 MuJoCo 的被动查看器（passive viewer）运行 MJPC GUI 应用程序。\n\n### Python API 安装问题\n如果安装失败或过早终止，我们建议删除 MJPC 构建目录并从头开始，因为构建很可能已损坏。此外，请删除 `python\u002F` 目录中安装过程中生成的文件。\n\n## 预测控制\n\n请参阅 [预测控制](docs\u002FOVERVIEW.md) 文档以获取更多信息。\n\n## 贡献\n\n请参阅 [贡献](docs\u002FCONTRIBUTING.md) 文档以获取更多信息。\n\n## 已知问题\n\nMJPC 不是生产级软件，它是一个 **research prototype (研究原型)**。其中可能存在缺失的功能和明显的 bug (错误)。如果您发现任何问题，请在 [issue tracker (问题追踪器)](https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fissues) 中报告。下面列出了一些已知问题，包括我们正在积极处理的项目。\n\n- 我们尚未在 Windows 上测试 MJPC，但原则上应该没有问题。\n- Task specification (任务规范)，特别是在 XML (可扩展标记语言) 中设置 norms (范数) 及其 parameters (参数)，有点笨拙。我们仍在迭代设计。\n- Gradient Descent (梯度下降) 搜索步长与 cost function (成本函数) 的尺度成正比，需要针对每个任务进行 tuning (调整) 才能良好工作。这不是 bug，而是 vanilla gradient descent (原始梯度下降) 的特性。或许可以通过某种 gradient normalisation (梯度归一化) 来改善这一点，但我们尚未彻底调查。\n\n## 引用\n\n如果您在工作中使用 MJPC，请引用我们伴随的 [preprint (预印本)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2212.00541)：\n\n```bibtex\n@article{howell2022,\n  title={{Predictive Sampling: Real-time Behaviour Synthesis with MuJoCo}},\n  author={Howell, Taylor and Gileadi, Nimrod and Tunyasuvunakool, Saran and Zakka, Kevin and Erez, Tom and Tassa, Yuval},\n  archivePrefix={arXiv},\n  eprint={2212.00541},\n  primaryClass={cs.RO},\n  url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2212.00541},\n  doi={10.48550\u002FarXiv.2212.00541},\n  year={2022},\n  month={dec}\n}\n```\n\n## 致谢\n\n使此仓库公开可用的主要工作由 [Taylor Howell](https:\u002F\u002Fthowell.github.io\u002F) 和 Google DeepMind Robotics Simulation (机器人仿真) 团队完成。\n\n## 许可证和免责声明\n\n所有其他内容版权所有 2022 DeepMind Technologies Limited，并根据 Apache License (Apache 许可证), Version 2.0 进行许可。此许可证的副本提供在本仓库顶层的 LICENSE 文件中。您也可以从 https:\u002F\u002Fwww.apache.org\u002Flicenses\u002FLICENSE-2.0 获取。\n\n这不是 Google 官方支持的产品。","# MuJoCo MPC (MJPC) 快速上手指南\n\nMuJoCo MPC (MJPC) 是由 Google DeepMind 开发的交互式应用程序和软件框架，用于基于 [MuJoCo](https:\u002F\u002Fmujoco.org\u002F) 的实时预测控制。它支持多种规划器（如 iLQG、Gradient Descent 和 Predictive Sampling），帮助用户轻松编写和解决复杂的机器人任务。\n\n## 环境准备\n\nMJPC 主要在 **Ubuntu 20.04** 和 **macOS-12** 上测试通过。\n\n### 系统依赖\n\n**macOS**\n需安装 [Xcode](https:\u002F\u002Fdeveloper.apple.com\u002Fxcode\u002F)，然后安装 `ninja` 和 `zlib`：\n```sh\nbrew install ninja zlib\n```\n\n**Ubuntu 20.04**\n安装编译所需的依赖包：\n```sh\nsudo apt-get update && sudo apt-get install cmake libgl1-mesa-dev libxinerama-dev libxcursor-dev libxrandr-dev libxi-dev ninja-build zlib1g-dev clang-12\n```\n\n### Python API 额外依赖（可选）\n如需使用 Python 接口，请确保安装 **Python 3.10** 及 MuJoCo：\n```sh\npip install mujoco\n```\n\n## 安装步骤\n\n### 1. 克隆仓库\n```sh\ngit clone https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\ncd mujoco_mpc\n```\n\n### 2. 创建构建目录\n```sh\nmkdir build\ncd build\n```\n\n### 3. 