Cyclo Lab
Cyclo Lab is the Isaac Lab-based environment for training AI Sapiens K1 policies in simulation. It provides the training and export workflow needed before deploying a policy with Sim2Real.
This guide is based on Cyclo Lab 3.0.0.
For robot-side deployment after training, use the Sim2Real guide.
Choose a Physics Backend
Cyclo Lab provides two physics backends, selected at runtime.
| Backend | Argument | Runtime |
|---|---|---|
| Newton (Default for K1) | physics=newton_mjwarp | Isaac Sim 6.2.0, Isaac Lab 3.0.0 |
| PhysX | physics=isaacsim_physx | Isaac Sim 6.2.0, Isaac Lab 3.0.0 |
Set Up the Docker Container
Run Cyclo Lab inside the provided Docker container. The container includes Isaac Sim, Isaac Lab, Cyclo Lab, and the Python dependencies used by the training scripts.
Install these on the host PC first:
- Docker and Docker Compose.
- NVIDIA Container Toolkit.
- NVIDIA GPU driver compatible with the Isaac Sim container image.
Clone Cyclo Lab with its submodules:
git clone https://github.com/ROBOTIS-GIT/cyclo_lab.git
cd cyclo_lab
git submodule update --init
Already cloned Cyclo Lab?
Initialize the submodules before starting the container:
git submodule update --init
Build, start, and enter the container:
./docker/container.sh build
./docker/container.sh start
./docker/container.sh enter
Inside the container, the repository is mounted at:
/workspace/cyclo_lab
Useful host-side commands:
| Command | Use |
|---|---|
./docker/container.sh build | Build the image. |
./docker/container.sh start | Initialize submodules and start the container. |
./docker/container.sh recreate | Recreate the container after rebuilding the image. |
./docker/container.sh enter | Open a shell inside the running container. |
./docker/container.sh stop | Stop the container. |
./docker/container.sh logs | Follow container logs. |
./docker/container.sh clean | Remove the container and image. |
After rebuilding, use recreate to apply the new image to an existing container.
The start script enables X11 forwarding when DISPLAY and xauth are available on the host.
Use the no-window option shown for your backend in the training guides when visualization is not required.
Next Steps
- Train and export a velocity policy with the Locomotion guide.
- Train and export a motion-imitation policy with the Mimic guide.
- Deploy the exported asset with How to Deploy Your Own Cyclo Lab-Trained Policy.