Characterization of acoustic detection efciency using a gliding robotic fsh as a mobile

receiver platform

Abstract Background: Autonomous underwater vehicles (AUVs) and animal telemetry have
become important tools for understanding the relationships between aquatic organisms and
their environment, but more information is needed to guide the development and use of
AUVs as efective animal tracking platforms. A forward-facing acoustic telemetry receiver
(VR2Tx 69 kHz; VEMCO, Bedford, Nova Scotia) attached to a novel AUV (gliding robotic fsh)
was tested in a freshwater lake to (1) compare its detection efciency (i.e., the probability of
detecting an acoustic signal emitted by a tag) of acoustic tags (VEMCO model V8-4H 69 kHz)
to stationary receivers and (2) determine if detection efciency was related to distance
between tag and receiver, direction of movement (toward or away from transmitter), depth,
or pitch. Results: Detection efciency for mobile (robot-mounted) and stationary receivers
were similar at ranges less than 300 m, on average across all tests, but detection efciency for
the mobile receiver decreased faster than for stationary receivers at distances greater than
300 m. Detection efciency was higher when the robot was moving toward the transmitter
than when moving away from the transmitter. Detection efciency decreased with depth
(surface to 4 m) when the robot was moving away from the transmitter, but depth had no
signifcant efect on detection efciency when the robot was moving toward the transmitter.
Detection efciency was higher when the robot was descending (pitched downward) than
ascending (pitched upward) when moving toward the transmitter, but pitch had no
signifcant efect when moving away from the transmitter. Conclusion: Results suggested that
much of the observed variation in detection efciency is related to shielding of the acoustic
signal by the robot body depending on the positions and orientation of the hydrophone
relative to the transmitter. Results are expected to inform hardware, software, and
operational changes to gliding robotic fsh that will improve detection efciency. Regardless,
data on the size and shape of detection efciency curves for gliding robotic fsh will be useful
for planning future missions and should be relevant to other AUVs for telemetry. With
refnements, gliding robotic fsh could be a useful platform for active tracking of acoustic tags
in certain environments. Keywords: Autonomous underwater vehicles, Underwater glider,
Gliding robotic fsh, Acoustic telemetry, Detection efciency © The Author(s) 2020. This article
is licensed under a Creative Commons Attribution 4.0 International License, which permits
use, sharing, adaptation, distribution and reproduction in any medium or format, as long as
you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons licence, and indicate if changes were made. The images or other third
party material in this article are included in the article’s Creative Commons licence, unless
indicated otherwise in a credit line to the material. If material is not included in the article’s
Creative Commons licence and your intended use is not permitted by statutory regulation or
exceeds the permitted use, you will need to obtain permission directly from the copyright
holder. To view a copy of this licence, visit [Link] [Link]/licenses/by/4.0/.
The Creative Commons Public Domain Dedication waiver
([Link] zero/1.0/) applies to the data made available in
this article, unless otherwise stated in a credit line to the data. Background Recent
improvements in acoustic telemetry technologies have advanced studies of spatiotemporal
ecology and behavior of aquatic organisms [1, 2]. In acoustic telemetry, animals are tagged
with acoustic transmitters Open Access Animal Biotelemetry *Correspondence:
ennasros@[Link] 1 Smart Microsystems Lab, Department of Electrical and Computer
Engineering, Michigan State University, East Lansing, MI 48824, USA Full list of author
information is available at the end of the article Ennasr et al. Anim Biotelemetry (2020) 8:32
Page 2 of 13 and their movements typically are tracked by stationary networks of
hydrophones or acoustic receivers that can identify acoustic-tagged animals in their vicinity
[3–6]. Acoustic telemetry enables more frequent observation (or sampling) of individuals
than what can be obtained by direct observation or sampling with traditional survey
methods (i.e., trawls, gill nets). Observations from such networks have already been used to
improve control and assessment of invasive species, gain new insights into spawning
behavior and habitat requirements of fsh, and describe movements of high-valued stocks [1,
6]. In the future, telemetry promises to address many critical uncertainties in fshery research
and aquatic ecosystems. However, logistical and economic constraints may preclude use of
stationary receiver networks to fll gaps in understanding of animal movements, especially in
large systems and extreme environments. Autonomous surface vehicles (ASVs) and
autonomous underwater vehicles (AUVs) are becoming more common telemetry assets as
receivers can be attached as a payload, or they can be integrated for real-time detection
while simultaneously measuring physical and biological properties of the aquatic
environment [7, 8]. For example, robotic sailboats were used to carry acoustic receivers to
quantify the spatial distribution of fshes [9], buoyancydriven gliders with integrated or
externally mounted receivers have been used to study sturgeon and shark habitats [10–12],
stereo-hydrophone acoustic receiver systems have been integrated into propelled AUVs to
track and follow leopard sharks [13], and receivers and hydrophones have also been
externally mounted and integrated into a propeller-driven AUV [14] and a wave glider for
real-time detections of tagged marine life [15]. Each of these vehicles has its unique
advantages and disadvantages. Propelled AUVs (e.g., REMUS-100 [14], Woods Hole
Oceanographic Institute, Woods Hole, MA, USA, and Iver2 [16], L3Harris OceanServer, Fall
River, MA, USA) can overcome large currents and surface waves, and are capable of traveling
at higher speeds than other unmanned vehicles. Te main drawback for these vehicles is the
power requirement needed for the propellers, limiting deployment duration in the feld.
Another popular class of AUVs are wave gliders, where a surface foat that uses wave energy
to move forward is attached to an underwater sub via a tether [8]. Te energy-efcient nature
of wave gliders, such as the Wave Glider (Liquid Robotics, Sunnyvale, CA, USA) used in [15],
allows them to be deployed for months at a time. Moreover, accurate positioning of the
surface foat is readily available through GPS, along with reliable communication for the
duration of the mission. In addition to their large size, a limitation of wave gliders is that the
environmental sensors deployed on these vehicles need to be tethered to the surface foat,
limiting the depth monitored by such sensors. Underwater gliders are another class of
energyefcient AUVs that travel by changing their buoyancy and center of gravity. Underwater
gliders, such as the Slocum glider (Teledyne Webb Research, North Falmouth, MA, USA) [17],
have been widely used in a variety of underwater applications [7, 10]. Like wave gliders, the
energy-efcient nature of these robots makes them ideal for long-duration missions.
