RCI 030 · Soft-actuator evidence audit

20 µm films. 1.2 mL selected fill. Zero MRI tests.

A new preprint discloses a useful S-HASEL expansion actuator and unusually specific self-sensing hardware. Its strongest evidence is still a single-actuator bench study—not a tested biopsy robot, MRI-compatible system or clinical device.

Published 2026-08-1015 min readResearch dataset v0.1.0
20 µmreported film thickness
1.2 mLwithin-study selected fill
4author-reported NRMSE values
0MRI-environment tests

The direct answer

What was actually tested? Individual S-HASEL expansion actuators made from 20 µm BOPP or Mylar 850 film, filled with silicone oil, driven up to 7 kV, and measured on force–stroke and laser-reference test benches. The authors also tested graphite and sputtered-gold electrodes for capacitive displacement self-sensing.

What was selected? BOPP with a 1.2 mL fill was the authors’ preferred configuration within a matrix of two films and three fill volumes. Gold electrodes reduced the reported Figure 7 NRMSE from 0.1639 to 0.0429; a later three-cycle validation reported 0.0126 under DC and 0.0078 under PWM.

What was not tested? Closed-loop position control, an integrated three-stack/two-DOF robot, needle insertion, tissue puncture, MRI operation, image artifacts, durability, efficiency and clinical performance. The article’s biopsy and MRI framing is an application target, not a completed application validation.

Strongest supported conclusion

The preprint is sufficiently detailed to establish a credible single-actuator prototype and bounded self-sensing result. It is not evidence that an MRI-compatible biopsy robot has been built or that the reported NRMSE is closed-loop targeting accuracy.

Where the evidence stops

StageStatusPublic evidence
Single S-HASEL actuator fabricatedauthor-testedBench photographs, fabrication description and material comparisons are present.
Single-actuator displacement self-sensingauthor-testedLaser-referenced DC and PWM validation cycles and NRMSE values are reported.
Closed-loop position controlnot-testedExplicitly described as future work.
Integrated three-stack, two-DOF robotconcept-onlyOnly a proposed architecture and concept rendering are published.
Needle insertion or tissue puncturenot-testedNo phantom, ex vivo, in vivo or human puncture test is reported.
MRI environment and image-artifact validationnot-testedThe authors state this will be tested and adapted subsequently.
Clinical usenot-supportedThe preprint cover warns against using preliminary reports to inform clinical practice.

This distinction matters because the paper includes a patient-mounted robot rendering and an MRI-guided workflow. Those visuals explain the intended architecture; they do not convert the actuator bench into a robot, puncture or MRI experiment.

What the paper discloses about geometry, materials and the bench

FieldPublished valueSource locationRCI boundary
proposed robot actuator stack count3 stacksSection 4.1; Figure 3The paper presents the three-stack arrangement as a robot concept; it does not report an assembled or tested three-stack robot.
proposed stack angular spacing120 degreesSection 4.1Angular spacing is a design statement, not a measured alignment or kinematic calibration result.
proposed needle positioning degrees of freedom2 DOFSection 4.1No integrated two-DOF trajectory, accuracy, repeatability or workspace result is reported.
actuator drawing outer height53 mmFigure 3 rightA two-dimensional figure dimension; no tolerance, thickness stack-up or released CAD is provided.
actuator drawing outer width36 mmFigure 3 rightA two-dimensional figure dimension; it does not establish the complete manufactured geometry.
central cutout shift4 mmSection 4.1; Figure 3 rightThe shift is reported as an air-breakdown mitigation choice; dielectric-breakdown testing for the geometry is not published.
high voltage tab offset50 degreesSection 4.1No tab tolerance, conductor width, creepage calculation or released drawing accompanies the angle.
tested film materialsBOPP; BoPET heat-seal film sold as Mylar 850 (DuPont)Section 4.3Mylar 850 is named; the BOPP grade, supplier, batch and surface treatment are not disclosed.
film thickness20 µmSection 4.3Nominal film thickness is reported without tolerance or independently measured thickness distribution.
dielectric fluidLow-viscosity silicone oil; 5 cSt; Carl RothSection 4.3The exact catalog number, lot, temperature and measured fluid properties are not provided.
tested fill volumes1.0; 1.2; 1.4 mLSection 5; Figure 8Only these three fill levels within the reported geometry were compared; this is not a global optimization over HASEL designs.
selected configurationBOPP with 1.2 mL fillAbstract; Section 5; Conclusion“Selected” and “optimal” apply to the two films, three fill levels and reported bench protocol, not to every material, volume or application.
material test replication statementThree S-HASEL actuators per material; three independent trials per fill volumeSection 5The text does not fully resolve actuator reuse across fill conditions, trial independence, batch allocation or error-bar definition.
figure 8 constant force levels0.2; 0.4; 0.6; 0.8 NFigure 8 legendsThese are plotted constant-load conditions, not maximum force, blocked force or an independently calibrated force envelope.
force test load cellKM40e 100N/010; ME-MeßsystemeSection 4.2No calibration certificate, calibration date, uncertainty or complete signal chain is published.

