1.8 KiB
Project guide
Start with Workflow when bringing an asset or policy into the platform. Use Agent setup to establish the environment and Usage for the CLI and configuration.
| Goal | Read |
|---|---|
| Inspect model readiness and choose a tier | Workflow |
| Set up native physics and verify a command | Tutorial and Hardware |
| Specify a keyboard or pad mapping | Controls |
| Train and measure a legged policy | Tutorial |
| Train on a Colab GPU | Colab |
| Package a native app or inspect a legacy bundle | Deploy guide and bundle format |
| Understand the backend seam and template lock | Design and Vendoring |
| Extend the platform | Contributing |
The checkout includes calibrated Go1, G1, six-legged and wheeled-biped examples,
covering all four locked families. G1 also has a separate compliant-gain variant.
Their asset READMEs distinguish execution checks, measured
learning and accepted policies; registration alone is not a performance or
hardware qualification. Add another robot with the onboarding tools to give it
a task ID. Run python scripts/train.py --list to list local registrations.
The development record indexes the measured example results
and generated applications from this refactor.
Native apps replay with MuJoCo and the Python requirements listed in the app. The legacy Rust bundle handles its own board, browser, and extension runtimes. Native applications currently cover model viewing, supported deterministic controllers, and compatible ONNX policies, including the wheeled-biped template's hybrid position and velocity actions. The legacy Rust runtime uses its position-action layout.