Cloud inference for your robot
Dropbear runs robot policies in the cloud while your local controller keeps ownership of sensing, actuation, and safety.Integrate with Python
Send one real robot observation and inspect the returned action chunk
without moving the robot.
Set up an SO-101
Configure, calibrate, diagnose, dry-run, and safely start the supported arm
workflow.
What Dropbear owns
- authenticated cloud sessions;
- policy worker allocation and startup;
- observation transport and action-chunk delivery;
- automatic QUIC-to-relay transport selection;
- session lifecycle and usage reporting.
What you own
- cameras and robot state;
- conversion into the selected model’s observation contract;
- action validation and robot-coordinate conversion;
- local watchdogs, limits, collision handling, and e-stop behavior;
- deterministic session cleanup.
Open sessions consume prepaid credits. Use a context manager or call
close() when work is complete. View current availability, balance, and
usage in the
dashboard.Choose your next step
- Have a controller already? Start with the Python quickstart.
- Need package or platform details? Read Installation.
- Need exact shapes and units? Open Model contracts.
- Something failed? Use Troubleshooting.