配置项目\n根据操作系统选择对应的配置命令：\n\n**macOS-12**\n```sh\ncmake .. -DCMAKE_BUILD_TYPE:STRING=Release -G Ninja -DMJPC_BUILD_GRPC_SERVICE:BOOL=ON\n```\n\n**Ubuntu 20.04**\n```sh\ncmake .. -DCMAKE_BUILD_TYPE:STRING=Release -G Ninja -DCMAKE_C_COMPILER:STRING=clang-12 -DCMAKE_CXX_COMPILER:STRING=clang++-12 -DMJPC_BUILD_GRPC_SERVICE:BOOL=ON\n```\n> **注意：** gRPC 依赖较大，首次下载可能需要 10-20 分钟。\n\n### 4. 编译\n```sh\ncmake --build . --config=Release\n```\n\n### 5. 安装 Python API（可选）\n如果需要使用 Python 接口，进入 python 目录并安装：\n```sh\ncd ..\u002Fpython\npython setup.py install\n```\n\n## 基本使用\n\n### 运行图形界面 (GUI)\n编译完成后，进入 bin 目录启动应用程序：\n```sh\ncd ..\u002Fbin\n.\u002Fmjpc\n```\n\n### 使用 Python API 示例\n如果已安装 Python 模块，可以运行示例脚本（例如倒立摆 GUI）：\n```sh\npython mujoco_mpc\u002Fdemos\u002Fagent\u002Fcartpole_gui.py\n```\n这将通过 Python 调用 MuJoCo 的 passive viewer 运行 MJPC GUI 应用程序。\n\n> **提示：** Python API 目前仍处于实验阶段，使用时需确保 Python 定义的模型与 C++ 定义的 MJPC 任务兼容。","某机器人算法团队正在研发一款四足机器人，需要为其设计能够在复杂崎岖地形上稳定行走的运动控制策略。\n\n### 没有 mujoco_mpc 时\n- 工程师需手动推导复杂的动力学模型公式，耗时数周且极易出现数学错误。\n- 传统强化学习训练周期漫长，调整参数后需重新训练数小时才能验证效果。\n- 缺乏交互式可视化工具，调试过程如同“黑盒”，难以直观定位平衡失效的具体原因。\n- 若要尝试 iLQG 等不同规划算法，需要从头编写底层代码，开发复用性极差。\n\n### 使用 mujoco_mpc 后\n- 基于 MuJoCo 物理引擎，无需手动推导公式，通过配置文件即可快速定义机器人任务。\n- 支持实时预测控制，参数调整后能立即在 GUI 界面中看到机器人步态的实时变化。\n- 提供直观的图形化调试界面，可实时观察受力分析和轨迹预测，快速锁定失衡点。\n- 内置多种高性能规划器（如 Predictive Sampling），直接调用即可实现复杂的运动合成。\n\nmujoco_mpc 将复杂的预测控制流程标准化，大幅降低了机器人运动策略的开发门槛与迭代周期。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgoogle-deepmind_mujoco_mpc_7562f0da.jpg","google-deepmind","Google DeepMind","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fgoogle-deepmind_06b1dd17.png","",null,"https:\u002F\u002Fwww.deepmind.com\u002F","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind",[83,87,91,95,99],{"name":84,"color":85,"percentage":86},"C++","#f34b7d",79,{"name":88,"color":89,"percentage":90},"Python","#3572A5",16.6,{"name":92,"color":93,"percentage":94},"CMake","#DA3434",3.7,{"name":96,"color":97,"percentage":98},"Jupyter Notebook","#DA5B0B",0.6,{"name":100,"color":101,"percentage":102},"Objective-C++","#6866fb",0.1,1591,255,"2026-04-05T08:37:41","Apache-2.0",4,"Linux, macOS","未说明",{"notes":111,"python":112,"dependencies":113},"Windows 系统未经测试但理论上可用。gRPC 依赖较大，首次下载需 10-20 分钟。Python API 处于实验阶段，需确保 Python 模型与 C++ 任务兼容。建议使用 VSCode 配合 CMake 工具构建。该软件为研究原型，非生产级质量。","3.10",[114,115,116,117,118,119,120],"cmake","ninja","clang-12","mujoco","zlib","gRPC","libgl1-mesa-dev",[13,54],[123,124,125,117,126,127],"model-predictive-control","mpc","mpc-control","predictive-control","motor-control","2026-03-27T02:49:30.150509","2026-04-06T05:27:14.737700",[131,136,141,146,151,156],{"id":132,"question_zh":133,"answer_zh":134,"source_url":135},4633,"在 Ubuntu 系统上编译 MJPC 遇到错误怎么办？","建议使用 Clang 12 编译器而不是 GCC。