You are on page 1of 13

IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 18, NO.

2, APRIL 2013 471


Design, Modeling, and Characterization of a
Miniature Robotic Fish for Research and
Education in Biomimetics and Bioinspiration
Vladislav Kopman, Student Member, IEEE, and Maurizio Porri, Senior Member, IEEE
AbstractIn this paper, we present the design of a biomimetic
robotic sh with a modular caudal n and analyze its perfor-
mance. The robots propulsion systemis experimentally character-
ized for different caudal n geometries by using an ad hoc thrust-
measurement system. The static thrust produced by the vibrating
tail is expressed in terms of the oscillatory Reynolds number and
compared with similar ndings in the literature. Nonlinear vibra-
tions of the propulsive tail are modeled using modal analysis and
classical results from the study of large vibrations of slender cylin-
ders in uids. This analysis allows for computing the oscillatory
Reynolds number in terms of the input parameters to the tail vi-
bration. Free-swimming experiments are performed to investigate
the performance and maneuverability of the robot and correlate
static thrust with terminal speed. This robotic platformis currently
being used in ethorobotics research for investigating collective be-
havior of gregarious sh species and in educational fun-science
activities for K-12 students.
Index TermsModeling, propulsion, robots, underwater vehi-
cles, vibrations.
I. INTRODUCTION
T
HE design of robotic sh continues to garner considerable
research attention for the performance benets associated
with sh-like swimming and the potential implementation of
such devices in eld studies [1], [2]. A novel application of such
devices lies in the emerging eld of ethorobotics, which involves
the integration of robots in animal systems for the purpose of
behavioral studies (see [3]). Specically, the possibility of mod-
ulating the behavior of a robotic sh within a school of live
sh may be used to address fundamental questions in collective
behavior, such as the emergence of leadership and follower-
ship [4], [5], and to provide technological aids to sh bypass
systems and pest control [6], [7]. In addition, robotic sh have
been proposed for use in mobile sensor networks and in studying
sh swimming (see, for example, [8]). Beyond their tremendous
potential for fundamental research, robotic sh have been shown
Manuscript received November 11, 2011; revised March 26, 2012; accepted
September 16, 2012. Date of publication November 15, 2012; date of cur-
rent version January 10, 2013. Recommended by Guest Editor M. Sitti. This
work was supported by the National Science Foundation under Grant CMMI-
0745753 and through a Graduate Research Fellowship to V. Kopman under Grant
DGE-1104522.
The authors are with the Department of Mechanical and Aerospace Engineer-
ing, Polytechnic Institute of New York University, Brooklyn, NY 11201 USA
(e-mail: vkopma01@students.poly.edu; mporri@poly.edu).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TMECH.2012.2222431
to be a valuable means of attracting the general public toward
science, technology, and mathematics elds when included in
temporary exhibits in public aquaria. For example, the carp-like
robots in [9] have attracted thousands of visitors during their
exhibits at places such as the London Aquarium.
Robotic sh are typically designed by using several mo-
tors attached in series for increased tail agility (see, for ex-
ample, [10][13]). Other techniques include the use of pumps
or power transmission devices such as cables, tendons, gears, or
belts attached to a single motor or hydraulic actuator, embed-
ded in the robots body, to achieve a tail-beating motion (see,
for example, [14][19]). Smart materials, such as ionic polymer
metal composites and shape memory alloys, are also considered
as promising actuators for underwater propulsion offering sev-
eral advantages, including silent operation and a high degree of
biomimicry (see, for example, [20][25]). Nevertheless, their
relatively high cost and their current performance as underwa-
ter propulsors are limiting factors for their implementation in
ethorobotics and educational applications. Another design strat-
egy consists of using a single motor attached to a exible/rigid
tail n that produces thrust (see, for example, [26][31]). This
latter strategy is relatively simple to implement and offers a vi-
able platformfor the realization of low-cost devices for rapid de-
ployment. In addition, platforms offering a small size, minimal-
cost, biologically inspired features, and interactive capabilities
are expected to serve well the needs of ethorobotics and K-12
education/outreach applications.
The robotic sh described in [26][29] offer bioinspired de-
signs of boxsh [26], [28] and knifesh [27], [29]. The robotic
sh presented in [30] comprises a rigid elliptical body with a
apping silicone tail at the rear. Moreover, the robot in [31]
resembles a submarine with a apping lunate n in place of
the propeller. With the exception of the platform in [28], the
size of these robots, ranging from 25 to 120 cm, poses practical
difculties in performing ethorobotics experiments in common
laboratory settings. For the purpose of studying collective be-
havior and leadership/followership responses, experiments may
include preference tests for gregarious sh in an instrumented
test-pool, or observations of sh and robots swimming together
in a water tunnel (see, for example, [32]). In such scenarios, the
robot is often required to closely resemble the target species in
its morphophysiology and/or modes of swimming.
The designs of the robots described in [26][29] include
multiple-part assemblies and application-developed custom
components. The use of fewer parts and widely available off-
the-shelf components would provide for a more cost-effective
1083-4435/$31.00 2012 IEEE
472 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 18, NO. 2, APRIL 2013
platform suitable for increasing the accessibility of such sys-
tems, especially toward K-12 educational/outreach activities.
Such activities (see, for example, [33]) linger for robots featur-
ing interactive components that allow participants to actively
learn through hands-on involvement in the platforms operation
and/or its design.
In this paper, we introduce a multipurpose, miniature, low-
cost, biomimetic robotic sh for possible use in ethorobotics
research and K-12 educational/outreach activities. Specically,
we present the hardware design and characterization of a novel
biomimetic robotic sh with a modular caudal n. This minia-
ture and low-cost robot utilizes a servomotor-based propulsion
system and is wirelessly operated via a hand-held remote con-
trol unit or a computer-based graphical user interface (GUI).
We characterize the robots thrust production through direct
measurements using an ad hoc experimental setup for a variety
of tail-beating frequencies and amplitudes, which serve as input
parameters, and for different caudal n geometries. We estimate
the coefcient of thrust fromthe data obtained in the experimen-
tal campaign to predict the static thrust production as a function
of the caudal n geometry and the oscillatory Reynolds number
associated with the tail vibration. Further, we develop a non-
linear modal model for the tail vibration using EulerBernoulli
beam theory to describe the oscillation of the tapered n. We
incorporate the effect of the encompassing uid by consider-
ing a time-varying distributed force acting on the tail, in the
form of the Morison equation [34] that is typically used to study
hydrodynamic loading on rigid slender columns due to oscilla-
tory ows. Two hydrodynamic coefcients are used to account
for uid-induced damping and added mass. Such coefcients are
experimentally identied fromdirect measurement of the n-tip
displacement using a vision-based acquisition system. We pro-
pose linear regression models to predict these two coefcients
in terms of the oscillatory Reynolds number and the Keulegan
Carpenter number. The modal model is used in conjunction with
the estimated thrust coefcient to compute the thrust production
as a function of the input parameters, that is, the frequency and
amplitude of the tail-beating. Through free-swimming exper-
iments, we characterize the robots velocity as a function of
tail-beating frequency and amplitude. We further correlate such
velocities with the static thrust data by introducing a coefcient
of drag.
The overarching modeling approach used to study the robots
locomotion shares similarities with the work in [18], [20],
and [21], as it seeks to estimate static thrust production by con-
sidering the forced vibrations of the robots tail in a quiescent
uid. Nevertheless, here we take a signicant departure from
these efforts by taking into consideration nonlinear damping in-
duced by the encompassing uid that plays a major role on low
frequency and moderately large vibrations of compliant beams,
as demonstrated in [35][37]. The accurate knowledge of static
thrust production as a function of the input parameters is one of
the main contribution of this paper and is a fundamental mile-
stone for developing control-oriented models of robotic sh as
well as informing robot design.
This paper is organized as follows. In Section II, we present
a description of the robotic platform and remote control archi-
Fig. 1. Biomimetic robotic sh. Length: 117 mm (without the caudal n);
height: 48 mm; width: 26 mm; mass: 70 g.
tecture. In Section III, we experimentally and analytically study
the propulsion systemof the robot. In Section IV, we investigate
the free-swimming of the robot and evaluate its performance.
Section V is left for conclusions and discussions.
II. SYSTEM DESCRIPTION
The system comprises a biomimetic robotic sh and a remote
control unit. The robot consists of a rigid acrylonitrile butadiene
styrene plastic body shell, tail section, and cap (see Fig. 1).
Amodular caudal n is attached to the tail section and may be
replaced with one of different size/geometry for produced-thrust
optimization. The robot is analogous to a three-link assembly,
where the body and tail sections are rigid links and the caudal
n is compliant. An actuator between the body and tail section
acts as an active hinge between these two sections and dictates
the angle between them. As this angle changes, the presence of
the encompassing uid induces the bending of the compliant
caudal n, which in turn allows for propulsion.
A remote control allows wireless operation of the robotic
sh. Parameters such as tail-beating frequency, amplitude, and
offset may be specied. A joystick, knob, and several switches
are included in the remote for manual control of the robots. In
addition, an optional connection of the remote to a computer
allows the use of a GUI for precise control-parameter selection
and autonomous operation of the robotic sh.
A. Robot Design
The robots body shell includes a payload compartment and
a motor compartment. The payload compartment contains the
control electronics, battery, and a counterweight to enhance
pitch and roll stability and achieve proper buoyancy. More
specically, buoyancy is set such that the robot is almost fully
submerged during operation. A cap provides a waterproof seal
for the payload compartment and extends toward the back of
the robot, partially covering the motor compartment. A toggle
switch hidden in the cap extension is used to turn the robot
ON or OFF. The motor compartment houses a Traxxas 2065
waterproof servomotor used for actuation, which connects to
the tail section via a modied servomotor horn. The caudal n
attaches to the tail section by snugly tting into a slot at the
tail sections free end. The robots plastic components are de-
signed in SolidWorks and printed on a Dimension Elite rapid
prototyping machine. The robot is approximately 117 mm long,
KOPMAN AND PORFIRI: DESIGN, MODELING, AND CHARACTERIZATION OF A MINIATURE ROBOTIC FISH 473
48 mm high, and 26 mm wide, without the modular caudal n,
and weights 70 g.
The onboard electronics include a microcontroller unit, a
wireless transceiver, power regulators, and a rechargeable bat-
tery. An Arduino Pro Mini microcontroller serves as the robots
processing unit. The microcontroller interfaces with a Nordic
nRF2401A transceiver chip for wireless communication with
the remote control. It is important to note that good wireless
communication in the water is achieved, mainly due to the
robots close proximity to the water surface. The remote con-
trol exchanges data with the microcontroller through data pack-
ets containing robot status information and control parameters,
namely, tail-beating frequency, amplitude, and offset. The con-
trol data are processed by the microcontroller and then used to
drive the servomotor. A single-cell lithium-ion polymer battery
procured from (www.powerstream.com) provides power to the
robot. The battery has an output of approximately 3.7 V with
a nominal capacity of around 180 mAh. The microcontroller
and servomotor are directly powered by the battery, while the
voltage is dropped down to 3.3 V by using a LD33V voltage
regulator for the transceiver. A charging port is situated at the
bottom-rear of the robot providing a means to externally charge
the onboard battery. We note that the charging port is inactive
when the robot is turned ON and is allowed to dry thoroughly
prior to charging. In this conguration, the robot may operate
continuously for approximately 1 h before recharging of batter-
ies is required, with a current draw ranging from 100 to 300 mA
during the servomotors actuation phases.
B. Robot Control
A remote control unit is devised to offer manual operation of
the robotic sh. The remote is comprised of an Arduino Duemi-
lanove microcontroller for processing, control input hardware,
and robot status indicators. A joystick provides steering control
with left/right motions and tail-beating frequency control with
up/down motions; a knob is used for the selection of tail-beating
amplitude; a toggle switch allows the selection of control by
joystick-mode or computer-mode; and two LEDs indicate
system ready and low battery status. Joystick-mode allows for
manual operation of the robot, while computer-mode offers con-
trol through a GUI for precise control parameter selection and
autonomous operation capabilities.
III. ROBOTIC TAIL ANALYSIS
Propulsion of the robot is achieved by undulating the rigid
tail section to which the exible caudal n is attached (see
Fig. 2). The static thrust production is characterized through
direct measurements for a variety of tail-beating frequencies
and amplitudes, which serve as input parameters, and for dif-
ferent caudal n geometries. An ad hoc experimental setup is
devised to measure the thrust produced by the undulating tail.
Thrust coefcients for each tail are estimated as a function of
the oscillatory Reynolds number from the data obtained in this
experimental campaign and the results are compared with the
literature. A tractable linear regression model for the thrust co-
efcients as a function of the oscillatory Reynolds number is
Fig. 2. Top view of the robotic sh representing the robots undulating tail.
Low friction linear motion guide
Thrust (Voltage)
Acrylic
Mount
Traxxas
2065
Servo
Passive
Caudal
Fin
60 liter glass
test pool
P
u
l
s
e

