This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 137.54.34.115

This content was downloaded on 07/11/2016 at 15:28

Please note that terms and conditions apply.

You may also be interested in:

Mechanical characteristics of harbor seal (Phoca vitulina) vibrissae under different circumstances

and their implications on its sensing methodology

H Hans, J M Miao and M S Triantafyllou

Hair flow sensors: from bio-inspiration to bio-mimicking—a review

Junliang Tao and Xiong (Bill) Yu

A flexible liquid crystal polymer MEMS pressure sensor array for fish-like underwater sensing

A G P Kottapalli, M Asadnia, J M Miao et al.

Crocodile-inspired dome-shaped pressure receptors for passive hydrodynamic sensing

Elgar Kanhere, Nan Wang, Ajay Giri Prakash Kottapalli et al.

Touch at a distance sensing: lateral-line inspired MEMS flow sensors

Ajay Giri Prakash Kottapalli, Mohsen Asadnia, Jianmin Miao et al.

Artificial lateral line with biomimetic neuromasts to emulate fish sensing

Yingchen Yang, Nam Nguyen, Nannan Chen et al.

Quantification of sensory data across the vibrissal array

Venkatesh Gopal and Mitra J Z Hartmann

Development of an artificial sensor for hydrodynamic detection inspired by a seal’s whisker

array

View the table of contents for this issue, or go to the journal homepage for more

2016 Bioinspir. Biomim. 11 056011

(http://iopscience.iop.org/1748-3190/11/5/056011)

Home Search Collections Journals About Contact us My IOPscience

Bioinspir. Biomim. 11 (2016) 056011 doi:10.1088/1748-3190/11/5/056011

PAPER

Development of an artificial sensor for hydrodynamic detectioninspired by a seal’s whisker array

WilliamCEberhardt1, Brendan FWakefield2, Christin TMurphy3, Caroline Casey4, Yousef Shakhsheer5,BentonHCalhoun5 andColleenReichmuth2

1 University of Virginia, School of Engineering andApplied Science, 351McCormick Road, POBox 400743, Charlottesville, VA 229042 University of California Santa Cruz, Institute ofMarine Sciences, 100 Shaffer Rd, Santa Cruz, CA 95060,USA3 WoodsHoleOceanographic Institution, 266WoodsHole Rd,WoodsHole,MA 02543,USA4 University of California Santa Cruz, Department of Ecology and Evolutionary Biology, 100 Shaffer Road, Santa Cruz, CA 95060,USA5 University of Virginia, Charles L. BrownDepartment of Electrical andComputer Engineering, 351McCormickRoad, POBox 400743,

Charlottesville, VA 22904,USA

E-mail: coll@ucsc.edu

Keywords: biomimetics, hydrodynamics, wake tracking, vibrissae, whiskers,mechanoreception, harbor seal

Supplementarymaterial for this article is available online

AbstractNature has shaped effective biological sensory systems to receive complex stimuli generated byorganismsmoving throughwater. Similar abilities have not yet been fully developed in artificialsystems for underwater detection andmonitoring, but such technologywould enable valuableapplications formilitary, commercial, and scientific use.We set out to design a fluidmotion sensorarray inspired by the searching performance of seals, which use their whiskers tofind and followunderwater wakes. This sensor prototype, called theWake InformationDetection andTrackingSystem (WIDTS), featuresmultiple whisker-like elements that respond to hydrodynamic disturbancesencounteredwhilemoving throughwater. To develop and test this system,we trained a captive harborseal (Phoca vitulina) towear a blindfoldwhile tracking a remote-controlled, propeller-drivensubmarine. Aftermastering the tracking task, the seal learned to carry theWIDTS adjacent to its ownvibrissal array during active pursuit of the target. Data from theWIDTS sensors describe changes inthe deflection angles of thewhisker elements as they pass through the hydrodynamic trail left by thesubmarine. Video performance data show that these detections coincide temporally withWIDTS–wake intersections. Deployment of the sensors on an actively searching seal allowed for the directcomparison of our instrument to the ability of the biological sensory system in a proof-of-conceptdemonstration. The creation of theWIDTS provides a foundation for instrument development in thefield of biomimetic fluid sensor technology.

1. Introduction

The development of movement detection systems influid environments poses technical challenges tocommercial, military, and scientific fields. Whilesignificant advances in underwater navigation, surveil-lance, and object tracking have been achieved in thelast century with the aid of projected stimuli such assonars, these active sensing systems often reveal thepresence and location of the emitter. The developmentof innovative passive systems for underwater sensingshould allow for expanded characterization of thesurrounding environment with increased stealth. Such

passive sensors have the potential to augment orpossibly replace existing active sensing systems, whileintroducing fewer environmental cues and distur-bances. In the last century, biomimetic approachessignificantly improved the performance of active sonarsystems for underwater object detection and charac-terization [1]. This was accomplished in part throughrefined experimental studies of echolocating dolphins(see e.g., [2–4]). Similar efforts are presently underwayto develop artificial fluid-motion sensors from biolo-gical models [5–12]. Because marine animals possessthe best known performance systems for the detectionof submerged wakes, this project turned to

RECEIVED

20November 2015

REVISED

12April 2016

ACCEPTED FOR PUBLICATION

22 June 2016

PUBLISHED

31August 2016

© 2016 IOPPublishing Ltd

biomimicry to guide the development of an under-water sensing systembased on the sense of touch.

The marine environment impedes visual percep-tion because of light absorption and scattering as wellas suspended particulate matter [13]. This constrainthas contributed to the evolution of specializedmechanoreceptive systems in some animals. A num-ber of aquatic species have demonstrated an ability tomonitor their surroundings by detecting hydro-dynamic signals independently of acoustic, visual, orchemical cues [14]. In particular, pinnipeds (seals, sealions, and walruses) are marine mammals that lackspecialized biosonar [15], but possess the ability todetect waterborne disturbances with their facial vibris-sae (sinus hairs or whiskers) [16–19]. Within thisgroup, true seals (family Phocidae) are particularlyinformative biological models for fluid sensing. Someseals hunt swimming prey in dark or murky watersand forage frequently at night when visual cues are oflimited use (e.g., [20]). Additionally, blind seals havebeen observed to survive in the wild [21] indicating asignificant role for non-visual sensory modalities. Theobservation that different seal species possess notablederived features in their vibrissae [22, 23] indicates ahigh degree of adaptive specialization—and heavyreliance on mechanoreceptive cues—within thistaxon.

The harbor seal (Phoca vitulina) is the best-studiedseal with respect to fluid motion sensing. Dehnhardtand colleagues [24] first tested the ability of a highlytrained harbor seal to detect and report minute waterdisturbances in the absence of other (non-tactile) sen-sory cues. This study revealed detection thresholds forfluid velocities as low as 245 μm s−1 in the 10–100 Hzrange. Subsequent experiments highlighted refinedhydrodynamic tracking performance in harbor sealsby demonstrating their ability to locate and followtrails generated by a remote-controlledminiature sub-marine [25] as well as trails generated by other seals[18]. Trained harbor seals have also been shown toreliably determine the direction ofmovement ofmuchsmaller objects, including synthetic fish fins [26], andto discern the size and shape of wake-generating sti-muli using only their whiskers to contact the wake leftbehind by amoving object [27].