However, unlike wave gliders, no tether is involved with these robots, allowing for sampling
of water environment at diferent depths up to the depth rating of the vehicle. Te biggest
challenges with these robots involve underwater localization and communication because
radio-frequency signals do not penetrate water efectively. Tus, vehicle position in between
surfacing events relies on dead reckoning, which is susceptible to large errors without the
aid of additional sensors such as the Doppler velocity log [18]. As the demand for AUVs to
study ecology of fsh and other aquatic animals increases, a need exists to understand how
design elements, operational characteristics, and environmental conditions infuence
detection efciency (defned here as the probability of detecting an acoustic signal emitted by
a tag) of telemetry-equipped AUVs. In this paper, we describe detection efciency of an
acoustic telemetry receiver mounted on a gliding robotic fsh, a novel type of underwater
AUV [19–21], during a series of feld trials in a freshwater lake conducted principally to
evaluate hardware and software changes during development. Like underwater gliders,
gliding robotic fsh (dubbed GRACE, for Gliding-Robot-ACE) achieve locomotion primarily
through buoyancy-driven gliding or spiraling. Tey are also equipped with an active tail fn that
can provide extra propulsion (e.g., “swimming” against current), act as a rudder to improve
steering during glide/spiral, and improve maneuverability via asymmetrical fapping (e.g.,
tight turns on the surface or during gliding). Consequently, gliding robotic fsh combine the
energy-efcient nature of underwater gliders with the high-maneuverability of robotic fsh (a
type of bio-inspired surface robots) and hence hold great potential in long-duration
monitoring of a broad spectrum of aquatic environments. During feld tests, we
opportunistically collected data on detection efciency using stationary transmitters and
receivers. Objectives were to (1) compare the detection efciency of the AUV-mounted
receiver to that of stationary receivers and (2) determine if the detection efciency was
related to distance between receiver and tag, direction of travel (i.e., toward vs. away from a
remote transmitter), robot depth, and pitch during gliding. Results were expected to inform
further development of gliding robotic fsh and improve the design of AUV-based telemetry
performance assessments. Ennasr et al. Anim Biotelemetry (2020) 8:32 Page 3 of 13
Methods Gliding robotic fsh An experimental prototype of GRACE, previously used to sample
crude oil and harmful algae [20, 22], was modifed to serve as a mobile platform for a
receiver for detecting acoustic tags. Te new robot design featured a carbon fber shell with
removable front section, and aluminum wings and tail. Sensor payloads consisted of a
dissolved oxygen and temperature sensor (In-Situ RDO Pro-X, Fort Collins, CO USA), an
underwater quantum sensor (LICOR LI-192, Lincoln, NE USA) for measuring
Photosynthetically Active Radiation, a Chlorophyll sensor (Turner Designs Cyclops-7F C, San
Jose, CA USA) and a freshwater Blue Green Algae sensor (Turner Designs Cyclops-7F P). Te
sensors were bundled and connected to the robot electronics through a waterproof
connector (MarcArney SubConn MCIL16-F/M, Esbjerg, Denmark). A GPS sensor (Garmin GPS
18×LVC, Olathe, KS USA) was used for robot localization when the robot was on the surface,
while a pressure sensor (Honeywell 40PC100G2A, Charlotte, NC USA) was used to measure
the robot depth underwater. Two linear actuators were used for buoyancy and pitch control;
in particular, one pushed water in and out of a tank for buoyancy control, and the other
moved a mass backward and forward to pitch the robot up or down. Tese actuators were
controlled, at a low level, by an embedded microcontroller (MCU) that operated at
7.37 MHz. Te same microcontroller handled the reading of sensor data. An XBee wireless
serial interface was used for communication with a laptop on a boat located within the
communication range (typically several hundred meters). Tis channel was used to send
commands to the robot or query data from the robot when it was on the surface (see
summary of robot components and specifcations in Table 1). Hardware and software design
of the robot underwent some evolution during the study period. In the 2016 trial, all robot
operations (i.e., control, communication, and data processing) were implemented on the
embedded MCU, and the robot sent its GPS position along with a UTC time-stamp over the
XBee channel every 5 s whenever it was on the surface. When the robot performed a dive, it
temporarily stored depth data that were time-stamped and sent through the XBee channel
upon surfacing. In 2017, the robot also temporarily stored orientation (yaw, pitch, and roll
angles) information whenever it was underwater. In 2018, we incorporated a Raspberry Pi
Zero W that performed high-level tasks such as communication and data storage. Tis
modifcation allowed us to complement the original broadcasted messages by storing all
available data onto an on-board SD card every 5 s. Tese data consisted of GPS coordinates
and UTC time, environmental sensor readings, orientation of the robot (yaw, pitch, and roll
angles), positions of each actuator, and battery level. Field tests A self-contained acoustic
receiver (VEMCO model VR2Tx; 69 kHz; Bedford, Nova Scotia, Canada) was attached to the
robot, and feld trials were conducted in Table 1 GRACE components and specifcations a Due
to fabrication and balancing imperfections, the magnitude of the pitch angle when pitching
down is smaller than that for pitching up Component Description Hull dimensions
103×20×30 cm (L x W x H) Tail-to-nose length 140 cm Wingspan 60 cm Weight 20 kg Robot
hull material Carbon fber Battery capacity 555 W H Li-ion polymer (approx. 5 days of
continuous actuation with 1400 dive cycles @ 5-min intervals) Processors Raspberry Pi Zero
W, Microchip dsPIC30F6014A MCU Communication XBee Positioning GPS and pressure
sensor Orientation VN100S IMU and attitude and heading reference system Buoyancy tank
volume 190 mL Pitch controla + 40° (pitching up), − 25° (pitching down) Average swimming
speed 25 cm/s Average glide speed 13 cm/s (as low as 5 cm/s against current and a high of
35 cm/s with current) Environmental sensors Dissolved oxygen and temperature,
photosynthetically active radiation, freshwater blue green algae, chlorophyll Ennasr et al.