W
i
d
t
h

M
o
d
u
l
a
t
i
o
n
GSO-10
Load
Cell
TMO-1
Signal
Conditioner
DEL Imaging
MotionPro Y-Series 3
Rigid Tail
Section
Linear Bearing
Load Stem
Fig. 3. Schematic representations of the thrust measurement experimental
setup. The servomotor with attached tail section and caudal n is xed to a
low-friction linear motion guide and connected to a load cell via a threaded
rod. A pulse-width modulation signal is applied to the servomotor and a video
camera setup is devised to capture the ns motion.
established. As observed in [18], [19], and [29], static thrust
values can be considered as appropriate lower bounds for the
dynamic thrust that is achieved during the actual motion where
vorticity shedding from the tail tip may provide further propul-
sion (see also [38]).
The n vibration is described by using the classical Euler
Bernoulli beam theory, and the effect of the surrounding uid
is incorporated through a time-varying distributed force, in the
formof the Morison equation. In this approach, the uid loading
is described by two hydrodynamic coefcients C
d
and C
a
that
account for damping and added mass forces, respectively. The
nonlinear distributed model is analyzed using a reduced-order
modal model where only the vibration along the fundamental
mode shape of the n in vacuum is retained. These coefcients
are experimentally identied by minimizing the error between
the n-tip displacement predicted by this model and data from
a vision-based experiment. A linear regression model is pro-
posed to approximate C
d
and C
a
as functions of the oscillatory
Reynolds number and the KeuleganCarpenter number.
A. Propulsion Experiments
The experimental setup, depicted in Fig. 3, is devised to quan-
titatively characterize the effect of tail-beating frequency and
474 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 18, NO. 2, APRIL 2013
amplitude on thrust production in a placid environment. The
setup is similar to the systems proposed in [20] and [39], to
analyze the thrust produced by vibrating ionic polymer metal
composites, and in [19] and [29] to study sh-like propulsion
of different robots. In this setup, the Traxxas 2065 servomotor,
used on the robotic sh with attached tail section and passive
caudal n, is set to ap in a 60 l freshwater glass test-pool,
75 30 32 cm in length, width, and height, respectively, at
room temperature. A Transducer Techniques GS-10 load cell
is used in this setup along with a TMO-1 amplier/conditioner
module for the force measurements. The load cell is not speci-
ed by the manufacturer to measure one-directional forces and
an inherent torque or moment during loading may provide un-
correlated readings. Therefore, a low-friction THK RSR-WZM
linear motion guide is utilized to restrict the translation of the
apping robotic tail to one dimension. The measurement di-
rection is xed along the longest dimension of the experimen-
tal tank to minimize wall effects (see Fig. 3). A custom-made
acrylic mount is used to attach the servomotor to the linear mo-
tion guide and connect to the load cell via a short precision
threaded rod. The linear motion guide with attached load cell is
securely rested on the top of the test-pool and xed to remain
above the water level during trials. The acrylic mount extends
down into the water to support the submerged servomotor with
attached tail section and caudal n.
A custom LabVIEW 8.6 virtual instrument GUI is developed
to acquire thrust data from the load cell via a National Instru-
ments USB-6221 data acquisition (DAQ) board, while driving
the servomotor. Data acquisition is performed at 100 Hz for
1000 samples. The servomotor is controlled by a pulse-width
modulation (PWM) signal generated to displace the servomotor
horns angular position sinusoidally. A high-speed DEL Imag-
ing Y-Series 3 camera is positioned above the setup to record
several periods of transverse motion and is synchronized with
the thrust measurement. The camera is set to capture 3 s of
motion at frame rates of 100 or 500 Hz. The selection of frame
rate is motivated by the need to improve motion clarity while
maintaining a manageable video le size.
Prior to experiments, the measurement setup is calibrated
with weights applied to the load cell. The loading conditions
present during the experiment are mimicked by applying force
to the load cell using a pulley system. A linear calibration curve
is extrapolated by using various precision weights and relates
the voltage output from the load cell to a corresponding force.
The experiment is performed by driving the servomotor to
undulate the passive caudal n, cantilevered in the tail section
of the experimental setup (see Fig. 3) in the placid aqueous
environment. Three caudal n geometries (see Fig. 4) are con-
sidered in this study and are all constructed out of transparent
Mylar and transparent tape. The effect of varying n geometries
on thrust production has been investigated in both the biology
and engineering literature, yet experimental results applicable
to compliant foils in a broad range of operating conditions along
with data on their vibration are scarce [40][42]. The thrust data
are acquired for tail-beating frequencies of 12 Hz at increments
of 0.1 Hz for an amplitude of 12

or of varying amplitudes from


10

to 20

in increments of 1

at a frequency of 1 Hz, totalling


24 mm
24 mm 44 mm 44 mm
24 mm 24 mm
5
0

m
m
B
i
o
i
n
s
p
i
r
e
d
T
r
a
p
e
z
o
i
d
a
l
R
e
c
t
a
n
g
u
l
a
r
Fig. 4. Schematic of the three caudal n geometries used in this study, each
50 mm long and nominally 250 m in thickness.
1 1.2 1.4 1.6 1.8 2
40
20
0
20
40
Time [s]
Thrust [mN] Fintip disp. [mm] Base disp. [mm] Base angle ref. [deg]
Fig. 5. Representative time trace of thrust, n-tip displacement, n-base dis-
placement, and n-base angle for a frequency of 1.9 Hz and an amplitude of
12

