The direct answer
What does the selected public evidence actually contain? All 18 records use some motion-quality language. Fifteen publish an explicit backlash statement: 13 give a numeric value and two HEBI H-Series rows say “None — harmonic-type reducer.” Three CubeMars records use qualitative positioning, low-backlash or high-precision language without a numeric backlash value. Four records publish a separate angular- or control-accuracy value.
What is missing? Across the same 18 records, the selected sources publish zero explicit lost-motion values, zero transmission-error values and zero repeatability values. None of the 15 explicit backlash statements supplies all seven fields needed to reproduce the measurement. That is a public-documentation result, not proof that a manufacturer lacks internal test data.
The values include ±0.25°, unsigned 0.25°, <0.1°, ≤10 arcmin and a categorical “None.” RCI converts arcminutes only for machine-readable unit arithmetic. It does not erase inequalities, assign zero to “None,” or rank records without a shared procedure.
Five familiar terms answer different questions
| Term | Question it answers | What it does not establish |
|---|---|---|
| Backlash | How much direction-dependent dead band is observed under a stated reversal and load procedure? | Full angle mapping, compliance, system accuracy or repeatability. |
| Lost motion | How much output motion is lost through reversal, hysteresis, friction and elastic effects under a defined method? | A universally interchangeable synonym for every vendor backlash row. |
| Transmission error | How far actual output angle departs from the ideal ratio across position, possibly without reversal? | Repeated positioning performance or end-effector accuracy. |
| Accuracy | How close a measured angle is to a defined physical reference under a stated condition? | Repeatability, resolution or low backlash by itself. |
| Repeatability | How closely a system returns to a target across a specified sequence? | Absolute accuracy or a complete robot-level pose budget. |
This separation is why the current component catalog adds a dedicated backlash field instead of reusing encoder resolution or accuracy. A joint may have low backlash and still have meaningful transmission error, torsional compliance or calibration error.
The 18-record evidence matrix
| Exact record | Statement boundary | Backlash as published | Separate accuracy value |
|---|---|---|---|
| HEBI Robotics R8-9+ R8-9+ column in the current R-Series+ family table | Numeric backlash statement | ±0.25° Numeric manufacturer backlash and angular-accuracy rows. | ±0.2° Manufacturer angular-accuracy row |
| ROBOTIS XH540-W270-R XH540-W270-T/R family row; R suffix uses RS-485 | Numeric backlash statement | 15 arcmin / 0.25° Numeric manufacturer backlash row. | Not published in selected source Kept separate from backlash |
| CubeMars AK80-9 V3.0 KV100 AK80-9 V3.0 KV100 current product-specific table | Numeric backlash statement | 15 arcmin Numeric manufacturer backlash row. | Not published in selected source Kept separate from backlash |
| MYACTUATOR RMD-X8-P9-32-R Current X8-32 V4 product page and its embedded Technical Parameters image | Numeric backlash statement | ≤10 arcmin Numeric upper-bound backlash and control-accuracy rows in an image-based table. | <0.01° Manufacturer control-accuracy row |
| HEBI Robotics T8-16+ T8-16+ column in the T-Series+ family table | Numeric backlash statement | ±0.25° Numeric family-wide backlash row. | Not published in selected source Kept separate from backlash |
| HEBI Robotics H25-90 H25-90 column in the H-Series family table | Categorical “none” statement | None — harmonic-type reducer Categorical no-backlash wording plus a numeric angular-accuracy row. | 0.02° Manufacturer angular-accuracy row |
| ROBOTIS PH54-200-S500-R PH54-200-S500-R specification row | Numeric backlash statement | <6 arcmin / 0.1° Numeric less-than backlash row. | Not published in selected source Kept separate from backlash |
| ROBOTIS XM540-W270-R XM540-W270-T/R family row; R suffix uses RS-485 | Numeric backlash statement | 15 arcmin / 0.25° Numeric manufacturer backlash row. | Not published in selected source Kept separate from backlash |
| CubeMars AK70-10 KV100 AK70-10 KV100 current product-specific table | Numeric backlash statement | 0.2° Numeric manufacturer backlash row. | Not published in selected source Kept separate from backlash |
| CubeMars AK10-9 V3.0 KV60 AK10-9 V3.0 KV60 row in the current AK-series comparison table | Qualitative language; numeric value unknown | Numeric value not published Dual encoders are presented in positioning-accuracy language, but no numeric backlash value is published in the selected boundary. | Not published in selected source Kept separate from backlash |
| HEBI Robotics T8-3+ T8-3+ column in the T-Series+ family table | Numeric backlash statement | ±0.25° Numeric family-wide backlash row. | Not published in selected source Kept separate from backlash |
| HEBI Robotics T25-20 T25-20 column in the T-Series+ family table | Numeric backlash statement | ±0.25° Numeric family-wide backlash row. | Not published in selected source Kept separate from backlash |
| HEBI Robotics H25-45 H25-45 column in the H-Series family table | Categorical “none” statement | None — harmonic-type reducer Categorical no-backlash wording plus a numeric angular-accuracy row. | 0.02° Manufacturer angular-accuracy row |
| ROBOTIS XH540-W150-R XH540-W150-T/R family row; R suffix uses RS-485 | Numeric backlash statement | 15 arcmin / 0.25° Numeric manufacturer backlash row. | Not published in selected source Kept separate from backlash |
| ROBOTIS XW540-T260-R XW540-T260-R specification row | Numeric backlash statement | 15 arcmin / 0.25° Numeric manufacturer backlash row. | Not published in selected source Kept separate from backlash |
| CubeMars AK60-6 V3.0 KV80 AK60-6 V3.0 KV80 current product page | Qualitative language; numeric value unknown | Numeric value not published The current page describes the actuator as low-backlash but provides no numeric value in the selected specification table. | Not published in selected source Kept separate from backlash |
| CubeMars AK80-64 KV80 AK80-64 KV80 current product page | Numeric backlash statement | 0.18° Numeric manufacturer backlash row. | Not published in selected source Kept separate from backlash |
| CubeMars AKA60-6 KV80 AKA60-6 KV80 current hollow-shaft product page | Qualitative language; numeric value unknown | Numeric value not published The current page uses high-precision control language but provides no numeric backlash value in the selected specification table. | Not published in selected source Kept separate from backlash |
Each row is tied to an exact catalog identity and a selected source location. The table does not silently borrow a value from a neighboring ratio, prior revision or family member. Qualitative-only rows become explicit U evidence in catalog v0.4.4 rather than blank cells.