Anim Biotelemetry (2020) 8:32 Page 4 of 13 Higgins Lake, Michigan, USA, during 2016–2018.
Te receiver was mounted at the bottom of the robot, facing forward (Fig. 1). Detection
efciency was investigated as a function of distance between a test transmitter, or ‘tag’, and
the receiver, robot direction (toward or away from tag), robot depth, and robot pitch. Each
receiver also measured temperature (internal), tilt (degrees from vertical), and
environmental acoustic noise (69 kHz) every 10 min, but those variables were not used in
analyses (see Additional fle 1) due to insufcient sampling frequency to describe within-dive
variation (temperature, noise, and tilt); inability to detect rapid changes during dives
(temperature); and high correlations (redundancy) with robot-integrated sensor data (tilt).
Tests (‘runs’) were conducted on 11 November 2016, 3 November 2017, and 14–15 June
2018 in Higgins Lake, Michigan, where water depths ranged between 10–20 m (Fig. 2;
Table 2). Tese locations were selected to facilitate feld testing of navigation and actuation
systems on a given day and not necessarily with detection efciency tests in mind. While
these tests were primarily focused on assessing the robot’s capabilities and identifying
necessary improvements, we used these opportunities to determine the parameters that
afect detection efciency for underwater gliders. During each trial, an acoustic transmitter
(VEMCO model V8-4H 69 kHz; source power level 147 dB re 1 Paµ @ 1 m) was deployed on
a stationary mooring (depth ranged 1.5–7.0 m among runs) and emitted a uniquely encoded
signal every 24 s. Transmitted signals were encoded using VEMCO’s 69-kHz pulse–position–
modulation scheme (VEMCO code space A69-1601). Once decoded, each signal’s unique ID
was recorded on the receiver along with time of detection. Tis tag type was used because it
provided a detection range amenable to the feld evaluation of the robot in the study system.
Specifc tag programming (24 s transmit interval) was selected to ensure that testing would
provide sufcient sample size for regression models and to capture relatively fne-scale
changes in variables of interest (e.g., depth, pitch). For every test, a target GPS waypoint was
sent to the robot from a laptop on a boat over the wireless communication channel, and the
robot was tasked with navigating to that location through a series of dives. A depth of 4.5 m
was used as the maximum diving depth for the robot as the depth rating had not been
established for the robot hull. Each dive was completed in approximately 3 min on average.
Robot heading was controlled by changes in tail position such that the tail acted as a rudder
during gliding to maintain course toward the target location. Between dives, the robot
remained on the surface for 20 s to ensure that a GPS lock was established, and a new GPS
position was obtained to calculate the desired heading angle. Mobile detection efciency
tests In 2016, the efect of distance between receiver and tag on detection efciency was
evaluated and compared between mobile and stationary receivers (objective #1) by
navigating the robot–receiver system along a line of fve stationary receivers (VEMCO model
VR2Tx 69 kHz). Stationary receivers were suspended 1.5 m below the surface (in the
hydrophone-up position) and moored via rope to weights on the lake bottom. Water depth
at stationary receiver sites ranged between 7.3 and 10.0 m. Stationary receivers were
arranged in a line with the transmitter such that the receivers were spaced 200, 400, 600,
800, and 1000 m from the transmitter, which was suspended 1.5 m below the surface in
11.0 m water depth. During the frst trial (run 16–1), the robot began navigating 996 m from
the transmitter, in a direction roughly parallel to the line of stationary receivers and toward
the transmitter. As the trial progressed, winds (approximately 4.9–5.8 m/s, based on a
regional model) moved the robot of-course while it attempted to obtain a GPS fx. Magnetic
disturbance, resulting from the internal electronics and actuators, which afected the stability
of the measured heading angle using the on-board Inertial Measurement Unit (IMU), also
contributed to the robot veering of-course. Consequently, the robot was removed from
water about 563 m from the transmitter, and then returned to the water for a second trial
(run 16–2). During the second trial, the robot started 411 m away from the tag, and was
removed from the water 288 m from the transmitter. Prior to the next feld test, robot
navigation problems observed in 2016 were addressed. First, the IMU was calibrated to
reject larger magnetic disturbances, which improved the stability of the measured heading
angle. Additionally, to combat high winds, the robot was tasked Fig. 1 Gliding robotic fsh.
The modifed version of GRACE with bottom-mounted telemetry receiver Ennasr et al. Anim
Biotelemetry (2020) 8:32 Page 5 of 13 Fig. 2 Lake tests. a Map of Higgins Lake, Michigan,
showing regions (red rectangles) where gliding robotic fsh GRACE was tested 11 Nov 2016
(b), 03 Nov 2017 (c), and 14–15 June 2018 (d–f) with stationary transmitter locations (green
circles), stationary receiver locations (yellow circles), and GPS locations (pink/red circles)
recorded by the gliding robotic fsh during each run. GPS points are shaded along a gradient
from start (pink) to end (red) of each run Ennasr et al. Anim Biotelemetry (2020) 8:32 Page 6
of 13 with “swimming” (by continuously fapping its tail) on the surface while waiting for a
GPS fx. After these corrections, a second feld trial was carried out in 2017 to further
investigate the efect of distance on the detection efciency for a mobile receiver (objective
#1). During this trial (run 17–1), the gliding robotic fsh navigated from 389 m to less than
1 m from the tag, in a direction roughly northwest. Te tag was suspended 3.0 m below the
surface in 19.8 m water depth. Stationary receivers were not deployed during the test in
2017 due to logistical constraints. While the windspeeds during this trial were close to those
in 2016 (approximately 4.2 m/s, based on regional model), the provisions we took with
surface swimming allowed the robot to complete its task. During navigation, the tag
transmitted 323 times and the robot repeatedly ascended and descended between the
surface and 4.5 m depth. A third set of feld trials were conducted in 2018 to investigate the
efect of the robot’s direction of travel, depth, and pitch on detection efciency (Objective #2).
In addition, four stationary receivers (R1–R4; Fig. 2) were deployed in roughly a square-
shaped pattern to explore the utility of obtaining fne-scale robot tracks using timediference-
of-arrival-based positioning (not evaluated in this paper). Stationary receivers were
suspended vertically (hydrophone up) 4.7–19.8 m below the surface in 9.3–24.0 m water
depth. A ffth stationary receiver was collocated with the stationary tag; both were
suspended vertically (hydrophone up) 4.6 m below the water surface in 9.1 m water depth.