The harbor seal’s ability to locate and precisely fol-low hydrodynamic signals holds significant promisefor the development of wake detection and trackingsensors. This biological system can inspire new fluidsensing applications for use in the underwaterenvironment. Here, we describe a sensor based on themechanoreceptive abilities of seals that is the first of itskind to be deployed during the active pursuit of a sub-merged, moving target. This interdisciplinary effortrequired the collaboration of animal trainers, biolo-gists, engineers, and computer scientists to support thesensor’s creation, development, and deployment. Thefinal proof-of-concept demonstration was conductedwith the instrument carried by a live seal trained to

perform a dynamic wake detection and following task,to determine if the instrument could detect hydro-dynamic disturbances used to guide the seal’s move-ments. In this paper, we first explore the initial designconcept, iterative revisions, the final construction, andtesting environments for the system. We finish withthe performance characteristics of the instrument andan evaluation of the study’s limitations and futureimplications.

2.Design concept for theWakeInformationDetection andTrackingSystem (WIDTS)

The conceptual design of the Wake InformationDetection and Tracking System (WIDTS) incorpo-rated certain aspects of the mechanoreceptive struc-tures of harbor seals into a self-contained unit withmultiple sensors that could be carried on a movingplatform. The performance goal of the instrument wasto detect relevant features present in underwaterwakes, such as those that influence the orientationbehavior of harbor seals. This goal was initiallyconsidered during previous laboratory work [28, 29]which developed and tested prototypes of individualsensors. The current WIDTS prototype was built aftermultiple iterative revisions of these artificial whiskersensors.

Harbor seals possess approximately 88 whiskers[30, 31] arranged in a complex array about the muzzle[31]. Seals have motor control over the array and holdthe whiskers into a protracted position during hydro-dynamic tracking, with the hair shafts nearly perpend-icular to the axis of the body [32, 33]. This increasesthe size of the sensor array, as illustrated by figure 1.Although the functional significance of the arrayarchitecture is not currently understood, we presumethat the spatial layout of the whiskers enables eachindividual sensor to receive a signal with minimalnoise interference from adjacent whiskers and to col-lectively cover the full area surrounding the animal’sface. Following this logic, the WIDTS (described insection 4) presents an array of eight sensors with sim-ple whisker-like extensions; the sensors are organizedin a tiered, radial pattern to offset each artificial whis-ker from the potential flow interference of upstreamwhiskers, while filling the space surrounding the cen-tralWIDTS body.

The whiskers of seals are composed of an externalkeratinous hair shaft that is tapered along its lengthand compressed on the dorsoventral axis, yielding anellipsoidal cross section [32, 34, 35]. Most seal species—including harbor seals—exhibit notable beadingalong the whisker length in a sinusoidal pattern [19].The hair shaft itself extends from a hard capsulebeneath the skin, where the neural elements are con-centrated [23, 35, 36]. When an individual whisker isexposed to water flow from swimming motions and

2

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

interaction with hydrodynamic signals, the displace-ments of the hair shaft are transmitted to the highlyinnervated follicle–sinus complex in the capsule [19].Seal whiskers are innervated and excited individuallyat the follicle, and the resulting neural stimulationlikely maps to well-organized areas of the primarysomatosensory cortex, as has been described for sealions [37]. The encapsulated structure of the seals’follicle–sinus complex informed the design of theindividual WIDTS whiskers (described in section 2).The sensing mechanism is local to each artificial whis-ker, the depth of the sensor’s capsule is similar to thatof the seals’ follicle–sinus complex, and the sensor isembedded within the WIDTS body. As in the animal,the whisker motion received at each sensor is con-verted to electrical signals that are routed to a centralprocessor for recording and integration.

The first step to creating this sensory array was todevelop a functional individual sensor, and prior workmade progress along these lines. Barbier et al [38]demonstrated the feasibility of a successful parallel-plate capacitance-based sensor to measure deflectionsof an artificial whisker. This sensor transcribed hydro-dynamic information, throughmechanicalmovementof the whisker column, into electronic signals bymeans of capacitance change at the base of the whis-ker. Stocking et al [28] progressed Barbier et al’s [38]

design by creating a cone-in-cone base for the sensor’scapacitance meter. Finally, Eberhardt [29] utilized thecone-in-cone design to create further revisions of asingle sensor that is the foundation for the individualsensor design used in this study.

3. Individual sensor design

The individual whisker sensors in the WIDTS weredesigned to measure both the direction and speed offluid movement in a saline, aquatic environment. Toaccomplish this, we used the same artificial whiskerelements from previous efforts in our laboratory[38, 39] and significantly improved the sensory mech-anism. The geometry of these whisker elements wasrelatively simple. In harbor seals, the length ofwhiskers varies from ∼4 to 10 cm and the corresp-onding length-to-diameter (L/d) ratios range from 20to 100; based in part on these measurements, Stockinget al [28] chose a cylindrical shape for the artificialwhisker with an initial length of 4 cm and a circulardiameter of 2 mm, yielding an L/d ratio of 20. Thewhisker elements did not reproduce more sophisti-cated features of seal vibrissae, including tapering ofthe shaft toward the tip, an elliptical or compressedcross section, or a beaded rather than smooth profile.We exploredmultiple iterations ofwhisker designwith

Figure 1.APacific harbor seal withwhisker array extended. Photograph byWalt Conklin.

3

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

a capacitive mechanism for converting motion of thewhisker into electrical signals that discriminate boththe direction andmagnitude of the whisker deflection.Initial prototypes with a flat base [38] showed a highsensitivity tomovement butwere delicate and sensitiveto false measurements from pressure on the base.There was a clear need for a more robust design of thewhisker base that improved the received signal quality.This was the primary focus of the current design effort.

After evaluating several concepts for the whisker,we settled on a cone-in-cone base design with con-ductive material creating parallel-plate capacitorsbetween the facing surfaces. These plates are onlyapproximately parallel to each other, as the nature ofthe cone-in-cone design prohibits a completely paral-lel arrangement of the plates. This design structureincreases the surface area of the capacitor plates toincrease signal magnitude. Figure 2 shows the cone-in-cone structure. The cones converge at their tips tocreate a fulcrumof rotation for the cantilever, allowingit to pivot. The inner cone is secured to the outer conewith silicone enabling fluid to completely fill and flow(surge) throughout the area between capacitor plateswhen the inner conemoves about the pivot point. For-ces exerted on the whisker change the distancesbetween the pairs of capacitor plates.

The system is excited by a sinusoidal signal, whichallows for whiskermotion to bemeasured as the outputvoltage across an electrical load impedance. The con-ductive plates on the outer cone are divided into fourquadrants to provide directional information for theflow through four distinct capacitance measurements.The base of eachwhisker is coveredby a thinmembraneof polydimethylsiloxane (PDMS), approximately200 μm thick, to dampen forces on the cylinder andrestore the whisker to its original resting orientationwhen these forces are withdrawn. Prior revisions of thiscone-in-cone sensor were developed by Stocking et al[28] and Eberhardt et al [39] which produced increas-ingly promising results, but suffered from issues related

to performance and stability. In these designs, the con-ductive silver epoxy covering the capacitive plates wascoated with a thin waterproofing layer. This water-proofing layer introduced a parasitic capacitance in ser-ies with the conductive plates that degraded the signalstrength, and the waterproofing layer also degradedover time to allow for corrosion of the underlyingmetalsurface that further degraded the signal. While thestructure of the sensor in the previous designswas effec-tive there was an apparent need to prevent corrosionand to improve signal strength in thewhisker sensor.

