図面ゲートというサービスを立ち上げました。
2DCADのファイル変換アプリで、DXF/DWG/JWW/SXF/DGN/DWF/SVG/PDFをサポートしてます。
既に同様のサービスは世にありますが、DGNをサポートしてたり、ビューアの見やすさにこだわってたりするので良ければ使ってみてください。
zumen-gate.com/
freecad-mcp
Create and edit FreeCAD models from Claude Desktop
FreeCAD MCP lets Claude Desktop and other MCP clients control FreeCAD. You can create and edit models, run Python scripts, inspect documents, and run FEM analyses.
Examples include a toy car and modelling from a 2D drawing.
It uses an addon inside FreeCAD and an MCP server launched by the client. Keep FreeCAD open with its RPC server running while you work.
github.com/neka-nat/freecad-…
Kenta Tanaka | WillRobotics | Physical AI retweeted
性能优化工程师连夜失业……GitHub 刚刚炸出一个逆天到离谱的工具!
不动你 95% 的老代码,一行命令直接给程序提速 116 倍!
REA → Rust 重写万物 → RAEA 刚把人看麻,今天直接杀出终极怪兽 fast-anything(加速万物)
别再傻傻拿 Rust 重写整个项目了,它只给最慢的那 5% 做全自动“心脏搭桥”:
· 自动揪出性能瓶颈函数,录制 500 次真实运行调用
· 丢给 Claude Code / Codex,只把这几个函数重写成 Rust
· 真实数据 + 极端边界全量回放,100% 结果一致且真变快了才放行
· 随时一键回滚,原项目零破坏
Python / TS / Go / PHP / Ruby / Java / C++ 全支持,原生自带 Claude Code 插件和 MCP
官方真实跑分直接看傻: Python 4.91s → 42ms(116×)PHP 55× | Go 21× | TypeScript 20×
下周这帮疯子还打算端出什么逆天“万物”?
开源仓库和一行安装命令扔一楼 👇
#AI #Rust #开源项目 #ClaudeCode #独立开发 #GitHub
不会吧……刚看完 REA,现在连 RAEA 都杀出来了
它干的事简单粗暴到不可思议:
让你直接用 React 和 TypeScript,在本地 GPU 上光速生成视频、3D 动效、甚至可玩的游戏。
~ 用代码做超高清视频。
~ 用代码搭 3D 场景和物理动画。
~ 用代码当场写出一个能玩的游戏。
~ 彻底踢掉浏览器,90 帧 1080p 视频只要 1 秒出片。
~ 让 AI 代理自己去写、自己抽帧检查 Bug、自己修复。
~ 一键导出视频,一键导出独立游戏。
说白了,你手里的 Claude Code 和编程 Agent,从今天起正式兼职成了你的个人皮克斯工作室 + 独立游戏引擎
想象一下,我们居然正在亲眼见证这一切发生……
开源仓库和一键安装命令方评论区
🌀 变态开源项目 FieldCraft,结构受力、传热和流体仿真一个软件全包,永久免费,结果还逐项对着理论值验过。
FieldCraft 出自 Tangency Labs,就是做开源 CAD 软件 SolidCraft 的那个团队,SolidCraft 在 GitHub 上 1.7 万 star。FieldCraft 10 月中才建仓,11 月 19 日刚发 0.8.1,nightly 版本几乎每天一个。
以前做一次仿真,得先在 CAD 里画零件,导到网格软件里划网格,再塞进求解器,尺寸一改,整条链路从头再走一遍。
FieldCraft 直接接 SolidCraft 的参数化零件,四面体、全六面体网格自动划,线性静力、大变形和塑性、热传导、热应力耦合、显式动力学都用自家求解器算,流体走格子玻尔兹曼方法。还想用老工具收尾的,也能导出 Abaqus、Nastran、LS-DYNA 的输入文件。桌面版、命令行和 AI agent(通过 MCP)驱动的是同一套操作,Mac、Windows、Linux 和浏览器都有版本。
最让人放心的是它不藏短板:README 里的验证套件 50 个算例、176 个量,过了 170 个,剩下 6 个是写明了的已知失败。
如果你也在做结构或热仿真,值得收藏试试。
GitHub:github.com/tangency-labs/fie…
Kenta Tanaka | WillRobotics | Physical AI retweeted
Most trackers lose sight of an object once it changes shape…
[👇Code & Dataset]
an apple turns into slices, a caterpillar into a butterfly, and the model just gives up.
Researchers at Cornell built a new system called Track Any State that does something different: it follows objects through their transformations while describing what actually changed.
> Tracks objects even after transformation
> Detects and describes new object states
> Recovers missing tracks with TubeletGraph
> State-of-the-art accuracy under transformations
It’s a step toward models that understand how things evolve over time… not just where they move.
📍Paper: “Tracking and Understanding Object Transformations” (NeurIPS 2025)
Code & Dataset: tubelet-graph.github.io
——
Weekly robotics and AI insights.
Subscribe free: 22astronauts.substack.com
Kenta Tanaka | WillRobotics | Physical AI retweeted
What if you could build your own autonomous mobile robot using open-source hardware, ROS 2 and micro-ROS?
OpenAMRobot is an open-source mobile robotics platform designed for autonomous navigation and industrial applications.