. Note that the thrust signal is the output after a 5-Hz cutoff low-pass lter.
22 measurements per each n type. From the time traces, the
peak value

T and the mean value

T of thrust are extracted.
The range of frequencies and amplitudes is selected according
to the actual range of operation implemented on the robot. We
note that higher frequencies produce considerable vibrations in
the system and compromise measurements. Raw data from the
load cell, via the signal conditioner, are ltered using a software
5-Hz cutoff low-pass lter to remove signal noise. This value
is selected to ensure a system bandpass greater than the twice
the largest characteristic frequency in the experiment (2 Hz).
The images captured during the experiment are analyzed us-
ing a commercially available image analysis and tracking soft-
ware Xcitex ProAnalyst (www.xcitex.com). The time histories
of the passive caudal ns tip and base transverse displacements
are extracted. Displacement signals are individually shifted to
have zero mean from the raw data. This information is used to
calculate the oscillatory Reynolds number for each input fre-
quency/amplitude pair, dened as [43]
Re
f
= 2fl/. (1)
Here, f is the tail-beating frequency, l is the length of the entire
actuator including the rigid tail section, is the maximumn-tip
displacement, and = 10
6
m
2
s
1
is the kinematic viscosity of
water at roomtemperature. The oscillatoryReynolds number has
a primary role on the ow physics in the vicinity of the tip of the
propulsive tail that controls vorticity shedding, which is, in turn,
responsible for thrust production [43][45]. Fig. 5 shows a rep-
resentative set of data, for an amplitude of 12

and a frequency
of 1.9 Hz, collected during experiments and demonstrates that
KOPMAN AND PORFIRI: DESIGN, MODELING, AND CHARACTERIZATION OF A MINIATURE ROBOTIC FISH 475
2 2.5 3 3.5 4 4.5
10
3
10
2
10
1
10
0
log
10
Re
f
A
v
e
r
a
g
e

t
h
r
u
s
t

c
o
e
f
f
i
c
i
e
n
t
Rectangle
Trapezoid
Bioinspired
Prince et al.
Peterson et al.
Aureli et al.
Abdelnour et al. ( = 0.2)
SBT (Rectangle)
SBT (Trapezoid)
SBT (Bioinspired)
2 2.5 3 3.5 4 4.5
10
3
10
2
10
1
10
0
log
10
Re
f
P
e
a
k

t
h
r
u
s
t

c
o
e
f
f
i
c
i
e
n
t
Rectangle
Trapezoid
Bioinspired
Aureli et al.
Abdelnour et al. ( = 0.2)
(a)
(b)
Fig. 6. Average and peak thrust coefcients as a function of the oscillatory
Reynolds number and n geometry. (a) Average thrust coefcient for experi-
mental results from this campaign (diamonds) are compared with experimental
ndings from[20], [44], and [45] (circles), numerical data for = 0.2 from[43]
and elongated body theory [46]. (b) Peak thrust coefcient for experimental re-
sults from this campaign (diamonds) are compared with experimental ndings
from [20] (circles) and numerical data for = 0.2 from [43].
the thrust produced by an undulating propulsor has twice
the frequency of the tail-beating frequency (see, for example,
[20], [43], and [44]. Inherent phase lags between the n-tip and
n-base displacements are due to viscous damping, while the
delay between the reference signal and actual n-base displace-
ment is due to system and servomotor dynamics.
Fig. 6(a) and (b), respectively, displays the average and peak
thrust coefcients, dened as

C
T
=
4

Tl

2
d
max
Re
2
f
(2a)

C
T
=
4

Tl

2
d
max
Re
2
f
(2b)
for each of the n geometries as a function of Re
f
. Here, d
max
is the maximum n width, at the trailing edge, and is the
uid mass density, set at 1000 kgm
3
. We note that the n-
tip displacement used for thrust coefcient map construction
is the displacement measured perpendicularly from the robotic
shs sagittal plane.
The experimental data in Fig. 6(a) and (b) indicate that the
thrust coefcient for the three different n geometries is com-
parable, yet differences can be observed. The limited scatter is
attributed to noise in the experimental thrust signals (see Fig. 5).
We comment that the effect of varying tails on thrust produc-
tion for different propulsion systems is also documented in [21]
and [47][50].
For design purposes, we compute individual linear regression
models capturing the scaling laws for the average and peak
TABLE I
THRUST COEFFICIENT FITTING PARAMETERS
Parameter Rectangular Trapezoidal Bioinspired

1
0.1105 746.90 8.1880

2
0.1214 -0.7600 -0.3296

1
6.9970 324.50 158.00

2
-0.2857 -0.6742 -0.6366
thrusts coefcients for each of the n geometries in the form

C
T
F I T
=
1
Re

2
f
(3a)

C
T
F I T
=
1
Re

2
f
. (3b)
Here,
1
,
2
,
1
, and
2
are parameters given in Table I.
The thrust coefcient data are in good agreement with the
coefcients derived from the numerical data on 2-D computa-
tional uid dynamics for 125 Re
f
3770 reported in [43]
and a maximum n-tip displacement to actuator length ratio
= /l = 0.2, that is,

C
T
N U M
= 0.06248 Re
0.07
f
and

C
T
N U M
=
1.208 Re
0.10
f
. Results on average static thrust are also in good
agreement with predictions from the classical elongated body
theory [46] that is also used in [18], [20], and [21] to estimate
the thrust production. In this case, using (2a) and (2b), the av-
erage thrust coefcient is

C
T
S B T
= d
max
/4l, where its value
corresponds to 0.1885 for the rectangular case and 0.3456 for
the trapezoidal and bioinspired cases.
1
As further evidence of the importance of the caudal n in the
mechanics of thrust production, we have collected thrust data in
the case when the caudal n is not present. As the oscillatory
Reynolds number Re
f
varies in the range of 400010 000, the
measured thrust production remains fairly constant within the
range of 1.232.79 mN and the peak and average thrusts gener-
ally overlap. These values are on the order of 120% of thrust
data obtained when using a compliant n, which illustrates the
importance and need of a passive caudal n in sh locomotion.
This element may also be important in education/outreach ac-
tivities when explaining to younger students the fundamentals
of swimming.
B. Vibration Modeling
Knowledge of the robots undulating propulsorthat is, its
rigid tail section and caudal n-tip displacement for a given
frequency and amplitudeallows the computation of thrust by
rearranging (3a) and (3b). The n-tip displacement is predicted
by modeling the vibrating propulsor as a stepped beam com-
posed of a rigid pendulum and a exible tapered beam (see
Fig. 7). The exible beam corresponds to the robots compliant
caudal n and the pendulum corresponds to the rigid tail section
between the caudal n and the servomotor. Fig. 7 presents a
schematic of the propulsor illustrating the rigid pendulum as the
dark blue rectangle and the exible beam as the light blue line.
The origin of the Cartesian xed frame {X, Y, Z} is placed at
the hinge and corresponds to the servomotor connection. Unit
vectors in the X, Y, Z directions are, respectively, called

i,

j,

k
1
For comparison, we note that (21) in [20] and (2) in [44] are scaled by a factor
of 2 as compared to this estimate, which is based on a sinusoidal oscillation.
476 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 18, NO. 2, APRIL 2013
n(t)

i
x
w(x, t)
( )
(t)
L

j

k
Y
X
Z
L L
0
L
Fig. 7. Schematic representation of the vibrating propulsor, modeled as a
rigid pendulum (dark blue rectangle) with a hinge on one end and a exible
beam (light blue line) on the other, showing relevant conventions and system
parameters.
and the unit vector in the XY plane perpendicular to the rigid
beam is given as n(t) = [cos (t)]

i + [sin (t)]

