Level-six engineering validation protocol · 01
A sixth-level engineering protocol for turning an assembled SO-101 pair into an identifiable, repeatable and reviewable research baseline.
Protocol scope
For one declared leader/follower pair used on a guarded bench for calibration, teleoperation and low-energy data collection. It excludes unattended operation, human-contact tasks and payload claims beyond the recorded test envelope.
Configuration baseline
Repository commit, exact SO-101 revision, leader/follower role, servo model and ID map, printed-part material, gripper variant and every substitution.
Reject whenDo not begin if the physical arm cannot be reconciled with the BOM, geometry and motor map.
LeRobot release or commit, dependency lock, operating system, device ports, robot and teleoperator identifiers.
Reject whenReject an environment that cannot be rebuilt from a clean machine.
Physical guarding, clearances, cable route, power isolation, maximum commanded speed and a reachable stop control.
Reject whenNo powered motion without a clear exclusion zone and stop owner.
Staged verification
Check fastener retention, free joint travel, cable pinch points, printed-part damage and leader/follower mechanical correspondence.
Signed inspection record, annotated photographs and deviations list.
All joints move through the intended manual range without interference.
Cracks, binding, unidentified parts, loose output elements or damaged insulation.
Read each servo identity at zero command, verify leader/follower mapping and bind persistent LeRobot identifiers to the mechanism.
Motor ID table, port map, configuration export and software hash.
Every commanded joint maps to one expected physical joint and role.
Duplicate IDs, swapped joints, intermittent communication or an unresolved firmware mismatch.
Calibrate leader and follower separately using their persistent IDs; save ranges and calibration files, then reload them in a fresh session.
Calibration files, joint ranges, offsets, session log and recovery copy.
Reloaded calibration reproduces the expected neutral pose and joint directions.
Unexpected direction, unstable neutral, inconsistent end range or calibration tied to the wrong mechanism.
Enable low-speed, small-amplitude leader-to-follower motion one joint at a time, then a short multi-joint sequence with relative-target limiting enabled.
Command/state trace, video, current and temperature samples, stop-response record.
Motion remains inside the declared envelope without divergence, collision or cable strain.
Unexpected motion, growing error, overheating, cable tension, collision or failure to stop on command.
Run one low-energy pick/place or pose-transfer task for a predeclared number of trials, including controlled failed attempts.
Trial matrix, task definition, interventions, failures, calibration identity and raw data manifest.
The measured claim is repeatable only for this mechanism, configuration, object set and environment.
Do not conceal failed trials, alter success criteria after testing or generalise beyond the recorded distribution.
Acceptance claims
A second operator reconstructs the software, ports, servo map and calibration identity from the record.
Claim boundaryIdentity does not prove mechanical durability.
A fresh session reloads the saved files and reproduces joint direction and bounded neutral pose.
Claim boundaryRecalibrate after mechanical, motor, firmware or identifier change.
Every trial links to configuration, raw signals, outcome, intervention and failure notes.
Claim boundaryNo claim extends beyond the tested objects, speeds or workspace.
Evidence pack & boundaries
FUURAA analysis
The valuable product of this protocol is not a successful motion clip. It is a portable evidence package that allows another engineer to identify the build, reproduce its calibration, inspect failures and decide whether the next envelope may be opened.
Primary sources