Sie sind auf Seite 1von 8

2013 IEEE International Conference on Robotics and Automation (ICRA)

Karlsruhe, Germany, May 6-10, 2013

A Ball-Shaped Underwater Robot for Direct Inspection of Nuclear


Reactors and Other Water-filled Infrastructure
Anirban Mazumdar, Aaron Fittery, Wyatt Ubellacker, and H. Harry Asada
Abstract In this paper we present a new type of spherical
underwater robot that is completely smooth and uses jets to
propel and maneuver. This robot is specifically designed for
the direct visual inspection of water-filled infrastructure such
as the inside of nuclear powerplants. The unique propulsion
architecture consists of a single bidirectional centrifugal pump
combined with two fluidic valves. The pump is used to produce
a high velocity jet while the valves are used to quickly switch the
jet between output ports. The spherical shape means that the
robot is simple to model and control, maneuverable, and robust
to collisions. The propulsion architecture is described in detail
along with a rigid body model for maneuvering control. A novel
valve PWM controller is used to achieve heading control, and
the controller performance is confirmed with both simulation
and experiments. Finally, experiments are used to illustrate the
turning and diving performance of the robot.

I. INTRODUCTION
Inspecting infrastructure has become an issue of increasing
relevance as the many plants and structures built in the 20th
century start to age. One particular area relates to water
filled infrastructure such as dams or ports. In addition water
piping systems are an area where improving maintenance
and reliability can directly impact the lives of many.
Another unique case where the inspection of water-filled
infrastructure is important is the inspection of the internal
components of boiling water reactor (BWR) power systems.
Inspections are generally carried out when the reactor is shut
down for refueling. The water filled pool that comprises
the reactor system requires thorough inspections and is a
growing area of robotics research.
The robotics community has responded quite well to these
new challenges, and new robots have been proposed and built
for the inspection of dams [1], ports [2], [3], water-filled
pipes [4], and nuclear power systems [5], [6]. Some key
challenges that still exist include accessing and inspecting
complex environments where small size and high maneuverability are required. In addition, some infrastructure such
as nuclear plants are subject to Foreign Material Exclusion
(FME) rules. This means that outside materials cannot be left
behind even if they are very small. Therefore the inspection
robots must be very robust to collisions.
A spherical shaped robot propelled by jets rather than
propellers would appear to be an intuitive approach. A
spherical shape can be designed to be quite strong, and
if no appendages such as fins are used, there exist no
A. Mazumdar, A. Fittery, and W. Ubellacker are with the Department
of Mechanical Engineering, Massachusetts Institute of Technology, 77
Massachusetts Ave., Cambridge, MA 02139 amazumda@mit.edu
H. Asada is with Faculty of Mechanical Engineering, MIT, 77 Massachusetts Ave., Cambridge, MA 02139 asada@mit.edu

978-1-4673-5643-5/13/$31.00 2013 IEEE

obvious stress concentrations. In addition, a spherical design


is attractive because the symmetry simplifies the dynamics
and control as well as turning ability. In addition, the Munk
moment which tends to destabilize elongated shapes such as
ellipsoids and torpedoes is eliminated for a spherical design
[7]. This simplifies vehicle control.
As a result of these positive attributes, there have been
several examples of spherical underwater robots. A few, like
MITs Eyeball ROV [8], and the University of Hawaiis
ODIN Robot [9] use propeller thrusters for propulsion.
This is the most efficient approach and is therefore the
best approach for many applications where size or loitering
ability is required. However, the presence of propellers can
complicate the outside shape and if not incorporated properly
can also present a risk of damage from collisions.


Fig. 1.

Photograph of the spherical robot prototype.

A different approach is to use water jets to provide the


propulsive forces to the vehicle. One variation is outlined
in [10] where the authors use mulitple pumps manipulated
by servo-motors to achieve multi-dof capability. This system
works well and is verified experimentally. A similar system
for space applications is the SPHERES testbed that has been
shown to achieve agile motions using gas-jet actuators [11].
We propose a different approach: using a central pump as
the main propulsor and then directing the output jet based
on the desired direction using high speed fluidic valves. Due
due to their simplicity, size, and high performance, fluidic
valves present an attractive alternative to additional pumps or
servomotors. Previous work from our group has shown how
fluidic valves can be carefully designed to maximize output
force [12]. In addition, preliminary experimental results have
illustrated that these pump-valve systems are capable of
achieving high performance closed loop control [13].

