Skip to main content
Cyberwave gives you five ways to control the robots in an environment. They are alternatives, not steps: pick the one that fits the task, and switch whenever you like. Each drives the same twins, in simulation or on hardware, and exactly one controller drives a given twin at any moment. Switching between them doesn’t stop telemetry or recording, so a session can start manual, hand over to a model, and fall back to a person if something goes wrong.

​
Assign a controller

  • In the dashboard: select the twin and use Assign Controller. A person can take over from the same panel at any time.
  • From Python: twin.policy.list() shows the controllers available to a twin, twin.policy.assign(policy) hands it over, and twin.policy.ensure_attached() lets your script drive it.
  • From an agent: the control agent plans an action or a skill run, and you approve it before it executes.
How controllers are routed, where each type runs, and how code controllers and RL checkpoints run on a controller host: Controllers and handover.

​
Where the models come from

Manual sessions, skills and workflows all produce recordings. Recordings become datasets, and datasets train VLA models and RL policies. See Data for training.

​
Next steps

Your first AI loop

A camera, a model and an SO-101 twin that picks and sorts.

Train a VLA on the SO-101

Record demonstrations, fine-tune, deploy the model as the controller.

Build a workflow

Triggers, models and robot commands on a canvas.

AI agents

Let an assistant plan the action, then approve it.