Vous êtes sur la page 1sur 10

GUIDANCE AND CONTROL OF AUTONOMOUS UNDERWATER

VEHICLES
Michael R. C. Andonian
Department of Mathematics
University of Hawaii at Manoa
Honolulu, HI 96822
Abstract
This paper discusses the design and practical use of efcient trajectories for an autonomous
underwater vehicle. The vehicle model that will be presented is based on the NASA funded Deep
Phreatic Thermal Explorer (DEPTHX) and the Omni-Directional Intelligent Navigator (ODIN).
The underlying mathematical framework is developed using geometric control theory and differ-
ential geometry. Briey, the control strategies are developed are based on integral curves dened
in the conguration space of the rigid body and kinematic reductions of rank one. Presented are
the control schemes for several legitimate mission scenarios and how pre-planned trajectories ex-
ploiting the inherent geometry of the vehicle and the environment overcome thruster failure.
Introduction
Autonomous robots not only represent the next great milestone for science, but their practical
uses span a wide range. Autonomous underwater vehicles (AUVs) are such an example. In partic-
ular, AUVs are being used to explore hostile environments far to hazardous for humans or manned
vehicles. These environments are of particular interest to researchers studying extremophiles, mi-
croorganisms capable of surviving in extraordinary circumstances. Such organisms may poten-
tially harbor secrets to the origins of life, both here on Earth and even plausibly extraterrestrial life.
For this reason, scientists involved in space exploration and the pursuit of extraterrestrial life have
turned their attention to AUVs. Projects such as the NASA funded rst generation Deep Phreatic
THermal eXplorer (DEPTHX) and the second generation, the Environmentally Non-Disturbing
Under-ice Robotic ANtarctiC Explorer (ENDURANCE) are examples of state-of-the-art AUVs
deployed to survey a given underwater environment in preparation for an anticipated opportunity
to explore Europa, a moon of Jupiter, which is believed to house oceans beneath its icy crust [6].
Other approaches to exploring hostile underwater environments include sending teams of AUVs
to survey hydrothermal vents [1], using sonar sensors to create maps, eventually to photograph
hydrothermal vent sites [9], and by predetermining trajectories for the AUV, while sampling the
water to choose a viable site to study [7]. Still, the fact remains that AUVs have to explore haz-
ardous environments, which may compromise the safe return of the vehicle, while still retrieving
satisfactory data such as a high resolution mapping of the surrounding region being studied. Pre-
cautionary techniques are thus initially implemented to improve the vehicles capability to return to
safely and intact in the event of unexpected damages which may occur during the mission. Based
on the reality of the situations AUVs may face, mission design and preparation are critical when
studying these dangerous areas and for the vehicle to return with minimal damage. Presented here
will be several strategies for practical missions based on realistic environments such as the under-
water volcano off the coast of the island of Hawaii, Loihi. These missions were developed with
the intent to keep the AUV close to walls for high resolution mapping and show a mission can still
be accomplished even after thruster actuation failure. The following presents an overview of the
model used for the design and simulation of trajectories which will also in turn be presented. The
model includes a brief introduction into the mathematics behind the wheel and a description of the
vehicle. One should note that the model and the vehicle description are separate entities, meaning
the model presented can be applied to any vehicle; we chose the hybrid DEPTHX-ODIN as the
vehicle the model would be applied to in order to illustrate this fact. Following the discussion of
the model will be two scenarios simulated based on the mathematics and realistic environments.
The conclusion will summarize the work done and emphasis will be placed on the versatility of the
model and its benets.
The AUV Model
To begin the discussion of the motion planning problem for an AUV, we consider a submerged
rigid boy in a viscous uid. The initial question is how will one express the location of the vehicle.
Typically, two right-handed, orthogonal, frames of reference are considered, an inertial (earth-
xed) frame and a body-xed frame. In addition, we chose the vertical axis to be positive in
the direction of gravity. For our purposes, we can naturally express the rigid body as a point
on the conguration space (manifold), Q, known as the special euclidean group, denoted SE(3).
Elements in SE (3) look like
_
R b
0
13
1
_
where b = (b
1
, b
2
, b
3
)
t
R
3
represents a position from
the origin of the inertial frame to the origin of the rigid body frame and R SO(3) is a 3 by 3
rotation matrix representing the orientation of the rigid body. Next, we denote = (
1
,
2
,
3
)
t
and = (
1
,
2
,
3
)
t
as the linear and angular velocities in the body-xed frame. Using two
different linear transformations (omitted here), one can transform between the linear and angular
velocities of the two reference frames. The resulting kinematic equations for the rigid body are

b = R (1)