Fifteen statements, seven reproducibility fields, all zero
A portable backlash claim needs a measurement protocol. RCI checked seven fields across the 15 sources with explicit statements:
| Method field | Explicitly disclosed | Why it matters |
|---|---|---|
| Applied load or torque | 0 / 15 | A reversal gap can change with preload, friction and elastic deformation. |
| Reversal or approach procedure | 0 / 15 | Backlash is direction-dependent; the motion sequence is part of the metric. |
| Output measurement point or reference | 0 / 15 | Motor encoder, reducer output and external flange are different boundaries. |
| Measurement instrument | 0 / 15 | Instrument resolution and uncertainty constrain the result. |
| Test temperature | 0 / 15 | Lubrication, preload and structure change with temperature. |
| Conditioning or cycle count | 0 / 15 | Run-in, wear state and sampling plan affect repeatability. |
| Metric definition or calculation | 0 / 15 | ±, unsigned, limit and dead-band conventions cannot be assumed equivalent. |
A value can still be useful for supplier screening while remaining an M-state manufacturer claim. The missing method prevents a fair cross-brand leaderboard and tells a buyer exactly what to request before relying on the number.
Why ±0.25°, 0.25°, <0.1° and “None” cannot share a ranking
- Sign convention: HEBI publishes ±0.25° while several other sources publish an unsigned magnitude. RCI does not assume whether one means total band, one-sided error or a different convention.
- Inequality boundary: <6 arcmin and ≤10 arcmin are limits, not point measurements. Sorting them beside an unsigned point value implies precision the source does not provide.
- Categorical wording: “None — harmonic-type reducer” is retained as text. Assigning numeric zero would manufacture a measurement and erase compliance and hysteresis questions.
- Metric scope: MYACTUATOR’s control accuracy and HEBI’s angular accuracy stay in an accuracy field. They are not substituted for backlash, lost motion or repeatability.
The machine-readable file includes a normalized magnitude only where unit conversion is defensible. The original value, unit, qualifier and editorial boundary remain beside it.
Three source-boundary lessons
HEBI’s family PDFs require layout-aware reading. The accuracy and backlash cells span model columns. RCI reviewed the rendered page rather than treating a PDF text extraction order as proof of model assignment.
MYACTUATOR’s current parameters are image-based. The embedded X8-32 table publishes rated torque, speed, mass, back-drive torque, backlash and control accuracy. Catalog v0.4.4 records those values with the image location and a missing-revision limitation. The earlier RCI015 torque dataset remains immutable because it intentionally captured a selected text boundary at release time.
One current CubeMars page conflicts with itself. AK70-10 shows 621 g in its product-specific specification table and 521 g in the comparison table. The current catalog marks the mass claim C with both locations. The 0.2° backlash row is recorded separately; one clean-looking field must not hide a conflict in another.
What to request before accepting a backlash claim
- Exact hardware: reducer type and ratio, motor, encoder location, brake, preload, lubrication, firmware and product revision.
- Metric definition: backlash, lost motion, hysteresis or torsional dead band; one-sided, total band, peak or statistical limit.
- Load sequence: applied torque, radial and axial load, reversal amplitude, approach direction, speed, dwell and settling rule.
- Measurement chain: output reference, external encoder or interferometer, resolution, calibration and uncertainty.
- Environment and state: temperature, mounting stiffness, warm-up, lubrication state, run-in cycles, wear and compensation settings.
- Production evidence: sample size, acceptance limit, distribution and whether the supplied unit or only a development sample was measured.
Method and limitations
RCI audited all 18 records in its integrated-joint actuator comparison group against the selected current manufacturer pages or PDFs on 2026-08-10. It preserved model columns, signs, inequalities, original units, categorical wording and source locations; converted arcminutes to degrees only as explicit unit arithmetic; and recorded seven method-disclosure flags.
This is a public-evidence audit, not a physical metrology campaign or market census. It does not establish that a value is wrong, that a qualitative product has more backlash, or that “None” means physical zero. It does not infer end-effector performance from a joint field. RCI-BLM-1 now publishes the common fixture disclosure, torque sequence, temperature, instrument, raw-data and uncertainty requirements; it still requires physical and inter-laboratory validation before RCI publishes cross-model measurements.
Download the record-level data
Download the immutable research dataset as JSON or CSV. Stable current aliases are /data/research/actuator-motion-error-claims.json and /data/research/actuator-motion-error-claims.csv. Current component evidence is in catalog v0.4.4 JSON and its claim CSV.
Suggested citation: Robot Component Index, “Robot Actuator Backlash Claims: 18 Records, 15 Statements, 0 Complete Test Methods,” RCI 016, research dataset v0.1.0 and component catalog v0.4.4, 2026-08-10.