The geometry is more than a generic HASEL reference: Figure 3 shows a 53 mm by 36 mm outline, a 4 mm cutout shift and electrode/contact regions, while the text adds a 50° high-voltage-tab offset. That is useful design evidence, but no full drawing, heat-seal recipe, tolerance stack or CAD is released.

The material result is also narrower than “BOPP is best.” The study compared BOPP with one BoPET heat-seal product, Mylar 850, at 1.0, 1.2 and 1.4 mL. The authors report significant Mylar batch variability—including batches with no actuation—and select 1.2 mL BOPP for the later sensing work.

The sensing and high-voltage chain is specific—but incomplete

FieldPublished valueSource locationRCI boundary
force test linear stageXR50P; ThorlabsSection 4.2The stepper motor, controller, stage settings, alignment and fixture compliance are not completely specified.
self sensing ground truth sensorHG-C1100 micro-laser distance sensor; PanasonicSection 4.2; Figure 5No calibration, target reflectivity error, alignment error, sampling synchronization or uncertainty budget is reported.
self sensing excitationDG 2021 A function generator; 20 V sinusoidal excitationSection 4.2The excitation frequency and complete component values for the sensing circuit are not stated in this paper.
self sensing data acquisitionPXI-6255; 200 kHz samplingSection 4.2The sample rate is not a demonstrated actuator bandwidth, control-loop rate or independent accuracy result.
dc training sweep0–7 kV in 200 V stepsSection 4.2Dwell time, ramp timing, cycle history, temperature and raw samples are not published.
self sensing modelSecond-degree polynomial (Poly 2) fit from RMS voltage to displacementSection 4.2Coefficients, training data, model files, preprocessing and code are not publicly released.
pwm logic componentsLTC6992-1 timer; SN74LVC2G86 XOR gate; optocoupler half-bridgeSection 4.2; Figure 6The optocoupler identities, passive values, schematic, PCB, firmware and safety design are not released.
pwm test condition150 Hz PWM; 7 kV fixed high voltage; 0–100% duty in 0.25% stepsSection 4.2This drive condition is not evidence of 150 Hz mechanical bandwidth, three-stack control or MRI-safe operation.
high voltage generator20HVA24-P2 UltraVolt; controllable by 10 V analog inputSection 4.2The paper does not publish output-current measurements, energy per cycle, efficiency, fault response or MRI compatibility for the supply.
graphite and first gold nrmse0.1639 graphite; 0.0429 gold NRMSESection 4.3; Figure 7The normalization definition, raw traces, model coefficients and repeated-sample statistics are not released; RCI did not recompute either value.
gold sputtering condition250 nm Au; 0.04 mbar argon; 25 mA; SCD 004 Balzers UnionSection 4.3Deposition time, thickness-uniformity measurement, adhesion test, mask geometry and batch yield are not reported.
final outer coatingThin dip-coated silicone layer; Ecoflex 0030; Smooth-OnSection 4.3Thickness, cure schedule, coverage, dielectric test and effect on mechanics are not quantified.
final self sensing validation protocolThree cycles; DC ramp 0–6 kV; PWM duty 0–100%Section 5; Figure 9The three-cycle test is short-duration author evidence; raw time series and actuator-to-actuator replication are not public.
final dc and pwm nrmse0.0126 DC; 0.0078 PWM NRMSEAbstract; Section 5; Figure 9NRMSE is not absolute positioning accuracy, closed-loop error, MRI performance or clinical targeting accuracy; RCI did not recompute it.
dc pwm displacement and relaxation comparisonDC displacement 0.18 mm larger; residual 0.043 mm at 0 kV versus 0.1 mm at 0% PWM; 9 s initial PWM plateauSection 5; Figure 9These values describe the reported validation sequence and must not be generalized to all HASEL designs, loads or controllers.

The paper names the Panasonic HG-C1100 laser, National Instruments PXI-6255 DAQ, RIGOL DG 2021 A generator, UltraVolt 20HVA24-P2 high-voltage source, Analog Devices LTC6992-1 timer and Texas Instruments SN74LVC2G86 XOR gate. It also states 200 kHz acquisition, 150 Hz PWM and a 0–7 kV training sweep.

Those identities make follow-up easier, but they do not form a reproducible electrical package. The optocouplers, passive values, sensing excitation frequency, complete schematic, RMS window, code, filter timing, PCB and high-voltage safety implementation are not public.