R = R

(2)
where : R
3
so(3). so(3) is called a Lie algebra and it is, in this case, correlated to SO(3)
as the space of skew symmetric matrices in SO(3)
Typically, one constructs the equations of motion for the situation. This construction is lengthy
and will not be presented, but for a classical treatment, see [8]. These equations of motion take
into account external forces and moments of a submerged rigid body in viscous uid and are given
by
M v+D()v+Cor ()v+g() = (t) (3)
= J ()v (4)
where M is the added mass and inertia matrix, D() is the drag forces, Cor () is the Coriolis
and centripetal forces, g() corresponds to the restoring forces and moments, J () corresponds to
the linear transformations (matrices) mentioned previously to transform linear and angular veloc-
ities between reference frames, thus is a vector representation of the position and orientation of
the rigid body in the inertial frame, (t) represent the external forces and moments acting on the
vehicle, and nally, v = (, )
t
. In a moment, we will see that (t) actually represents the input
controls for our system in order to realize a desired motion. However, there exists a geometric
representation of these equations that is coordinate invariant. The derivation is very long but can
be found in [11], which also provide a detailed treatment of the mathematics. Here, we will briey
mention an analog of these equations.
The model is further developed by considering the representation of the kinetic energy by
a metric, which is a unique Riemannian metric, G, on Q. Furthermore, there is a Levi-Civita
connection that is unique and is associated with G. These last two statements are useful theories
in differential geometry. This afne connection provides one with an appropriate means for
discussing accelerations on Q. In fact, the connection is a projection of elements in the tangent
space of Q onto itself. This means everything we need to discuss exists naturally on Q. So, given
a curve (t) on Q, which corresponds to congurations (position and orientation) as a function of
time, there are

(t), which represent the linear and angular velocities of the congurations, and
there exists

(t)

(t), which represents accelerations, given as

(t)

(t) =
_
+M
1
(M)

+J
1
(J+ M)
_
(5)
where, M and J are the mass and inertial matrices with added mass and inertia included.
To consider the submerged rigid body, gravitational, buoyancy, and viscous dissipative (drag)
forces are considered. The model continues its development as seen in [4], [11], and [12].By
reparametrizing the curves (t) so the vehicle will be at rest initially and between kinematic mo-
tions, a concatenation of these motions is practical. Ultimately, this leads us to an equation for the
necessary controls,
m

i=1

i
(t)I
1
i
( (t)) =
_

(t)
_
2

V
V ( (t)) +

(t)V ( (t)) (6)


where
i
represents the kinematic control vector (corresponding to control signals to be sent to
thrusters for desired motions), I
1
i
is the ith column of the inverse of the generalized inertia matrix
I, (t) is the reparameterization of the curve (t) such that the vehicle begins and stops with
zero velocity, V =
m

i=1
h
i

I
1
i
, h
i
R, represents a vector eld, where the set
_

I
1
i
_
m
i=1
, 1 m6, is
a given set of input control vector elds and a subset of
_
I
1
i
_
6
i=1
, and

is the modied connection
of the original connection dened to take into account the uid dynamics. Here, we will simply
note that decoupling vector elds (DVFs) are a kinematic reduction of rank one. Consider

(t)

(t) =
6

i=1

i
(t)I
1
i
( (t)) (7)
which is an analogous, coordinate invariant, geometric interpretation of equation 5. Equation 6
is the dynamic control system called the afne connection control system (ASSC). Thus, speaking
roughly, a kinematic reduction of the dynamic system is when every trajectory of the kinematic
system,

