RCI 029 · Actuation-system evidence audit

Six motor-cable units. Three axis MAEs. Zero released dataset files.

A new RAVEN II preprint reports a clever way to apply external forces while the surgical robot moves. The public package supports the architecture and three preliminary axis-wise error values—not an independently reproducible accuracy claim.

Published 2026-08-1013 min readResearch dataset v0.1.0
6author-reported motor-cable units
3reported axis MAEs
10official source-bundle files checked
0released dataset files found

The direct answer

What was built? The authors describe six motor-cable units arranged around a RAVEN II surgical robot. Each cable terminates at the end-effector; motor torque changes cable tension so the system can command an external force while the robot moves.

What performance number is public? The paper reports mean absolute errors of 0.80 N on X, 0.79 N on Y and 0.44 N on Z. These are three author-reported, axis-wise MAEs. They are not a public sample-level error distribution, a three-dimensional vector-error statistic or an RCI reproduction.

Can another lab reproduce the result from the arXiv package? Not yet. RCI found no raw dataset, controller or analysis code, CAD, BOM or schematic in the ten-file source bundle. Exact motors, load cells, cable mechanics, calibration, force reference, force range, synchronization and trial definition are also missing.

Strongest supported conclusion

The paper is a useful frontier architecture report: six parallel motor-cable units, two feedback levels, SLSQP tension allocation and three preliminary per-axis MAEs. The public record is not yet a build-ready hardware release or an independently recomputable force-accuracy result.

What the public paper discloses about the system

FieldPublished valueSource locationRCI boundary
motor cable unit count6 unitsAbstract; System; Figure 1 captionUnit count was read from the paper; RCI did not inspect or count the physical apparatus.
unit compositionDC motor; cable reel; load cell; fixing frameHardware subsectionThe text gives functional part classes, not exact purchasable models or a complete BOM.
force application pathSix controlled-tension cables connect to the RAVEN II end-effectorAbstract; SystemNo independently checked force envelope, collision envelope or cable-elasticity model is public.
software frameworkROS; described as compatible with the RAVEN II software controllerControl Software subsectionROS distribution, package versions, repository, commit and launch configuration are not disclosed.
tension optimizerSLSQPControl Software subsectionObjective, constraints, tolerances, initialization, solve rate and failure handling are not published.
feedback levels2 levelsControl Software subsectionThe paper describes force-command and cable-tension feedback but does not publish gains, loop rates or stability results.
minimum cable tensionMaintained continuouslyControl Software subsectionNo numeric minimum tension or selection rule is given.
load cell driver measurement rate320 HzAssociated Electronics subsection; Figure 3 captionThe paper does not establish whether 320 Hz is a per-channel output rate, bus transaction rate, timestamp cadence or closed-loop bandwidth.
current motor control pathRight arm of RAVEN II with modified control softwareAssociated Electronics subsectionIndependent motor drivers and microcontrollers are described as future work, not as the tested configuration.
motor unit localizationMicroScribe MX measurements mapped to the robot frame with the Kabsch algorithmLocalization of Motor Units subsectionPoint count, residuals, repeatability, coordinate uncertainty and calibration files are not published.
recorded dataset descriptionRobot trajectories with position and external force, intended for future neural-network trainingPreliminary ResultsThe dataset is described but no public download, manifest, schema, sample count or license is linked.

The control software is described as ROS-based and compatible with the RAVEN II controller. It calculates cable directions from the motor locations and current end-effector position, then uses SLSQP to allocate desired cable tensions while maintaining a minimum tension. The paper describes upper force-command feedback and lower cable-tension feedback.

Motor-unit positions are measured with a MicroScribe MX, then mapped from the measurement-device frame into the robot frame with the Kabsch algorithm. The named method is valuable, but the public package does not include the point set, fitted transform, residuals or coordinate-uncertainty record.

“Below 1 N” means three axis-wise MAEs here

AxisReported statisticValueEvidence stateWhat it does not establish
XMean absolute error0.80 Nauthor-reported resultAxis-wise MAE only; raw samples, force range, trial count and computation procedure are not public, and RCI did not recompute it.
YMean absolute error0.79 Nauthor-reported resultAxis-wise MAE only; it is not evidence that every Y-axis sample error was below 1 N.
ZMean absolute error0.44 Nauthor-reported resultAxis-wise MAE only; it is not a three-dimensional vector-error metric or uncertainty statement.