3415

In this paper we present a new type of spherical underwater robot that uses only a single pump and two fluidic valves
to achieve three dimensional motions. Previous works such
as [14] and [15] have focused on ellipsoidal vehicles that
require careful feedback controller design for stable motions.
In addition, these vehicles used multiple pumps or propellers
for multi-DOF performance. The new spherical design and
analysis presented in this paper illustrates the potential of
a different approach to underwater vehicle design. Singlepump architectures have the potential to greatly reduce size
and complexity. In addition, spherical designs offer superior
capability due to their symmetry, minimal drag, and reduced
added mass.
As illustrated in Fig. 1, this robot is completely spherical and has no external protrusions or appendages. In the
following we will describe the design techniques used to
specifically tailor the fluidic valves for use in a small robot
system. A spherical vehicle design that uses only a single
pump is proposed and design tradeoffs will be discussed.
A planar maneuvering model will be used to design a
heading controller for the robot and its performance will be
evaluated. Finally, the prototype robot is described in detail
and experimental data is used to illustrate its performance.
II. P UMP J ET P ROPULSION S YSTEM
The goal of our design approach is to achieve forces
in multiple directions while minimizing size, weight and
complexity. Forces in two directions can be achieved by
cleverly designing a pump impeller and its housing. The
output direction of the pump is controlled by changing the
direction of the impeller. Pumps or bow thrusters with this
configuration are commercially available and have been used
for underwater robots such as [10].
The use of a bidirectional pump enables the generation
of jets in two directions. This means that additional systems
are needed to achieve a multi-DOF robot. Several choices
exist. First, multiple such pumps can be added. This is an
intuitive approach and simplifies control over the vehicle.
However, minimizing the number of pumps is an important
challenge. The pumps described in this work only provide
about 50 percent of the maximum force of a similarly sized
propeller. Therefore to achieve comparable forces, pump-jet
systems must be larger than comparable propeller systems.
In addition, with their seals, bearings, and controllers, pumps
can be heavy, bulky, and expensive.
A second approach is to use thrust vectoring. This involves
using a second actuator (usually a servomotor) to change
the output force direction by moving the output nozzle. This
approach is challenging because the jet vectoring mechanism
can become bulky and heavy. Servomotors by themselves can
be complex and cumbersome to fit into small spaces.
A. Bistable Wall Attachment Amplifier
Our approach is different from either of these. We draw
inspiration from fluidic technologies to achieve multidirectional forces. Specifically we focus on a device known as
a bistable wall attachment amplifier [16] which uses the

Coanda Effect to bend jets along curved surfaces. This


technology has existed since the 1970s and similar devices
and principles have been used for a range of applications
from limb system identification ([17], [18]) and gas jets [19]
to thrust vectoring systems for aircraft [20]. The key principle
is outlined in Fig 2. The device sits in the ambient fluid
and control ports C1 and C2 open and close to the ambient
fluid. A high velocity jet enters the device, and, based on the
configuration of the control ports, bends towards either Exit
E1 or Exit E2 . If C1 is closed and C2 is open, the jet bends
towards E1 due to the Coanda Effect. Similarly, if C2 is
closed and C1 is opened, the jet will bend towards E2 . For
more details on selection of geometry and miniaturization
see [12], [16], [17].
This approach has several key attributes. First, the jet
switches direction very quickly. Experimental results have
illustrated that the valve can switch the jet direction in
about 0.11s. This is substantially faster than the dynamics of
centimeter scale underwater vehicles. For the robot described
in this paper, a characteristic time for turning maneuvers is
approximately 1.4s.
In addition, switching the jet direction is very simple;
it only requires covering and uncovering two small holes.
A very simple mechanism can be fabricated using a single
micro DC motor. This switching motor can be very small
and can be controlled using simple on-off techniques.
Finally, the high speed switching of the valve can be used
to compensate for potentially challenging pump dynamics
such as nonlinear performance or dead zones. The idea is
that the pump could be run continuously at a constant speed
while the valve would used to modulate the force using
PWM. Previous work has used this idea to achieve improved
feedback control performance [13].







W





W


/Y


W



W


/Y

Fig. 2. Diagram illustrating how a fluidic device can be used to switch a


high velocity jet.

B. Propulsion Architecture
Combining a bidirectional pump with two fluidic valves
gives us the ability to produce jets in four directions. Figure
3 illustrates how we achieve three dimensional motions using
just one pump. We divide the motions into two classes. The
first type of motions are related to vertical motions (diving
and surfacing). We plan to design the vehicle to be neutrally
buoyant, so we use Valve 2 to provide forces in the z

3416

direction. Jet 3 is used to surface and Jet 4 is used to dive.