Te robot was tasked with navigating toward the tag during two runs (runs 18–1, 18–3) and
away from the tag during one run (run 18–2). Like the trials in 2017, the robot was tasked
with swimming while awaiting a GPS fx on the surface to counteract surface waves that
could push the robot away from its desired course. Water current and wind data (e.g.,
velocity and direction) were not measured at the site during testing, but regional wind data
were obtained from NCEP’s Global Forecast System [23] using the R package RWind.
Modelbased estimates of mean wind speed and direction during 3-h intervals (e.g., 00:00–
02:59 UTC, 03:00–05:59 UTC) at 10 m above land surface were obtained for two locations
about 24 km west (85.0° W, 44.5° N) and 17 km east (84.5° W, 44.5° N) of the study site.
Mean speed and direction between the two locations were used to represent conditions at
the study site during the 3-h interval that contained each run. Data analysis Transmitter
detection efciency curves were estimated for stationary and mobile receivers using
generalized additive models (GAMs) with binomial error structure. GAMs were used because
they relax distributional assumptions and allow greater fexibility in modeling non-linear
responses than generalized linear models [24]. Times of missed detections (i.e., transmitted
but not detected) were estimated by sequencing every 24 s through each detection data set
to identify time stamps that were not in the detection fle for the receiver. A binary indicator
variable was used to represent detection (1) or non-detection (0). Variables describing robot
location and orientation (i.e., latitude, longitude, depth, pitch) and water temperature (see
Additional fle 1) were Table 2 Summary of mobile detection efciency trials with GRACE
at Higgins Lake in 2016, 2017, and 2018 (Tag location refers to sites identifed in Fig. 2) Mean
heading refects the bearing from the frst GPS measurement toward the last GPS
measurement in each run. Wind speed and direction (heading; note that this is the direction
the wind vectors are following, not “blowing from” as is customary) are coarse-scale regional
estimates based on the NCEP’s Global Forecast System, as described in methods Run ID Date
time (UTC-4) Tag location Distance from tag (m) Mean heading (deg.) Wind speed (m/s)
Wind heading (deg.) Start End Start End 16–1 11/11/16 10:47 11/11/16 12:38 T1 996 563
219° (SSW) 5.8 187° (S) 16–2 11/11/16 12:52 11/11/16 13:30 T1 411 288 214° (SSW) 4.9
182° (S) 17–1 11/3/17 12:45 11/3/17 14:54 T1 389 8 303° (WNW) 4.2 148° (SE) 18–1
6/14/18 18:06 6/14/18 19:00 T6 360 11 191 (S) 4.1 163° (SSE) 18–2 6/14/18 19:20 6/14/18
21:13 T6 44 361 3° (N) 1.7 157° (SE) 18–3 6/15/18 10:06 6/15/18 11:09 T7 408 10 8° (N) 1.9
324° (NW) Ennasr et al. Anim Biotelemetry (2020) 8:32 Page 7 of 13 estimated at time of
each detection or non-detection using time-based linear interpolation over measurements
recorded by the robot (Figs. 3, 4). Detection efciency curves were frst described separately
for each run and for stationary and mobile receivers during each run, by ftting a GAM to data
from each run separately, except that runs 16–1 and 16–2 were combined. Tese GAMs
simply estimated the probability of detection (binomial response) as a function of distance
between tag and receiver (Fig. 5). Specifcally, each model estimated the log-odds (logit) of
the probability of detecting each transmitted tag signal as a function of the predictors:
where b0 is the intercept; srt_dist:dir is a smoothed function of distance for each direction
and run, and ε is error assumed from an independent draw from a normal logit(p) = b0 +
srt_dist + ε, distribution with mean 0 and variance σ2 . All models were ft to data using the
‘gam’ function in the mgcv library (v. 1.8.31) [25] in R (v. 3.6.2) [26]. Te basis for smoothing
functions was a cubic regression spline (bs=“cr” in mgcv) and 10 knots. A GAM was ft to the
data from runs 17–1, 18–1, 18–2, and 18–3 to estimate the detection efciency as a function
of distance between receiver and transmitter, direction of robot relative to transmitter
(toward, away), robot depth (meters below water surface), and robot pitch (degrees from
horizontal). Runs 16–1 and 16–2 were not included in the model because (1) little overlap
occurred in distances covered between those runs and others (Fig. 5); (2) they lacked pitch
data; and (3) depth data were incomplete during those runs. Specifcally, the Fig. 3 Time-
based dive profles. Depth (a), pitch (b), and temperature (c) profles of gliding robotic fsh
GRACE during a 10-min segment of run 18–1 in Higgins Lake. Depths, pitch, and temperature
were recorded by the on-board sensors (open symbols) and were estimated at the time of
tag signal transmissions (red symbols). Horizontal broken line in a shows depth of the
stationary transmitter at site T6. The black dot in c shows a single temperature
measurement by the stationary transmitter at site T6 and the broken line connects that
observation to the previous and next temperature measurements at that site Fig. 4 Distance-
based dive profles. Depth profles of gliding robotic fsh GRACE during a selected 100- to 300-
m segment of each run (a–d) in Higgins Lake, with locations of test tag transmissions that
were detected (closed symbols) or not detected (open symbols) by the on-board receiver.
