> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dropbear.dreamscalelabs.com/llms.txt
> Use this file to discover all available pages before exploring further.
> ## Agent Instructions
> Treat the Dropbear Python SDK and CLI as the supported public integration surfaces.
> Prefer context-managed sessions and model-specific observation helpers.
> Do not infer or recommend internal control-plane APIs.
# SO-101
> Configure, calibrate, diagnose, dry-run, and safely confirm motion on an SO-101.
This is Dropbear's turnkey hardware path. It configures one SO-101 follower arm,
a side camera, and a distinct wrist camera before running a dry prediction loop.
Motor commands remain disabled until the dry run finishes and the operator
explicitly confirms motion.
Keep the robot inside a clear workspace with a reachable physical power stop.
Stay close enough to stop it, but keep people, cables, and loose objects
outside its swept volume. A keyboard stop supplements the physical stop; it
does not replace it.
## Requirements
* Python 3.12 or 3.13;
* an SO-101 follower arm supported by LeRobot 0.5.1;
* one side RGB camera and one distinct wrist RGB camera;
* a terminal attached to a TTY for the keyboard e-stop;
* a Dropbear account with available credits.
## 1. Install and authenticate
```bash theme={null}
uv tool install "dropbear[so101]==0.1.0a2"
dropbear login
dropbear doctor
```
`dropbear login` initializes motion as disabled on a new machine.
## 2. Prepare the workspace
Before connecting:
1. Secure the arm base and both cameras.
2. Route the wrist-camera cable so no joint can pull or trap it.
3. Remove tools and unused props from the swept volume.
4. Test the robot's physical power stop.
5. Place the terminal where `Space`, `q`, or `Ctrl-C` is immediately reachable.
### Side camera
Mount the side camera 60–80 cm from the table, approximately side-on to the
arm, with the whole workspace and arm visible. Keep exposure, white balance,
and the mount stable during a run.
### Wrist camera
Use a second physical camera mounted on the arm. Keep the gripper and immediate
workspace visible, secure the cable through the full joint range, and verify
that it is not the same device index as the side camera.
## 3. Detect the arm and cameras
```bash theme={null}
dropbear robots so101 setup
```
The setup flow:
1. verifies your API key;
2. detects or confirms the serial port;
3. previews and saves the side camera;
4. previews and saves a distinct wrist camera;
5. starts warming the SO-101 policy worker while you finish setup.
Configuration is written to `~/.dropbear/config.toml`. If setup starts a cloud
session and you stop here, close it with:
```bash theme={null}
dropbear sessions list
dropbear sessions stop
```
## 4. Calibrate
Clear the table, then run the guided LeRobot calibration:
```bash theme={null}
dropbear robots so101 calibrate
```
Follow the prompts to place the arm near the middle of its range and sweep each
joint. Dropbear writes the LeRobot 3.0 calibration format to
`~/.dropbear/calibration/so101.json`.
Do not continue if the calibration exits early or the saved joint ranges look
wrong. Re-run calibration with the workspace clear.
## 5. Pass hardware diagnostics
```bash theme={null}
dropbear doctor so101
```
The command checks:
* Python and LeRobot versions;
* the configured serial port;
* calibration presence and format;
* distinct side and wrist camera capture;
* side-camera framing against the reference;
* the current network path when a warmed session is available.
Expected result:
```text theme={null}
All hardware checks passed. Next: dropbear run pick-place
```
Fix every failure and re-run the command. A skipped network preflight is
acceptable when no warm session exists; connection performs the check again.
## 6. Run without motion
Stage a red block and target square, keep the workspace clear, and run:
```bash theme={null}
dropbear run pick-place --duration 60
```
On a newly configured machine, the terminal must show:
```text theme={null}
motion: dry-run
DRY RUN (no motion until you enable it)
```
The arm is observed but not commanded. Inspect the `would move` lines for:
* six finite values in the documented joint order;
* plausible targets near the current pose;
* stable camera framing;
* no unexpected discontinuities.
Press `Space`, `q`, or `Ctrl-C` to test the software e-stop during the dry run.
Do not approve motion if the stop path, cameras, calibration, or targets are
wrong.
## 7. Confirm one live run
After a successful dry run, the same command asks:
```text theme={null}
Dry run complete. Enable real motion? [y/N]
```
Before entering `y`, confirm again that:
* the physical stop works and is reachable;
* one operator is watching the entire workspace;
* no person or object is inside the swept volume;
* calibration and diagnostics passed in this setup;
* the dry-run targets were plausible.
Enter `y` only when all checks pass. Dropbear then persists motion enablement
and continues on the same warmed session. The turnkey loop clamps each
per-tick joint change, checks the keyboard e-stop every tick, and freezes stale
actions after a disconnect.
Persisted motion means later `dropbear run pick-place` invocations can start
live after diagnostics. To return to a safe default, set
`enabled = false` under `[motion]` in `~/.dropbear/config.toml` before the
next run.
## Stop and recover
* Press `Space`, `q`, or `Ctrl-C` for the software e-stop.
* Use the physical power stop for unexpected motion or any loss of confidence.
* After a disconnect or timeout, leave the arm stopped, inspect
`dropbear sessions list`, and run `dropbear doctor so101` before retrying.
* If UDP is unavailable, the default command may use the hosted relay and will
report reduced responsiveness. Use `--no-fallback` when the run must refuse
that path.
* If a session remains open, terminate it with
`dropbear sessions stop `.
For controller-owned integrations, return to the
[Python SDK](/sdk/overview). For help with a failed safety check, email
[team@dreamscalelabs.com](mailto:team@dreamscalelabs.com).