Note that there is no control over pitch or roll. We assume
that the center of mass is placed slightly below the center
of the vehicle which adds a stabilizing moment against pitch
and roll.
For the most part, gravity will not act to provide stabilizing
moments for yaw. This means that the vehicle must have the
ability to correct its yaw angle using the jets. We achieve
this by using Valve 1 to create two jets that both exit the
rear of the vehicle. Jets 1 and 2 both act to push the vehicle
forward, but they also can be used to adjust the yaw angle.
Note that this means that in order for the vehicle to travel
straight, Valve 1 must switch symmetrically between Jets 1
and 2 at high speed. This is one example of how we exploit
the high speed nature of the fluidic valve. Also note that
with this architecture the vehicle has no control authority in
the x direction. We address this problem by demonstrating
the ability of the vehicle to turn 180 degrees and reverse
direction very quickly.
>&

:

s
 

s
:

:
:

>&
ds

CD =

FD
1
A
c CD u|u|
2

(1)

The expression for the drag coefficient is taken from [24].


The expression equates the drag moment, MD , to the drag
moment coefficient, CM , based on the fluid density, the body
diameter, D, and the yaw rate, r.
CM =

MD
1
2 CM


D 5
r|r|
2

(2)

We develop the dynamic equations using the techniques


outlined in [22]. We use the same xyz body fixed coordinate
frame outlined in Fig. 3. In accordance with ocean engineering conventions, we use u, v, w to denote translational
velocities in the x, y, and z axes respectively. Similarly, we
use p, q, r to denote rotational velocities about the x, y, and
z axes respectively. The translational equations of motion are
provided in eqs. 3-5.

s

s

The spherical shape also simplifies the analysis. The


vehicle is assumed to have mass m, and there is only a single
moment of inertia, I. Similarly, the added mass, ma , is the
same for all directions and there is no added mass associated
with rotations [23]. There are only two drag coefficients to
identify, the one associated with translations, CD , and the
one associated with rotations, CM . We use the common
quadratic drag models. An expression for CD is provided
based on the drag force, FD , the fluid density, , the cross
sectional area, Ac , and the surge velocity, u.

Zs


du
1
du
+ qw rv = FJ 1 + FJ 2 ma
Ac CD u |u|
dt
dt
2
(3)


dv
dv
1
m
+ ru pw = ma
Ac CD v |v|
(4)
dt
dt
2


dw
dw
1
m
+ pv qu = FJ 4 FJ 3 ma
Ac CD w |w|
dt
dt
2
(5)
m

Fig. 3. Diagram illustrating how fluidic valves can be combined with a


bidirectional pump to achieve multiple jets.

III. V EHICLE M ANEUVERING M ODEL


A. Full Nonlinear Model
Before examining the vehicle dynamics, we describe several key assumptions used to simplify the nonlinear equations.
1) We assume the dominant hydrodynamic forces are
from added mass and drag.
2) We assume that the dominant drag is quadratic due to
a large Reynolds number ( 10, 000).
3) The center of mass is located at the geometric center
of the vehicle.
4) Pump and valve dynamics are sufficiently fast to not
interfere with vehicle dynamics.
In reality, the center of mass is not exactly positioned at
the geometric center of the vehicle. We use the center of
mass to add stability against roll and pitch by placing it
slightly below the geometric center. We assume this offset is
very small compared to the length scale of the vehicle and
therefore treat the center of mass and the geometric center as
coincident. This assumption greatly simplifies the governing
dynamics.

Similarly, the rotational equations of motion are provided


in eqs. 6-8. Note the absence of the Munk moment, a moment
that acts to destabilize streamlined vehicles. The symmetry
of the spherical shape means that the Munk moment is zero
which simplifies control.
 5
dp
1 D
=
CM p |p|
dt
2
2
 5
dq
1 D
I
=
CM q |q| FJ4 LF, x + FJ3 LF,x
dt
2
2
 5
dr
1 D
I
= (FJ1 FJ2 ) LF,y
CM r |r|
dt
2
2

(6)
(7)
(8)

B. Linearized Planar Model


Since the robot is completely smooth and symmetric, it
does not have a preferred direction. This means that unlike
vehicles with tail fins, this robot will not maintain a heading.
This means that designing a feedback controller for heading

3417

is the first priority. As illustrated above, controller design


is complicated by the nonlinear and coupled nature of the
governing dynamic equations. We choose to linearize about
the equilibrium state where the vehicle is moving forward
with some cruising speed, Uc . All other velocities (v, w, p, q,
r) are treated as very small. The forces required to maintain
this cruising state are FJ1,0 and FJ2,0 . For the sake of clarity
we use to denote the small perturbations that comprise the
linearized dynamics.
FJ1 = FJ1 FJ1,0
FJ2 = FJ2 FJ2,0

(9)
(10)

The linearized equations of motion are provided in eqs.