j. Here, (t) is
the angle formed between the rigid beam and the Y axis, act-
ing as an input to the propulsor. The angle (t) prescribed by
the servomotor is of the form
(t) = Bsin(2ft +) (4)
where B is the amplitude, f is the frequency, and is a phase
lag that can be discarded upon selection of the initial time of
observation. Further, L
0
= L = 50.0 mm are the lengths of the
rigid pendulum and exible beam, respectively, and w(x, t) is
the transverse displacement eld of the exible beamin the n(t)
direction, with x as the material abscissa along the exible beam
axis.
In-plane vibrations of the exible beam are described us-
ing classical EulerBernoulli beam theory. The exible beam
is assumed to undergo small deformations, and the rotational
inertia of its cross sections is disregarded. We further assume
that the exible beam has constant thickness equal to h and that
it is composed of a homogenous and isotropic material. The
displacement of points on the exible beam is given by
u(x, t) = u
R
(x, t) +w(x, t) n(t) (5)
where u
R
(x, t) = [(L
0
+x) sin (t)]

i [(L
0
+x) cos (t)]

j
accounts for the rigid body motion and w(x, t) n(t) for the dis-
placement due to the beams vibration. Similar problems arise
in the analysis exible arms driven by shafts (see, for exam-
ple, [51] and [52]. Taking the derivative of (5) with respect to
time twice and projecting it onto n(t) yields the acceleration of
the exible beam

u(x, t) n(t) = w(x, t) w(x, t)

2
(t) + (L
0
+x)

(t) (6)
where a superimposed dot denotes derivatives with respect to
the time variable t.
The forced underwater vibration of the exible beam is de-
scribed by the following partial differential equation:
[K
M
(x)w

(x, t)]

= F
f
(x, t) (x)[

u(x, t) n(t)] (7)


where a prime denotes differentiation with respect to x and a
dot denotes vector inner product. Here, K
M
(x) =
1
12
Eh
3
d(x)
is the exible beams stiffness, where E is Youngs modulus
and d(x) is the n width. Further, the exible beams mass per
unit length is given as (x) = md(x)/A, where m is the total
n mass and A =
_
L
0
d(x)dx is the total n area, their values
TABLE II
SYSTEM PARAMETERS
Parameter Rectangular Trapezoidal Bioinspired
m(mg) 356 496 387
A(m
2
) 1.20 10
3
1.70 10
3
1.47 10
3
Eh
3
(Nm) 1.59 10
2
2.15 10
2
1.11 10
2

1
(rads
1
) 94.02 91.38 98.09
are given in Table II. In addition, F
f
(x, t) is the force due to
the uid acting on the beam during its vibration. The boundary
conditions for (7) are
w(0, t) = 0; w

(0, t) = 0 (8a)
w

(L, t) = 0; w

(L, t) = 0. (8b)
Equation (8a) imposes a clamped condition on the exible beam
at the junction with the rigid pendulum (x = 0), while (8b)
imposes a free-end condition at the tip of the exible beam
(x = L).
For typical robot operation, the caudal n vibrates in its rst
vibrational mode. Therefore, an approximate solution for the
propulsors motion is obtained by projecting the equation of
motion onto the fundamental mode shape of in-vacuum vibra-
tions
1
(x) by using the classical Galerkin Method (see, for
example, [53]).
We numerically compute the fundamental mode shape of
this tapered beam from the mode shapes of a cantilever beam
with uniform width
s
(x), where s = 1, 2, . . .. Specically,
we consider (7) with F
f
(x, t) = 0, (t) = 0, and w(x, t) =

p
j=1

j
(x)e
t
, where is the imaginary unit, is the ra-
dian frequency, and p is the number of modes retained in the
expansion. Multiplying (7) by
i
(x) with i = 1, . . . , p and inte-
grating between 0 and L, we obtain the following stiffness and
mass matrices:
K
ij
=
Eh
3
12
_
L
0
_
d(x)

j
(x)

i
(x)dx (9a)
M
ij
=
_
L
0
(x)
j
(x)
i
(x)dx (9b)
for i, j = 1, 2, . . . , p. The eigenvalue equation for in-vacuum
vibration becomes
_
K
2
s
M

c
s
= 0 (10)
where
s
is the sth natural radian frequency for in-vacuum
vibrations and c
s
= [c
s1
, c
s2
, . . . , c
sp
]
T
is the corresponding
eigenvector, with s = 1, . . . , p. Rearranging the eigenvalues in
ascending order, the propulsors fundamental mode shape is

1
(x) =
p

l=1

l
(x)c
1l
. (11)
We normalize
1
(x) to a unitary n-tip displacement, that is,

1
(L) = 1 and use the rst ten mode shapes of the homogenous
cantilevered beam (p = 10) to nd
1
(x).
In this analysis, the rectangular, trapezoidal, and bioinspired
caudal n widths are, respectively, dened by
d(x) = 0.024 (12a)
d(x) =
2
5
x + 0.024 (12b)
KOPMAN AND PORFIRI: DESIGN, MODELING, AND CHARACTERIZATION OF A MINIATURE ROBOTIC FISH 477
d(x) =
_

_
2
5
x + 0.024, 0 x <
_
2
5
x + 0.024
_
(x), x L.
(12c)
Here, all numbers are in meters and (x) =
2
_
R
2
(x x
c
)
2
, with R = 0.035 being the radius of
the circle centered at x
c
= 0.0772, identies the cutaway at the
end of the bioinspired n and =
2
5
x + 0.024. We note that the
edges of the bioinspired n are approximated as straight lines.
By using (11), we numerically compute the fundamental mode
shape
1
(x) and the eigenvalue
1
= Eh
3
/
2
1
. We nd Eh
3
for
each n by experimentally identifying the fundamental natural
frequency
1
. Such identication is performed from the time
trace of the n-tip displacement in a set of impulse-response
experiments executed on the same setup used for the forced
vibration analysis. The experiments are conducted in air
and free-vibration of the n is induced by impact. The time
history of the n-tip displacement is captured at 1000 frames/s
and compared with a second-order single degree of freedom
damped response. More specically, the damping ratio is
obtained by considering the exponential decay of the transverse
displacement of the n-tip and used in conjunction with the
damped natural frequency to nd the fundamental natural
frequency in vacuum
1
. These natural frequencies and the
values for each ns Eh
3
are given in Table II.
By setting w(x, t) =
1
(x)q(t) in (7), with q(t) being the
caudal ns tip displacement, multiplying both sides by
1
(x),
and integrating between 0 and L, we nd the following one
degree of freedom modal model:
q(t) +
_
K
M

2
(t)
_
q(t)
=
1
M
_
L
0
F
f
(x, t)

1
(x)
(x)
dx
G
M

(t) (13)
where
M =
_
L
0

2
1
(x) dx (14a)
K =
_
L
0
[K
M
(x)

1
(x)]


1
(x)
(x)
dx (14b)
G =
_
L
0
(L
0
+x)
1
(x) dx. (14c)
The force acting on the exible beam due to the encompassing
uid is given by Morisons formula (see, for example, [54])
F
f
(x, t) =
1
2
v(x, t)|v(x, t)|d(x)C
d

1
4
v(x, t)d
2
(x)C
a
(15)
where v(x, t) =

u(x, t) n(t) = w(x, t) + (L
0
+x)

(t). Here,
C
d
and C
a
describe damping and added mass effects including
nonlinearities arising in moderately large amplitude oscillations
(see, for example, [35] and [36]). We note that modeling vibra-
tions of compliant robotic tails generally discard the presence
of nonlinear damping exemplifying by C
d
in (15). For example,
0.00s
0.07s
0.13s 0.40s
0.33s
0.27s
0.20s 0.73s
0.67s
0.60s
0.53s
0.47s
~1cm
Fig. 8. Series of consecutive snapshots showing the motion of the undulating
robotic tail, with attached rectangular caudal n, for a frequency of 1.9 Hz and
an amplitude of 12

.
Chen et al. [21] use the approach proposed by Sader [55] to
describe the unsteady Stokes ow generated by the vibrating
beam, which leads to a linear formulation. In [18] and [19],
only added mass effects are taken into consideration and lin-
ear damping is described through a standard viscous effect. For
the problem at hand, where the beam is very thin and the sur-
rounding medium has a comparable mass density to the beam
material, the added mass plays a fundamental role on the sys-
tem dynamics by dominating inertia effects. Indeed, the second
summand in (15) effectively corresponds to an increased mass
per unit length that is analogous to the mass per unit length of a
water cylinder of diameter d scaled by C
a
.
The displacement data obtained from the image analysis por-
tion of the experiment (see Fig. 8) are used to extract the propul-
sors tip displacement q(t), measured from the exible beams
center of oscillation, for comparison with the aforementioned
modal model. For each combination of f and B in (4), C
d
and
C
a
are taken as constants during the identication process. We
note that the n-tip displacement q(t) obtained from this anal-
ysis corresponds to the sole vibration of the exible beam (see
Fig. 7).
The built-in differential equation solver NDSOLVE of Math-
ematica (www.wolfram.com/mathematica) is used to solve (13).
More specically, test values for C
d
and C
a
, prescribed to be
in the range 03.00 in increments of 0.01, are used in (13). A
least squares t of the predicted n-tip displacement q(t), based
on the values of C
d
and C
a
, to experimental data is performed
for each frequency/amplitude pair within the normal operating
range of the robot; that is, for frequencies of 12 Hz and am-
plitudes of 1020