The public evidence supports that the authors reported three per-axis mean absolute errors below 1 N. It does not show that every sample error was below 1 N, does not publish a three-dimensional vector MAE or RMSE, and does not provide the data required for RCI to recompute any error metric. The three axis MAEs must not be combined as though they were one measured vector statistic.

Do not manufacture a fourth metric

Taking the square root of the sum of the three squared MAEs would not produce a measured vector MAE or RMSE. Axis-wise absolute errors lose the sample-level signs and joint distribution needed for that conversion.

Figure 5 normalizes position and force by their ranges for visualization. The normalized plot therefore cannot reveal the underlying commanded force range, time series or worst-case error. Those missing quantities matter because the same MAE can describe very different operating envelopes.

320 Hz is a measurement-rate statement—not a closed-loop bandwidth

The associated-electronics section says the load-cell drivers provide measurements at 320 Hz for cable-tension feedback. It does not specify whether this is per channel or aggregate, what clock timestamps the samples, how six channels are scheduled, what filtering occurs, or how often the lower and upper controllers execute.

RCI therefore records 320 Hz exactly as an author-reported load-cell-driver measurement rate. It must not be rewritten as 320 Hz force-control bandwidth, synchronized six-axis sampling, or end-to-end command latency.

Useful electronics labels, but not a BOM

The text does not list exact electronics, but Figure 3 visibly labels several items. RCI records those labels because they are useful search and follow-up leads while keeping their evidence status narrow.

Visible labelSourceWhat remains unknown
LTC4311Figure 3A visible label, not a complete schematic, ordering code or independently inspected assembly.
NAU7802 BreakoutFigure 3Board vendor, revision, configuration and load-cell interface calibration are not disclosed.
Arduino MegaFigure 3Exact board revision, firmware and timing implementation are not disclosed.
I2C MuxFigure 3Exact multiplexer model and bus topology are not specified in text.
RJ45 / SSTP cablingFigure 3The label does not define electrical pinout, cable length, shielding termination or data integrity.

A visible “Arduino Mega” or “NAU7802 Breakout” label is not enough to establish board vendor, revision, firmware, pinout, calibration or procurement identity. No new component-catalog record is created from these figure labels.

Eighteen disclosure checks: one complete, several partial, the reproduction core absent

Disclosure fieldStatusPublic evidenceReproduction impact
system architecturedisclosedSix motor-cable units, end-effector cable path and two feedback levels are described.Supports a functional block diagram, not a build-ready implementation.
exact dc motor model and specsnot-disclosedThe text says DC motor only.Torque, speed, inertia, electrical limits and procurement identity cannot be reproduced.
exact load cell model range accuracynot-disclosedLoad cells are described functionally; no exact model, range or accuracy is stated.Cable-tension resolution, saturation and uncertainty cannot be reconstructed.
cable reel mechanical definitionnot-disclosedNo cable material, diameter, length, stiffness, reel radius, bearing or frame drawing is published.Elasticity, torque-to-tension mapping and force envelope remain unknown.
electronics identitypartial-figure-labelsFigure 3 visibly labels LTC4311, NAU7802 Breakout, Arduino Mega, I2C Mux and RJ45/SSTP cabling.Useful lead-level identity, but not a schematic, complete BOM, ordering-code list or configuration record.
software and firmware releasenot-disclosedROS and modified RAVEN II software are named; no repository, commit, ROS distribution, package lock or firmware image is linked.The controller cannot be rebuilt or version-matched.
slsqp definition and ratepartialSLSQP is named, but objective, constraints, tolerances, initialization and solve rate are absent.Cable-tension allocation and real-time behavior cannot be reproduced.
measurement rate semanticspartialLoad-cell drivers are said to provide measurements at 320 Hz.Per-channel rate, bus rate, timestamp cadence and closed-loop bandwidth remain distinct unresolved quantities.
synchronization latency filteringnot-disclosedNo timestamp source, synchronization error, end-to-end latency, digital filtering or channel-skew result is given.The force-error time alignment cannot be independently checked.
load cell and force calibrationnot-disclosedNo calibration procedure, coefficients, reference instrument or residuals are published.Traceability and measurement uncertainty remain unknown.
actual applied force referencenot-disclosedThe paper reports desired versus actual applied force but does not fully define how actual Cartesian force is constructed or independently referenced.The ground-truth quantity behind the MAE cannot be independently reconstructed.
force range and trajectory protocolnot-disclosedThe figure is normalized; numeric force range, trajectory coordinates, speed, trial count and duration are not reported.Operating-envelope coverage and result comparability are unknown.
mae computation and aggregationpartialThree axis MAEs are printed, but sample selection, weighting, alignment, filtering and aggregation are not defined.The three numbers cannot be recomputed from public evidence.
uncertainty dispersion and worst casenot-disclosedNo standard deviation, confidence interval, percentile, maximum error or uncertainty budget is reported.Axis MAE does not establish worst-case or metrological accuracy.
raw dataset releasenot-releasedThe paper says a dataset was recorded; the arXiv page and ten-file source bundle contain no dataset file or dataset link.The trajectory, force range, timestamps and errors cannot be independently inspected.
analysis and controller code releasenot-releasedThe arXiv record and source bundle contain no analysis or controller code link or file.Neither the controller nor the MAE analysis can be rerun.
cad bom and schematic releasenot-releasedNo CAD, BOM or schematic file is present in the ten-file source bundle.The apparatus is not build-ready from the public package.
independent replicationnot-foundNo independent reproduction is linked from the arXiv record or paper.The reported system behavior remains author evidence only.

