Skip to main content
Robotics is in beta. Model APIs and integrations may change. Check each model’s release notes before updating your integration.
Run robot policies or generate visual observations on Reactor’s GPU cloud. To use an action policy, send camera images, measured robot state, and a task instruction through the Reactor API, and receive predicted actions for your controller. Your Python client runs alongside your robot or simulator; your local controller validates and executes the returned actions. The action policies below support the DROID embodiment: a Franka arm and gripper. Praxis-1 also supports a bimanual YAM rig. Each model has its own camera names, state encoding, and prediction protocol. Start with How the API works, then choose a model.

Choose an action policy

Each embodiment has its own variant guide, which defines its camera setup, observation and action contract, and integration requirements. Matching tensor dimensions alone does not establish compatibility with a different robot.
More action policies are coming soon.
Prices are in USD per minute of session time, metered per second. See Pricing & Billing for billing and session termination details.

Try it with your robot or simulator

Start with your chosen policy’s quickstart to install the Python client and send an example request. The example uses test inputs so you can try the API before connecting cameras or a robot. Once you can send inputs and receive actions, evaluate a task in simulation or on a physical robot using observations from your setup. For simulation, follow your chosen model’s simulation guide. Each guide explains the simulator setup and whether it includes runnable connection code or describes the adapter you need to build. When you’re ready to use a physical robot, follow the model’s integration guide to send camera images and robot state, then pass the returned actions to your controller. Your application then controls when the robot moves and enforces its safety limits. The model’s API reference covers the exact inputs, action format, and settings you’ll use along the way.

Simulation

These models generate visual observations; the action policies above predict robot actions. See the Simulation guides to get started. Prices are in USD per minute of session time, metered per second—not per minute of generated video. See Pricing & Billing for billing and session termination details.

Bring your own model

Have your own model code and weights? Reactor Serve is in private beta. Request an invitation to deploy your model on Reactor.