MODELING AND DESIGN METHODOLOGY OF AN EFFICIENT

UNDERWATER PROPULSION SYSTEM

Pablo Valdivia y Alvarado and Kamal Youcef-Toumi
Department of Mechanical Engineering, Massachusetts Institute of Technology
Room 3-348, 77 Massachusetts Avenue, Cambridge, MA 02139
USA
Email: [email protected], [email protected].

ABSTRACT
This paper addresses the issues of energy efficiency in
mechanisms for underwater locomotion and its
accomplishment using a more robust class of mechanical
structures. We present a design methodology for
fabricating a continuous mechanical fish body such that
its kinematic behavior matches parametric models of fish
swimming with minimum energy cost. It is concluded that
this new design has advantages over classic manipulator
type designs. The limitations of this passive design are
also discussed and possible solutions proposed.

KEY WORDS
Design, energy-efficiency, robustness, underwater
vehicles, non-uniform beam forced vibrations.

1. INTRODUCTION
One of the main obstacles for increasing the range of
action of mobile robots is the problem of energy
efficiency. An autonomous robot has to carry its own
power supply. However, if it can use energy efficiently
while accomplishing its tasks it will be able to run longer
missions with the same supply or alternatively, if weight
is an issue, use a smaller supply for the same mission. In
addition, mechanism robustness is always paramount if
missions involve hazardous terrains. However, most
autonomous robots use mechanical elements such as
electrical motors, gearboxes, linkages, cams, pulleys, etc,
to transmit power internally even though these have
limited efficiencies. The combination of these elements
always yields overall low system efficiencies.
Furthermore, the nature of current mechanical elements is
not robust enough to withstand harsh environments
without sophisticated and complex mechanical protection.
As a result most autonomous robots designs are
expensive.
Scientists have studied fish swimming as an energy
efficient alternative to current underwater propulsion
methods. Several robots have been built in order to prove

this theory [1,2,3]. However, most of them still use hyperredundant manipulator type designs with classical low
efficiency machine elements to transmit power internally.
Therefore, the low internal energy transfer might diminish
energy gains brought by a more efficient external
momentum transfer when using fish swimming modes.
Kumph [3] built a tethered 0.82 m long robotic pike with
a drive system comprising 5 servomotors. The robot
achieved a maximum forward velocity of 0.09 m / s at an
actuation frequency of 1 Hz , with a supply of 17 V
drawing 0.2 A . The estimated system efficiency was
13.5%.
In this paper we address the problem of accomplishing a
more efficient mechanical implementation of such
swimming modes in a robot. We assume that the
kinematic behavior for forward motion described in [1] is
a given optimal behavior in terms of achieving efficient
momentum transfer to the environment. Then the problem
we address is how to design a machine where the
implementation of such kinematic behavior (fish
swimming modes) is more efficient and requires the least
power consumption. Our proposed design has the
additional feature of being intrinsically robust and simple.
Previous work on alternative mechanisms for achieving
fish-like underwater motions is not very extensive.
Mcletchie [4] casted flexible fish shapes with a stiffer
spine embedded in the models and chose the material
modulus so that the wave speed and body amplitude of
the actuated fish would match a parametric model of fish
swimming. A rod attached to the frontal part of the bodies
and driven by an external motor actuated the models by
applying a moment. Therefore, models had to be towed.
The vibrational response of the models was found
numerically, thus there was a lack of an analytical
solution to which a designer could relate for further
analysis and development. Maximum overall system
efficiency was estimated to be close to 30% at an
actuation frequency of 6.5 Hz , and an actuation angle of
7 o . The net thrust at this efficiency was close to 0.2 N .
Davidson and Julian [5] built a radio controlled
maneuverable fish with a flexible actuated tail. However,

no publications were found with details of their design

methodology and performance measurements.
Blickhan et al. [6] did studies on the benefits of elastic
energy storage in steady swimming. Based on dolphin
swimming data they calculated optimal spring
compliances. The values found were similar to measured
tendon compliances.
Harper et al. [7] investigated the potential benefit of
elastic energy storage in the propulsion of a swimming
robot. They concluded that the addition of linear springs
to the system provides a means to reduce energy required
of the power supply.
The current work presents the design methodology for a
potentially more efficient class of underwater propulsion
mechanisms using body compliance and natural modes of
vibration to eliminate internal transmission mechanisms
in order to achieve the desired kinematics. We illustrate
the processes of design and fabrication. Finally we
present experimental performance measurements of the
resulting robot fish.