17 of the 18 checks are partial, absent or not released. The main bottleneck is not the three-page length by itself; it is that the paper says a dataset was recorded while publishing no data link, schema, manifest, sample count, timestamps or reuse license.

A minimum follow-up package would publish exact motor and load-cell identities, the mechanical drawings and cable parameters, a schematic and BOM, controller/software versions, calibration and synchronization records, the full trial protocol, raw desired/actual force data, and the code that calculates the axis errors.

Audit the facts without mirroring the paper

The official arXiv record labels v1 with the arXiv non-exclusive distribution license, not a Creative Commons reuse license. That license lets arXiv distribute the work; RCI does not treat it as a blanket right to mirror the PDF, TeX source bundle or figures.

RCI instead links the official PDF and source route, records cryptographic hashes and file counts, and publishes independently written factual observations and disclosure analysis. Figure labels are transcribed only as minimal facts; no figure is reproduced.

What RCI independently checked

  1. Resolved arXiv:2608.06488 v1, its submission date, authors, affiliations and license route.
  2. Downloaded the 3-page PDF, recorded 2,151,930 bytes and SHA-256 f6e546bcea4b5f8676ce158432064c051f167ccc9ce081764e98b852cf14775f, then rendered and visually checked pages 1, 2, 3.
  3. Downloaded the official source bundle, recorded 2,300,418 bytes and SHA-256 a4b4ebfa51a46afc5d8fcc7bf70fb7e4c4a6d1acc1e9a04238c0278f2bddcb6a, then enumerated all 10 files.
  4. Separated fifteen paper observations from eighteen RCI disclosure checks and treated every performance number as author-reported.
  5. Confirmed that the source bundle contains 5 figures but zero dataset, code, CAD and BOM files.
  6. Created a fact-check packet for exact hardware, rate semantics, calibration, force reference, trial design, metric computation and artifact release; no outreach is sent until the editorial-identity gate is complete.

Limits that stay attached

  • RCI audited public files and disclosures only; it did not inspect the apparatus, run the controller or perform a force test.
  • The three axis MAEs are author-reported preliminary results and were not independently recomputed.
  • The absence of a public dataset, code or build package is bounded to arXiv v1, its ten-file source bundle and linked surfaces checked on 2026-08-10; private or later artifacts may exist.
  • Figure-visible electronics labels do not establish exact ordering codes, revisions, firmware, calibration or a complete BOM.
  • The 320 Hz statement is a reported load-cell-driver measurement rate, not a demonstrated six-channel closed-loop bandwidth or end-to-end latency.
  • No claim is made about clinical safety, surgical effectiveness, sterilization, regulatory status or suitability for use on patients.

Download the evidence audit

Download the immutable JSON release and record-level CSV. Stable aliases are current JSON and current CSV.

Suggested citation: Robot Component Index. “RAVEN II Force Actuation Evidence: 6 Motor-Cable Units, 3 Axis MAEs, 0 Released Dataset Files.” RCI 029, research dataset version 0.1.0, 2026-08-10. https://robotcomponentindex.com/research/raven-ii-force-actuation-evidence-audit/