In the present effort, we developed a fourth revi-sion of the whisker sensor (that makes several keyimprovements to previous work [28]) which was usedin the tested WIDTS. Most importantly, the silverepoxy capacitor plates are gold-plated to prevent cor-rosion and allow for the removal of the separate water-proofing layer that previously covered the plates. Thisreduces the parasitic capacitance in series with the gapbetween the plates and substantially improves the sig-nal strength. Further improvements include addingshielded wiring to reduce electromagnetic noise in thewires. Significantly, the depth of the inner cone(18 mm) places the fulcrum of the artificial whisker atthe samemean depth of a harbor seal vibrissa (18 mm)[31]. Figure 3 shows the final revision sensor prior tothe application of the PDMS membrane and beinginserted into the inner cone (left) and after being inte-grated into the WIDTS ring (see section 4) with thePDMSmembrane (right).

Measurements of the current whisker sensorshowed substantially improved longevity and resolu-tion over the version from Eberhardt [39]. Extendedtesting in water over several months showed no decre-ment in signal quality or range, showing that the goldplating effectively removed the previous corrosionissues. The gain range for the sensor was measured at100 Hz for loads ranging from 2Ω to 100 kΩwith dis-tilled water in the gap. The gain range of the whiskerwas increased by 50% over previously tested high-

Figure 2.Diagram of the basic cone-in-cone sensor design fromStocking et al. Reprintedwith permission from [28]Copyright 2016IEEE. The PDMSmembrane (left) provides damping and restoring forces on thewhisker, while the four quadrants of the parallel-plate capacitor (right) generate directional information.

4

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

frequency excitation signals (over 1MHz). Thisimproved gain range provided the sensor greater reso-lution. The measured sensor output matched thetheoretical values within 0.5% across this full range ofloads. In addition, the changes to the sensor allowedfor peak signal quality at much lower input fre-quencies (in the range of 100 Hz). This provided sig-nificant power savings and lower sampling rates tocapture thewhiskermotion.

4. Individual sensor testing in awaterflume

We tested the ability of the single sensor to detect thefrequency content generated by a hydrodynamic wakeand to detect wake crossings. The sensor performedwell when detecting controlled wakes generated byupstream objects in a water flume. For these measure-ments, we used a calibrated Rolling Hills ResearchCompany Model 1520 water flume (test section152 cm long, 28 cm wide, and 48 cm deep) withvelocity uniformity outside the wall boundary layer of±2%. Cylinders with diameters of 2, 6, and 8.9 cmwere mounted to a sting positioned 19.7 and 51.2 cmdirectly upstream of the sensor. Laminar flow speedsranged from 15 cm s−1 to 61 cm s−1. Estimating thevon Karman vortex shedding frequency (assuming aStrouhal number of 0.21), the cylinder wakes testedhere should demonstrate frequency characteristicsranging from 0.34 to 6.03 Hz. The sensor response wasmeasured on two orthogonal quadrants (associatedwith stream-wise and cross-stream motion) with aTektronix 2014C oscilloscope.

The cross-stream quadrant consistently capturedthe frequency content from the wake. Example cases

are shown in figure 4. When the 2 cm cylinder wasplaced upstream in a 61 cm s−1

flow, the sensor detec-ted a distinct peak frequency (identified by FFT analy-sis) at 6.8 Hz. This closely matches the predictedshedding frequency of 6 Hz for the cylinder underthese conditions. Similarly, for the case of an 8.9 cmcylinder, placed upstream from the whisker in a15 cm s−1

flow, the sensor’s response showed a peak at1.36 Hz, which closelymatches the predicted sheddingfrequency of 1.1 Hz. Across 12 test conditions, com-posed of the cylinder diameter, distance to the sensor,and flume speed, the sensor measured a peak fre-quency within 0.1 to 0.8 Hz of the predicted frequencyfor that scenario. These tests demonstrated that thewhisker-like sensor could reliably detect predictedshedding frequencies from awake-generating object.

Once the sensor was demonstrated to detect thepresence and frequency content of a hydrodynamicdisturbance, the next step was to simulate a wakedetection task. Here, the sensor remained in the mid-dle of the water column attached to the sting, while acylinder was swept across the width of the tankupstream from the sensor. The sensor initially sat inthe unobstructed free-stream flow and then wasexposed to the wake from the cylinder as it crossedupstream from the sensor. The strongest signalsoccurred across the quadrant under compressionowing to the inverse relationship to capacitance anddistance, and the strongest response was always eli-cited in the stream-wise voltage. There was a clearincrease in the stream-wise voltage associated with thedetection of the 6 and 8.9 cm cylinders. While thewake from the 2 cm cylinder was more difficult todetect, the sensor was still capable of resolving

Figure 3. (Left)Photo of cone-in-cone sensor with gold plating and no explicit waterproofing layer over the capacitor plates. Theimage shows an individual sensor with the inner cone removed andwithout the PDMSmembrane. Thewhisker is 40 mm longwith a2 mmdiameter. The inner cone is 18 mm in length and 23 mmwide. The outer cone is 18 mm in length and 26 mmwide. The gapbetween the cones is 1.5 mmand isfilledwith distilledwater when the sensor is in use; the PDMSmembrane (not shown) is 0.2 mm.(Right)The same sensor fully integrated into theWIDTS ring (see figure 5)with the PDMS cover andwires routed through to theinner cavity of theWIDTS.

5

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

interception of the wake from 50 cm away in30.5 cm s−1

flow (i.e., from a distance of 25 cylinderwidths). When the cylinder passed in front of the sen-sor, a change in the cross-stream voltage occurredwithan increase in stream-wise voltage. These observationsdemonstrated that the sensors were capable of detect-ing an object crossing their path via sensor rod deflec-tion from the obstruction’s wake. These trials showedthat the sensor would be able to perform in the plan-ned proof-of-concept demonstration, in which wakefeatures could be detected in a realistic tracking task.

5.WIDTS construction and deployment

The WIDTS device enables the detection of hydro-dynamic features with multiple whisker sensors in a

context and configuration similar to those encoun-tered by a swimming seal’s whisker array. The designfeatures a slightly buoyant, torpedo-like body witheight individual artificial whiskers and their associatedcone-in-cone sensors.

Figure 5 shows an exploded view of the completeWIDTS sensor array. The WIDTS body has a conicalhead and octagonal form factor down the length of thebody to provide eight flat surfaces, one for each cone-in-cone sensor. The eight whisker-like sensors are dis-tributed radially around the body and embedded intwo octagonal rings. The rings are hollow, and a shal-low cavity near the inner edge of the side of the ringallows for O-rings between each component of theWIDTS tomaintain awaterproof interior. The octago-nal rings carry four whiskers each, and each individualwhisker occupies one face of an octagonal ring (see

Figure 4. Frequency response of the sensor output to the two extreme conditions tested in the laminar flowof thewaterflume. (Left)FFTmagnitudemeasured from the cross-streamquadrant when the sensor was exposed to an upstream cylinderwith a 2 cmdiameterin a 61 cm s−1

flow. The Strouhal number predicted awake shedding frequency of 6 Hz and the peak capturedwas 6.8 Hz. (Right) FFTmagnitudemeasured from the cross-streamquadrant when the sensor was exposed to an upstream cylinderwith an 8.9 cmdiameterin a 15 cm s−1

flow. The sensor detected a frequency of 0.49 Hz compared to a predicted frequency of 0.34 Hz.