在 Ubuntu 20.04 上，可以尝试以下命令进行编译：\n1. `sudo apt-get install clang-12`\n2. `mkdir build && cd build`\n3. `CXX=clang12++ CC=clang12 cmake ..`\n4. `make -j 5`\nGCC 编译器可能在编译到 70% 时出错，Clang 12 是主要测试过的配置。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fissues\u002F280",{"id":137,"question_zh":138,"answer_zh":139,"source_url":140},4634,"Python 版本安装卡住或耗时过长如何解决？","请确保按照更新后的 README 说明进行操作。最新版本的安装脚本文件名为 `setup_direct.py`。如果是安装 estimator，请先运行 `python python\u002Fsetup_estimator.py install`。直接使用 `pip install` 可能会遇到问题，建议检查构建文件夹是否正常增长，并参考最新的安装指南。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fissues\u002F140",{"id":142,"question_zh":143,"answer_zh":144,"source_url":145},4635,"在 VSCode 中构建项目或运行测试失败怎么办？","1. 确保使用 Clang 14 编译器。\n2. 在 VSCode 中将 CMake 构建 variant 设置为 Release 模式（通过 Shift+Ctrl+P 选择）。\n3. 运行测试时，请进入 `mujoco_mpc\u002Fbuild\u002Fmjpc\u002Ftest` 目录运行 `ctest`，或者使用命令 `ctest -C Release --output-on-failure .`。\n4. 确保克隆的是最新仓库并正确配置了 CMake VSCode 扩展。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fissues\u002F125",{"id":147,"question_zh":148,"answer_zh":149,"source_url":150},4636,"编译时遇到 -Werror 警告错误或结构体赋值错误如何修复？","1. 对于警告视为错误的问题，可以尝试在 `cmake\u002FMpcOptions.cmake` 文件中移除 `-Werror` 选项，允许项目在有警告的情况下编译。\n2. 对于结构体赋值错误（如 `mjuiDef_`），可以在 `mjpc\u002Fagent.cc` 的相关行（如 366, 386, 389 行）添加 `(struct mjuiDef_)` 进行强制类型转换。\n3. Mac M1 用户也可能遇到类似问题，若目标 mjpc 能编译运行，部分警告可忽略。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fissues\u002F109",{"id":152,"question_zh":153,"answer_zh":154,"source_url":155},4637,"如何为四足机器人（如 A1）设置外部扰动（Perturbation）？","需要在源代码 `app.cc` 中设置 `xfrc_applied` 变量。具体参考代码行号（如 L327 和 L353）进行修改。不建议直接在任务文件中修改此参数，需在应用程序逻辑中施加外力以实现扰动效果。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fissues\u002F132",{"id":157,"question_zh":158,"answer_zh":159,"source_url":160},4638,"项目是否支持除 A1 以外的其他机器人模型？","项目主要基于 A1 机器人，但理论上支持类似结构的机器人。例如，可以参考 Issue #206 使用 Cassie 模型与 MJPC 配合。不过用户反馈 Cassie 行走调试较为困难。对于 Unitree Go1 等相似机器人，可能需要自行适配模型参数。","https:\u002F\u002Fgithub.com\u002Fgoogle-deepmind\u002Fmujoco_mpc\u002Fissues\u002F116",[162,167],{"id":163,"version":164,"summary_zh":165,"released_at":166},104123,"v0.1.0","A periodic update of the pinned version.\r\n\r\nThis version depends on a feature that hasn't yet been included in a release of MuJoCo at this time.","2024-02-23T13:51:34",{"id":168,"version":169,"summary_zh":170,"released_at":171},104124,"v0.0.1","To make it easier to refer to MJPC versions across dependent projects, we will occasionally tag commits we consider stable with a version number.\r\n\r\nWe intend to increase the version number when the version of MuJoCo we depend on is updated, and when dependent projects are likely to require changes to continue building (e.g. when the task definition API is modified).","2024-01-15T11:31:54"]