Bio-Inspired Underwater Whisker Sensors

• Bio-inspired underwater whisker sensors are tactile and hydrodynamic systems that mimic harbor seal whiskers to detect subtle wake flows and upstream object features.
• They employ innovative designs like undulating, tapered, and compliant structures to suppress self-noise and amplify flow-induced vibrations for clearer signal detection.
• Advanced transduction methods using strain gauges, Hall sensors, and FBG arrays enable precise measurement of bending moments and effective wake signal decoding.

Searching arXiv for recent and foundational papers on bio-inspired underwater whisker sensors and related wake sensing. Bio-inspired underwater whisker sensors are hydrodynamic and tactile sensing systems that emulate selected features of mammalian vibrissae—most notably the harbor seal’s wake-sensing whiskers and, in related work, sea lion and terrestrial whisker mechanics—to convert flow disturbances or contact into measurable mechanical signals. In current research, the term encompasses several non-equivalent architectures: undulating elliptic whiskers that suppress self-induced vortex-induced vibration (VIV) while locking to external wakes, tapered compliant structures that encode flow into distributed vibration modes, follicle-like compliant bases instrumented with silicon gauges, mechano-magnetic probes with sealed electronics, and optical Fiber Bragg Grating (FBG) arrays for underwater contact tracking (Beem et al., 2015, Jin et al., 7 Oct 2025, Hang et al., 27 Nov 2025, Wang et al., 2023, Li et al., 2024).

1. Biological model and sensing objectives

The canonical biological reference is the harbor seal. Reported capability includes tracking vortex wakes of animals and objects that passed by up to about $30$ s earlier, with wake velocities as low as approximately $1$ mm/s, while swimming forward and rejecting self-generated flow noise. The same literature also attributes to seals some ability to infer characteristics of the upstream object, including aspects of its size and shape, from the wake signature. Complementary experiments with a 3D-printed sea lion head showed that whiskers encountering a von Kármán street exhibit a jerky stick–slip response, and that the Strouhal frequency of the upstream wake is more clearly decoded in the time derivative of the whisker response than in displacement alone (Beem et al., 2015, Muthuramalingam et al., 2018).

These biological examples motivate several distinct sensing objectives. One is passive wake detection: identifying the presence of an upstream body by frequency lock-in and amplitude amplification. A second is inverse inference: estimating a characteristic upstream length $D$ from the measured frequency $f$ using the Strouhal relation

$$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$

A third is local source seeking, in which the sensor or a whisker-equipped vehicle estimates the local direction of information propagation from spectral phase and steers against it toward the source of a periodic disturbance (Beem et al., 2015, Colvert et al., 2018).

The operational envelope of the field is broader than wake following alone. Recent devices have been aimed at distinguishing attached flow, Kármán streets, asymmetric shedding, and broadband turbulence; detecting canonical dipole disturbances in darkness and turbidity; estimating ambient flow speed on mobile platforms; and tracking underwater contact location along a whisker without relying on precise robot proprioception (Jin et al., 7 Oct 2025, Hang et al., 27 Nov 2025, Wang et al., 2023, Li et al., 2024).

2. Structural archetypes and design taxonomy

A useful unifying description decomposes a whisker sensor into five functional components: the whisker element (WE), compliant element (CE), sensing element (SE), support structure (SS), and data acquisition module (DAQ). This taxonomy is deliberately modality-agnostic and makes it possible to compare sensors that transduce deflection through pressure, magnetic flux, image features, strain, or optical wavelength shift while still referencing the same underlying mechanical quantity, namely bending at the whisker base (Routray et al., 19 Sep 2025).

The most direct morphological abstraction of harbor seal vibrissae uses a $2{:}1$ elliptic cross-section with sinusoidal spanwise undulations in both major and minor axes, with the upstream and downstream undulations out of phase. In the artificial wake sensor reported in 2015, this geometry was fabricated at $20\times$ scale with diameter $d_w = 1.06$ cm and span $L/d_w = 30$, and mounted as a cantilever on a compliant flexure allowing in-line and crossflow vibration. A later amplification-oriented design retained the wavy whisker but introduced a spiral-perforated rectangular base, with the whisker mounted at the center and the sensing point on the opposite surface; by tuning thickness, turns, growth rate, and slot width, the sensitive band was shifted to frequency ranges associated with animal-induced wakes (Beem et al., 2015, Wu et al., 4 Oct 2025).

A second structural family emphasizes stiffness grading rather than seal-like cross-sectional undulation. The tapered-spring turbulence sensor uses a conical/tapered helical spring fabricated by selective laser sintering of nylon PA12, with total height $122$ mm, base outer diameter $1$0 mm, tip diameter $1$1 mm, and uniform wire diameter $1$2 mm. Three embedded IMUs placed along the spring exploit the taper-induced stiffness gradient so that the bottom node emphasizes mid/high bands, the top node emphasizes low frequencies, and the middle node shows broadband intermediate sensitivity. This is presented as structural encoding analogous to position-dependent mechanoreceptor activity in biological vibrissae (Jin et al., 7 Oct 2025).