Its hardware repository provides mechanical CAD, electrical schematics, wiring diagrams, component lists and manufacturing files.
The reference build combines a Raspberry Pi 5, Teensy 4.0, BLDC motors, wheel encoders, IMU and LiDAR.
The interesting part is how the project connects embedded motor control with ROS 2, SLAM, Nav2 and autonomous docking.
It can be used for warehouse automation, indoor logistics, research labs and mobile robot development.
図面ゲートのAPIの提供を開始しました。
2DCADではJw_cad形式/DWG・DWF/SXF形式などをサポートしており、3DCADではSolidworkds/Inventor/eDrawingsなどの形式をサポートしてます。
図面の内容の調査、レンダリング、変換といったAPIが使用できます。
zumen-gate.com/developers
図面ゲートというサービスを立ち上げました。
2DCADのファイル変換アプリで、DXF/DWG/JWW/SXF/DGN/DWF/SVG/PDFをサポートしてます。
既に同様のサービスは世にありますが、DGNをサポートしてたり、ビューアの見やすさにこだわってたりするので良ければ使ってみてください。
zumen-gate.com/
2
324
SolidCraft 0.40でカーネルがSolidCoreに置き換わった模様。
SolidCraftの開発チームが早速、Nvidiaが出したSMLibのRust移植を行っている。
対応が速いのと、すごい勢いでCADのオープン化が進んでいる。
github.com/tangency-labs/sol…
5
350
Today we’re open-sourcing Steel, our internal physics solver built with @NVIDIA.
It runs simulations up to 13× faster than PhysX while using ~3× less GPU memory.
Steel unlocks scaleable simulation for manipulation, for everyone.
Kenta Tanaka | WillRobotics | Physical AI retweeted
苹果研究团队给一台宇树 G1 后背驮了台 Mac mini。
钻桌底、侧身过窄门、自己找椅子坐,默认由本地的 Qwen3.6-35B-A3B 指挥,不用联网。
苹果自己的机器,背包里跑的却是千问。
真实厂商工业 CAD 碰撞 Three.js 的极客美学。
开发者将常见电子元器件做成 3D 解构库,最克制的是整个工程被压成单一个 .html 文件,零依赖完全离线渲染元器件爆炸图。
体验:breakboard.art
原推:xitter.cf/hckmstrrahul/status/21…
Collection of beautiful electronic components with their teardown anatomy → breakboard.art
Built using ThreeJS, with real manufacturer drawings and CAD. The .html file works even offline on any WebGL 2 browsers.
This was made possible by combining 3 distinct skills and some custom refinements for accuracy (open sourced):
• /teardown by @jakubkrehel → Technical illustrations that explain physical objects and how their parts work together.
• /anatomy by @wheresryan22 → Detailed, interactive isometric illustrations of machines and complex systems.
• /teardown by @elvar_official → Guidance for exploded views and component-level technical illustrations.
Link to these skills are in the Github (link below)
Kenta Tanaka | WillRobotics | Physical AI retweeted
Can reinforcement learning solve inverse kinematics precisely?
Introducing 𝗣𝗼𝗹𝗶𝗰𝘆𝗜𝗞 🦾, a goal-conditioned RL policy for high-precision inverse kinematics.
On a real arm, it reaches median errors of 𝟬.𝟱𝟬 𝗺𝗺 𝗮𝗻𝗱 𝟬.𝟭𝟮° and holds the goal pose on 𝟮.𝟱× 𝗺𝗼𝗿𝗲 𝘁𝗮𝘀𝗸𝘀 than the best classic IK controller.
🌐 tinaliu0123.github.io/Policy…
🧵👇
CUDAやHIPのプログラムをCPU上で動かすことができるGPUエミュレータ。
GPUが無い環境での開発やCIとかに使えそう。
github.com/pantheongpu/panth…
15
941
Kenta Tanaka | WillRobotics | Physical AI retweeted
In 2022, our first cuRobo prototype took 100 ms and consumed the whole GPU.
Today, we’re open-sourcing cuRoboV2-Fast under Apache 2.0: collision-free, torque-limited motion planning in 4.4 ms.
Collaborating with researchers across robotics and GPU architecture brought us to 50 ms, then 30 ms. B-splines, GPU time optimization, and faster kernels brought us to 4.4 ms while producing motions that execute faster.
As planning got faster, perception, grasping, and control became the bottlenecks. We built a 7 ms perception pipeline (SceneFields): from stereo RGB to depth, voxels, and robot and object segmentation, and accelerated GraspGenX to 1.6 ms. We also integrated wrote a cpp controller for YAM and built a low-latency CUDA-GL visualizer for an interactive live view of the task.
We deploy the stack as a two-process application: one dedicated to real-time control, the other handling perception, grasping, planning, and visualization.
On a real YAM arm, pick-and-place cycles averaged 6 seconds, 34.6% shorter than cuRoboV2.
A long journey, made possible by amazing collaborators. The rest of the stack is coming soon.
research.nvidia.com/labs/lpr…
Joint work with @Adithya_Murali_ , @BirchfieldStan
Prompt your agent with :
Use cuRobo cpp_runtime for collision-free motion planning: github.com/NVlabs/curobo/blo…





