Two rates that must stay separate

The 200 kHz number is the DAQ sampling rate. The 150 Hz number is the PWM carrier. Neither is a measured mechanical bandwidth, closed-loop update rate or end-to-end control latency.

Four NRMSE values do not establish robot accuracy

ComparisonReported NRMSEWhat it supportsWhat it does not support
Figure 7 graphite electrodes0.1639Graphite-sensor validation in the reported setupAbsolute accuracy in millimetres, cross-actuator generalization, closed-loop accuracy, MRI performance or biopsy targeting accuracy. RCI could not recompute the metrics without raw traces, normalization definition, coefficients and code.
Figure 7 sputtered gold electrodes0.0429Lower reported normalized error after electrode change
Figure 9 final DC validation0.0126Three reported actuator displacement cycles
Figure 9 final PWM validation0.0078Three reported PWM displacement cycles

The public evidence supports that the authors fabricated and bench-tested individual S-HASEL actuators, selected BOPP with 1.2 mL fill within a bounded material matrix, and reported improved displacement-estimation NRMSE with gold electrodes. It does not establish a force rating, mechanical bandwidth, lifetime, closed-loop accuracy, integrated robot performance, needle-puncture capability, MRI compatibility or clinical usefulness.

The final comparison also reports 0.18 mm more displacement under DC, residual displacement of 0.043 mm at 0 kV versus 0.1 mm at 0% PWM, and a 9 s initial PWM plateau. These are useful warnings about charge/drive behavior; they are not a full dynamic model or lifetime characterization.

Twenty-five disclosure checks: enough for an audit, not enough for a rebuild

Disclosure fieldStatusPublic evidenceReproduction impact
item identity version and licensedisclosedThe official Research Square record and covered PDF identify DOI v1, posted date, authors and CC BY 4.0.The exact source version and item-level reuse terms can be cited and revisited.
peer review statusdisclosed-preprintResearch Square states that the work is a preprint and has not been peer reviewed by a journal.Every result remains preliminary author evidence.
actuator geometrypartialFigure 3 gives a 53 mm by 36 mm outline plus selected offsets and electrode regions.No complete dimensioned drawing, tolerance stack, seal geometry or CAD is released.
materials and fluidspartialFilm families, nominal thickness, Mylar 850 identity, silicone-oil viscosity/vendor and Ecoflex 0030 are stated.The BOPP grade, catalog numbers, lots and measured material properties remain unknown.
heat seal fabrication processnot-disclosedThe actuator is described as heat sealed and prior work is cited, but this paper gives no complete sealing recipe.Temperature, pressure, dwell, tooling, cut sequence and yield cannot be reproduced from v1.
electrode fabricationpartialGraphite stencil coating and gold sputtering conditions are described; the gold layer is reported as 250 nm.Masks, deposition time, resistance maps, adhesion, alignment tolerance and yield are absent.
material test protocolpartialTwo films, three fill volumes, four plotted loads and voltage sweeps are visible or stated.Dwell, ramp rate, order, preconditioning, ambient state and rejection rules are not fully defined.
sample allocation and statisticspartialThe paper says three actuators per material and three independent trials per fill volume, with mean values and error bars.Reuse across fill levels, batch assignment, independence and the error-bar statistic are not fully specified.
instrument calibration and uncertaintynot-disclosedExact load-cell and laser models are named, but no calibration record or uncertainty budget is provided.Force, stroke and NRMSE traceability cannot be independently quantified.
material test raw datanot-releasedFigure 8 shows means and error bars; the official item exposes one PDF and no data file.Plot values, variability, exclusions and alternative analyses cannot be independently checked.
self sensing circuitpartialThe Voltage Method, omitted 1:2 stage, instruments and a high-level block diagram are described.No complete schematic, passive values, excitation frequency, PCB or noise characterization is released.
model training artifactsnot-releasedA local second-degree polynomial training workflow is described.Training traces, coefficients, model files and MATLAB code cannot be rerun or version matched.
nrmse definition and recomputationpartialFour NRMSE values are printed, but the normalization denominator and calculation code are not supplied.The metrics cannot be independently recomputed or safely compared with differently normalized studies.
timing filtering and latencypartialDAQ sampling is 200 kHz and PWM is 150 Hz; a 9 s PWM plateau is reported.Clocking, filtering, RMS window, latency and mechanical bandwidth remain unresolved.
maximum force stress and work densitynot-reportedFigure 8 uses constant loads up to 0.8 N, but no blocked-force, maximum-force, stress or work-density result is reported for S-HASEL.The actuator cannot be compared as a force-rated component from this paper alone.
dynamic response and bandwidthnot-reportedThe paper plots slow validation cycles but gives no step response, frequency response or bandwidth.The 150 Hz electrical PWM rate must not be treated as actuator bandwidth.
power energy and efficiencynot-reportedThe high-voltage source is identified, but electrical current, energy per stroke and efficiency are absent.Power-system sizing and energy comparison remain unsupported.
durability breakdown and cycle lifenot-reportedThree final validation cycles are reported; no endurance, failure distribution, leakage-current or breakdown-voltage study is published.Reliability, maintenance interval and safe operating life cannot be inferred.
closed loop position controlfuture-workThe conclusion says the self-sensing approach will serve as the basis for a position-control system.The paper validates estimation only; it does not demonstrate closed-loop positioning.
integrated three stack robotconcept-onlyA patient-mounted three-stack robot is shown as a concept and described as the objective of future development.No assembled robot, workspace, kinematics, trajectory or two-DOF accuracy result is published.
needle puncture and tissue testnot-testedThe paper motivates needle biopsy but reports actuator bench tests only.No needle force, targeting, phantom, tissue, animal or human result supports the application title.
mri compatibility testfuture-workThe conclusion states that the complete system will subsequently be tested and adapted to MRI conditions.Zero MRI-environment tests, image-artifact measurements or MR-safety results are reported.
code cad bom and schematic releasenot-releasedThe official item exposes one manuscript PDF and no supplementary data, code, CAD, BOM or schematic file.The actuator, sensing electronics, PWM driver and test workflow are not build-ready from the public package.
independent replicationnot-foundNo independent reproduction of this S-HASEL geometry and result set is linked by the v1 record or manuscript.The reported behavior remains author evidence from one research group.
clinical safety sterility and regulatory evidencenot-applicable-at-this-stageThe item is a preclinical actuator-development preprint with no human or vertebrate-subject study declared.No clinical effectiveness, patient safety, sterilization, biocompatibility or regulatory claim can be made.