A third family focuses on the follicle analog. One implementation uses a straight cylindrical PLA whisker of diameter $1$3 mm and length $1$4 mm inserted into a PDMS square-prism base of $1$5 mm footprint and $1$6 mm height. Four silicon piezoresistive strain gauges are embedded on the sidewalls, with two aligned with the principal bending axis and two orthogonal, so that the dominant compliance is concentrated in the base rather than the shaft. Another mechanically decoupled architecture places the wetted whisker and compliant suspension on one side of a seal, and a rotating magnet plus Hall sensor in a dry cavity on the other side; in this mechano-magnetic arrangement only the whisker and spring are exposed to water (Hang et al., 27 Nov 2025, Wang et al., 2023).

Optical and tactile branches add further structural variation. The underwater FBG whisker sensor uses thin pre-curved superelastic Nitinol wires of diameter $1$7 mm and length about $1$8 mm, each anchored in a 3D-printed base with two orthogonal FBGs and a strain-isolated temperature-compensation FBG. The optical fiber is routed in an undulating groove with minimum bend radius about $1$9 mm and encapsulated in soft silicone. Across these examples, “bio-inspired” does not denote a single morphology: some devices replicate the harbor seal’s undulated elliptic section, some replicate conical stiffness grading, and some replicate only the compliant follicle-like root (Li et al., 2024).

3. Hydrodynamic mechanisms of self-noise suppression and wake responsiveness

The foundational hydrodynamic design principle is asymmetry between clean-flow behavior and disturbed-flow behavior. In uniform flow, the harbor-seal-inspired undulating elliptic whisker exhibits very low vibration, with open-water transverse amplitude peaking at approximately $D$0 over the tested reduced-velocity range. Dye visualization showed an incoherent wake forming further downstream rather than the coherent $D$1 shedding typical of circular and plain elliptical cylinders. The proposed mechanism is that the out-of-phase wavy elliptic cross-section breaks spanwise coherence of Kármán shedding, introduces streamwise vorticity, and prevents lock-in to the Kármán frequency. Once the same whisker enters an upstream wake, however, it undergoes large-amplitude oscillations and passively locks to the wake frequency by “slaloming” between alternating vortices; near each low-pressure vortex core, suction drives the whisker toward one side and then toward the next vortex on the opposite side (Beem et al., 2015).

This passive wake lock-in is distinct from the signal variable emphasized in sea lion array experiments. There, the informative event is not a large smooth oscillation but a sequence of short slip and stick episodes caused by alternating vortex-core encounters. In that work, the wake frequency was decoded most clearly by differentiating the whisker response in time, effectively using a jerk-like signal as a proxy for the time derivative of the bending moment. For weak wakes generated by a $D$2 mm cylinder at $D$3 m/s, the reported signal-to-noise ratio in the $D$4–$D$5 Hz band increased from approximately $D$6 in bending/displacement to approximately $D$7 in jerk, a roughly $D$8 amplification (Muthuramalingam et al., 2018).

The dynamical picture is not limited to classical vortex shedding. A separate fluid–structure study at subcritical Reynolds number showed self-sustained whisker oscillation without classical Kármán shedding. In air at $D$9, the whisker synchronized with its second bending mode at about $f$0 Hz, with drag and streamwise motion sharing that frequency, lift and cross-flow motion sharing the same frequency, and oval-shaped trajectories resulting from equal streamwise and cross-flow frequencies. In prior water results summarized in the same work, synchronization occurred in the first mode, and figure-eight trajectories appeared because streamwise frequency was twice cross-flow frequency. This broadens the mechanistic landscape from wake-induced vibration to synchronization with shear-layer dynamics more generally (Heydari et al., 2022).

A 2026 cyber-physical study reframed wake sensitivity in terms of nonlinear fluid damping. Comparing a circular cylinder, a smooth elliptical cylinder, and an undulating vibrissa model under prescribed virtual mass, stiffness, and damping, it found that reduced-aspect-ratio bodies have minimal VIV in free flow but pronounced wake-induced vibration (WIV) when forced by the wake of a pitching–heaving hydrofoil. Ringdown analysis showed that a Van der Pol damping model describes the data better than quadratic drag, and that the fitted Van der Pol coefficient magnitude $f$1 is systematically lower for the vibrissa than for the smooth ellipse. The vibrissa also showed lower damping ratio and lower fitted fluid-damping coefficients across tested structural frequencies, implying higher sensitivity to weak wakes. A plausible implication is that undulation is not merely a VIV-suppression device; it also reduces effective fluid damping in the sensing regime (Erickson et al., 23 Mar 2026).

