Skip to main content
Cyberwave sits between AI models and physical machines. AI agents sit on top: they build the environment, plan the actions and write the workflows. Below them, models perceive through a twin’s sensors, a controller turns decisions into motion, and every run becomes data for the next model. The twin is the same object whether the robot is simulated or real, so what you test in simulation runs on hardware without code changes.

​
How it works

​
The four parts

AI agents

Assistants that build the environment, control the robot and write workflows from a plain-language request. You approve every plan.

AI models

Models that perceive (a cup at x = 212 px, “stop”), reason (plan steps from a goal) or act (VLA and RL policies that drive the robot). Any task, any modality, on the edge or in the cloud.

Control

Choose who drives the robot, from a person with a gamepad to a policy that learned the task: manual input, skills, workflows, VLA models, RL policies.

Learn

Every run can be recorded. Recordings become datasets, datasets train the next VLA or RL policy, and the policy deploys back to the same twin.

​
Where to start

New here? The quickstart has an agent build a scene and move a robot in about 10 minutes. To use your own coding agent, see Connect your AI tool. To write the loop in Python, follow Your first AI loop.

​
Where it runs

Models can move real hardware. Keep a person one click away from taking control, approve agent plans before they run, and test in simulation first.

​
Next steps

Ways to control a robot

The five kinds of controller and when to use each.

How hardware works

One twin model for arms, quadrupeds, drones and cameras.