11-13.
du
(m + ma ) = (FJ1 + FJ2 ) Ac CD Uc u (11)
dt
dv
(m + ma ) = mUc r
(12)
dt
dr
I
= (FJ1 FJ2 ) LF,y
(13)
dt
To decouple the dynamic equations we use two new inputs.
The average jet increment, FJ,av , and the difference in the
jet increments, FJ .
FJ1 + FJ2
FJ,av =
(14)
2
FJ = FJ1 FJ2
(15)
Using this approach, we can decouple the surge (x) dynamics from the sway-yaw dynamics. The surge or forward
motion of the vehicle can be controlled using FJ,av , while
FJ can be used to control yaw.
We then turn our focus to the sway (y) and yaw (r)
dynamics. State space methods such as full state feedback are
complicated by the need to measure the sway velocity, v.
This value is small and is non trivial to measure or estimate.
Instead we will focus on designing a controller that only
requires feedback on the yaw angle , and the yaw rate,
r. Since we want to stabilize the heading, we are left with
the following dynamics for yaw.
2

I (s) s = FJ (s)

C. Feedback Control Using a Single Jet


We model the pump force, FP , as being controlled by a
voltage VP which can be approximated using a linear gain.
FP = GF VP

The jet switches between FJ1 and FJ2 based on Valve 1.


The planar motions of this vehicle are predicated around the
ability of Valve 1 to switch between Jets 1 and 2 very quickly.
Our preferred technique for exploiting the high speed valve
switching is to use PWM (Pulse Width Modulation) on the
valve. This involves using a square wave at a set frequency
and then modulating the signal using the duty cycle, d. We
choose this approach over the simpler relay control (on-off
control) because using a square wave of fixed frequency
reduces high frequency switchings (chattering).
We employ this basic technique to make the robot move
forward. The pump is run at some nominal voltage (voltage
is proportional to jet force), VP 0 , associated with the desired
cruising speed, Uc . To go perfectly straight the valve is
switched between Jets 1 and 2 with a 50 percent duty cycle.
The question that arises is how to achieve turning motions.
The duty cycle, d, is used to create turning motions by
modulating the value of FJ per cycle. The value for d
ranges from 0 to 1. As Fig. 4 shows, a duty cycle of 0.5
(50 percent) creates a zero value for FJ while increasing
or decreasing d can be used to linearly modulate FJ .
The advantage to this approach is that pump dynamics and
nonlinearities can be ignored because the pump simply runs
at a constant speed.

mUc
v(s)
=
(17)
(s)
m + ma
This means that when the vehicle changes angle while
moving forward it will develop a constant sway velocity. The
negative sign means that as the vehicle turns it gets pushed
radially outwards due to centrifugal effects. This is damaging
for precision navigation tasks. The model performance is
complicated by the quadratic drag terms disappearing during
the linearization. Physical intuition tells us that the sway
velocity will eventually go to zero due to drag. This will
be discussed further in the simulation section.

FP

Pump

GFVP0
VP0

Valve

d
FJ

dTPWM

(16)

The dynamics for are relatively simple, and a PD controller can be designed to get a suitable response. However,
the sway dynamics are less well behaved.

(18)

TPWM
sWtD

/ 

Fig. 4. Illustration of Valve PWM (a), and the input output mapping for
control (b).

IV. PWM PARAMETER S ELECTION


A prototype vehicle was designed and modeled. The
vehicle properties are provided in Table I. We choose to
operate at Uc = 0.15m/s which corresponds to roughly
90 percent of the maximum speed. Based on this speed we
can use the techniques in [13] to estimate the peak to peak
oscillation amplitude (A ). We first declare a few variables
to simplify the expressions. We use BZ to represent the drag
coefficient for yaw rotations, and N0 to represent the input

3418

TABLE I

moment. The input moment is dependent on the cruising


speed Uc because the same jets are used for forward motion
as well as for turning motions.

SUMMARY OF PHYSICAL PROPERTIES FOR PROTOTYPE


ROBOT.