The robot navigated towards the tag in 17–1, 18–1, and 18–3, and navigated away from the
tag in 18–2 Ennasr et al. Anim Biotelemetry (2020) 8:32 Page 8 of 13 model estimated the
log-odds (logit) of the probability of detecting each transmitted tag signal as a function of
the predictors where b0 is the intercept; dir is the fxed efect of direction; srt_dist:dir and
srt_dist:run are smoothed functions of distance for each direction and run; sdepth:dir and
spitch:dir logit(p) = b0 + dir + srt_dist:dir + srt_dist:run + sdepth:dir + spitch:dir + ε, are
smoothed functions of depth and pitch for each direction; and ε is error assumed from an
independent draw from a normal distribution with mean 0 and variance σ2 . Direction-
specifc smoothers were included to estimate the partial efects of each level of each
continuous predictor on the response (detection efciency). Te run-specifc smoother for
distance was included to account for run-specifc variability in detection efciency not
explained by other predictors. To eliminate confounding between run and direction, efects
of direction were limited to contrast between runs 18–1 and 18–2, which occurred on the
same day, and was accomplished by treating runs 18–1 and 18–2 as a single run (“18–1 &
18–2”) in the model. Tus, we assumed that diferences between runs 18–1 and 18–2 were
attributed to change in direction relative to tag and no other variables or conditions. Tis
assumption was supported by similar detection efciency curves as functions of distance for
stationary receivers during those runs (Fig. 5c, d). Run-specifc variation could have been
treated as a random efect in a mixed-efects model, but we anticipated no beneft of a mixed
model with only four unique runs. An added advantage of a run-specifc smoother is that it
allows exploration of the unique shape of each curve that might inform future hypotheses
about variables infuencing the shape of the curve during each run. While it is possible that
some across-run variation could be explained by wind, ambient noise, or water
temperatures, our observational data set did not have replicate runs (wind, water
temperature) or sufcient within-predictor contrast (ambient noise) and thus we did not
include those variables in the model. Rather, we used those observations to generate
hypotheses from our descriptive analyses. Te basis for all smoothing functions was a cubic
regression spline with shrinkage (bs=“cs” in mgcv) and 10 knots. “Shrinkage smoothers”
allow identifcation of non-signifcant smoothers that essentially carry no weight in the model
by reducing the efective degrees of freedom to a value as small as zero [24]. Tus, shrinkage
smoothers are an alternative to stepwise model selection procedures. All smoothers were
estimated using restricted maximum likelihood. Prior to model ftting, data were checked for
evidence of collinearity (correlations among predictors). Model ft was evaluated by checking
for concurvity (non-independence) among smoothers and evidence of non-normality among
residuals. Signifcance of each smoothing term was determined based on estimated degrees
of freedom and approximate p-value for the null hypothesis that the smoothing term was
zero. Te signifcance level for all tests was 0.05. Fig. 5 Detection efciency as a function of
distance for mobile and stationary receivers. Estimated detection efciencies for acoustic
receivers afxed to stationary moorings (red symbols are ftted values; red shaded regions are
GAM-based 95% confdence regions) and gliding robotic fsh GRACE (black lines are ftted
values; gray-shaded regions are GAM-based 95% confdence regions) in Higgins Lake,
Michigan, during feld tests in 2016 (a), 2017 (b), and 2018 (c–e). Vertical gray bars show
distances between robot and the transmitter when each coded signal was detected (1) or
not detected (0) Ennasr et al. Anim Biotelemetry (2020) 8:32 Page 9 of 13 Partial efects plots
were used to assess the infuence of each smoother on the log-odds of detection efciency at
each level of the predictor. Plots of ftted values were used to assess the infuence of each
predictor on detection efciency during each run on the probability scale. Results Estimated
detection efciency of the robot-mounted receiver was lower than concurrently operated
stationary receivers during all three runs where stationary receivers covered the full range of
mobile runs (Fig. 5 a, c, d). Scale and shape of detection efciency curves, when plotted
against distance between transmitter and receiver, varied among runs for both stationary
and mobile receivers. During 2016, when detection efciencies were higher at longer
distances for both stationary and mobile receivers than during any other run, detection
efciencies for the mobile receiver were only slightly lower than the stationary receivers up to
600 m from the transmitter, but substantially lower at 800 m and 1000 m. Te shape of the
curve for the stationary receivers was unexpected (e.g., higher efciency at 800 m than 600
m), and suggested that these stationary receivers were afected by processes that were not
quantifed. During 18–1, estimated detection efciency was similar between stationary and
mobile receivers for distances up to 200 m but difered at larger distances due to mobile
detection range declining much faster than stationary detection range. Te greatest diference
between stationary and mobile detection efciency was observed during 18–2 when the
robot was moving away from the transmitter. During that run, mobile detection efciency for
any distance was markedly lower, even at 50 m–the closest distance between receiver and
transmitter during that run. Te GAM model explained 53.2% of the null deviance. Detection
efciency difered by direction of travel (Table 3) with more than a fvefold increase in the
logodds of detection efciency when moving toward the transmitter than away from the
transmitter. Signifcance of distance-based smoothers suggested that variation in the shape
of range curves was attributed to direction of movement and other run-specifc variables not
included in the model. Pitch was signifcant when the robot was moving toward the
transmitter (with higher detection efciency for lower pitch value, regardless of pitch
direction), but not when the robot was moving away from the transmitter (Table 3a, Fig. 6c).