. For each pair of f and B, an error vector


is dened by taking the difference between the values of the
measured n-tip displacement at the sampling instant and the
model prediction samples at that instant. This vector accounts
for the motion in the second to last vibration period of the 3 s
recording window. Notably, its length depends on f and the
478 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 18, NO. 2, APRIL 2013
TABLE III
C
d
AND C
a
AS A FUNCTION OF CAUDAL FIN GEOMETRY, f, AND B
f(Hz) B(deg) C
R
d
C
R
a
C
T
d
C
T
a
C
B
d
C
B
a
1.00 10.0 2.79 1.07 3.15 0.77 2.85 1.00
1.00 11.0 2.53 1.01 3.15 0.71 2.65 0.99
1.00 12.0 2.44 0.94 3.12 0.66 2.56 0.98
1.00 13.0 2.16 0.94 2.86 0.62 2.44 0.94
1.00 14.0 2.07 0.93 2.64 0.60 2.32 0.88
1.00 15.0 1.88 0.88 2.39 0.59 2.28 0.90
1.00 16.0 1.80 0.86 2.34 0.55 2.13 0.84
1.00 17.0 1.79 0.85 2.23 0.54 2.09 0.86
1.00 18.0 1.70 0.78 2.08 0.52 2.01 0.83
1.00 19.0 1.62 0.78 2.11 0.49 1.92 0.79
1.00 20.0 1.63 0.72 1.99 0.47 1.88 0.76
1.00 12.0 2.64 1.04 2.82 0.67 2.58 0.95
1.10 12.0 2.47 0.88 2.90 0.55 2.48 0.83
1.20 12.0 2.46 0.78 2.76 0.49 2.41 0.72
1.30 12.0 2.35 0.64 2.61 0.41 2.33 0.65
1.40 12.0 1.94 0.59 2.50 0.38 2.19 0.63
1.50 12.0 2.06 0.54 2.48 0.35 2.18 0.57
1.60 12.0 2.10 0.48 2.39 0.30 2.09 0.54
1.70 12.0 1.99 0.46 2.31 0.30 2.04 0.52
1.80 12.0 2.35 0.43 2.28 0.27 1.95 0.49
1.90 12.0 1.91 0.40 2.07 0.26 2.01 0.48
2.00 12.0 1.73 0.38 2.25 0.23 1.92 0.46
0 0.5 1 1.5 2 2.5 3
-0.04
-0.02
0
0.02
0.04
Time [s]
D
i
s
p
l
a
c
e
m
e
n
t

[
m
]
Experimental data Model approximation
Fig. 9. Comparison of the experimental and modal model-predicted
time traces of the caudal ns-tip displacement for a representative fre-
quency/amplitude pair 1.9 Hz and 12

of the rectangular n. In this case, the


least squares error is 1.66 mm.
associated video-frame rate. For example, a case where f =
2 Hz and B = 12

with an associated video frame rate of


500 Hz, the vector length is 1/f frame rate = 250 samples.
The vectors Euclidean norm is minimized to nd C
d
and C
a
;
these values are reported in Table III. The resulting least squares
error averaged over the entire set of studied cases is 2.18 mm,
which is within a few percent of the vibration amplitude.
Table III shows that C
d
and C
a
are generally monotonically
decreasing as f or B are increased. In addition, both C
d
and
C
a
decrease as the KeuleganCarpenter number, dened as
KC = 2/d
max
, increases. The variations of these quantities
with the KeuleganCarpenter number motivate the need for the
use of Morisons formula versus the linear approaches presented
in [19] and [21]. Notably, C
a
is taken equal to 1 in [19] and
variations of C
a
as a function of Re
f
from the unsteady Stokes
problem in [55] are considered in [21].
Fig. 9 offers a comparison between the model with exper-
imental data of the propulsors tip-displacement for a repre-
sentative frequency of 1.9 Hz and an input angle of 12

of the
rectangular caudal n. The red points represent experimental
TABLE IV
FITTING PARAMETERS FOR C
d
AND C
a
Parameter Rectangular Trapezoidal Bioinspired

1
-0.22 -0.41 -0.40

2
-0.71 -0.39 -0.29

3
1.87 2.39 2.25

1
-1.36 -1.51 -1.11

2
0.80 0.86 0.68

3
5.02 5.65 4.25
data from vision acquisition, while the black line represents the
model approximation. As demonstrated by Fig. 9, the model
well approximates the steady-state n vibration. Higher fre-
quencies reduce the effects of imperfections with the clamping
conditions and possible defects in n fabrication. We note that
the input angular amplitudes B [see (4)] for this analysis are
taken to be the values achieved by the servomotor during the
experiment; these values are determined through vision acqui-
sition. Although the input angular amplitude B is maintained at
12

, for the varying-frequency condition, the servomotor output


angular amplitude falls in the range of 812

. This may be due


to the discretizatized sinusoidal input signal to PWMconversion
within the virtual instrument GUI.
Both C
d
and C
a
can be expressed as functions of Re
f
and
KC through the linear regression models
C
d
= Re

1
f
KC

2
10

3
(16a)
C
a
= Re

1
f
KC

2
10

3
(16b)
where
1
,
2
,
3
,
1
,
2
, and
3
are reported in Table IV. These
parameters are obtained by tting (16a) and (16b) with the
experimental data presented in Table III. We note that the tting
is performed independently for each of the ns. Fig. 10(a)(c)
illustrates the correlation of the ts presented in (16a) and (16b)
with experimental data. These ndings show that C
d
and C
a
are
well approximated by a regression model of the form in (16a)
and (16b). With reference to Table III, the fact that
1
and
2
are
negative indicates that, for all the considered n geometries, C
d
decreases as Re
f
and KC increase. Similarly, the fact that
1
is
negative and
2
is positive, yet close to 1, demonstrates that an
increasing Re
f
and decreasing KC leads to an increase in C
a
.
Fig. 11(a) and (b) reports the comparison of the values of C
d
and C
a
obtained herein with those presented in [35][37], [54],
and [56]. As seen in Fig. 11(a), the decrease in C
d
with increas-
ing KC is in line with the ndings of [35], [37], [54], and [56].
When comparing the data in the literature as in Fig. 11(a) and
(b), it is important to note that the dependence on the Reynolds
number is discarded leading to large scatter in the data and sig-
nicant discrepancies between the experimental results. For ex-
ample, data in [35] consider Reynolds numbers that differ from
the ones considered in this study by 1 or 2 orders of magnitude.
C. Predictions of Thrust Production
Simulations are performed to predict the values of the n-tip
displacement, and therefore the average thrust, for varying input
parameters. Due to the dependence of C
d
and C
a
on , a set of
KOPMAN AND PORFIRI: DESIGN, MODELING, AND CHARACTERIZATION OF A MINIATURE ROBOTIC FISH 479
1 1.5 2 2.5 3 3.5 4
1
2
3
4
Re
f

