Compute
Choose where to execute a saved training experiment. Local execution, an existing SSH host, and the initial RunPod integration have different setup and lifecycle requirements.
| Option | Current scope |
|---|---|
| Local · MuJoCo | Run the joint-feedback trainer with the runtime installed through Settings → Tools. |
| Cloud · SSH host | Run the same CPU workload on an existing host configured with SSH and the required Python environment. |
| Cloud · provision RunPod GPU | Provision and monitor a remote job through Compute. The current trainer still runs on CPU. |
Start locally
Use First robot experiment to validate the robot, task, and environment before choosing remote execution.
Remote execution
SSH execution requires a prepared host and a connected desktop while results return. RunPod provisioning requires a provider account, a session API key, and review of the job configuration and estimate. Retrieve results before terminating a pod; stopped pods can retain billable storage.
Follow the desktop compute setup guide for prerequisites and lifecycle details. Keep cloud-job files containing worker access tokens out of shared project copies.
Boundaries
The initial RunPod integration has not been validated with a live funded account. Provisioning a GPU does not accelerate the current CEM trainer. Isaac Sim, GPU-accelerated training, NVIDIA Brev, and live remote simulation streaming remain planned.
Edit this page on GitHub