Depth was signifcant when the robot was moving away from the transmitter (with better
detection efciency at shallower depths), but not when the robot was moving toward the
transmitter (Table 3b, Fig. 6d). Over the ranges of depth and pitch observed, the efect of
depth on detection efciency moving away from the transmitter was greater than the efect of
pitch moving toward the transmitter (Fig. 7). Overall, the highest detection efciency was
observed when the robot was navigating towards the tag with near-zero pitch, while the
lowest detection efciency was observed when the robot was navigating away from the tag
and was at large depths. However, results from data checking (collinearity and concurvity;
Additional fle 2), variable selection (shrinkage smoother degrees of freedom and Table 3
Summary of parametric coefcients (linear terms) and smoothing terms from GAM used to
determine if detection efciency was related to distance from transmitter (tag distance,
in meters), direction of robot travel relative to transmitter (toward or away), robot depth
(depth, in meters from water surface), or robot pitch (pitch, in degrees from horizontal)
Included for each parametric coefcient is the estimate, standard error (SE), test statistic (Z),
and p-value for the null hypothesis that the corresponding parameter is zero. Included for
each smoothing term is the estimated degrees of freedom (EDF), test statistic (χ2 ), and
approximate p-value for the null hypothesis that the smoothing term is zero. Italicized p-
values are signifcant at signifcance level of 0.05 Linear terms Estimate SE Z p-value
(Intercept) −4.282 0.581 −7.363 1.79E−13 dir=toward 5.322 0.594 8.965 3.11E-19 Smoothing
terms EDF χ2 p-value s(rt_distance): dir=away 6.48E−04 4.85E−04 0.265 s(rt_distance):
dir=toward 1.642 4.345 5.84E−04 s(rt_distance): run=17–1 2.152 6.651 3.07E−04
s(rt_distance): run=18–1, 18–2 2.466 44.738 4.25E−13 s(rt_distance): run=18–3 1.118 2.699
5.27E−03 s(pitch): dir=away 0.570 1.087 0.152 s(pitch): dir=toward 2.416 33.366 3.15E−09
s(depth): dir=away 1.212 9.468 8.55E−04 s(depth): dir=toward 0.209 0.255 0.247
Ennasr et al. Anim Biotelemetry (2020) 8:32 Page 10 of 13 p-values), and model selection
(AIC) suggested that the relationship among depth and pitch was complex. Terefore, we
caution against any strong conclusions about individual efects of pitch and depth from these
results. Discussion Understanding detection efciency of telemetry receivers is essential for
designing and conducting a successful animal tracking project [4]. Although an animal
location can be derived from detection time diferences among multiple hydrophones [27],
the location of a receiver at the time of detection is often used to represent the general
location of a tagged animal at time of detection for presence–absence data. Data describing
the efect of distance between transmitter and receiver on detection efciency are useful for
interpreting spatial ambiguity or uncertainty around detection locations, but is not
commonly collected [28]. Moreover, receivers are often assumed omnidirectional, though
several processes contributing to directional or non-uniform detection areas have been
described [29]. Decreased detection efciency when the robot was moving away from the
transmitter was not unexpected due to potential shielding of the signal based on position
(bottom-mounted) and orientation (forward-facing) of the receiver on the robot and the
importance of line-of-site to acoustic detection. Te magnitude of the efect, however, has
important implications for the ability of an AUV in this confguration to detect tagged fsh and
for inferences of fsh locations based on detection data. Our results suggested that both
depth and pitch can infuence detection efciency, even under a narrow Fig. 6 Partial efects of
robot depth and pitch on detection efciency. Estimated partial efects of robot depth and
pitch on the log-odds of detection efciency when the robot was moving away from (a, b) or
toward (c, d) a stationary transmitter in Higgins Lake, Michigan. Black lines are ftted values
from the model. Shaded regions are 95% confdence regions Ennasr et al. Anim Biotelemetry
(2020) 8:32 Page 11 of 13 scope of environmental conditions. Depth and pitch are critical
control parameters of AUVs that are carefully programmed to ensure mission success in the
face of environmental and energetic constraints. When the objective is to detect acoustic-
tagged fsh, operational adjustments may be needed to achieve favorable balance among
detection efciency and other operational processes (e.g., navigation, communication).
Vertical gradients of environmental variables known to afect acoustic signals in water (e.g.,
salinity, temperature, suspended particulates, entrained air) exist in most aquatic systems,
but we were not aware of any such gradients during this study. Although lack of
environmental heterogeneity may limit application of these results to other systems, it may
have improved our ability to identify relatively small efects of depth and pitch on detection
efciency by minimizing background variability during our study. Moreover, observed
variability that is not environmentally driven may be related to variables that can be
controlled or modifed. We hypothesize that decreasing detection efciency from the surface
to 4 m depth when moving away from the transmitter was driven by shielding of the signal
by the body of the robot at depth. As described above, this explanation seems reasonable
based on position and orientation of the receiver on the robot. If true, then the depth
variable used in our analysis may be a proxy for the diference in depth between the
hydrophone and transmitter. Although the range in depth diference between receiver and
transmitter in our tests (receiver ranged 0 to 4 m above the tag) may be representative of
some shallow environments, potential clearly exists for much greater vertical separation
between transmitter and receiver in many aquatic systems. For an AUV with bottom-
mounted receiver, shielding efects are expected to be largest when the vehicle is deeper
than a tagged fsh and smallest when the vehicle is shallower than the tagged fsh. Tus, efects
of shielding may be minimized by operational parameters (e.g., depth range) based on
knowledge of the ecology of the target organisms (i.e., remaining near the surface for
pelagic fsh) or structural changes to the vehicle (i.e., positioning the receiver on top of the
AUV for surface-oriented fsh). Although the mobile receiver on the gliding robotic fsh did not
perform as well as stationary receivers throughout the entire range tested, detection
efciency over shorter distances (300 m or less) are likely still adequate for some active
tracking needs. In practice, the distance at which a transmitter is reliably detected can vary
considerably due to hardware and software diferences and environmental conditions. Future
work should seek to determine if diferences between the robot-mounted receiver and
stationary receivers are caused by characteristics of the robot (e.g., electrical or mechanical
noise) or interaction of the robot with the environment (e.g., turbulence of fow over the
hydrophone). For example, future work might seek to determine if low detection efciency
while ascending (pitched up) was associated with noise from the tail motor or increased
turbulence on the receiver during tail actuation. Regardless of future improvements,
however, knowledge of the robotmounted detection efciencies will be useful for planning
future missions, including active tracking with a network of AUVs. While direct comparison
of these results to similar work using the Slocum glider in [7] and Wave Glider in [8] in
saltwater is difcult, the detection efciency curves obtained here can further improve our
understanding of such systems. Our obtained detection efciency results when the
hydrophone was facing towards the tag over short distances of 100 m (>80%) are close to
those reported in [8] (50–90%), and are higher than those reported in [7] (40–0%). Tese
results highlighted the signifcance of alignment between transmitter and receiver for a
forward-facing receiver, and can help explain the Fig. 7 Detection efciency as a function of
distance for diferent depth and pitch values. Estimated detection efciencies for acoustic
receivers afxed to gliding robotic fsh GRACE in Higgins Lake, Michigan, during feld tests in
which evidence (GAM model results) suggested that detection efciency was infuenced most
by depth when the robot was moving away from the transmitter (c) and pitch when the
robot was moving toward the transmitter (a, b, d). Lines are ftted values from the model
with depth=2 m (a, b, d) and pitch=0° from horizontal (c) Ennasr et al. Anim Biotelemetry
(2020) 8:32 Page 12 of 13 lower detection efciency observed in [7], as it could be due to
occasional shielding as well as other environmental diferences. Tese result also supported
the observed improvement of detection efciency for the vertically mounted receivers
compared to the forward-facing receiver in [7], as forward-facing receivers are more
susceptible to shielding of acoustic signals. Although our results add to knowledge of AUV
performance as receiver platforms, much variation in detection efciency remained
unexplained in our analysis. Unfortunately, we were not able to account for the infuence of
the environmental variables (e.g., wind, water temperature, ambient noise) on the results
because availability of these data were limited. Our observational data set did not have
sufcient replication (multiple runs) over environmental variables (e.g., wind, ambient noise,
thermal stratifcation) to explore this aspect. Future work should seek to obtain a balanced
study design with replicate runs over a range of environment variables, so that one can
attribute variation in the detection efciencies to those variables. Conclusion We considered
parameters that afect the detection efciency of mobile receivers mounted on AUVs through
a series of trials using a gliding robotic fsh, in which a forward-facing acoustic receiver was
fxed to the robot’s hull. While the detection efciency of the robotmounted receiver was
expected to be lowered when the receiver was pointed away from the tag, the impact of this
efect was signifcant. Results suggested that much of the observed variation may be related
to shielding of the signal by the robot body depending on the positions and orientation of
the hydrophone relative to the transmitter. Tese results inform hardware, software, and
operational changes to gliding robotic fsh that will improve detection efciency. As an
example, tag-tracking controllers should consider the directionality of the receiver, as well as
the relative position of the tag with respect to the receiver. Other options to mitigate the
shielding efect include the use of vertically mounted receivers, pointing upwards or
downwards [7, 12], or using two bidirectionally mounted receivers (forwardand rear-facing).