4. Transduction, modeling, and signal interpretation

Base-centric transduction remains the dominant engineering approach because biological mechanoreception is concentrated at the follicle analog. In the seal-whisker wake sensor, tip deflection was sensed by strain gauges in a Wheatstone bridge mounted on a compliant flexure, with calibration giving linear voltage–deflection curves of $f$2. In the PDMS-root architecture, each silicon gauge is used in a quarter-bridge configuration, and the piezoresistive relation is written as

$f$3

with measured primary-axis sensitivities $f$4 $f$5N and $f$6 $f$7N for the tensile/compressive pair. Using $f$8, $f$9 $$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$0, and $$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$1 $$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$2N, the reported limit of detection was $$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$3 mN (Beem et al., 2015, Hang et al., 27 Nov 2025).

Magnetic transduction seeks the same base-moment information while isolating electronics from water. In the mechano-magnetic sensor, a permanent magnet rotates with the whisker through a compliant spring suspension, and a 3-axis Hall sensor measures the resulting magnetic flux in a sealed cavity. The later modular study generalized this idea by explicitly calibrating pressure, magnetic, and visual outputs into a common representation,

$$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$4

so that heterogeneous whisker sensors could be compared in terms of bending moment at the base rather than raw signal type. For Hall sensing, the mapping is written as $$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$5; in one calibration dataset, $$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$6 deflection samples were used to fit $$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$7 (Wang et al., 2023, Routray et al., 19 Sep 2025).

Optical FBG systems push the same principle into distributed, EMI-immune sensing. The underlying equations are

$$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$8

In the underwater contact-tracking system, two FBGs per whisker anchor measured orthogonal bending moments and a third FBG provided temperature compensation. A Gaussian Process Regression mapping with a Thin-Plate kernel converted real wavelength measurements into simulated base torques, after which a transformer decoder, “WhiskerNet,” predicted contact location from torque history. This formulation is explicitly a sim-to-real calibration of sensing modality rather than a purely geometric calibration (Li et al., 2024).

Signal interpretation has expanded from spectral peak finding to lightweight inference on mechanically encoded dynamics. The tapered-spring sensor defines the instantaneous $$\mathrm{St} = \frac{fD}{U}, \qquad D \approx \mathrm{St}\,\frac{U}{f}.$$9-D reservoir state as

$2{:}1$0

and uses multinomial logistic regression

$2{:}1$1

as the readout. For vortex peak extraction, the same work used mean removal, a $2{:}1$2–$2{:}1$3 Hz fourth-order Butterworth band-pass, Welch PSD with Hamming window and $2{:}1$4 overlap, and summary features including the dominant low-frequency peak, relative spectral energy within $2{:}1$5 Hz of that peak, and temporal coefficient of variation from STFT. It also used Shannon entropy,

$2{:}1$6

to quantify the transition from coherent vortex streets to turbulence (Jin et al., 7 Oct 2025).

5. Quantitative performance and operational use

Across the literature, quantitative performance is highly task-specific. Wake sensors, turbulence classifiers, dipole detectors, contact trackers, and mobile flow estimators are not optimized for the same bandwidth, geometry, or operating medium. Even so, several recurring benchmarks are visible: suppression of self-noise in clean flow, frequency fidelity under hydrodynamic forcing, low-power operation, and robustness under repeated loading.

Task	Representative quantitative result	Reference
Wake detection by undulating seal-whisker geometry	Open-water baseline $2{:}1$7; wake amplitude $2{:}1$8 at $2{:}1$9; $20\times$0 baseline at $20\times$1; max $20\times$2	(Beem et al., 2015)
Turbulence and vortex-regime classification with tapered spring	Peak identification errors generally $20\times$3; overall accuracy $20\times$4; inference time per sample $20\times$5 ms	(Jin et al., 7 Oct 2025)
Dipole-flow sensing with PDMS-root silicon gauges	LoD $20\times$6 mN; stability after $20\times$7 cycles; offset drift $20\times$8; frequency tracking from $20\times$9 to $d_w = 1.06$0 Hz	(Hang et al., 27 Nov 2025)
Flow-speed estimation on a remote-controlled boat	Velocity RMSE $d_w = 1.06$1–$d_w = 1.06$2 m/s versus survey ground truth	(Wang et al., 2023)
Underwater contact tracking with FBG whiskers	$d_w = 1.06$3 mm RMSE for most tested objects in water	(Li et al., 2024)
Band-limited amplification for animal-wake monitoring	$d_w = 1.06$4 RMS displacement enhancement in the $d_w = 1.06$5–$d_w = 1.06$6 Hz band for the tuned Whisker 1/base pair	(Wu et al., 4 Oct 2025)