2. DESIGN METHODOLOGY
We want to design a robot that swims in the manner of a
fish. Two important fish swimming modes that are well
understood by the current art are named anguilliform and
carangiform. The former is an oscillatory mode in which
the entire body participates but amplitude increases
towards the rear end of the body. The latter mode is such
that oscillations become significant only in the posterior
half or even third part of the fish length, the front of the
fish almost does not move. Studies point to the idea that
carangiform swimming mode is more efficient and can
achieve higher speeds [8]. Therefore, we chose to design
a robot that uses the carangiform swimming mode. Our
design methodology consisted of fabricating a robot with
a rigid anterior part and a flexible posterior part, which
from now on we will refer to as tail. The tail was modeled
as a non-uniform flexible beam and the forced vibrations
response was derived and used to design a tail whose
dominant modes of vibration corresponded closely to a
kinematic model of body deformation in a fish. According
to [1], one such model of body deformation is a traveling
wave that increases in amplitude from the nose to the tail
of a fish and is given by:

yb (xb , t ) = c1 xb + c2 xb

)[sin(K x )cos( t ) + cos(K x )sin( t )] (1)

b

where yb is the lateral displacement of the fish body at a

distance xb from the fish nose. We will assume that the
values of the linear wave amplitude envelope c1 , the
quadratic wave amplitude envelope c2 , the body wave
number K , and the body wave frequency are known

for a given fish (or application). Our thesis is that the

actuation of a mechanism whose principal modes of
vibration correspond to the desired motions in (1) has the
potential of requiring less energy than actuating a
manipulator type design such as in [1,3] where motions of
each link are controlled to achieve the desired kinematics.
We argue that required trajectories can be achieved
using only one actuator instead of a discrete number in the
case of a manipulator type design. Furthermore, internal
transmission elements are eliminated. In the next section
we develop a dynamic model of the tail that will be used
for the design.

3. TAIL MODEL
Rigid

Flexible

( )

y x, t

lr
Figure 1: Schematic of robot body with associated reference frame for
modeling.

Figure 1 shows a diagram of the basic configuration of the

robot together with the reference frame used for the
development of the model. Since our design exploits the
carangiform swimming mode, the posterior third of the
body is stiff whereas the anterior two thirds of the body
are flexible, lr denotes the length of the rigid part of the
robot body ( x = xb lr ). The manner in which we achieve
this will be further explained in section 4. The robots tail
can be modeled as a Bernoulli-Euler beam with varying
cross-sectional area. We consider this model appropriate
since for most fish the tail amplitude of motion is not very
big compared with their length (the excursion of the tip of
the tail is approximately 0.14 of the overall body length)
[9]. Therefore, all the mass elements in the flexible tail
can be assumed to move in pure translations. In addition,
most of the potential energy is due to bending rather than
shear.
l

y
2r

(t )

( )

y x, t
a
Figure 2: Tail model

Figure 2 shows the tail modeled as a beam with variable

cross-section. Parameters l , and a , denote the length of
the tail and the distance from the tail base at which a timevarying actuation moment M acts. The forced vibrations
of a non-uniform beam moving steadily in a fluid are
described by the following partial differential equation

A(x )

2 y 2
2 y
+ 2 EI (x ) 2 = f (x, t ) (w) ,
2
t
x
x

where y is the lateral displacement of a tail element

relative to a reference frame fixed to the tail as displayed
in figure 2. The density and the modulus of elasticity
E of the tail are taken to be constant throughout the tail
continuum. The cross-sectional area A and the second
moment of cross-sectional area I of the tail vary with x .
A distributed force f is used to generalize the possible
actuation forces. The resistive forces (w) due to
interaction with the fluid environment depend on the
normal component w of relative velocity between the tail
and the fluid environment. For an undulating body with
body wave speed V and mean swimming speed
U relative to the stream, w is given by
y U
y
w=
.
1 =
t V
t
The resistive forces dependence on w can be
approximated to be linear at very low Reynolds numbers
and is generally taken to be quadratic at high Reynolds
numbers. We are interested in the case of swimming at
high Reynolds numbers thus we should model the
resistive forces as some sort of form drag. However, the
resulting PDE would be nonlinear and finding a closed
form solution becomes more difficult. Thus, we
approximate the resistive forces by a linear dependence
(viscous drag). This underestimation of the resistive
forces will tend to yield results that are overestimations of
the actual behavior, but it would allow us to find a closed
form solution for the predicted transverse motions of the
tail. Therefore, if we assume that the flow around the fish
is steady ( = cst.) , we can model the resistive forces by
a term analogous to viscous damping, with a damping
coefficient that varies with x since the area of the tail
changes with x as well:
y
2 y 2
2 y
A(x ) 2 + 2 EI (x ) 2 = f (x, t ) C (x ) . (2)
t
t
x
x

