Robot stack
Middleware, navigation, integration, and reproducible environments.
Physical AI skills for coding agents
Robium is an open-source robotics development harness that helps your coding agent build robotics projects in minutes. It combines tested skills, common troubleshooting steps, and runnable reference applications.
Install Robium
npx robium-ai setup npx robium-ai setup --agent claude npx robium-ai setup --agent codex npx robium-ai setup --agent gemini npx robium-ai setup --agent cursor Requires Node.js 18+, Git, and a supported coding agent. Choose a folder during setup; it includes editable skills and reference apps.
Your workspace, your code
Pick any folder name or location. By default, setup creates:
~/robium/ ├── robium/ # skills and plugin └── robium-apps/ # reference examples
Or choose your own parent folder:
npx robium-ai setup --dir ~/projects/my-robotics Use it: open your agent in your own project and choose a prompt below. Robium remembers where the examples live. App environments and models are set up when needed.
Claude Code installs Robium for your user account, so its skills are available across projects. Restart your agent after setup; access to files outside a project may require permission.
Make it yours: create a Git branch in either checkout, edit, and test. Contribute a pull request whenever you’re ready.
Find your folders from anywhere:
npx robium-ai workspace Check for updates without changing your working files:
npx robium-ai update --check When you’re ready, npx robium-ai update updates clean main branches from Robium’s official repositories. Your branches and local edits stay untouched. Quiet checks happen at most daily during example discovery, with notices at most weekly—not on every prompt. Set ROBIUM_UPDATE_CHECKS=0 to turn them off.
Choose a prompt and start building.
Start with a working robot example, then make it your own.
Help me build a simulated robot assistant that understands what it sees and follows natural-language instructions.
Simulation · Docker · Gemini API accessHelp me run a pretrained policy that transfers a cube between two simulated robot arms.
Simulation · Pretrained model · No training requiredSkills for every part of your robotics ecosystem
Focused skills for the way you build: your robot platform, your models, and your compute. Pick the skills that fit your project.
Simulation
Build and test the robot before hardware enters the loop.
AI & ML & Data
Use public robot datasets and state-of-the-art models to train and evaluate robot policies.
Visualization
Inspect live robots, simulations, rollouts, and recorded data.
Testing & deployment
Test your application, then run it on your robot, in simulation, or on a remote server.
Continuous learning & publishing
Capture what works, improve the skills, and publish proven applications.
Why use Robium?
Already working with a coding agent? Add Robium skills to give it the robotics know-how to make better-informed choices and spend less time troubleshooting.
Grounded in real applications
A growing collection of runnable reference applications gives your agent practical examples to learn from and build on.
Less time troubleshooting
Equip your agent with known issues, diagnostic steps, and practical fixes for common robotics problems.
From architecture to deployment
Give your agent focused skills to plan, build, simulate, train, test, and deploy your robotics project.
Works with your agent
Use the same skill library with Claude Code, Codex, Gemini CLI, or Cursor.
Use common robotics patterns
Give your agent proven approaches for environments, integrations, simulators, and compatible versions.
Open source with growing support
A growing library of skills and reference applications supports more robot platforms, models, simulators, and tools.
How Robium works
Robium loads the right skills into your coding agent and grounds its work in your project context. Your agent builds and tests against the robot, simulation, and cloud targets you choose, while Robium captures lessons for future work.
Explore interactive demos
Try robotics applications built with Robium skills directly in your browser.
Robot Navigation
Drive a TurtleBot3 through Gazebo, build a map, and send it to waypoints with ROS 2 and Nav2.
Pi0.5 VLA pick and place
Explore Pi0.5 rollouts of a Panda arm following a language instruction to put a bowl on a plate.
Quadruped locomotion
Compare recorded checkpoints as a Unitree Go2 learns to walk, from early attempts to a trained policy.
Reference applications, ready to build on
Clone, test, and adapt complete reference applications for your own robotics project.
Robot Navigation
Map a Gazebo world and tune navigation with ROS 2 and Nav2. Runs in Docker; no GPU or local ROS installation required.
ACT ALOHA Cube Transfer
Replay cube transfers and tune the action horizon in MuJoCo. Runs with uv on macOS; no NVIDIA GPU required.
PushT with Diffusion Policy
Change block layouts and denoising steps, then rerun a pretrained policy. Local setup uses uv and ffmpeg; no NVIDIA GPU required.
Latest from the blog
Practical tutorials and engineering notes from building robotics applications.
Mapping and navigation with ROS 2, from one app
Map a simulated TurtleBot3 with SLAM Toolbox, reuse the map with AMCL, save waypoints, and send Nav2 goals from a browser.
Running a Pi0.5 VLA pick-and-place policy with a Franka Panda
See how Physical Intelligence's Pi0.5 turns camera views, robot state, and a language instruction into actions for a simulated Panda arm.
Building a mobile home robot with Gemini Robotics ER 2 and ROS 2 in under an hour
A TurtleBot 4 sends camera and robot state to Gemini Robotics ER 2, while a deterministic guard, ROS 2, and Nav2 stay in charge of motion.
Training and running an ALOHA cube-transfer policy with ACT
Run the official ACT cube-transfer policy locally, see why action chunking mattered, and use the same workflow as a starting point for your own demonstrations.
Running Diffusion Policy on PushT from a laptop
Run a compact, verified robot-learning experiment on a Mac, then use it to understand when ACT, Diffusion Policy, or a VLA fits.
Training a Unitree Go2 to walk with Isaac Lab, checkpoint by checkpoint
A recovered Isaac Lab run turns policy training into a recorded progression, backed by a fresh capability-scoped GPU smoke.
Common questions
Which coding agents does Robium work with?
Claude Code, Codex, Gemini CLI, and Cursor. One command,
npx robium-ai setup, detects the agents on your machine
and sets up each one (or target one: --agent codex). The
skills follow the open Agent Skills format, so any agent that speaks
it can read them.
Do I need a robot or a GPU?
No. Many reference applications can run on a local CPU or on a Mac using MPS. If you don't have robot hardware, you can still explore, test, and build with simulations. For workloads that need more compute, you can use a remote GPU, with Robium skills to help you deploy your application to the cloud.
What exactly is a skill?
Think of a skill as a practical guide for your coding agent. Each one covers a specific robotics task with proven approaches, troubleshooting steps, references, and runnable examples.
Why not just ask my agent directly?
Coding agents know a lot, but they still need reliable context for your robotics stack. Robium gives them tested guidance, common pitfalls, and working patterns so they can make better choices and spend less time guessing.
How do skills stay correct as the ecosystem moves?
Robium includes a learning engine that captures what your agent learns during build sessions, including problems and fixes. You can review those learnings and use them to update the relevant skills, keeping the guidance current and useful.
Can it capture my team's own knowledge?
Yes. Capture hooks log what breaks and what works during your builds, and the skill-author workflow turns that into skills for your own stack (your conventions, your hardware, your infra), hardened by the same loop that maintains the public catalog.
Is it free?
Yes. Robium is MIT-licensed. The plugin, full skill catalog, and CLI are open source in the Robium repository, and the reference applications are open source in the Robium Apps repository.
Can I contribute a skill?
Yes, and the contribution unit is deliberately small: one skill. Copy the template, pass the validator, open a PR.