1KC

210

3
C
d
0 0.5 1 1.5 2
0
1
2
Re
f

1KC

210

3
C
a
1 1.5 2 2.5 3 3.5 4
1
2
3
4
Re
f

1KC

210

3
C
d
0 0.5 1 1.5 2
0
1
2
Re
f

1KC

210

3
C
a
1 1.5 2 2.5 3 3.5 4
1
2
3
4
Re
f

1KC

210

3
C
d
0 0.5 1 1.5 2
0
1
2
Re
f

1KC

210

3
C
a
(a) (b) (c)
Fig. 10. Experimental data compared with the linear regression model for the (a) rectangular, (b) trapezoidal, and (c) bioinspired ns.
10
1
10
0
10
1
10
2
10
1
10
0
10
1
10
2
KC
(a)
(b)
C
d
Rectangular
Trapezoidal
Bioinspired
Graham
Aureli and Porfiri
Bidkar et al.
Aureli et al.
Falcucci et al.
10
1
10
0
10
1
10
2
10
1
10
0
10
1
KC
C
a
Rectangular
Trapezoidal
Bioinspired
Graham
Aureli and Porfiri
Bidkar et al.
Aureli et al.
Falcucci et al.
Fig. 11. Comparison of (a) C
d
and (b) C
a
obtained in this study with exper-
imental [35], [54] and numerical [35][37], [56] data from the literature. Note
that the data in [54] refer to the case of a xed rigid at plate of constant width
in an oscillating ow, where the damping and inertia coefcients are measured
(rather than the added mass coefcient).
test n-tip displacements
1
, . . . ,
P
is selected and values of
C
d
and C
a
are computed via (16a) and (16b) for use in (13),
where P is the set cardinality. The values of
1
, . . . ,
P
are
taken in the range 5.0035.0 mm in increments of 0.10 mm, for
a total of P = 301 values. This range corresponds to possible
values of the n-tip displacement, as observed through vision
acquisition. For a given f and B pair, (13) is numerically solved
for all the pairs of values of C
d
and C
a
and P n-tip displace-
ments
1
, . . . ,
P
are obtained. The n-tip displacement for a
1 1.2 1.4 1.6 1.8 2
0
0.02
0.04
Rectangular Fin
Experiments Theory
1 1.2 1.4 1.6 1.8 2
0
0.02
0.04
Trapezoidal Fin
A
v
e
r
a
g
e

t
h
r
u
s
t

[
N
]
1 1.2 1.4 1.6 1.8 2
0
0.02
0.04
Bioinspired Fin
Frequency [Hz]
Fig. 12. Experimental values for average thrust compared with theoretical
predictions.
given f and B pair is taken as the test n-tip displacement
i
that is the closest to the corresponding
i
for some i between 1
and P.
By taking into the account each ns maximumwidth, Fig. 12
presents the comparison of average thrust data (red points), ob-
tained from the experiments presented in Section III-A, with the
theoretical values (black points). The error bars in Fig. 12 ac-
count for the motors tendency not to exactly adhere to the 12

angular amplitude input signal, as described in Section III-B.


Specically, angular amplitudes of 812

are used in the simu-


lations. Fig. 12 demonstrates that the model well approximates
the average thrust produced by the robotic sh for the input
parameters featured in the experimental campaign. We note
that the model does not use any tuning parameters and builds
from the actual inputs provided to the motor. As mentioned in
Section III-A, the scatter of the experimental data is attributed to
noise in the experimental thrust signals seen in Fig. 5. We also
comment that data are reported here in the linear scale, unlike
the logarithmic scale shown in Fig. 6 to display the large varia-
tion of thrust in the considered range of Reynolds numbers. The
480 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 18, NO. 2, APRIL 2013
0.00 s 2.07 s 4.14 s 6.21 s 10.35 s 12.41 s 8.28 s
11 cm
x
y
Fig. 13. Consecutive snapshots of a robotic sh swimming in straight trajec-
tory for a trapezoidal n vibrating at 1.4 Hz.
remaining discrepancies may be due to the simplistic regression
models for the thrust and C
d
and C
a
parameters as well as the
fact that C
d
and C
a
are taken as constants that depend only on
rather than time-varying functions of the n motion.
IV. FREE-SWIMMING
The performance of the robot is assessed through a set of
free-swimming experiments in a large test-pool. This setup in-
cludes an overhead camera for capturing the free-locomotion
of the robot, controlled wirelessly according to the discussion
in Section II. Trials are performed for a set of tail-beating fre-
quencies ranging from 1 and 2 Hz, with a constant tail-beating
amplitude of approximately 12

, for each of the three caudal n


geometries.
The robot is set to swim from one end of a 1 m-diameter
circular test environment to the other in a straight trajectory,
starting from rest. The top of the robot is tagged with markers
for tracking its motion throughout the trials. Fig. 13 shows a
detailed set of consecutive snapshots demonstrating the robots
biomimetic motion. More specically, it highlights the oscilla-
tory swimming motion of the robot in response to the beating of
its tail. This is due to the reaction of the rigid body shell to coun-
teract the forces and moments applied by the undulating tail at
the servomotor attachment point; for more detailed analysis of
this effect, see [20]. This inherent effect facilitates a greater de-
gree of biomimetic locomotion of the robot and is thus sought in
the design. In addition, under certain hydrodynamic conditions,
this motion offers drag reduction over nonundulating rigid body
swimming (see [14] and [57]). ProAnalyst is utilized to track the
motion and trajectory of the robot in order to extract its velocity.
Following, for example, [21], the force of drag for small angles
of attack is expressed as
F
D
=
1
2
v
2
SC
D
. (17)
Here, v is the robots swimming velocity, S is its wetted surface
area, and C
D
is a nondimensional coefcient of drag. The robots
wetted surface area is directly extracted from the CAD model
and is found to be approximately 136 10
4
m
2
, including the
bioinspired caudal n. We note that this value is used without
loss of generality as the variability in area for the different ns
is less than 3% of S.
1 1.2 1.4 1.6 1.8 2
0
10
20
Rectangular Fin
Experiments Theory Videler
1 1.2 1.4 1.6 1.8 2
0
10
20
Trapezoidal Fin
T
e
r
m
i
n
a
l

v
e
l
o
c
i
t
y

[
c
m
/
s
]
1 1.2 1.4 1.6 1.8 2
0
10
20
Bioinspired Fin
Frequency [Hz]
Fig. 14. Experimentally achieved terminal velocity of the free-swimming
robot (red points), for the three caudal n geometries, compared with theo-
retical predictions (black points). The dashed line represents the velocity of a
live sh of similar motility [58].
Avalue for C
D
is obtained by considering the case of terminal
velocity of the robot and using the conservative estimate offered
by the static thrust, that is, setting F
D


T. Thus, by rearranging
(17), we nd
C
D
=
2
S

T
v
2
. (18)
By combining the thrust data obtained in Section III-A and the
velocity data extracted through vision acquisition, (18) provides
a linear relationship between average thrust

T and the square
of the velocity v
2
. Through least squares tting, we nd the
drag coefcient C
D
= 0.45, with a coefcient of determination
R
2
= 0.382. The value of C
D
for similar robots is found to be
0.21 in [20] from simplied 2-D computational uid dynamics
analysis, and 0.12 in [21] from an experiment where the robot
is pulled through water at a constant velocity, while measuring
the force of drag using a spring scale. The higher value for C
D
found herein may be due to the additional drag contribution of
the tail section during swimming. More specically, our robot
exploits a considerably larger part of its body for propulsion
than the robots developed in [20] and [21]. During swimming,
the oscillatory motion of the robot may cause the tail to act as
a bluff body in the ow, and due to the larger wetted surface
area, this effect may be more drastic for our robot. Yet, the
relatively low coefcient of determination and the underlying
simplications in the identication of this coefcient should be
taken into consideration when drawing such conclusions.
Fig. 14 reports the predicted terminal speed of the robot (black
points), computed by rearranging (18) into the form
v =