(t), also satises the dynamic system. The primary reason for going through this work
is because there is well develop theory on the kinematic, 1st order system, bu not so much for the
2nd order system. As a result, integral curves of DVFs correspond to kinematic motions which
we concatenate to develop our full trajectories. Thus, using the DVFs to solve for allows one
to simulate a desire trajectory by solving for v and in equations 3 and 4. Frankly, an example
of DVF would be V
1,3
= h
1
I
1
1
+h
3
I
1
3
+h
4
I
1
4
, which would correspond to a surge-heave-pitch
motion. When the vehicle is fully actuated, every vector eld is decoupling, meaning any trajectory
can be realized. However, when the vehicle is under-actuated, it is difcult to know which vector
elds are decoupling. [11] gives a list of the DVFs for any under-actuated situation.
Lastly, the vehicle model we used to showcase the above theory was a hybrid of the ODIN and
the DEPTHX. This is to show that the theory is exible to any vehicle. Here are the parameters of
the model vehicle:
m : 1360 kg j

1
b
: 1008 kgm
2
m

1
f
: 3750 kg j

2
b
: 1008 kgm
2
m

2
f
: 3750 kg j

3
b
:1114 kgm
2
m

3
f
: 6222 kg j

1,2,3
f
: 0 kgm
2
Diameter (m) : (1.9, 1.9, 1.9)
Center of Gravity (C
G
): (0, 0, 7) mm Center of Buoyancy (C
B
): (0, 0, 7) mm
B = gV : 1215.8 N W = mg :1214.5 N
drag 1 (D
1
): 500.0 kg/m
2
drag 2 (D
2
): 500.0 kg/m
2
drag 3 (D
3
): 777.8 kg/m
2
drag 4 (D
4
): 280.0 kg/m
2
drag 5 (D
5
): 280.0 kg/m
2
drag 6 (D
6
): 318.2 kg/m
2
Table 1: AUV DEPTHX-ODIN model parameters
Moreover, we assumed the vehicle to have a thruster conguration similar to ODINs, as seen
in the following gure. Since the rigid body is expressed in SE (3), which is 3 dimensional, it has
six degrees of freedom (DOF). To have motion in all six DOF from thruster actuation is considered
fully actuated and allows the rigid body to realize any kinematic motion. In addition, this model is
over-actuated, since there is redundancy with the thruster controls. A simple linear transformation
is performed on the control signals from the eight thrusters to yield the six dimensional control
signal, .
Figure 1: Thruster conguration for and DEPTHX-ODIN vehicle. Such a conguration gives the
rigid body motion in the six DOF and is thus fully actuated.
Missions
Once everything desired is known, masses, inertias, the DVFs for a desired trajectory, etc.,
simulations were made in Mathematica (and MATLAB) script. For a user dened DVF, the above
parameters, and the correct function (t) that has the vehicle start and stop with zero velocity, any
desired trajectory was realized through a computer simulation by computing the necessary controls,
, and solving numerically equations 3 and 4 for v = (, )
t
and . To show the application of
mathematics presented and the versatility of the code and model, we simulated various missions.
These mission designs were based on legitimate situations an AUV may face when exploring
underwater caves and the Loihi seamount, an active underwater volcano off the coast of the Island
of Hawaii. In addition, minimization of energy and the event of under-actuation were also kept in
mind during the design process to emphasize the practicality of these trajectories.
Mission 1: Cave Exploration