Figure 5.Exploded viewofWIDTS design showing arrangement of body, sensors, and circuit board. The inset (above) provides thewire frame schematic of the nose section (above left) and a sensor array ring (above right).

6

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

also figure 3). The two rings are offset by 45 degrees sothe whiskers do not overlap in the front view and thewires connected to the capacitive plates on the outerand inner cones of each whisker run to the inside ofthe ring via a waterproofed hole. A bite plate attachedto the back of the instrument enabled a trained seal tocarry the device clenched firmly in its mouth anddirectly in front of its ownwhisker array.

A 5 cm×13 cm custom printed circuit board(PCB) inside the middle of the WIDTS body is held inplace by a slot on the inside of the ring sections of theWIDTS. The PCB’s custom-designed electronicsexcite all eight sensors with a sine voltage waveform,acquire returning output signals, sample the peak volt-age values from each whisker quadrant, and store theoutput data to memory. The PCB, which is poweredby a 3.6 V lithium-ion battery, excites all whiskers witha 790 Hz sine wave, terminates the received signal withan on-board resistor, and extracts the gain rangewith apeak detector. The signal is sampled and digitized at arate of 512 Hz and stored tomemory.

The selected sampling rate was constrained by themicrocontroller and the absolute sampling rate of themicrocontroller was constrained by the number ofADC inputs. Data were later verified by comparingsensor output with the overhead video tracks of theswimming seal. Thismitigated some of the issues asso-ciated with undersampling; however, the low sam-pling rate remains a limitation of the electronics. Thedata collected by the WIDTS can be transmitted wire-lessly over Bluetooth radio to an external base station(laptop) via a custom Java computer program. Impor-tantly, the board also includes three gyroscopes andthree accelerometers to correlate the orientation of theboard and head movement artifacts to measured dataacquired at the whisker sensors, since head motionmight cause the whiskers to experience motion causedby drag through thewater.

It was necessary to deploy the WIDTS adjacent tothe seal’s own vibrissae to enable field testing of theWIDTS in a realistic environment used by the biologi-cal sensory system. The bite plate allowed the trainedseal to hold the WIDTS directly in front of its ownwhisker array, exposed to the same hydrodynamicinformation the seal would receive when tracking amoving target. However, the size and placement of theWIDTS raised concerns that it would obstruct theseal’s sensory abilities and thus tracking performance.To examine the impact of theWIDTS body on the flowfield received by the animal, we created a 3D-printedmodel of a juvenile harbor seal head from available CTscan images (the model head was smaller than that ofthe adult test subject, representing a ‘worst-case’ dis-turbance scenario) and evaluated the flow dynamicssurrounding both theWIDTS and the seal’s head.

Particle image velocimetry (PIV)measurements ofthe flow field around models of the WIDTS and theseal head were obtained in the water flume describedin section 2. These were used to verify flow field

modeling in ANSYS CFX. Summary results of themodeling are shown in figure 6 (for details see [29]).The ANSYS CFX flow-field simulations revealed anarea of low-speed recirculation between the WIDTSand the seal nose. Fluid flow directly downstream ofthe WIDTS edge was slowed near the seal’s muzzleuntil encountering the contours of the seal’s head.Flow velocities were compared at points near to(<1 cm) and far from (>1 cm) the seal’s nose. The farpoint (located in the field of the seal’s whisker array,figure 6) did not experience changes with the intro-duction of the WIDTS. Because the vibrissae act ascantilever beams with an end fixed at the seal, the for-ces acting further out along the whisker have a greaterinfluence on the whisker’s deflection. As a result, thefar field velocitymagnitude is likely to play a larger rolein the information gathered by the seal than that of thenear location point. With this in mind, we predictedthat the instrument would not substantially block thehydrodynamic information received by the seal. Wedid not explicitly model the potential disturbance cre-ated by the artificial whiskers on the seal’s ownwhiskerarray.

After the design and modeling phase of theWIDTS was complete, we 3D-printed the WIDTSnose cone and cast the ring sections in epoxy. Fourwhisker sensors were hand-assembled in each ringsection with their outer cones covered by PDMS andwith distilled water in the gap between the cones. Dueto this design, occasional refilling with distilled waterwas required; future iterations of the sensors wouldaim to reduce evaporative water loss. The base plateand bite plate of theWIDTS were cut from plastic, andthe bite plate was covered in neoprene for the seal’scomfort and extra grip. The entire assembly wassecured by bolts that ran from the base plate throughthe ring sections to a nut in the nose cone (figure 2).

6. Biological performance of a trained seal;test subject and environment

To enable a proof-of-concept demonstration of theWIDTS, we worked closely with a captive seal thatcould track moving objects under water using onlyhydrodynamic trails. The subject for this study was amale Pacific harbor seal (P. vitulina) named Sprouts(identification NOA0006602). The seal was 22 yearsold at the start of the study and had extensive priorexperience in performing trained behaviors forresearch purposes. This individual was housed in anoutdoor, circular seawater-filled pool (1.8 m depthand 7.6 m diameter)with adjacent haul-out decking atLong Marine Laboratory in Santa Cruz, CA. Allhusbandry and research-related tasks relied on stan-dard operant trainingmethods, using positive reinfor-cement. Training for this task occurred between 2010and 2013, during which time the seal participated inwake-following sessions between two and five times

7

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

perweek. The seal received about one-third of his dailydiet as reward for cooperative behavior during thesetraining sessions, which typically included 10 to 20trials. No aspects of the research were invasive orinvolved restraint or food deprivation.

The research activities were conducted with theapproval and oversight of the Institutional AnimalCare and Use Committee at the University of Cali-fornia at Santa Cruz, with federal authorization fromthe National Marine Fisheries Service of the UnitedStates (marinemammal research permit 14535).

The seal was gradually trained to find and followwakes while simultaneously wearing a blindfold andcarrying the WIDTS instrument firmly in his mouth.The seal was initially taught to swim and eat comfor-tably while wearing a blindfold made of visually opa-que neoprene. He also learned to detect and follow aturbulent underwater wake created by one of two dif-ferent moving objects in his pool. The wake-generat-ing object used during the initial stages of training wasa simple sphere (racquetball, 7 cm diameter) draggedthrough the water by a 2 cm-diameter rigid steel pole.Later, we introduced a self-propelled, remote-con-trolled submersible with physical characteristics sui-table for the eventual WIDTS field tests. Thesubmersible was a Thunder Tiger Neptune SB-1

submarine (length 78 cm, height 21 cm, beam 18 cm,propeller size 5 cm, 12 V motor propulsion) whichoperated at a speed of 0.5 m s−1 and amaximumdepthof 1 m. The top speed of the submarine was less thanthe normal swimming speed of harbor seals, but theseal was trained to slow his swim tomatch the speed ofthe submarine. The time delay between the start ofwake generation and the release cue for the trackingbehavior was progressively increased, from very briefdelays (0.5 to 2 s) to delays as long as 15 s. Although theseal could have performed the task with much longerdelays [25], we used a maximum 15 s delay to avoidoverlap in hydrodynamic paths in the pool at the startof each trial. In all cases, the seal was rewarded for fol-lowing the path of the wake-generating object at afixed distance of ∼0.3 m, with continuous trackingdurations ranging from a few seconds to more than aminute. The final component of training was for theseal to carry a mock WIDTS instrument in his mouthwhile performing the task.