T
SC
D
(19)
compared with experimental data (red points) of the swim-
ming robot. Additionally, Fig. 14 includes the predicted
KOPMAN AND PORFIRI: DESIGN, MODELING, AND CHARACTERIZATION OF A MINIATURE ROBOTIC FISH 481
velocity (dashed line) of a live sh of similar motility, expressed
as v
Videler
= 0.63f (see [58]), with being the length of the
sh. As in Fig. 12, the error bars in Fig. 14 account for the
variations in the motors angular displacement.
Despite the good agreement between the predicted velocities
through (19) and experimental data, such equation should only
be taken as a rst-order estimate that is usable for qualitative
assessment and initial design. Indeed, among other approxima-
tions, it neglects oscillations of the forward velocity and sway
and yaw motions of the robot and utilizes the static thrust rather
than the dynamic one. For example, by extrapolating the range
of applicability of (19) to higher frequencies beyond 2 Hz using
the methods in Section III-C, we nd substantial deviations be-
tween this coarse estimate and empirical observations (see [20]).
V. CONCLUSION
In this paper, we presented a multipurpose, miniature,
biomimetic, and low-cost robotic sh with possible applica-
tion to ethorobotics research and K-12 educational/outreach
activities.
The robotic sh presented in this study utilizes widely avail-
able, low-cost, and resilient off-the-shelf components that make
it ideal for rapid and standardized assembling in laboratories
interested in ethorobotics. More specically, the entire robot
and remote control system costs less than 100 USD, on a
limited production basis. Furthermore, production beyond in-
laboratory fabrication including the utilization of manufactur-
ing techniques such as injection-molding for the plastic compo-
nents; the use of nonwaterproof servomotors housed in plastic-
waterproof casings; and customization of printed circuit board
electronics would potentially allow the system cost to be in the
range of 2530 USD.
The comparable velocity of the robot and its live counterpart
for low frequencies shown in Fig. 14 suggests that the robot can
be well integrated in a group of sh of similar size, swimming at
approximately one body length per second. In addition, the robot
design includes a unique modular feature for easily changing
the size, shape, and material of the robots caudal n to obtain
different vorticity patterns for live-sh robot interactions and
altering the robots morphophysiology. This feature also offers
an interactive component to the robot, fostering its use in K-12
educational/outreach activities. Specically, it allows users to
contribute to the design of the robotic sh by constructing their
own caudal ns and testing them on the swimming robot to
observe their effect on thrust.
We presented an integrated modeling framework for predict-
ing the robots static thrust production, based on the frequency
and amplitude of the beating tail, by combining reduced-order
linear and nonlinear modeling with experimental methods. This
set of tools accounts for important hydrodynamic effects such
as damping and added mass and shows their dependence on
the oscillatory Reynolds and KeuleganCarpenter numbers. The
robots velocity, as a function of the tail-beating frequency and
amplitude, was characterized via free-swimming experiments,
and these data were correlated with static thrust data.
0.00 s
2.52 s
4.90 s
7.34 s
9.34 s
11.66 s
17.03 s
18.90 s
20.86 s
22.41s
24.55 s
26.69 s
Fig. 15. Robotic sh maneuvering with a gure-eight trajectory.
Future work will include a detailed experimental characteri-
zation of the robots hydrodynamic properties and will further
explore differences between static and dynamic thrust. Exper-
imental data on the robots locomotion will be used to derive
reduced-order models based on proper orthogonal decomposi-
tion following [59]. An extension of this study will comprise
evaluation of the robots maneuverability through additional
free-swimming experiments and closed-loop autonomous con-
trol, utilizing vision tracking techniques such as those presented
in [60], for in-plane trajectory tracking. Preliminary observa-
tions have already been performed on the robot for several sim-
ple trajectories. For example, Fig. 15 demonstrates the robot
performing a gure-eight maneuver.
ACKNOWLEDGMENT
The authors would like to gratefully acknowledge Dr. N.
Abaid for her collaboration in developing the educational activ-
ities around the robotic sh, Dr. M. Aureli for useful discussion,
Dr. A. Rizzo and J. Laut for reviewing the manuscript, and
C. Xu for his assistance with the initial experimental setup.
REFERENCES
[1] P. R. Bandyopadhyay, Trends in biorobotic autonomous undersea vehi-
cles, IEEE J. Ocean. Eng., vol. 30, no. 1, pp. 109139, Jan. 2005.
[2] J. E. Colgate and K. M. Lynch, Mechanics and control of swimming: A
review, IEEE J. Ocean. Eng., vol. 29, no. 3, pp. 660673, Jul. 2004.
[3] J. Krause, A. F. T. Wineld, and J.-L. Deneubourg, Interactive robots in
experimental biology, Trend. Ecol. Evol., vol. 26, no. 7, pp. 369375,
2011.
[4] J. Krause and G. D. Ruxton, Living in Groups. London, U.K.: Oxford
Univ. Press, 2002.
[5] D. J. T. Sumpter, Collective Animal Behavior. Princeton, NJ: Princeton
Univ. Press, 2010.
[6] C. R. Schilt, Developing sh passage and protection at hydropower
dams, Appl. Animal Behav. Sci., vol. 104, no. 34, pp. 295325, 2007.
[7] R. Goldburg and R. Naylor, Future seascapes, shing, and sh farming,
Front. Ecol. Envir., vol. 3, no. 1, pp. 2128, 2005.
[8] X. Tan, Autonomous robotic sh as mobile sensor platforms: Challenges
and potential solutions, Marine Technol. Soc. J., vol. 45, no. 4, pp. 3140,
2011.
[9] H. Hu, Biologically inspired design of autonomous robotic sh at Essex,
presented at the IEEE SMC UK-RI Chapter Conf. Adv. Cybernet. Syst.,
Shefeld, U.K., 2006
[10] J. Guo, A waypoint-tracking controller for a biomimetic autonomous
underwater vehicle, Ocean Eng., vol. 33, no. 17/18, pp. 23692380,
2006.
482 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 18, NO. 2, APRIL 2013
[11] J. Liu and H. Hu, Biological inspiration: From carangiform sh to multi-
joint robotic sh, J. Bio. Eng., vol. 7, no. 1, pp. 3548, 2010.
[12] M. Malec, M. Morawski, and J. Zajac, Fish-like swimming prototype
of mobile underwater robot, J. Autom. Mob. Robot. Intell. Syst., vol. 4,
no. 3, pp. 2530, 2010.
[13] J. Shao, L. Wang, and J. Yu, Development of an articial sh-like robot
and its application in cooperative transportation, Control Eng. Pract.,
vol. 16, no. 5, pp. 569584, 2008.
[14] D. S. Barrett, M. S. Triantafyllou, D. K. P. Yue, M. A. Grosenbaugh, and
M. J. Wolfgang, Drag reduction in sh-like locomotion, J. Fluid Mech.,
vol. 392, pp. 183212, 1999.
[15] K. H. Low and C. W. Chong, Parametric study of the swimming per-
formance of a sh robot propelled by a exible caudal n, Bioinspirat.
Biomimet., vol. 5, no. 4, p. 046002, 2010
[16] K. Morgansen, B. Triplett, and D. Klein, Geometric methods for mod-
eling and control of free-swimming n-actuated underwater vehicles,
IEEE Trans. Robot., vol. 23, no. 6, pp. 11841199, Dec. 2007.
[17] P. C. Streing, A. M. Helium, and R. Mukherjee, Modeling, simulation,
and performance of a synergistically propelled ichthyoid, IEEE/ASME
Trans. Mechatronics, vol. 17, no. 1, pp. 3645, Feb. 2012.
[18] P. Valdivia y Alvarado, Design of biomimetic compliant devices for
locomotion in liquid environments, Ph.D. dissertation, Massachusetts
Inst. Technol., Cambridge, MA, 2007.
[19] P. Valdivia y Alvarado and K. Youcef-Toumi, Design of machines with
compliant bodies for biomimetic locomotion in liquid environments, J.
Dyn. Syst., Meas., Control, vol. 128, no. 1, pp. 313, 2006.
[20] M. Aureli, V. Kopman, and M. Porri, Free-locomotion of underwater
vehicles actuated by ionic polymer metal composites, IEEE/ASMETrans.
Mechatronics, vol. 15, no. 4, pp. 603614, Aug. 2010.
[21] Z. Chen, S. Shatara, and X. Tan, Modeling of biomimetic robotic sh
propelled by an ionic polymermetal composite caudal n, IEEE/ASME
Trans. Mechatronics, vol. 15, no. 3, pp. 448459, Jun. 2010.
[22] S. Guo, T. Fukuda, and K. Asaka, A new type of sh-like underwater
microrobot, IEEE/ASMETrans. Mechatronics, vol. 8, no. 1, pp. 136141,
Mar. 2003.
[23] M. Mojarrad and M. Shahinpoor, Noiseless propulsion for swimming
robotic structures using polyelectrolyte ion-exchange membranes, in
Proc. SPIE North Amer. Conf. Smart Structur. Material., vol. 2716, San
Diego, CA, 1996, pp. 183192.
