GATO documentation#
These pages are also published as the reference tier of the project site, https://a2r-lab.org/GATO/docs/, with the API reference and the current receipt.
Start with the quick start, then run the numbered examples. Build only the module needed by your example; the full receipt profile is for validation and research.
Question |
Reference |
|---|---|
What works, and what is experimental? |
|
Which module variant carries which feature, and what does an unsupported call raise? |
|
How do I use the Python API or migrate old code? |
|
How do I add limits, cones, collision or foot rows? |
|
How do I install from a wheel and use my own URDF? |
|
How do I use torch or JAX arrays? |
|
How do I use the solver from C++/CUDA? |
|
How do I add a robot or read the dynamics layer? |
|
How do I build, test, sign a receipt or time a change? |
|
What block size and launch does the merit kernel run at, and why? |
|
How do I set an orientation target, a robust tracking term, a cost on any linear feature or a centre-of-mass target? |
|
How do I move a tracking target every control step without re-applying the menu? |
|
Which linear solver runs by default, and why the direct solve from 256 knots? |
|
What does the solver’s |
|
Which integrator does a module run, and how do I get the RK4 step? |
Consumer contract: the integrator is a module property, example 20 |
When does Gauss-Newton fall short, and what does the exact-Hessian path add? |
|
Which results reproduce the paper? |
|
What do the current-code figures show? |
|
How is the project page updated? |
|
Where did historical datasets come from? |
Dated archaeology, history rather than current API guidance |
The project website presents the published experiments and the latest current-code results, each labeled. It is not a live benchmark dashboard. A green correctness receipt does not certify runtime speed, arbitrary robot models or safety on physical hardware.