Parameter
D
Ac
m
I
ma
CD
CM
LF,x
LF,y
Kc
fP W M
Uc

1
CM D5
(19)
2
1
N0 = CD Ac Uc2 LF,y
(20)
2
We then combine these expressions and integrate the
nonlinear differential equation governing r dynamics twice
over a quarter period. This can be used to approximate the
size of the vehicle angle oscillations in response to a PWM
input.



N 0 BZ T P W M
2I
A
log cosh
(21)
BZ
I
4
BZ =

We can use eq. 21 to generate a family of curves that


illustrate the speed-oscillation tradeoff. If maximum speed is
desired, reducing oscillations is challenging, and the PWM
frequency must be relatively high. Similarly, oscillations can
be reduced by reducing the cruising speed and operating
at a lower frequency. This combination is perhaps best
for efficient, long distance travel. The relationship between
oscillation amplitude, A , cruising speed, Uc , and PWM
frequency fP W M , is illustrated graphically in Fig. 5.

A [Deg]

16

Uc = 0.1 m/s

14

Uc = 0.125 m/s

12

Uc = 0.15 m/s

10
8
6
4

K

2
0
0.5

1.5

fPWM [Hz]

2.5

Fig. 5. Performance curves illustrating the speed-oscillation tradeoffs. Note


our choice of operating frequency based on speed and oscillation level.

Based on this analysis, fP W M , was chosen to be 1.4Hz


in order to achieve 2 peak to peak oscillations. The choice
of 2 peak to peak was arbitrary, higher frequencies would
result in smaller oscillations but also greater power consumption. Future studies will focus on the tradeoffs between
oscillations, speed, and power consumption.
V. S IMULATION
The modeling techniques described previously were used
to generate a simulation environment in M ATLAB. The
vehicle parameters were chosen to match those of the design.
A PD controller was designed using the linear model to
stabilize yaw, and relatively low gains were chosen to avoid
issues with actuator saturation later. This controller was
simulated using both the linear model as well as a full

Value
0.106[m]
0.0088[m2 ]
0.626[kg]
7 104 [kgm2 ]
0.313[kg]
0.75
0.75
0.03[m]
0.01[m]
0.002[N m/V ]
1.4[Hz]
0.15[m/s]
1000[kg/m3 ]

nonlinear simulation. The full nonlinear simulation includes


actuator dynamics and saturation as well as the quadratic
drag and couplings that were in the models described in
eqs. 3-8. In addition, the nonlinear simulation includes a
model for the Valve PWM control approach and includes
the switching dynamics associated with this mode. Therefore,
the nonlinear simulation serves as a useful predictive tool for
examining the system performance quickly.
The results of the simulations are provided in Fig. 6.
The linear simulation matches the full nonlinear performance
quite well. The full nonlinear system is a bit slower due to
actuator saturation and drag. Also note the oscillations in the
steady state response. The oscillations are approximately 2
in peak to peak amplitude. These oscillations stem from the
switching of the valve as discussed previously.
Figure 6-b illustrates the response of the sway velocity, v.
For the linear case, the sway velocity remains constant even
as the angle settles. This was discussed in the linearization
section and his behavior is undesirable because it means the
vehicle is drifting. Fortunately, as the nonlinear simulation
illustrates, the drag serves to counter this motion and slowly
drives v to zero. This corresponds to our physical intuition.
Work is underway to improve performance. A dedicated y
thruster is required for controlling the sway, but achieving
this in a compact design remains challenging.
VI. V EHICLE P ROTOTYPE
A prototype vehicle was designed and fabricated. The
robot is 106mm in diameter and has a mass of 636g in water.
A photograph of the outside of the vehicle is provided in
Fig. 1. Due to the curved shapes and complex contours of
the fluidic valves, 3D printing was used extensively for the
fabrication of the prototype. As Fig. 1 illustrates, two 3D
printing techniques were used for fabrication. The bottom
(white) half of the robot was printed using a Dimension
SST 1200es FDM style 3D printer. This printer uses ABS
plastic and the parts are quite strong. However, due to the
3D printing technique, the parts are porous and therefore
unsuitable for any housings that need to be waterproof. The
top (silver / clear) houses the electronics, microcontroller,

3419

-0.05

60
40

-0.1

-0.15
Full Nonlinear Simulation
d(t)

20
0
0

Linear Simulation
2
3
4
Time [s]

-0.2
0



Full Nonlinear Simulation


Linear Simulation
2
3
4
5
Time [s]