The seal was trained at first to carry only the biteplate while performing the wake-following task, andthen the bite plate was fitted with a mock-up of theWIDTS (i.e., without electronics). We found that fol-lowing practice with the instrument, the seal’s wake-following ability was not notably impeded by the body

Figure 6.ANSYSCFX flow-field simulations of the area surrounding a harbor seal’smodeled headwithout theWIDTS (upper left)andwith theWIDTS positioned in front of the seal (upper right). The images show the steady-state velocity contour plot (of themid-plane passing through the seal’s head) in 0.5 m s−1

flow. There is an area of low-speed recirculation between theWIDTS and seal nose.The fluidmotion passes along theWIDTS in the downstreamdirection until turningwhen it encounters the head. The lower panelsprovide a closer look at the area around the seal’s head and between the nose andWIDTS. Velocity vectors colored for velocitymagnitude (lower left) show the area of recirculation between theWIDTS and harbor seal’s head. Theflow is slowed around the sealmuzzle in the regionwhere vibrissae are present, but there is only a small difference in theflowfield a short distance away from thehead (>1 cm) in the area thewhiskers extend out (lower right). Given that the seals’whiskers extend>4 cm from themuzzle, itappears that the vibrissae should still be capable of reaching the free-stream flow and capturingfluidmotions.

8

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

of the WIDTS. Further, despite holding the bite platefirmly between his teeth, the protraction of his whiskerarray was not different with and without the WIDTS.By the conclusion of training, the seal had learned toreadily perform the wake-following task while wearingthe blindfold and carrying the mock WIDTS securelyin front of his own whiskers (supplementary video 1).He had also learned to carefully return the WIDTS tohis trainer at the end of each trial before receiving a fishreward.

During experimental trials the blindfolded sealcarried the WIDTS while finding and following thetrail generated by the self-propelled submarine. Thesubmarine was controlled remotely to follow pre-determined paths ranging from simple circuits aroundthe edge of the pool to more complex variations invol-ving one ormore direction changes. Sessions were var-ied, alternating between test trials with the sealcarrying the WIDTS and behavioral maintenancetrials that involved the seal carrying the mock instru-ment or no instrument. The experimental trials lastedfrom 15 to 75 s, and themaintenance trials lasted from1 to 75 s. All trials were recorded by a wide-angle videocamera suspended directly above the center of the poolto enable time-synched performance evaluation ofboth the seal and the instrument.

The blindfolded seal was able to find and then fol-low the generated wake on all experimental trials. Thatis, the seal would find the path of the wake-generatingobject, and thenmaintain a consistent following beha-vior that reflected the movement patterns of theobject. Despite a delay of up to 15 s between the onsetof the submarine’s movement and the release of theseal by the trainer, the seal could accurately detect thewake as soon as he crossed its path. Additionally, hecould resolve the direction of the moving object fromthe available hydrodynamic cues, altering his courseimmediately upon intersecting the wake. Further, hecould follow the wake vertically as the submarine’spath was varied between the water’s surface and themaximum depth tested (1 m). We often observed theseal sweeping his head from side to side while per-forming the task, which is consistent with descriptionsof find and follow behavior in seals [18, 25]. This beha-vior was most evident when the seal was released tolocate a hydrodynamic trail, but also occurred whenthe seal was following a stimulus that was turning ordipping in an unpredictable pattern. In contrast, theseal rarely moved his head in an up–down motion.There was no notable difference in head-weaving intrials where the seal carried or did not carry theWIDTS.

While we did not use a deliberate acoustic maskerto cover potential sound localization cues provided bythe moving objects, we are confident that the seal usedhydrodynamic cues to find and follow the hydro-dynamic trails. During testing and training, there wassteady water flow into the tank that generated broad-band noise that would have diminished available

acoustic cues. More importantly, it was obvious thatthe blindfolded seal never used dead reckoning to findthe wake-generating stimulus; rather than taking theacoustic path (the shortest path), the seal would searchslowly in a general area until he detected the hydro-dynamic trail, and then turn quickly and immediatelyin the correct direction to follow the path (supplemen-tary video 1).

7. Field testing and analysis ofWIDTSperformance

Our objective for the field test was to determinewhether the ability of the WIDTS sensors to detectsubmerged wakes in the lab would translate to adynamic testing environment, similar to that used bythe biological model. We hoped to show that some ofthe features of the hydrodynamic path that guided theseal’s tracking behavior were also detectable by theWIDTS. Our approach for this proof-of-conceptdemonstration was to observe howWIDTS detectionsrelated to changes in the harbor seal’s wake-trackingbehavior. By coupling the WIDTS to the harbor sealvia the bite plate held securely in his mouth, thesensors were exposed to a hydrodynamic environmentsimilar to that encountered by the free-swimming seal.This is illustrated by figure 7, which shows theblindfolded seal following the path of a dragged objectwhile carrying the complete instrument package.

Before conducting experimental trials, we eval-uated the wake produced by the moving submarine inthe water flume described in section 2, and then ver-ified that theWIDTS could detect the submarine in thetesting pool used by the seal.

As with the seal head modeling effort, we obtainedPIV measurements in the flume to characterize thewake produced by the radio-controlled submarine.This evaluation provided useful but limited informa-tion due to the large size of the submarine relative tothe size of the water flume. The diameter of the sub-marine blocked 37.8% of the flume cross-sectionalarea, which accelerated water flow and constrained thewidth of the wake left by the submarine’s propeller.Therefore the flume measurements provided only aminimum estimate of the width of the wake. PIVsampling (five sections of 18 by 30 cm with 500 imagepairs in 15 cm increments as the submarine wasmoved forward in successive steps) allowed the flowfield 70 cm behind the sub to be captured with goodresolution. The results indicated that the wake shouldcover an area at least as large as the submarine’s dia-meter and that the wake is likely to persist for severalbody lengths with measurable velocities (for details,see [29]). Visualization of the surface wake visible inthe seal’s testing pool confirmed these minimumestimates.

To demonstrate that the WIDTS could detect thewake of the submarine in the test pool, the submarine

9

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

was held stationary and the WIDTS was manuallyswept behind it and through the propeller stream at arange of 0.5 m. Overhead video footage indicatedwhen in time the WIDTS was directly behind the sub-marine. Figure 8 shows the measured outputs of theWIDTS’s accelerometer and of the horizontally posi-tioned whiskers on the left and right sides of theWIDTS during one trial. As the WIDTS passed from

left to right, the right sensor reported a detection justbefore the main WIDTS body was directly behind thesubmarine, and the left sensor detected flow immedi-ately after. The opposite happened when the WIDTSswept back through the wake from right to left; the leftsensor detected the wake just before the body wasdirectly behind the submarine, followed by the rightsensor. The accelerometer data show the change in

Figure 7.The harbor seal Sprouts tracking an underwater wakewhile wearing a blindfold and holding theWIDTS firmly in hismouth.Note that the straight artificial whiskers extending from the body of theWIDTS are positioned directly in front of the seal’s whiskerarray.