[24] A. Villanueva, C. Smith, and P. Priya, A biomimetic robotic jellysh
(robojelly) actuated by shape memory alloy composite actuators, Bioin-
spirat. Biomimet., vol. 6, no. 3, p. 036004, 2011.
[25] S.-W. Yeom and I.-K. Oh, A biomimetic jellysh robot based on ionic
polymer metal composite actuators, Smart Mater. Struct., vol. 18, no. 8,
p. 085002, 2009.
[26] A. Crespi, D. Lachat, A. Pasquier, and A. J. Ijspeert, Controlling swim-
ming and crawling in a sh robot using a central pattern generator,
Autonom. Robot., vol. 25, no. 1/2, pp. 313, 2008.
[27] O. M. Curet, N. A. Patankar, G. V. Lauder, and M. A. MacIver, Me-
chanical properties of a bio-inspired robotic knifesh with an undulatory
propulsor, Bioinspirat. Biomimet., vol. 6, no. 2, p. 026004, 2011.
[28] P. Kodati, J. Hinkle, A. Winn, and X. Deng, Microautonomous
robotic ostraciiform(MARCO): Hydrodynamics, design, and fabrication,
IEEE/ASME Trans. Mechatronics, vol. 24, no. 1, pp. 105117, Feb. 2008.
[29] F. Liu, K.-M. Lee, and C.-J. Yang, Hydrodynamics of an undulating n
for a wave-like locomotion system design, IEEE/ASME Trans. Mecha-
tronics, vol. 17, no. 3, pp. 554562, Jun. 2012.
[30] R. L. McMasters, C. P. Grey, J. M. Sollock, R. Mukherjee, A. Benard,
and A. R. Diaz, Comparing the mathematical models of Lighthill to the
performance of a biomimetic sh, Bioinspirat. Biomimet., vol. 3, no. 1,
p. 016002, 2008.
[31] T. Wang, L. Wen, J. Liang, and G. Wu, Fuzzy vorticity control of a
biomimetic robotic sh using a apping lunate tail, J. Bionic Eng., vol. 7,
no. 1, pp. 5665, 2010.
[32] S. Marras and M. Porri, Fish and robots swimming together: attraction
towards the robot demands biomimetic locomotion, J. Roy. Soc. Interface,
vol. 9, no. 73, pp. 18561868, 2012.
[33] N. Abaid, V. Kopman, and M. Porri, The story of a Brooklyn outreach
program on biomimetics, underwater robotics, and marine science for
K-12 students, IEEE Robot. Autom. Mag., 2012, to be published
[34] J. M. J. Journ ee and W. W. Massie, Offshore Hydromechanics. Delft,
The Netherlands: Delft Univ. Technol., 2001.
[35] M. Aureli and M. Porri, Low frequency and large amplitude oscilla-
tions of cantilevers in viscous uids, Appl. Phys. Lett., vol. 96, no. 16,
p. 164102, 2010.
[36] R. A. Bidkar, M. Kimber, A. Raman, A. K. Bajaj, and S. V. Garimella,
Nonlinear aerodynamic damping of sharp-edged exible beams oscil-
lating at low Keulegan-Carpenter numbers, J. Fluid Mech., vol. 634,
pp. 269289, 2009.
[37] G. Falcucci, M. Aureli, S. Ubertini, and M. Porri, Transverse harmonic
oscillations of laminae in viscous uids: A lattice Boltzmann study,
Philos. Trans. Roy. Soc. Lond. A, vol. 369, no. 1945, pp. 24562466,
2011.
[38] J. N. Newman and T. Y. Wu, A generalized slender-body theory for sh-
like forms, J. Fluid Mech., vol. 57, no. 4, pp. 673693, 1973.
[39] K. J. Kim, W. Yim, J. W. Paquette, and D. Kim, Ionic polymer-metal
composites for underwater operation, J. Intell. Material Syst. Struct.,
vol. 18, no. 2, pp. 123131, 2007.
[40] G. V. Lauder, Function of the caudal n during locomotion in shes:
Kinematics, ow visualization, and evolutionary patterns, Integr. Com-
par. Biol., vol. 40, no. 1, pp. 101122, 2000.
[41] J. A. Walker and M. W. Westneat, Performance limits of labriform
propulsion and correlates with n shape and motion, J. Experim. Biol.,
vol. 205, no. 5, pp. 177187, 2002.
[42] M. J. Wolfgang, J. M. Anderson, M. A. Grosenbaugh, D. K. P. Yu, and
M. S. Triantafyllou, Near-body ow dynamics in swimming sh, J.
Exp. Biol., vol. 202, no. 17, pp. 23032327, 1999.
[43] K. Abdelnour, E. Mancia, S. D. Peterson, and M. Porri, Hydrodynam-
ics of underwater propulsors based on ionic polymer metal composites:
A numerical study, Smart Mater. Struct., vol. 18, no. 8, pp. 111,
2009.
[44] S. D. Peterson, M. Porri, and A. Rovardi, A particle image velocimetry
study of vibrating ionic polymer metal composites in aqueous environ-
ments, IEEE/ASME Trans. Mechatronics, vol. 14, no. 2, pp. 474483,
Aug. 2009.
[45] C. Prince, W. Lin, J. Lin, S. D. Peterson, and M. Porri, Temporally-
resolved hydrodynamics in the vicinity of a vibrating ionic polymer metal
composite, J. Appl. Phys., vol. 107, no. 9, p. 094908, 2010.
[46] M. J. Lighthill, Note on the swimming of slender sh, J. Fluid Mech.,
vol. 9, no. 2, pp. 305317, 1960.
[47] B. Hobson, M. Kemp, I. Le Goff, and A. Leonessa, Oscillating n
thrusters for multi-view classication maneuvering on MCM UUVS,
in Proc. MTS/IEEE Oceans, Sep. 2003, vol. 4, pp. 21672169.
[48] N. Kato, Control performance in the horizontal plane of a sh robot with
mechanical pectoral ns, IEEE J. Ocean. Eng., vol. 25, no. 1, pp. 121
129, Jan. 2000.
[49] G. Lauder and P. Madden, Fish locomotion: kinematics and hydrodynam-
ics of exible foil-like ns, Exper. Fluids, vol. 43, no. 5, pp. 641653,
2007.
[50] K. H. Low, C. W. Chong, C. Zhou, and G. Seet, An improved semi-
empirical model for a body and/or caudal n (BCF) sh robot, in Proc.
IEEE Int. Conf. Robot. Autom., May 2010, pp. 7883.
[51] O. Kopmaz and K. S. Anderson, On the eigenfrequencies of a exible arm
driven by a exible shaft, J. Sound. Vibrat., vol. 240, no. 4, pp. 679704,
2001.
[52] K. H. Low, Eigen-analysis of a tip-loaded beam attached to a rotating
joint, J. Vibrat. Acoust., vol. 112, no. 4, pp. 497500, 1990.
[53] R. C. Batra, M. Porri, and D. Spinello, Electromechanical model of elec-
trically actuated narrow microbeams, J. Microelectromech. Syst., vol. 15,
no. 5, pp. 11751189, 2006.
[54] J. M. R. Graham, The forces on sharp-edged cylinders in oscillatory
ow at low Keulegan-Carpenter numbers, J. Fluid Mech., vol. 97, no. 2,
pp. 331346, 1980.
[55] J. E. Sader, Frequency response of cantilever beams immersed in viscous
uids with applications to the atomic force microscope, J. Appl. Phys.,
vol. 84, no. 1, pp. 6476, 1998.
[56] M. Aureli, E. Basaran, and M. Porri, Nonlinear nite amplitude vibra-
tions of sharp-edged beams in viscous uids, J. Sound. Vibrat., vol. 331,
no. 7, pp. 16241654, 2012.
[57] I. Borazjani and F. Sotiropoulos, Numerical investigation of the hy-
drodynamics of carangiform swimming in the transitional and iner-
tial ow regimes, J. Exper. Biol., vol. 211, no. 10, pp. 15411558,
2008.
[58] J. J. Videler, Fish Swimming. London, U.K.: Chapman & Hall, 1993.
[59] B. F. Feeny and A. K. Feeny, Complex modal analysis of the swimming
motion of a whiting, in Proc. ASME Int. Des. Eng. Tech. Conf. Comput.
Inf. Eng. Conf., 2011, pp. 111.
[60] C. Rossi, M. Abderrahim, and J. C. Diaz, Tracking moving optima using
Kalman-based predictions, Evolut. Comput., MIT Press, vol. 16, no. 1,
pp. 130, 2008.
KOPMAN AND PORFIRI: DESIGN, MODELING, AND CHARACTERIZATION OF A MINIATURE ROBOTIC FISH 483
Vladislav Kopman (S09) was born in Odessa,
Ukraine, in 1986. He received the B.Sc. and M.Sc.
degrees in mechanical engineering from the Poly-
technic Institute of New York University, Brooklyn,
in 2008 and 2010, respectively, where he is currently
working toward the Ph.D. degree in mechanical en-
gineering as an NSF Graduate Research Fellow.
In 2008, he was a recipient of the NSF GK-12
fellowship for his Masters studies at the Polytechnic
Institute of New York University. His research inter-
ests include robotics, underwater robotics, dynamics,
and control.
Maurizio Porri (M06SM12) was born in Rome,
Italy, in 1976. He received the M.Sc. and Ph.D. de-
grees in engineering mechanics from Virginia Poly-
technic Institute and State University (Virginia Tech),
Blacksburg, in 2000 and 2006, respectively, the Lau-
rea degree (Hons.) in electrical engineering from the
University of Rome La Sapienza, Rome, Italy, in
2001, and the Ph.D. degree in theoretical and ap-
plied mechanics from the University of Rome La
Sapienza and the University of Toulon, Toulon,
France, in 2005 within a dual degree program.
From 2005 to 2006, he was a Postdoctoral researcher with the Department
of Electrical and Computer Engineering, Virginia Tech. He has been a member
of the Faculty of the Mechanical and Aerospace Engineering Department of the
Polytechnic Institute of New York University Brooklyn, since 2006, where he is
currently an Associate Professor. He is involved in conducting and supervising
research on dynamical systems theory, multiphysics modeling, and underwater
robotics.
Dr. Porri received an NSF CAREER Award (Dynamical Systems program)
in 2008. He was included in the Brilliant 10 list of Popular Science in 2010
and received the Outstanding Young Alumnus Award from the College of En-
gineering of Virginia Tech in 2012.

You might also like