^

Fig. 6. Simulation results for the heading controller. The angle tracking (a)
and sway response (b) are shown for both the linearized and the nonlinear
cases.

and battery and therefore must be waterproof. This half was


fabricated using a Viper Stereolithography technique.
A powerful TCS M400S submersible micropump is used
to generate the jet. The two fluidic valves were designed
and 3D printed by our group, and each is controlled using a
small DC motor. A microcontroller with custom written code
is used to control the robot, and angle sensing is provided
by a small inertial measurement unit (IMU). The IMU angle
estimate does have small amounts of drift (usually about 1
degree of error for a 20s trial). An onboard compass can be
used to compensate for drift but requires careful calibration.
For the purpose of experiments, a wireless radio link is
used to communicate with the robot. Radio communication
would clearly be challenging if not impossible for many
applications and this is an area of active research by our
group as well as many others [21].
Figure 7 provides a photograph of the inside of the robot.
The two valves and the pump are clearly visible. The robot
also contains an onboard camera. This camera can record to
flash memory which can then be recovered later. The robot
is powered using a small lithium polymer battery (LiPo), and
has approximately 30 minutes of lifetime.

Several experiments were performed to validate both the


vehicle design and control system design. The prototype
robot was tested in the small test tank at MITs dArbeloff
Laboratory. The first tests were performed to evaluate the
basic performance of the vehicle. Two key tests were the
straight test and the disturbance test.
The straight test involved simply commanding the robot to
start from rest and move straight along the desired heading.
First the robot was controlled without feedback. As the angle
results in Fig. 8-a illustrate, the smooth spherical shape does
not maintain a heading well without closed loop control. A
PD Valve PWM controller was designed and the gains were
chosen using the techniques described previously. The gains
for the final implementation were approximately 50 percent
of those used in the simulations. As Fig. 8-a illustrates, the
closed loop controller does a far better job of maintaining a
heading, with maximum steady state errors of approximately
4 . These errors are primarily due to the oscillations from
the PWM controller.
The disturbance test consisted of disturbing the robot
with an external force while it was moving. In this case
disturbances were applied by the researchers using their
hands to bump the robot. As Fig. 8-b illustrates, the controller
is able to counter the disturbance and maintain its heading.
80
60

50

Open Loop
d(t)

d(t)

40

Closed Loop

(t)

30

40

[Deg]

80

VII. E XPERIMENTAL R ESULTS


A. Stability Tests

[Deg]

v [m/s]

[Deg]

100

20

20
10

0
-20
0

4
6
Time [s]

^d

-10
0

4
Time [s]

Z

Fig. 8. Experimental data illustrating the results of the (a) straight test,
and (b) disturbance test.

K
s
W

s

Fig. 7.

Photograph of the robot prototype.

B. Turning Tests
One of the potential benefits of using a spherical shape
is the low resistance to turning motions. We examined
this by subjecting the robot to several turning tests. In
order to remain within the context of our controller design
the vehicle was commanded to turn while maintaining its
forward velocity. The first test was a 180 turn. This is an
important study because this robot design does not have the
ability to reverse thrust. Therefore, the only way to avoid
collisions or stop very quickly is by turning 180 degrees.
Figure 9-a provides video trajectory tracking of the robot
performing such a maneuver. Notice how the robot turns
fairly quickly and is able to reverse direction and return to
a position approximately 10cm from where it began.
3420

0.8

X [m]
Y [m]

s:

Distance [m]

0.6

z
y
d

0.4
0.2
0
-0.2
0

sW

Time [s]

10

sd

15

dd

Fig. 9. Illustration of the vehicle performing a 180 turn. Part (a) shows
the XY trajectory from the video, and part (b) shows the time trajectories.

The heading tracking of the controller during such turns


is also important. To this end the vehicle commanded to
perform both 90 and 180 turns while traveling at a cruising
speed. As the results in Fig 10 illustrate, the PD Valve PWM
controller is able to track the heading command and achieve
very small steady state errors.
120
100

200
150
[Deg]

[Deg]

80
60

Fig. 11. Video data illustrating diving capability. A small fish tank was
used for this experiment in order to film properly.

would likely create challenges for any pilot or user. These


oscillations can be reduced substantially by increasing the
PWM frequency or by moving at a slower speed.
A different perspective is that the oscillations may not
be entirely negative. Oscillations can perhaps increase the
coverage area of the camera during inspections, and they
may also have fluid dynamic effects. For example, an area
of further exploration is studying whether vortices can be
created along the trailing surfaces of the robot due to the
oscillatory motions. This could have ramifications in terms
of improving hydrodynamic efficiency.

