LEROBOT / SO101 TELEOPERATION

A functioning leader–follower link is the beginning of a robot-learning experiment, not its conclusion.

Teleoperation turns human movement into repeatable robot-action demonstrations. The OpenRGD SO-101 work sessions confirmed working native LeRobot leader/follower motion, but did not yet establish quantitative trajectory tracking or policy autonomy.

MATURITYOPERATOR-CONFIRMED FIELD NOTE
LAST VERIFIED16 Oct 2026
SOURCE VERSIONLeRobot real-world robot workflow + OpenRGD operator notes · reviewed 16 Oct 2026
EVIDENCE BOUNDARYThe teleoperation result is an operator report. Public time series, latency distributions, controller error and independent replay traces are not yet archived.

DOCUMENTED LAB STATE

What our first smoke sessions actually established.

On 30 August 2026, the project operator reported that the leader arm drove the follower after configuring and calibrating the twelve bus servos. A second working teleoperation confirmation followed on 16 September 2026 after troubleshooting.

The work sessions also reported a local single-episode no-video recording and a successful replay. The evidence ledger records these separately. None of these facts is proof that a learned policy was trained or that the robot reached a measured autonomous task-success rate.

LE ROBOT PIPELINE

Native LeRobot gives us a real data-acquisition path.

The upstream LeRobot interface connects the follower robot and leader teleoperator with separate configs. The robot ID determines which calibration record is loaded. When cameras are configured, an observation stream can be added, but the historical OpenRGD smoke capture was explicitly no-video.

Use the current LeRobot CLI documentation as the version-specific source for commands. Do not assume a command copied from an older repository branch remains valid after framework upgrades.

01Calibrated leader→
02Human joint movement→
03Teleoperator sample→
04Follower action command→
05Physical joint state→
06Optional episode recording

WHAT TO MEASURE

Visual following is not yet a tracking benchmark.

To quantify leader/follower fidelity, record calibrated leader position, follower target, follower observed position, timestamps and connection faults. Compare per-joint lag, sign consistency, overshoot, steady-state error, dropped commands and reset behavior.

Feedback acquisition timestamps must be distinguished from timestamps at which software receives values. A read-back that arrives late is not a direct measurement of motor onset or mechanical response.

01

Correspondence

Are the leader and follower reporting comparable joint coordinates under matching calibration?

02

Lag

What is the distribution from observed leader motion to measured follower movement?

03

Tracking error

How close is the physical follower trajectory, not just the sent action?

04

Fault handling

What happens on port loss, unresponsive servo, bad calibration or safe-stop?

OPENRGD RUNTIME

Native LeRobot teleoperation is not the same as an OpenRGD physical runtime sign-off.

LeRobot acts as the native robot-interface and data-acquisition layer in this experiment. OpenRGD candidate contracts describe a separate semantic handoff through ActionIntent, somatic translation, capability grounding and safety gating.

The canonical OpenRGD repository is non-actuating. Passing a native LeRobot teleoperation smoke test does not automatically validate an independently governed embodied runtime or a Body Adapter.

PRIMARY SOURCES