so-arm101-actuator 0.2.1

SO-ARM101 Actuator Protocol driver for robot-md-gateway. RPN-000000000002.

so-arm101-actuator

Actuator Protocol driver for the SO-ARM101 arm. RPN-000000000002 · pip install so-arm101-actuator

from so_arm101_actuator import SOArm101Actuator

actuator = SOArm101Actuator.from_default_port()  # /dev/ttyACM0 @ 1 Mbps
actuator.home()
actuator.move({"shoulder_pan": 0.3})
print(actuator.read_state())

Capabilities

• move(joint_positions, *, timeout_s=5.0) — drive named joints to target radians; blocks until in tolerance or timeout.
• home(*, timeout_s=10.0) — move to the configured zero pose.
• read_state() — snapshot positions + best-effort motor temperatures.

What this is

This is the first real actuator driver registered against the post-2026-05-09-reset RobotRegistryFoundation. It bridges Hugging Face / The Robot Studio's SO-ARM101 hardware to robot-md-gateway's Actuator Protocol via entry-point dispatch.

For a step-by-step install + register + run walkthrough, see the robot-md cookbook beat 8 (will land in Phase 3 of the roadmap).

For one operator's full setup story, see the Bob case study (also Phase 3).

Examples

python -m so_arm101_actuator.examples.wave
python -m so_arm101_actuator.examples.move_to_home
python -m so_arm101_actuator.examples.read_state

Architecture

Two layers:

• protocol.py — pure SCS/STS wire protocol on a serial.Serial-like.
• actuator.py — Actuator Protocol implementation; depends only on protocol.py.

This separation lets you swap the protocol layer (alternative servo lib, simulator) without touching capability code.

Hardware tests

SO_ARM101_HARDWARE=1 pytest tests/test_hardware.py -v

Skipped by default. Run on real hardware only.

Operator overrides (v0.2.0+)

Two env vars allow operator-specific tuning without code changes. Both are JSON, partially merged with defaults.

# Override HOME_POSE_RAD for one joint (e.g., gravity-load on shoulder_lift)
export SO_ARM101_HOME_POSE_RAD='{"shoulder_lift": 0.10}'

# Tighten SAFE_RANGE_RAD for shoulder_pan
export SO_ARM101_SAFE_RANGE_RAD='{"shoulder_pan": [-0.5, 0.5]}'

HOME_POSE_RAD can also be set via the home_pose_rad= constructor kwarg (kwarg overrides env).

Sweep CLI (v0.2.0+)

A developer-facing range-of-motion test that drives all 6 joints with random poses and prints a live status table.

# 100 iterations, default seed
python -m so_arm101_actuator.sweep

# Dry-run (no hardware): print 10 planned poses to JSONL
python -m so_arm101_actuator.sweep --dry-run --iterations 10 --seed 42 --out /tmp/poses.jsonl

The sweep CLI is for hands-on hardware bring-up + demos. Cert evidence (bob.local/FULL-SWEEP-100) does NOT come from this CLI; it comes from opencastor-ops/scripts/hil/ orchestrating pre-signed envelopes through robot-md-gateway.

Calibration (v0.2.1+)

The default SAFE_RANGE_RAD and MOVE_TOLERANCE_RAD values reflect calibration measurements taken on a specific SO-ARM101 rig (Bob) on 2026-05-10. Different physical rigs will have different mechanical stops and steady-state precision. Operators should:

# Override SAFE_RANGE_RAD per joint after measuring your own rig's reachable range
export SO_ARM101_SAFE_RANGE_RAD='{"elbow_flex": [-0.30, 0.95]}'

# Loosen MOVE_TOLERANCE_RAD if your arm has larger steady-state error (gravity-loaded joints)
export SO_ARM101_MOVE_TOLERANCE_RAD=0.05

See the sweep CLI (python -m so_arm101_actuator.sweep --iterations 5 --dry-run) to generate a sample pose list for your rig.

License

Apache-2.0.