100

40

VIII. C ONCLUSIONS
50

20
0
0

(t)
d(t)
5
Time [s]

Fig. 10.

10

0
0

(t)
d(t)
5

10
Time [s]

15

Yaw angle data for a 90 and a 180 turn.

C. Diving Tests
Preliminary data was also obtained on the diving capability
of the robot. For the purposes of these experiments, the robot
was allowed to be slightly positively buoyant. Nonetheless
the vertical jets (Jets 3 and 4) were sufficiently powerful
to allow the robot to dive and surface. Figure 11 provides
an example of video data obtained from such an experiment.
While the diving mode does not cause the vehicle to pitch or
roll substantially, it does induce small yaw moments which
causes the vehicle to rotate. Since the Jets 1 and 2 are
inactive during a dive or surface motion, the robot can only
be reoriented once it reaches the desired depth. This is one
clear downside to the simple 1 pump architecture.
The video submitted along with this paper provides more
visual illustrations of the motions described in this paper.
D. Discussion
While the experimental results are for the most part
promising, there are clear areas that require further exploration. One area is the oscillations that are caused by
the Valve PWM control. These oscillations are generally
viewed as a negative; they cause errors in the heading and

This paper has described the design, control and performance of a new type of spherical underwater inspection
robot. The robot is completely smooth and is therefore
maneuverable, simple, and robust. Moreover, the robot uses
only one bidirectional pump for propulsion, thereby increasing simplicity. High speed fluidic valves are used to route
the jets in order to achieve motions in multiple directions
including surge, heading control, and diving. The spherical
shape greatly eases modeling, and a linear heading controller
was designed. The controller was implemented using a Valve
PWM control algorithm, and experimental results illustrate
the efficacy of this approach. Experimental results also
confirm the turning performance and diving capability of this
robot design. Future work will focus on designing trajectories
to take into account the advantages and limitations of this
propulsion architecture.
The scaling of this approach is an important consideration.
While the fluidic valves can be scaled to both larger and
smaller sizes, scaling upwards is the most practical for
underwater robots. Initial research has already explored using
larger pumps with such valves. Such designs hold particular
promise because the size and weight of the valves do not
scale in the same way as pumps. Specifically, the switching
motor is the heaviest component and can be used on larger
valves. This means that by combining fluidic valves with
larger pumps, significant size and space improvements could
be achieved when compared to multi-pump designs.
Finally, the techniques and technologies outlined in this

3421

paper can be applied to different pump-valve architectures.