We hypothesize that such confgurations could still sufer from reduced detection efciency
due to shielding, and knowledge of the ecology of the target organisms should be
considered to guide the positioning of the receivers. Data on the size and shape of detection
efciency curves for gliding robotic fsh will be useful for planning future missions and should
be relevant to other AUVs for telemetry. While the detection efciency of the mobile receiver,
when facing the tag, matched that of stationary receivers for distances of less than 300 m,
these distances are still adequate for several active tracking applications using networks of
these AUVs. Future work should seek to determine if diferences between the robot-
mounted receiver and stationary receivers are due to characteristics of the robot or
interaction of the robot with the environment. Finally, future studies should use a balanced
study design with replicate runs over a range of environment variables and compare the
detection efciency using diferent AUVs. Supplementary information Supplementary
information accompanies this paper at [Link] org/10.1186/s40317-020-00219-7.
Additional fle 1. Summaries of environmental noise, water temperature, and receiver tilt
data collected during feld tests of gliding robotic fsh in Higgins Lake, 2016–2018. Additional
fle 2. Data and model checking methods and results for generalized additive model ft to
detection data collected during feld tests of gliding robotic fsh in Higgins Lake, 2016–2018.
Acknowledgements We would like to thank Chris Wright and Henry Thompson from USGS
for their help during the feld trials in Higgins Lake. This work was supported in part by the
U.S. National Science Foundation (ECCS 1446793, IIS 1848945), the Great Lakes Fishery
Commission (2014_Tan_44058), and the United States Geological Survey Innovation Center
for Earth Sciences. This work also was supported in part by funds from the Great Lakes
Fishery Commission by way of Great Lakes Restoration Initiative appropriations (GL-
00E23010). This paper is Contribution 66 of Great Lakes Acoustic Telemetry Observation
System (GLATOS). Any use of trade, frm, or product names is for descriptive purposes only
and does not imply endorsement by the US Government. Authors’ contributions OE and CH
jointly wrote the initial manuscript. OE, DC, PS, JT and XT developed the robotic platform
and its navigation algorithms. CH analyzed the acoustic detection data. DWH and CK assisted
with project development and feld trials. All authors read and approved the fnal manuscript.
Funding This work was supported in part by the U.S. National Science Foundation (ECCS
1446793, IIS 1848945), the Great Lakes Fishery Commission (2014_ Tan_44058), and the
United States Geological Survey Innovation Center for Earth Sciences. This work also was
supported in part by funds from the Great Lakes Fishery Commission by way of Great Lakes
Restoration Initiative appropriations (GL-00E23010). This paper is Contribution 66 of Great
Lakes Acoustic Telemetry Observation System (GLATOS). Availability of data and materials
The data sets used and analyzed during the current study are available from the
corresponding author on reasonable request. Ethics approval and consent to participate Not
applicable. Consent for publication Not applicable. Competing interests The authors declare
that they have no competing interests. Ennasr et al. Anim Biotelemetry (2020) 8:32 Page 13
of 13 Author details 1 Smart Microsystems Lab, Department of Electrical and Computer
Engineering, Michigan State University, East Lansing, MI 48824, USA. 2 U.S. Geological
Survey, Hammond Bay Biological Station, 11188 Ray Rd, Millersburg, MI 49759, USA. 3 Great
Lakes Science Center, U.S. Geological Survey, 1451 Green Rd, Ann Arbor, MI 48105, USA. 4
Center for Systems Integration and Sustainability, Michigan State University, East Lansing, MI
48824, USA. Received: 6 June 2020 Accepted: 12 October 2020 References 1. Hussey NE,
Kessel ST, Aarestrup K, Cooke SJ, Cowley PD, Fisk AT, Harcourt RG, Holland KN, Iverson SJ,
Kocik JF. Aquatic animal telemetry: a panoramic window into the underwater world. Science.
2015;348(6240):1255642. 2. O’Dor RK, Stokesbury MJ. The Ocean Tracking Network--adding
marine animal movements to the global ocean observing system. In: Tagging and Tracking of
Marine Animals with Electronic Devices, Springer, 2009, pp. 91–100. 3. Welch DW, Boehlert
GW, Ward BR. POST – the Pacifc Ocean salmon tracking project. Oceanol Acta.
2002;25(5):243–53. 4. Heupel MR, Semmens JM, Hobday AJ. Automated acoustic tracking of
aquatic animals: scales, design and deployment of listening station arrays. Mar Freshw Res.