This mission saw many changes throughout the semester. The following gure shows the
evolution of the mission.
a) b) c)
Figure 2: Trajectories for cave exploration. Left most image (a) is a spiral motion down with a
constant radius, (b) is a spiral motion down with an expanding radius, and (c) is a concatenation of
motions.
The rst mission, seen in Figure 2(a), was developed to explore the walls of an underwater
cave (or basin) with cylindrical features. Here, the dimensions of the cave are 40 m in radius and
150 m in depth. If the intent of the mission is to successfully map the cave, staying close to the
walls to get a high resolution image is necessary. This choice of trajectory allows the vehicle to
remain close and closely inspect the wall surface. The trajectory depends on solely on motions
corresponding to sway, heave, and yaw (the kinematic motion of such is generated by the linear
combination of the vector elds I
1
2
, I
1
3
, and I
1
6
, respectively). In other words, the vehicle may
lose the ability of motion in the other DOF, but it must have the necessary thruster conguration to
perform a sway, heave, and yaw. Since the DEPTHX-ODIN model has the thruster conguration
corresponding to full actuation every vector eld is decoupling and as such, it can perform this
trajectory, even in certain under-actuated scenarios. For example, the vehicle may lose actuation
in the thrusters which produce the sway motion, but with a simple rotation, the thrusters originally
used to produce a surge can be used for a sway. This is a direct exploitation of the symmetry of the
vehicle. If the vehicle were to be xed with an exceptional array of sensors and a SLAM algorithm,
the detailed 3-D map of the basin could be acquired in a few attempts.
The second mission seen in Figure 2(b) is very similar to the previous mission. The goals of
minimizing energy and staying close to the wall of the basin were kept as well. In this scenario, the
basin is similar to a cone, with the top, initial radius of 20 m and expands to 60 m at 100 m deep.
Like the previous mission, the vehicle must have actuation in sway, heave, and yaw to successfully
accomplish this motion.
The last mission involving cave exploration can be seen in gure 2(c). In this scenario, we
continue to make use of the inherent mathematical framework provided and consider a mission
where we lose actuation in surge an pitch. The vehicle begins its descent into a basin using the
helical motion previously described. Upon reaching the bottom of the 110 deep, 15 m in radius
sinkhole, it continues the exploration by entering a tunnel 5 m in diameter and 15 m in length with
a slight arc. Exploration of the tunnel is accomplished by following a parabolic-like curve using
a surge-heave-pitch motion (the kinematic motion of such is generated by the linear combination
of the vector elds I
1
1
, I
1
3
, and I
1
5
, respectively). However, once inside the tunnel, the vehicle
loses actuation of four thrusters resulting in the loss of motion corresponding to surge and pitch.
The thruster loss can be seen in Figure 3.
Figure 3: ODIN-like vehicle with loss of four thrusters (grey). An AUVs health monitor can
detect loss of thrusters.
With the loss of surge and pitch, every vector eld is no longer decoupling. Thus, our set of
input vector elds becomes