Figure 8.Unfiltered time series plots depicting theWIDTS beingmanually swept behind the submarine’s propeller stream in thetesting pool. The red lines indicate the temporal instanceswhen theWIDTS arraywas directly behind the submarine. The horizontalaccelerometer data (top) indicate the sweepmotion from left to right and back again. The sensor plots of themeasured voltage(middle, bottom) show that the submarinewake strikes the right sensor of theWIDTS before the left one on thefirst pass, and thepattern is reversed as theWIDTS ismoved back to its initial position, with the left sensor responding prior to the right.

10

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

direction of theWIDTS in between passes through thewake. These data confirm that the WIDTS sensorscould detect the submarine’s wake in the testing pool.Further, the data gathered by the accelerometers andgyroscopes confirmed the horizontal movement pat-terns in the body of theWIDTS.

Experimental pursuit trials with the seal wererecorded by the wide-angle video camera mounteddirectly above the testing pool. A custommotion ana-lysis program made in Matlab identified both theWIDTS and the submarine in the video footage bytheir color, and tracked the center of these objects asthey traveled around the pool. The plotted trackscould be viewed as animations of the progressing trial,or as a composite image of the entire trial. A review ofthe seal’s performance data across the test trials indi-cated that he closely followed the path left by the sub-marine, whether he was carrying the WIDTS or not.Figure 9 shows a composite track of a simple trial,while supplementary video 1 shows an example of thesealfinding and then following amore complex path.

On all trials, the seal was able to find and then fol-low the hydrodynamic path with a high degree of acc-uracy, typically remaining within 0.3 m of the pathtaken by the submarine. The WIDTS did not interferewith his performance, and although the effect of theartificial whiskers had not been modeled, and he hadnot practiced with artificial whiskers on the mockinstrument, he had no difficulty in performing the taskwith the full WIDTS unit. On most trials, some weav-ing in and out of the wake was observed, as seen in themovement of the red tracking line in figure 9. In gen-eral, the seal appeared to have a preference to stay juston the outside of the submarine track. It is notable thatthe motion analysis software plots the center of the

submarine’s location and not the wake location; as thetail of the submarine points outward from the curva-ture of the track, and the propeller displaces water inthis direction (to the outside of the path), it is assumedthat the outer path taken by the seal is actually alongthewake trail left by the submarine.

The overhead video was correlated to the datatransferred from the WIDTS after each trial to exam-ine the extent to which the artificial sensors respond tothe same stimuli as the seal. Figure 10 depicts aninstance where the WIDTS sensor detections corre-spond to the location of the submarine wake during anexperimental trial. At 27.2 s into the pursuit, the sealcarried the WIDTS just to the left of where the sub-marine had passed (the former location of the sub-marine indicated with a red ellipse). He crossed thewake and was on the right side of the wake at 29 s.These times are marked with red lines on the right andleft sensor output data. Each line coincides with adetection spike; the right sensor detects at 27.2 s (whentheWIDTSwas on the left of the submarine wake) andthe left sensor detects at 29 s (when theWIDTS was onthe right of the submarine wake). These data are repre-sentative of the wake crossings observed during thetrials. While head motion, rather than wake detection,could be a possible explanation for these detectionevents, movement data from the accelerometers andgyroscopes, as well as the time-synched overheadvideo, indicate that this was unlikely to be the case.

8.Discussion

This work accomplished two major objectives. First,it built upon a previously fabricated concept for a

Figure 9.A simple experimental trial illustrating the path of the center of the submarine (yellow) and the subsequent path of the centerof theWIDTS (red), which is held in the seal’smouth. Several wake crossings are evident, as is amild bias of the seal’s path toward theoutside boundary of the submarine’s path.

11

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

biomimetic, whisker-like, individual fluid motionsensor by improving its mechanical design anddeveloping its sensitivity to identify specific wakecharacteristics. We created an array of eight indivi-dual, capacitance-based, cone-in-cone cantileversensors, each offset by 45 degrees, into a completeWIDTS that is capable of identifying hydrodynamicwakes generated by simple and complex objects.Second, this work provided a proof-of-conceptdemonstration that the WIDTS could detectencounters of underwater disturbances on a movingplatform, which represented realistic environmentsfor sensor application. The field testing conditionswere similar to the environment encountered bythe imitated biosensory system; while swimming, atrained seal carried the experimental device immedi-ately adjacent to its own vibrissal array. Observationsof the field trials give every indication that theWIDTS successfully detected the same hydrody-namic disturbances that triggered changes in theseal’s tracking behavior, providing further evidencethat artificial whisker arrays can achieve underwatersensing capabilities.

To our knowledge, this is the first time such anarray of hydrodynamic sensors has been used in a rea-listic environment to detect a freely moving sourceobject.Whilefield testing in the current scenario reliedon data from only two of the instrument’s eight sen-sors as a result of time limitations and some technicalissues, the design of the instrument allows for morerobust, spatially integrated measurements in thefuture. Additional refinements can further improvethe sensitivity and accuracy of this sensor system. TheWIDTS will certainly benefit from developments thatreduce electrical noise and increase the resolution ofoutput data. Such improvements will enhance theability of the instrument to detect and identify subtlewake characteristics.

The ability of an artificial sensory system to detectand report fine-scale attributes of hydrodynamic sti-muli will lead to improved identification of sourceobjects in the presence of environmental flow noise.Different object shapes, and objects traveling at vary-ing speeds, leave different ‘footprints’ in water, and itis crucial to understand how these stimulus differencesinfluence a wake’s signature. Such research has

Figure 10.Example ofWIDTSwake detection during field trials with the harbor seal. (Top) Screenshots from the overhead video of ananalyzed trial (top), taken at 27.2 s and 29 s. The red ovals indicate the prior positioning of the submarine relative to the seal. Right-side sensor (middle) and left-side sensor (bottom)WIDTS data are shownwith the screenshot times indicated by red lines. The right-side sensor shows a peak at 27.2 s as the right side of theWIDTS passes into the submarine’s wake; the left-side sensor shows a peak at29 s as theWIDTSmoves out to the right side of the submarine’s path.

12

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

recently been conducted to describe the hydro-dynamic characteristics of the wakes produced byswimming fish, which in turn has provided insightinto the structure and function of the seal’s tactile sen-sory array [40–42]. This work in the biological domainunderscores the importance of accurately defining thecharacteristics of target wakes. Improving knowledgeof wake characteristics unique to source objects willsupport decision-making for whatever operator oralgorithm is interpreting theWIDTS output data.

An additional need is for the continued refinementof the sensor design to more thoroughly evaluate andmimic the unique morphology of seal vibrissae (seee.g., [43]). Pinnipeds possess the most highly specia-lized vibrissae of any animal group, and the uniquemorphology of these structures is thought to enabletheir unsurpassed ability to extract complex informa-tion fromhydrodynamic flow fields [19]. For example,the elliptical cross-sectional profile of the whiskers ofseals and other pinnipeds (in contrast to the circularprofile of those of terrestrial mammals)may play a rolein noise reduction, thereby affording an advantageover a structure with a circular profile [32]. In addi-tion, most species of seals have a repeating series ofundulations or ‘beads’ along the surface of their vibris-sae that do not occur in any other animal group (Gin-ter et al 2012, 2010). The fine-scale structure of theseundulations acts to suppress vortex-induced vibra-tions that would otherwise be generated by the ani-mal’s movement through the water [33]. Thesefeatures of seal whiskers that apparently serve toincrease signal-to-noise ratios have been shaped byevolutionary pressures; such design features could beevaluated and incorporated into the WIDTS toadvance the capabilities of the artificial sensors tomore closely simulate those of the biological system.

