Prerequisites
Python 3.13+
Required for all installation methods
Rust toolchain
Required for source installation only
uv (optional)
Faster package management
System requirements
- Python: 3.13 or 3.14
- Operating System: Linux, macOS, or Windows (Linux recommended for hardware)
- Rust: 1.70+ (source installation only)
Quick install
For most users, install the published package from PyPI:- Core Python package (
rfx) - Rust extension for real-time control
- Essential dependencies (tinygrad, numpy, PyYAML)
- CLI tool (
rfxcommand)
Why uv? uv is a fast Python package installer written in Rust. It’s significantly faster than pip and handles dependency resolution more reliably.Install uv:
curl -LsSf https://astral.sh/uv/install.sh | shOptional dependencies
rfx has several optional feature groups. Install only what you need:Simulation backends
sim-genesis: torch, genesis-world ≥0.4.0 (GPU recommended)sim-mjx: torch, mujoco ≥3.0, jax[cuda13], warp-langsim-mock: torch only (runs on CPU)
Teleoperation and data collection
- opencv-python ≥4.8 (camera capture)
- torch ≥2.2
- mcap ≥1.1.0 (logging)
LeRobot integration
- Direct export to LeRobot datasets
- HuggingFace Hub push/pull
- Compatible with
lerobotpackage
LLM agent integration
- anthropic ≥0.25.0 (Claude)
- openai ≥1.0.0 (GPT)
Development tools
- pytest ≥7.0, pytest-asyncio
- ruff ≥0.1 (linting and formatting)
- mypy ≥1.0 (type checking)
- pre-commit ≥3.6
All optional features
Install everything:Install from source
For development or bleeding-edge features:.venv/requirements-dev.txtpip install -e)Custom virtual environment path
Override the default.venv location:
Manual source installation
If you prefer to install manually:Run without installation
Useuv run to execute rfx commands without installing:
- Quick testing without environment setup
- Running examples from the repo
- CI/CD pipelines
Verify installation
Check that everything is working:not available for optional packages is normal. Only install what you need.Troubleshooting
Rust extension not found
Ifrfx doctor reports rfx._rfx missing:
Backend probe fails
If GPU backends (CUDA, Metal) don’t appear:- Check that you have appropriate drivers installed
- Install tinygrad runtime support:
- Verify with:
Import errors
If you seeModuleNotFoundError: No module named 'rfx':
- Ensure you activated the virtual environment
- Reinstall in editable mode:
uv pip install -e . - Check that you’re in the repo root directory
Serial port permissions (Linux)
For hardware access, add your user to thedialout group:
Development setup
For contributors or those modifying rfx source:Enable git hooks
Setup pre-commit hooks for quality gates:pre-commit: Rust format check + Ruff on staged filespre-push: Full test suite (Rust + Python)
Use Moon task runner
rfx uses Moon for monorepo task orchestration:.moon/workspace.yml for configuration.
Manual quality checks
Without Moon, use the Python checks script:Next steps
Quickstart
Get your first policy running in minutes
SO-101 Setup
Configure your SO-101 robotic arm
Simulation
Set up simulation backends
Python SDK
Explore the API reference