I
1
1
= I
1
2
,

I
1
2
= I
1
3
,

I
1
3
= I
1
4
, and

I
1
4
= I
1
6
, i.e., we can only
sway, heave, roll and yaw. From [10] and [11], the DVFs are V =

h
2

I
1
2
+

h
4

I
1
4
(heave and yaw)
and V =

h
1

I
1
1
+

h
2

I
1
2
+

h
3

I
1
3
(sway-heave-roll).
As a result, the vehicle exists by use of a roll-heave motion. The vehicles thrusters roll the
vehicle to the desired angle and then apply thruster required to heave. Geometrically, the path
exiting the tunnel is same as the path entering. Now, to make even more use of the mathematics,
upon exiting the tunnel, instead of using a pure heave motion to ascend, we designed a more
complex concatenation of motions. The kinematic motions concatenated were sway-heave-roll
motions seen in red in Figure 2(c). Here, the descent took one hour and required 58 kJ of energy,
the tunnel exploration fully actuated took 3 minutes and required 2616 J, the exit of the tunnel
under-actuated also took three minutes but 2254 J were needed, and nally the ascent took one hour
and required 10.6 kJ. Of course, lengthening the mission time would also decrease the amount of
energy required. This mission and a more detailed discussion of the mathematics can be found in
[3].
Mission 2: Loihi Seamount
Most recently, we have begun the development of a simulation of a mission to map the summit
of Loihi, an active underwater volcano off the southeast coast of the Island of Hawaii. Naturally,
this mission would take a long time to accomplish with the current parameters of the vehicle
model we have used for the exploration of the basin presented above. The vehicle must be larger
and faster in order to successfully accomplish the mission within a reasonable time frame and
also be unaffected by ocean currents. However, the reader should be reminded that what is being
presented here is a model, meaning things such as vehicle size, weight, thruster conguration,
etc, can easily be modied and accounted for. That being said, our current simulations do use
the same DEPTHX-ODIN hybrid. However, due to external perturbations that may compromise
the vehicles planned trajectory, ocean currents for instance, we have opted to add an additional
feedback controller into the simulations. Details about the feedback controller implemented can
be found in [5]. This mission involves a survey of a region of the summit referred to as Peles
Pit. The full expedition concatenates various trajectories into a single mission. Unfortunately, the
simulations for this mission have yet to be complete, so visuals of the trajectories used are not yet
available, but the mission is as follows.
Initially, the vehicle is fully actuated as it descends 1000 m just below the summit of Loihi. the
vehicle then arcs over the ridge and heads straight for the crater labelled (a) in the following Figure
4. Using the helical trajectory, this time a surge-heave-pitch motion, the vehicle successfully can
map the walls of this region. The exit of this crater is a pure heave motion. Next, the vehicle moves
towards the ridge labelled (b) below. Upon reaching this ridge, it experiences thruster failure and
loses two of its horizontal thrusters, those corresponding to the sway motion. The vehicle still
manages to arc over the ridge (not using pure motions) and heads towards the crater labelled (c)
in Figure 4. At this point, the surge-heave-pitch helical motion cannot be realized. However, due
to the symmetry of the vehicle, a simple rotation can cause the dead thrusters to now correspond
to sway, meaning we revive the surge-heave-pitch helical motion. To remain in the spirit of the
problem, we continue to consider the surge motion as lost, but instead note that we are still
actuated in sway and roll. Thus, we use the original helical motion described previously. Finally,
the vehicle simply leaves the seamount. Currently, we are developing this mission further to the
point where the vehicle becomes so under-actuated, only pure motions are possible.
!
"
#
Figure 4: Loihi Seamount. Modications made to original image. The red dot is the starting point
of the mission. Numeric values are depth in fathoms. Image from [10]
Conclusion
The missions designed here can certainly be implemented on vehicles, to an extent, such as
the ODIN or the DEPTHX. The idea is to place these pre-planned trajectories into the software
architecture of an AUV as mission plans or contingency plans. Since the DEPTHX is considered
under-actuated (roll and pitch have been neutralized), the vehicle cannot perform some of the
more complex motions such as the spiral. On the other hand, the DEPTHX also does not fully
exploit its capabilities. By use of a method termed iterative exploration, the vehicle must descend
and ascend repeatedly in order to successfully map an environment. In their defense, the scheme
works remarkably well if there is no prior information of the environment. However, if a team has
a rough idea of the geography of the environment prior to the mission, an adequate implementation
of the presented control strategy would be more useful in two crucial ways. The rst is the fact
that a geometric pre-planned trajectory can allow a vehicle such as the DEPTHX to remain close
to the walls. In return, any array of sensors (in particular sonars) would yield high quality data.