Much of what we observed through this work rai-ses questions regarding exactly how seals (and otheranimals dependent on fluid-dynamic information)perceive wakes in their aquatic environment. Animalbehavior can be a valuable window into nature’sadvanced stimulus detection systems, and analysis ofthe wake-tracking behavior of harbor seals in con-trolled conditions can inform effective tracking algo-rithms andmethods. For example, the seal often swepthis head from side to side while finding and followingsubmerged wakes. This movement could be inter-preted as search behavior, or repeated edge detectionof the hydrodynamic path, although additionalresearch is needed to describe the pattern of headmotion relative to wake diameter. These observationsmay be helpful in developing algorithms for adaptivewake-tracking technology. Such seal-inspired algo-rithms could achieve correction strategies for theWIDTS or WIDTS-like sensors such as ‘if driftingright, veer left’, and could easily work multi-dimensionally.

In addition to providing a proof-of-concept of theinstrument design, the field tests revealed several

weaknesses in the prototype WIDTS. Waterproofingthe WIDTS body and sensor membrane thresholdpresented a constant challenge, as uneven surfaces atthe O-ring interfaces caused slow leaks during testing.The uneven surfaces likely arose from the 3D printingprocess used for the nose cone, so it is possible that adifferent manufacturing approach could prevent theleakage. Flooding resulted in the loss of one PCB.While the sensors were excited at 790 Hz, aliasingresulted from undersampling the output data at 16 Hzwhen all thirty-two quadrants, three gyroscopes, andthree accelerometers were recorded simultaneously.Sharp oscillations in the data, initially attributed tonoise, were deemed a result of aliasing. Finally, ablackening of the gold-covered capacitor places wasnoticed during testing in seawater, which had notoccurred until the PCB was used to excite the sensors.The blackening of the capacitor plate surface appearsto further increase noise in the sensor output and ishypothesized to have resulted from higher currentsgenerated by the PCB than the function generator usedin individual sensor prototyping.

The quality and correlation of video data, move-ment data of the instrument, and detection data fromthe instrument could also be improved. For example,poor image quality from the video camera oftenreduced our software’s ability to follow the seal and thesubmarine in less-than-ideal lighting. The object-tracking software groups pixels using highly contrast-ing areas, and algorithms follow these areas fromframe to frame. Often, direct sunlight brightly illumi-nated the pool bottom and wind textured the surfaceof the water; both of these conditions made it harderfor our software to continuously and accurately trackthe trained harbor seal and the submarine. During theproject, we did not attempt to merge the data streamfrom the video recording with the data streams fromthe WIDTS in a dynamic manner. A correlated inter-face for data evaluation would be useful for advanceddata interpretation.

The submarine used during field testingwas not anideal wake generator; it was relatively large, slow, andbulky, and was limited to a radio-controlled depthrange of 1 m in seawater.While we tracked themotionof the submarine from its center, the wake was pro-duced from the rear of the object and water was typi-cally displaced to the outside of the circular pool; thedeviations of the seal from the submarine’s path infigure 9 may reflect this difference. Smaller objectswith more constrained flow fields would be useful totest with theWIDTS. During training, the seal showedexcellent dynamic wake-tracking behavior with muchsmaller objects. For example, the blindfolded sealcould easily discriminate between the sphere it wastrained to follow and the rigid pole that dragged thesphere though the water column (a difference of5 cm). Identifying a smaller and faster wake-generat-ing option that could create distinctive hydrodynamic

13

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

paths in a large area would be helpful for the biomi-metic comparison.

To view this effort in a broader context, behavioralevidence has clearly demonstrated that seals use theirvibrissae to determine the size, shape, and movementdirection of submerged stimuli from hydrodynamictrails. Although the underlying functioning of the ani-mal’s sensory system is not fully understood, biologistshave begun to explore the mechanisms supporting itsadvanced capabilities and to identify aspects of incom-ing signal characteristics that support wake detection.Our attempt to develop an artificial sensor array thatcould detect hydrodynamic events relevant to thewake information received and used by seals hasresolved several technical challenges and identifiedmore. Further development of hydrodynamic sensorscan continue to incorporate emerging knowledge ofthe seal’s sensory system, so that these artificial sys-tems can progress beyond basic detection of hydro-dynamic events towards improved characterization ofwake-generating stimuli. The interdisciplinary natureof this project drew inspiration from animal behaviorand research on sensory capabilities to build and test anovel hydrodynamic wake-sensing system. The designof the WIDTS and the results herein advance our abil-ity to replicate the seal’s remarkablemechanoreceptivesensory abilities.

Acknowledgments

This project was the vision of Dr Joseph (Pepe)Humphrey at the University of Virginia who devel-oped the original concept, found initial funding for thestudy, and brought the team together prior to hispassing. Most of the work described here is containedin the dissertation of W C E who conducted most ofthe research and development tasks. The project wassupported by the Office of Naval Research (awardN00014-09-1-0468) and we thank Dr Ron Joslin ofONR’s Turbulence and StratifiedWakes and Submar-ine Maneuvering and Control Programs for hissupport. The sensor components were fabricated byJohn Paulus andMike Appleby of Mikro Systems, Inc.The CT data for the modeling of the harbor seal headwere provided by Dr Eric Montie of the University ofSouth Carolina Beaufort. Chris Gregg, MichaelLandau, and Jonathan Stocking of UVA providedessential technical assistance that enabled the successof this project. Many dedicated students, interns, andvolunteers at LongMarine Laboratory (UCSC) assistedus with the project, especially Amy Bernard, MeganConnolly Sadou, andAndrewRouse.

Author contributionsWC E conceived the project, designed the instru-

ment, and collected and analyzed the data. B FWassis-ted with data collection and wrote themanuscript. C TM assisted with data collection and manuscript prep-aration. C C trained the seal, and wrote the

manuscript. Y S assisted with instrument testing andinterpretation of data. B H C was involved in allaspects of the study and was responsible for funding.C R conceived the project, supervised animal research,andwrote themanuscript.

References

[1] Müller R andKucR 2007 Biosonar-inspired technology: goals,challenges and insightsBioinsp. Biomim. 2 S146–61

[2] Nachtigall P E, AuWWL,RoitblatH L and Pawloski J L 2000Dolphin biosonar: amodel for biomimetic sonarsProc. 1st Int.Symp. onAqua Bio-Mechanisms pp 115–21

[3] ZhangY et al 2014A biomimetic projector with highsubwavelength directivity based on dolphin biosonarAppl.Phys. Lett. 105 123502

[4] Pailhas Y, CapusC, BrownKandMoore P 2010Analysis andclassification of broadband echoes using bio-inspired dolphinpulses J. Acoust. Soc. Am. 127 3809–20

[5] FanZ, Chen J, Zou J, BullenD, LiuC andDelcomyn F 2002Design and fabrication of artificial lateral lineflow sensorsJ.Micromech.Microeng. 12 655

[6] DijkstraM, van Baar J J,Wiegerink R J, Lammerink T S J,de Boer JH andKrijnenG JM2005Artificial sensory hairsbased on theflow sensitive receptor hairs of crickets JMicromech.Microeng. 15 S132–8