2006;57(1):1–13. 5. Cooke SJ, Iverson SJ, Stokesbury MJ, Hinch SG, Fisk AT, VanderZwaag DL,
Apostle R, Whoriskey F. Ocean Tracking Network Canada: a network approach to addressing
critical issues in fsheries and resource management with implications for ocean governance.
Fisheries. 2011;36(12):583–92. 6. Krueger CC, Holbrook CM, Binder TR, Vandergoot CS, Todd
HA, Hondorp DW, Nate N, Paige K, Riley SC, Fisk AT, Cooke SJ. Acoustic telemetry observation
systems: challenges encountered and overcome in the Laurentian Great Lakes. Can J Fish
Aquat Sci. 2018;75(10):1755–63. 7. Oliver MJ, Breece MW, Haulsee DE, Cimino MA, Kohut J,
Aragon D, Fox DA. Factors afecting detection efciency of mobile telemetry Slocum gliders.
Animal Biotelemetry. 2017;5(1):14. 8. Cimino M, Cassen M, Merrifeld S, Terrill E. Detection
efciency of acoustic biotelemetry sensors on Wave Gliders. Animal Biotelemetry.
2018;6(1):16. 9. Mordy CW, Cokelet ED, De Robertis A, Jenkins R, Kuhn CE, Lawrence-Slavas
N, Berchok CL, Crance JL, Sterling JT, Cross JN, Stabeno PJ. "Advances in ecosystem research:
Saildrone surveys of oceanography, fsh, and marine mammals in the Bering Sea.
Oceanography. 2017;30(2):113–5. 10. Oliver MJ, Breece MW, Fox DA, Haulsee DE, Kohut JT,
Manderson J, Savoy T. Shrinking the haystack: using an AUV in an integrated ocean
observatory to map Atlantic Sturgeon in the coastal ocean. Fisheries. 2013;38(5):210–6. 11.
Breece MW, Fox DA, Dunton KJ, Frisk MG, Jordaan A, Oliver MJ. Dynamic seascapes predict
the marine occurrence of an endangered species: Atlantic Sturgeon Acipenser oxyrinchus
oxyrinchus. Methods Ecol Evol. 2016;7(6):725–33. 12. Haulsee DE, Breece MW, Miller DC,
Wetherbee BM, Fox DA, Oliver MJ. Habitat selection of a coastal shark species estimated
from an autonomous underwater vehicle. Mar Ecol Prog Ser. 2015;528:277–88. 13. Clark
CM, Forney C, Manii E, Shinzaki D, Gage C, Farris M, Lowe CG, Moline M. Tracking and
following a tagged leopard shark with an autonomous underwater vehicle. Journal of Field
Robotics. 2013;30(3):309–22. 14. Grothues TM, Dobarro J, Ladd J, Higgs A, Niezgoda G,
Miller D. Use of a multi-sensored AUV to telemeter tagged Atlantic sturgeon and map their
spawning habitat in the Hudson River, USA. In: Autonomous Underwater Vehicles, 2008.
AUV 2008. IEEE/OES, 2008. 15. Carlon R. Tracking tagged fsh using a wave glider, In: OCEANS
2015 - MTS/IEEE Washington, Washington, DC, 2015 16. D. Shinzaki, C. Gage, S. Tang, M.
Moline, B. Wolfe, C. G. Lowe and C. Clark, "A multi-auv system for cooperative tracking and
following of leopard sharks. In: 2013 IEEE International Conference on Robotics and
Automation, Karlsruhe, Germany, 2013. 17. Webb DC, Simonetti PJ, Jones CP. SLOCUM: An
underwater glider propelled by environmental energy. IEEE J Oceanic Eng. 2001;26(4):447–
52. 18. Woithe HC, David B, Ulrich K. Improving Slocum Glider dead reckoning using a
Doppler Velocity Log. In: OCEANS’11 MTS/IEEE KONA, Waikoloa, 2011. 19. Zhang F, Tan X.
Gliding robotic fsh and its tail-enabled yaw motion stabilization using sliding mode control.
In: ASME 2013 Dynamic Systems and Control Conference, Palo Alto, 2013. 20. Zhang F, Wang
J, Thon J, Thon C, Litchman E, Tan X. Gliding robotic fsh for mobile sampling of aquatic
environments: In: 2014 IEEE 11th International Networking, Sensing and Control (ICNSC),
Miami, 2014. 21. Zhang F, Ennasr O, Tan X. Gliding robotic fsh: an underwater sensing
platform and its spiral-based tracking in 3d space. Marine Technol Soci J. 2017;51(5):71–8.
22. Zhang F, Ennasr O, Litchman E, Tan X. Autonomous sampling of water columns using
gliding robotic fsh: algorithms and harmful-algae-sampling experiments. IEEE Syst J.
2016;10(3):1271–81. 23. "Global Forecast System (GFS)," National Oceanic and Atmospheric
Administration (NOAA). [Link] -data/model-
datasets/global-forcast-system-gfs. 24. Zuur AF, Ieno EN, Walker NJ, Saveliev AA, Smith GM.
Mixed efects models and extensions in ecology with R, Springer Science & Business Media,
2009. 25. Wood SN. Fast stable restricted maximum likelihood and marginal likelihood
estimation of semiparametric generalized linear models. J Royal Stat Soc Ser B.
2011;73(1):3–36. 26. R Core Team, R. A language and environment for statistical computing.
Vienna: R Foundation for Statistical Computing; 2019. 27. Espinoza M, Farrugia TJ, Webber
DM, Smith F, Lowe CG. Testing a new acoustic telemetry technique to quantify long-term,
fne-scale movements of aquatic animals. Fish Res. 2011;108(2–3):346–71. 28. Kessel ST,
Cooke SJ, Heupel MR, Hussey NE, Simpfendorfer CA, Vagle S, Fisk AT. A review of detection
range testing in aquatic passive acoustic telemetry studies. Rev Fish Biol Fisheries.
2014;24(1):199–218. 29. Kessel S, Webber HND, Gruber S, Young J, Smale M, Fisk A. Close
proximity detection interference with acoustic telemetry: the importance of considering tag
power output in low ambient noise environments. Animal Biotelemet. 2015;3(1):5.
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in pub