Wake sensing remains the most biologically direct application. In the model problem of a whisker downstream of a circular cylinder, the oscillation frequency matched the theoretical wake frequency with $d_w = 1.06$7 across speeds at $d_w = 1.06$8, and the whisker continued to lock over a broad range at $d_w = 1.06$9. In simultaneous pressure–strain sensing, the maximum cross-correlation coefficient between wake pressure and whisker strain exceeded $L/d_w = 30$0 for all tested speeds and cylinder sizes. Because

$L/d_w = 30$1

the locked-in frequency can be used to estimate upstream size, while deviations of $L/d_w = 30$2 from cylinder-like $L/d_w = 30$3 may indicate a different bluff geometry or altered wake state (Beem et al., 2015).

Recent work has extended this operational logic from passive sensing to autonomous tracking. A local control strategy for periodic hydrodynamic signals represents the sensed field by its spectral magnitude $L/d_w = 30$4 and phase $L/d_w = 30$5, estimates the direction of information propagation as $L/d_w = 30$6, and uses the lateral projection of that direction to turn a mobile sensor according to

$L/d_w = 30$7

In simplified radial traveling-wave fields the static-gain and inverse-gain versions are unconditionally convergent, while proportional gain is conditionally convergent; in oscillating-airfoil wakes, the proportional-gain law gave tighter centerline tracking, whereas inverse gain suffers from practical singularity as $L/d_w = 30$8. This suggests that whisker arrays are naturally paired with phase-based control rather than amplitude thresholding alone (Colvert et al., 2018).

6. Limitations, misconceptions, and research directions

A common misconception is that a bio-inspired underwater whisker sensor must reproduce the harbor seal’s exact whisker morphology. The literature shows a broader and less uniform picture. Some devices copy the seal’s out-of-phase undulating elliptic cross-section; some copy only conical taper and stiffness grading; some copy mainly the compliant follicle-like base; and some are straight cylindrical whiskers coupled to alternative transducers. This diversity is not merely terminological. It affects VIV suppression, directional discrimination, calibration invertibility, packaging strategy, and the degree to which conclusions about one architecture transfer to another. For example, the PDMS-root silicon-gauge sensor explicitly notes that the device does not implement the wavy/elliptic cross-section of seal whiskers, while the modular architecture paper does not cover underwater sealing or underwater testing at all (Hang et al., 27 Nov 2025, Routray et al., 19 Sep 2025).

Hydrodynamic and dynamical limitations remain substantial. Far wakes can reorganize, shifting dominant frequencies and limiting precise shape inference at long ranges; the 2015 wake-sensing work observed that continuous alignment across speed narrows in the far wake because of wake reorganization and “wake stiffness” effects. The tapered-spring turbulence study did not report exact node spacing, precise hydrodynamic environment parameters, or sensor–fin spacing, and its accelerometer bandwidth and chosen band-pass constrain which regimes can be discriminated. The cyber-physical vibrissa study was restricted to single-DOF heave and did not directly image the flow field, so its damping interpretations remain indirect. The subcritical synchronization study emphasizes that linear Euler–Bernoulli coupling is insufficient to reproduce sustained oscillation in some regimes, implying that reduced-order design rules can miss essential weak nonlinearities (Beem et al., 2015, Jin et al., 7 Oct 2025, Erickson et al., 23 Mar 2026, Heydari et al., 2022).

Current future directions are comparatively consistent across papers. They include field trials with multiple whiskers and synchronized pressure or velocity sensing at biologically relevant wake speeds; robust packaging and long-term deployment tests addressing biofouling, fatigue, and impact resistance; adaptive filtering or probabilistic inference for environmental variability in $L/d_w = 30$9 and $122$0; arrays tuned across multiple frequency bands for broader coverage; expanded libraries of $122$1 versus $122$2 for varied shapes; 3D wake mapping by distributed whisker arrays and data fusion; and tighter coupling between whisker sensing and local phase-gradient control for wake following and source localization (Beem et al., 2015, Jin et al., 7 Oct 2025, Wu et al., 4 Oct 2025, Colvert et al., 2018).

In aggregate, the field has converged on a clear engineering logic. A successful underwater whisker sensor should be quiet in clean flow, strongly responsive to externally imposed disturbances, instrumented close to the base or along mechanically informative locations, and analyzed in a frequency-aware manner that respects the distinction between self-generated dynamics and forced wake response. The precise route to that objective remains architecture-dependent: seal-like undulated geometry emphasizes passive hydrodynamic selectivity, tapered reservoirs emphasize structural computation, compliant follicle analogs emphasize packaging and robustness, and optical arrays emphasize multiplexed tactile precision.