The transformation leading from (2) to (4) is explained

in [10,11]. It involves essentially a change of variable
u = y , where the functions are polynomials in x of
at least second order. In our analysis we assumed a
quadratic variation of the tail cross section since this
assumption is not terribly unrepresentative of real fish
profiles. Thus, we use:

x
l

Where A0 , I 0 , and C0 are the cross sectional area, the

second moment of cross-sectional area, and the damping
coefficient respectively at x = 0 . Equation (4) can be
solved using the method of virtual work. The transverse
deflection response, as expressed with the change of
variable, is expanded in series
u=

y
EI (x ) 2 = 0
x x =l

y
EI (x ) 2 = 0.
x
x x =l
2

(6)

We define a virtual displacement for the ith mode to be

ui = i X i . The sum of all virtual works in the system
should add up to zero as explained in [12]. The response
of the ith vibrational mode can be found by using the
Duhamel integral [12]. Reverting the change of variables
we find the response to be
c0

y ( x, t ) =
i =1

X i X ia e 2 A
A0 pi

c0

2 A
e M a ( )sin pi (t )d .
t

(7 )

In our actuation scheme, the moment M a applied at a

distance a from the tail base is chosen as
M a (t ) = M sin ( t )
(8)
Combining equations (7) and (8) , and evaluating the
integral we get the following expression
y ( x, t ) =

f (x )e [g (t ) + e h (t )]

Nt

Nt

(9)

i =1

f i (x ) =
g i (t ) =

(3)

Finding a closed form solution for the problem defined by

equations (2) and (3) is not straightforward. However,
closed form solutions exist for the uniform counterpart of
the PDE given by
2u
4u
u
A0 2 + EI 0 4 = f (x, t ) C0
(4)
t
x
t

(t )X (x )
j =1

where,
The boundary conditions for this case (clamped-free) are,
y
( y )x=0 = 0
=0
x x =0

= 1 , A = A0 , I = I 0 2 , C = C0 . (5)

hi (t ) =

X i X ia M
,
A0 pi

N=

C0
2 A0

(
) + 2(N

2 Npi cos( pi t ) + N 2 pi + 2 sin ( pi t )

(N

+ pi

2 2

pi + 4
2

2 Npi cos( t ) + pi N + pi 2 sin ( t )

(N

+ pi

) + 2(N

2 2

pi + 4
2

The steady state response is then

y ss (x, t ) =

f (x ) h (t )
i

(10)

i =1

For our boundary conditions (clamped-free), the

normalized eigenfunctions and angular frequencies are
given by

X i = ai [cos(k i x ) cosh (k i x )] + bi [sin (k i x ) sinh (k i x )]

pi = k i

EI 0
A0

where the ratio of coefficients ai and bi is known, and

1
ki i , [12]. Now for design purposes we would
2 l
like to match the terms in equations (1) and (10) , so we
rewrite equation (10) as

X X
y ss (x, t ) = i ia [Qi cos(t ) + Ri sin (t )]

i =1
where,
2 N M
Qi =
2 2
2
2
A0 N + pi + 2 N 2 pi 2 + 4

((

Ri =

((

(N

A0 N + pi
2

+ pi M
2

) + 2(N

2 2

pi 2 + 4
2

Therefore, for the transverse vibrations of the tail to

match the parametric model of equation (1) , the
following conditions must be met (for x 0 )

X X
2
i ia Qi = c1 xb + c2 xb sin (K xb )
i =1

i =1

X i X ia

(
(

)
)

Ri = c1 xb + c2 xb cos(K xb )
2

(11)

The speed of convergence of the series in

equations (11) depends on the values of the parameters.
Depending on the dimensions and the complexity of the
parametric model that needs to be matched, more or less
terms (modes) might be needed for an approximation to
match the desired kinematics.

4. ROBOT DESIGN AND FABRICATION

4.1 DESIGN
The previous section described the dynamics of motion of
a general continuum whose shape resembles a fish tail.
The tails modes of vibration can be made to match the
parametric model of fish swimming presented in
equation (1) by using the constraints given by
equations (11) in the design of the continuum. The third
equation in (11) is the only explicit condition since it
prescribes the actuation frequency to the body wave
frequency of the fish being modeled. The series on the
left hand side of the first two equations in (11) depend on
nine design parameters: a , A0 , I 0 , , l , E , , M , . If
we want to preserve the geometry of the fish we are

modeling, the parameters A0 , I 0 , , and l are known.