For example, current work has focused on developing a
similar spherical vehicle using two pumps. One pump will
be for z axis motions while the other will be for xy motions.
This robot would be capable of more degrees of freedom and
would be able to maintain heading even during dives.
IX. ACKNOWLEDGMENTS
This material is based upon work supported by the Electric
Power Research Institute (EPRI). In addition, the authors
gratefully acknowledge the National Science Foundations
support in the form of a Graduate Research Fellowship.
Finally, the authors thank our colleagues at MITs Bioinstrumentation Laboratory for their assistance with rapid
prototyping.
R EFERENCES
[1] P. Ridao, M. Carreras, D. Ribas, R. Garcie, Visual Inspection of
Hydroelectric Dams Using an Autonomous Underwater Vehicle,
Journal of Field Robotics, vol. 27(6), 2010, pp 759-778.
[2] J.A. Ramirez, R. Vasquez, L. Gutierrez, D. Florez, Mechanical/Naval
Design of an Underwater Remotely Operated Vehicle (ROV) for
Surveillance and Inspection of Port Facilities, Proc. of the 2007 ASME
International Mechanical Engineering Congress and Exposition, 2007.
[3] J-K. Choi, H. Sakai, T. Tanaka, Autonomous Towed Vehicle for
Underwater Inspection in a Port Area, Proc. of the 2005 IEEE
International Conference on Robotics and Automation, 2005, 188-193.
[4] A. Halme, M. Vainio, P. Appelqvist, P. Jakubik, T. Schonberg, A.
Visala, Underwater Robot Society Doing Internal Inspection and Leak
Monitoring of Water Systems, SPIE, Vol. 3209, 1997, pp 190-199.
[5] K. Koji, Underwater inspection robot -AIRIS 21, Nuclear Engineering and Design, Vol., 188, 1999, 367-371.
[6] B-H. Cho, S-H. Byun, C-H. Shin, J-B. Yang, S-I. Song, J-M., Oh,
KeproVt: underwater robotic system for visual inspection of nuclear
reactor internals, Nuclear Engineering and Design, Vol. 231, 2004,
pp 327-335.
[7] E. Lewandowski, The Dynamics of Marine Craft, World Scientific,
2004, pp 38-39.
[8] I. Rust, H.H. Asada, The eyeball ROV: Design and control of a
spherical underwater vehicle steered by an internal eccentric mass,
Proc. of the 2011 IEEE International Conference on Robotics and
Automation, 2011, pp 5855-5862.
[9] H.T. Choi, A. Hanai, S.K. Choi, J. Yuh, Development of an Underwater Robot: ODIN III, Proc. of the IEEE International Conference
on Intelligent Robots and Systems, 2003, pp 836-841.
[10] X. Lin, S. Guo, K. Tanaka, S. Hata, Development of a Spherical
Underwater Robot, Proc. of the IEEE/ICME International Conference
on Complex Medical Engineering, 2003, pp 662-665.
[11] D. Miller, A. Saenz-Otero, J. Wertz, A. Chen, G. Berkhowski, C.
Brodel, S. Carlson, D. Carpenter, S. Chen, S. Cheng, D. Feller,
S. Jackson, B. Pitts, F. Perez, J. Szuminski, S. Sell,SPHERES: A
Testbed For Long Duration Satellite Formation Flying in MicroGravity Conditions, Proceedings of the 2000 AAS/AIAA Space Flight
Mechanics Meeting, AAS 00, vol. 110, 2000, pp 23-36.
[12] A. Mazumdar, M. Lozano, A. Fittery, H.H. Asada, A Compact,
Maneuverable, Underwater Robot for Direct Inspection of Nuclear
Piping Systems, Proc. of the 2012 IEEE International Conference on
Robotics and Automation, 2012, pp 1544-1550.
[13] A. Mazumdar, H. Asada, Valve-PWM Control of Integrated PumpValve Propulsion Systems for Highly Maneuverable Underwater Vehicles, Proc. of the 2012 American Controls Conference, 2012.
[14] A. Mazumdar, A. Fittery, M. Lozano, H. Asada, Active Yaw Stabilization for Smooth Highly Maneuverable Underwater Vehicles, Proceedings of the 2012 ASME Dynamic Systems and Controls Conference.
[15] A. Fittery, A. Mazumdar, M. Lozano, H. Asada, Omni-Egg: A
Smooth, Spheroidal, Appendage Free Underwater Robot capable of
5 DOF Motions, Proceedings of the 2012 IEEE/MTS OCEANS
Conference.
[16] J.M. Kirshner, Design Theory of Fluidic Components, Academic
Press, New York, NY; 1975.

[17] Y. Xu, I.W. Hunter, J.M. Hollarbach, D.J. Bennett, A Portable Air
Jet Actuator Device for Mechanical System Identification, IEEE
Transactions on Biomedical Engineering, vol. 38(11), 1991, pp 11111122.
[18] J. Belden, W. Staats, A. Mazumdar, I. Hunter, An Airjet Actuator
System for Identification of Human Arm Joint Mechanical Properties,
Review of Scientific Instruments, vol. 82, 2011.
[19] R. Chen, Q. Huang, G.G. Lucas, Theorectical and experimental study
of a fluidic device as a fuel injector for natural gas engines, Proc.
Instn Mech Engrs, vol. 212, 1998, pp 215-226.
[20] M. S. Mason, W.J. Crowther, Fluidic Thrust Vectoring of Low
Observable Aircraft, CEAS Aerospace Aerodynamic Research Conference, 2002.
[21] I. Rust, H.H. Asada, A dual-use visible light approach to integrated
communication and localization of underwater robots with application
to non-destructive nuclear reactor inspection, Proc. of the 2012 IEEE
International Conference on Robotics and Automation, 2012, pp 24452450.
[22] M. Triantafyllou, F. Hover, Maneuvering and Control of Marine
Vehicles, MIT Course Notes, Department of Ocean Engineering, MIT,
2003, MIT.
[23] J. Newman, Marine Hydrodynamics, MIT Press, 1977.
[24] M. Zastawny, G. Mallouppas, F. Zhao, B. van Wachem, Derivation of
drag and lift force and torque coefficients for non-spherical particles
in flows, International Journal of Multiphase Flow, vol. 39, March
2012, pp. 227-239.

3422

Das könnte Ihnen auch gefallen