[7] Ju F and Ling S-F 2014 Bioinspired active whisker sensor forrobotic vibrissal tactile sensing SmartMater. Struct. 23 125003

[8] Tao J andYuX (Bill) 2012Hairflow sensors: frombio-inspiration to bio-mimicking—a review SmartMater. Struct.21 113001

[9] Humphrey J AC, Barth FG, ReedMand SpakA 2003Thephysics of arthopodmedium-flow sensitive hairs: biologicalmodels for artificial sensors Sensors and Sensing in Biology andEngineering pp 129–44

[10] ZhuL, Zeng L, ChenX, LuoX and Li X 2015Abioinspiredtouching sensor for amphibiousmobile robotsAdv. Robot. 291437–52

[11] Ju F and Ling S-F 2015Amicrowhisker transducerwithsensorlessmechanical impedence detection capability for fluidand tactile sensing in space-limited applications SensorsActuatorsA 234 104–12

[12] BeemHRandTriantafyllouMS 2015Wake-induced‘slaloming’ response explains exquisite sensitivity of sealwhisker-like sensors J. FluidMech. 783 306–22

[13] Kröger R 2008The physics of light in air andwater SensoryEvolution on the Threshold: Adaptations in Secondarily AquaticVertebrates ed JGMThewissen and SNummela (Berkeley, CA:University of California Press) pp 113–9

[14] KalmijnA J 1988Hydrodynamic and acoustic field detectionIn Sensory Biology of Aquatic Animals ed J Atema et al (NewYork: Springer) pp 83–130

[15] SchustermanR J, KastakD, LevensonDH,ReichmuthC J andSouthall B L 2014Why pinnipeds don’t echolocate J. Acoust.Soc. Am. 107 2256–64

[16] HankeW,Wieskotten S,Marshall C andDehnhardt G 2013Hydrodynamic perception in true seals (Phocidae) and earedseals (Otariidae) J. Comparative Physiol.A 199 421–40

[17] GläserN,Wieskotten S,Otter C, DehnhardtG andHankeW2011Hydrodynamic trail following in aCalifornia sea lion(Zalophus californianus) J. Comparative Physiol.A 197 141–51

[18] Schulte-PelkumN,Wieskotten S,HankeW,DehnhardtG andMauck B 2007Tracking of biogenic hydrodynamic trails inharbour seals (Phoca vitulina) J. Exp. Biol. 210 781–7

[19] HankeWandDenhardt G 2015Vibrissal touch in pinnipedsScholarpedia 10 6828

[20] Frost K J, SimpkinsMA and Lowry L F 2001Diving behaviorof subadult and adult harbor seals in PrinceWilliamSound,AlaskaMar.Mammal Sci. 17 813–34

[21] NewbyT,Hart F andArnold R 1970Weight and blindness ofharbor seals J.Mammalogy 51 152

14

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al

[22] Ginter CC,DeWitt T J, Fish F E andMarshall CD 2012 Fusedtraditional and geometricmorphometrics demonstratepinnipedwhisker diversity PLoSOne 7 1–10

[23] Ling J K 1977Vibrissae ofmarinemammals FunctionalAnatomy ofMarineMammals edR JHarrison (London:Academic) pp 387–415

[24] DehnhardtG,Mauck B andBleckmannH1998 Seal whiskersdetect watermovementsNature 394 235–6

[25] DehnhardtG,Mauck B,HankeWandBleckmannH2001Hydrodynamic trail-following in harbor seals (Phoca vitulina)Science 293 102–4

[26] Wieskotten S, Dehnhardt G,Mauck B,Miersch L andHankeW2010Hydrodynamic determination of themovingdirection of an artificialfin by a harbour seal (Phoca vitulina)J. Exp. Biol. 213 2194–200

[27] Wieskotten S,Mauck B,Miersch L,Dehnhardt G andHankeW2011Hydrodynamic discrimination ofwakes causedby objects of different size or shape in a harbour seal (Phocavitulina) J. Exp. Biol. 214 1922–30

[28] Stocking J B, EberhardtWC, Shakhsheer YA, CalhounBH,Paulus J R andApplebyM2010A capacitance-basedwhisker-like artificial sensor forfluidmotion sensing Proc. IEEE Sensorspp 2224–9

[29] EberhardtWC2012Development of a Bio-InspiredWhisker-Like Sensor for FluidMotion TrackingDoctoral DissertationUniversity of Virginia

[30] DehnhardtG andKaminski A 1995 Sensitivity of themystacialvibrissae of harbour seals (Phoca vitulina) for size differencesof actively touched objects J. Exp. Biol. 198 2317–23

[31] SadouMC, BeltranRB andReichmuthC2014 Short note: acalibration procedure formeasuring pinniped vibrissae usingphotogrammetryAquaticMammals 40 213–8

[32] MurphyCT, EberhardtWC,CalhounBH,MannKA andMannDA2013 Effect of angle onflow-induced vibrations ofpinniped vibrissae PLoSOne 8 69872

[33] HankeW et al 2010Harbor seal vibrissamorphologysuppresses vortex-induced vibrations J. Exp. Biol. 213 2665–72

[34] Ginter CC, Fish F E andMarshall CD 2010Morphologicalanalysis of the bumpy profile of phocid vibrissaeMar.MammalSci. 26 733–43

[35] HyvärinenH1989Diving in darkness: whiskers as senseorgans of the ringed seal (Phoca hispida saimensis) J. ZoologyLondon 218 663–78

[36] Marshall CD, AminH,Kovacs KMand LydersenC 2006Microstructure and innervation of themystacial vibrissalfollicle-sinus complex in bearded seals, Erignathus barbatus(Pinipedia: Phocidae)Anat. Rec.A 288A 13–25

[37] Ladygina T, PopovV and SupinAY 1992Micromapping of thefur seal’s somatosensory cerebral cortexMarineMammalSensory Systems (Berlin: Springer) 107–17

[38] Barbier C, Paulus J, ApplebyMandHumphrey J AC 2007Design, fabrication, and testing of a bioinsired hybrid hair-likefluidmotion sensor arrayASMEProc. 8th Int.MechanicalEngineering Congress and Exposition pp 1319–24

[39] EberhardtWC, Shakhsheer YA, CalhounBH, Paulus J R andApplebyM2011A bio-inspired artificial whisker forfluidmotion sensingwith increased sensitivity and reliability Proc.IEEE Sensors pp 982–5

[40] HankeW, BrückerC andBleckmannH2000The ageing of thelow-frequencywater disturbances caused by swimminggoldfish and its possible relevance to prey detection J. Exp. Biol.203 1193–200

[41] HankeWandBleckmannH2004The hydrodynamic trails ofLepomis gibbosus (Centrarchidae), Colomesus psittacus(Tetraodontidae) andThysochromis ansorgii (Cichlidae)investigatedwith scanning particle image velocimetry J. Exp.Biol. 207 1585–96

[42] Niesterok B andHankeW2013Hydrodynamic patterns fromfast-starts in teleostfish and their possible relevance toPredator–prey interactions J. Comparative Physiol.A 199139–49

[43] BeemH,HildnerMandTriantafyllouM2012Characterization of a harbor seal whisker-inspired flow sensorOcean 2012MTS/IEEEHarnessing PowerOcean p 014012

15

Bioinspir. Biomim. 11 (2016) 056011 WCEberhardt et al