Then the remaining unknown design parameters are the
actuation point a , the modulus of elasticity E , the
density , and the magnitude of actuation moment
M (since is prescribed by the third equation in (11) ).

From the first two equations in (11) we can get the

following condition

X XQ
i

ia

i =1

X i X ia Ri

= tan (K xb )

(12)

i =1

We can see from equation (12) that for an acceptable

approximation, for this particular parametric model of
swimming, at least the first two terms of the series must
be considered since otherwise the left hand side of the
equation would be a constant. Also, the right hand side of
the equation has discontinuities at xb = n 2 K (with
n = 1,3,5,... ). The problem is reduced then to finding a set
of parameters to match the models in the two sides of
equation (12) , achieving the smallest error possible. This
fitting can be done numerically in a least squares sense.
We included the constraint of minimizing the value
for M . The designer can prescribe some of the unknown
variables if he chooses to do so. Also by including the
constraint of minimizing the value of the magnitude of the
actuation moment M we are effectively minimizing the
energy consumption of the design that matches the
kinematic behavior of a fish, since the input power is
Pin = M . The parameters chosen for our design are
listed in table 1. Our design is radio controlled for testing
purposes so the actuation frequency can be set to any
value during operation. However, the results presented in
section 5 indicate that an optimal performance is achieved
using a frequency close to the value prescribed.
Parameter
a
A0

I0

l
E

Value
0.05 [m]

3.04 10 3 [m 2 ]
4.9 10 7 [m 4 ]
0.42
0.1 [m]
0.04 [MPa]

1080 [Kg m 3 ]
1.4 [Nm]

Table 1: Design Parameters

4.2 FABRICATION
An appropriate material can be chosen once we know the
required material properties. The next step is to fabricate a
tail that corresponds to the required geometry. In addition

it is convenient to have rigid parts embedded in the

continuum, which give a good support for actuation. We
use simple casting techniques and shape deposition
manufacturing (SDM) principles, which are explained in
[13]. For most of the parts we use a machinable base,
such as a block of solid wax, where the mold of a given
parts profile can be machined. Next, parts such as
actuators, or sensors, are positioned inside the empty
mold and the required material is poured in resin form.
Once the material cures, some machining can still be done
to complete what becomes a single continuous piece.
Figure 3 shows the tail design. Flexible and stiff sections
are stacked together forming a single continuum. Stiff
sections are used for actuation support, and flexible
sections provide a compliant structure that transmits
momentum to the environment. The harmonically varying
moment input is applied by actuating the stiff plate
embedded inside the tail at a distance a from the tail base.
A servomotor actuates the plate by pulling two
inextensible cables attached to the plate. It should be
noted that while we are still using a classical actuator, in
this case a servomotor, no classical transmission elements
(pulleys, gear boxes, etc.) are present between the
servomotor and the end-effector, in this case the tail. Also
since the design is monolithic no sealing is required for
that section of the body.

Figure 4: Tail fabrication and testing. Clockwise from top left: Rigid tail
components stacked in mold, Components bound by polymer, Top view
of tail reaching maximum deflection, Top view of tail and servo
assembly.
Receiver

Batteries

Figure 5: Complete fish.

Flexible Polymer

5. EXPERIMENTAL SETUP & RESULTS

Cables

Stiff Polymer
Servomotor

In order to measure propulsion performance parameters

such as trust and speed, several experiments were
performed. The experiments where made in a towing tank
where models can be towed at controlled velocities by a
carriage and also swim freely. The tank is 30 m long,
2.5 m wide, and 1.2 m deep. The carriage is suspended
over the testing tank and driven by a flat belt. Output from
sensors can be recorded by the tank data acquisition
system. The sampling frequency used was 200 Hz .

Figure 3: Continuous fish tail design

The main advantages of this type of design are that the

compliance is implemented by the structure itself, and the
mechanism is self-contained. The goal is to eventually use
a more efficient type of actuator. Figure 4 shows the
procedure for the tail fabrication and the actuation of the
finished tail. The tail is casted using a two-part mold. The
rigid parts, made from IE-70D polyurethane, are
sandwiched inside the mold, and IE-10AH polyurethane
(both from Innovative Polymers, Inc., St. Johns, MI) is
poured in between them. The maximum tail bit amplitude
is close to 0.1 m in air, with an input torque of 1.4 Nm .
Figure 5 displays the entire robot body and the location of
some of its components.