23 of the 25 checks remain partial, absent, untested, future work or not released. The public item exposes one manuscript PDF and no supplementary data, code, CAD, BOM or schematic file.

A minimum reproduction package would add the complete actuator drawing and seal recipe, exact BOPP and fluid ordering identity, masks and electrode files, full sensing/PWM schematics, calibration records, raw force–stroke and laser traces, NRMSE formula and scripts, model coefficients, test sequence and environmental conditions. Application claims require separate integrated-robot, puncture and MRI protocols.

Item-level CC BY 4.0 is confirmed

The official Research Square v1 record and the first page of the covered PDF both identify a CC BY 4.0 license. This is item-level evidence, not merely an aggregator’s generic “open” label.

RCI could reuse or adapt licensed material with attribution. This release deliberately publishes original audit text and structured facts without reproducing a source figure, so verification stays attached to the official record and no separate assumption is made about embedded third-party visual elements.

What RCI independently checked

  1. Resolved DOI 10.21203/rs.3.rs-10623292/v1 to the official Research Square v1 item and verified authors, posted date, preprint status and item-level license.
  2. Downloaded the only public manuscript file, recorded 2,514,822 bytes and SHA-256 cc9da81a1e63410140d4e0451b4a5c6a089cb91f6473d5cdb47e6791e4a29c94.
  3. Extracted the complete text, rendered all 16 covered-PDF pages and visually checked pages 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16.
  4. Separated 30 author-source observations from 25 RCI disclosure checks and seven application-stage boundaries.
  5. Checked the official item’s file list: one manuscript PDF and zero supplementary, raw-data, code, CAD, BOM or schematic files.
  6. Treated all NRMSE, displacement and material-selection results as author-reported; RCI did not digitize plots or perform a physical/software reproduction.

Limits that stay attached

  • RCI audited the public Research Square v1 item and its only public PDF file; it did not inspect laboratory notebooks, private data, hardware or later versions.
  • All performance values are author-reported. RCI did not digitize the plots or recompute the four NRMSE values.
  • The absence counts for data, code, CAD, BOM, schematic and supplementary files are bounded to the official public item surfaces checked on 2026-08-10; private or later artifacts may exist.
  • The selected 1.2 mL BOPP configuration is best only within the stated study matrix and protocol, not a universal HASEL optimum.
  • The 200 kHz DAQ rate and 150 Hz PWM carrier are electrical/acquisition conditions, not measured mechanical or closed-loop bandwidths.
  • No claim is made about needle-biopsy effectiveness, MRI safety, image artifacts, sterilization, biocompatibility, clinical safety or regulatory status.

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. “S-HASEL Actuator Evidence: 20 µm Films, 1.2 mL Fill, Zero MRI Tests.” RCI 030, research dataset version 0.1.0, 2026-08-10. https://robotcomponentindex.com/research/s-hasel-actuator-evidence-audit/