Flexure transmission

Load cell
Attachment plate

Figure 6: Test setup and experimental fixture details.

Figure 6 displays a photograph of the mechanical fish

attached to the carriage mast and a solid model diagram
displaying the important features of the setup. A flexural
mechanism was designed to hold the fish underwater
while transmitting the forces acting onto it to a tension
compression load cell without applying any shear to it.

Experiments were made to measure both the thrusts and

velocities achievable with our design. Figure 7 shows
measured static thrusts and average forward velocities as
a function of the actuation frequency. A maximum is
noted in both graphs at an operational frequency close
to 4 Hz . This most likely reflects a combined effect of
reaching the optimum point of the servomotor efficiency
and being close to the prescribed frequency by which the
fish swimming mode is more closely mimicked.

not decrease the overall energy efficiency, which would

defeat the purpose of the design.

7. ACKNOWLEDGEMENTS
The authors greatly appreciate the support of the
Singapore-MIT Alliance program. The authors would also
like to acknowledge Schlumberger Ltd. for the financial
support and the personnel of the MIT ocean engineering
towing tank for allowing us to use their facilities for
experimentation.

REFERENCES

Figure 7: Thrust and average velocities versus actuation frequency.

6. DISCUSSION AND CONCLUSIONS

Measurements of the actual tail mechanism energy
efficiency are still in progress. Once these are finished we
will be able to draw a more conclusive comparison
between our design and classical manipulator designs in
terms of mechanical efficiency. Meanwhile, we have
presented a dynamic model for the forced vibrations of a
damped non-uniform beam, representing the tail of a
robotic fish, and applied it to the design of an alternative
underwater propulsion mechanism. The resulting design
was built and tested showing good performance behavior
compared to the present art. A maximum thrust close to
0.4 N was achieved at an operating frequency of 4 Hz .
Also the peak average forward velocity of 0.095 m / s was
reached at a similar frequency.
The present design eliminates the need for classical
internal transmission elements between the actuator and
the end-effector, which in this case is the whole tail.
Furthermore, the fact that the mechanism is a single
monolithic part eliminates the need for sealing in that
section, and also makes is more robust against failure.
Finally, one of the main limitations of this mechanism is
that it is passive in nature. It is designed to perform under
certain specific environment conditions assumed in the
model. Therefore, as it is, the design cannot react against
disturbances. Hence, future work should address the
incorporation of some active mechanisms inside the tail
continuum. Nevertheless, any mechanism added should

[1] D.S. Barrett, Propulsive efficiency of a flexible hull

underwater vehicle. PhD thesis, Massachusetts Institute of
Technology, 1996.
[2] J.M. Anderson, Vorticity control for efficient
propulsion. PhD thesis, Massachusetts Institute of
Technology, 1996.
[3] J.M. Kumph, The design of a free swimming robot
pike. B.S. thesis, Massachusetts Institute of Technology,
1996.
[4] K.M. Mcletchie, Drag reduction of an elastic fish
model. B.S. thesis, Massachusetts Institute of
Technology, 2002.
[5] N. Davidson, N. Julian, www/me.berkeley.edu/hel/
calibot.htm
[6] R. Blickhan, J.-Y. Cheng. Energy storage by elastic
mechanisms in the tail of large swimmers a reevaluation. Journal of Theoretical Biology, 168:1994,
315-321.
[7] K.A. Harper, M.D. Berkemeier, S. Grace, Decreasing
the energy costs of swimming robots through passive
elastic elements, Proc. IEEE Int. Conf. On Robotics and
Automation, Albuquerque, NM, 1997, 1839-1844.
[8] J. Lighthill, Mathematical Biofluiddynamics
(Philadelfia, PA: Society for Industrial and Applied
Mathematics, 1975).
[9] J.J. Videler, Fish swimming (London: Chapman &
Hall, 1993).
[10] H.P.W. Gottlieb, Comments on Vibrations of Nonuniform Beams and Rods, Journal of Sound and
Vibration 195, 1996, 139-140.
[11] S. Abrate, Vibration of Non-Uniform Rods and
Beams, Journal Sound and Vibration, 185(4), 1995, 703716.
[12] S. Timoshenko, D.H. Young, W. Weaver Jr,
Vibration problems in engineering (John Wiley & Sons,
1974).
[13] J. G. Cham, B.L. Pruitt, M.R. Cutkosky, M. Binnard,
L.E. Weiss, G. Neplotnik, Layered manufacturingwith
embedded components: Process planning issues, Proc. Of
ASME Design Engineering Technical Conference, Las
Vegas, NV, 1999.