Second, and as a consequence of the previous fact, the number of attempts to obtain a satisfactory
map would be minimized to an extent, thus resulting in energy conservation. But the DEPTHX
is a special vehicle, and it is thus important to remark again that the paper presents a model. In
other words, this type of geometric control planning can be applied to and used for any vehicle.
Moreover, any vehicle design can use this theory to discuss the event of under-actuation on their
vehicle, rather than implementing an abortion system which has the vehicle surface immediately.
In fact, these schemes become necessary when deciding if the vehicle can even return safely.
As seen in the mission designs and simulations above, even in the event of under-actuation, the
vehicle was capable of performing tasks and, if conditions are nice, complete a given mission.
This is a critical benet of the theory presented here. In addition, since these trajectories are
global path planners, they will not be deceived by the geometry of a region. For example, if
an AUV were to enter an enclosed dome relying on sensors, it is highly plausible the vehicle
may get trapped. However, there are some drawbacks to pre-planning these trajectories. These
trajectories (and indirectly the control scheme in its entirety) are feed-forward controllers and
do not account for uncertainties. Thus, some means of tracking is necessary, which is why the
Loihi mission included the feedback controller. In spite of this, the benets of the techniques
presented do continue to accumulate. When coupled with a feedback controller, the system overall
becomes very robust. As stated previously, the AUV can implement the techniques presented in
this paper to analyze its health and environment to decide what trajectories are can be realized
due to thruster actuation or the geometry/geography of the environment. If the vehicles sensors
become useless, due to common scenarios such as murky water or electrical failures, the reliability
of the feedback controller falls but the trajectories developed by this control scheme can still be
highly useful for dead reckoning. In addition, the AUV can diagnose an unfavorable situation
and determine if a mission can be completed or if (and how) it can return to the surface safely.
Finally, these simulations and the our associated geometric control strategy approach can allow
one to determine the pros and cons of a certain thruster conguration. By using our code in the
process of developing an AUV, one can visually see what the vehicle can and cannot do, i.e., one
has the knowledge of permissible motions for the AUV. This is very benecial for designing an
AUV without sophisticated and expensive sensors.
Acknowledgements
Id like to thank NASA and the Hawaii Space Grant Consortium for allowing me this opportu-
nity to extend my education far beyond the classroom. Because of this project, I have seen and felt
rst-hand what research is like, not to mention the acquiring of valuable public speaking skills. Id
like to thank Dr. Ed Scott and Dr. Luke Flynn for their genuine words of encouragement, even if
it was only in passing. A special thank you to Marcia Rei Sistoso, who was always exceptionally
helpful and understanding, and the rest of the staff in POST 501, Joana Choy, Amber Imai, Bianca
Soriano, and Lauren Kamei for their additional help. A tremendous thanks to Dario Cazzaro, Ser-
gio Grammatico and Luca Invernizzi; none of these simulations would have been possibly without
their high quality knowledge as engineers and computer programmers. Thanks to Dr. George
Wilkens for supplemental help regarding the mathematics involved with this entire project. Most
of all, my gratitude and thanks goes to my mentor Dr. Monique Chyba, especially for offering me
this opportunity in the rst place. Her guidance and concern as a professor and educator has dra-
matically raised my level as a student and undergraduate mathematician. Thanks again to everyone
for their enthusiasm, support, and this wonderful opportunity.
References
[1] Caiti A., Munafo A., Viviani R. (2007) Adaptive on-line planning of environmental sampling
missions with a team of cooperating autonomous underwater vehicles. International Journal
of Control 80/7, 1151-1168.
[2] Chyba M., Cazzaro D., Invernizzi L., Andonian M. (2010) Trajectory Design for Autonomous
Underwater Vehicles for Basin Exploration. 9th International Conference on Computer and
IT Applications in the Maritime Industries, 139-151. Gubbio, Italy.
[3] Chyba M., Grammatico S., Cazzaro D., Invernizzi L., Andonian M. (2010) Geometric control
for autonomous underwater vehicles: overcoming a thruster failure. 49th IEEE Conference
on Decision and Control, pending. Atlanta, Georgia.
[4] Chyba M & Smith R.N. (2008) A rst extension of geometric control theory to underwater
vehicles. Proceedings of the 2008 IFAC Work- shop on Navigation, Guidance and Control of
Underwater Vehicles 2/1. Killaloe, Ireland.
[5] Conte G. & Serrani A. (1999) Robust nonlinear motion control for AUVs. Robotics & Au-
tomation Magazine IEEE 6/2, 33-38.
[6] Faireld N., Kantor G.A., Jonak D., Wettergreen D. (2008) DEPTHX Autonomy Software:
Design and Field Results. tech. report CMU-RI-TR-08-09, Robotics Institute, Carnegie Mel-
lon University.
[7] Ferri G., Jakuba M.V., Yoerger D.R. (2008) A novel method for hydrothermal vents prospect-
ing using an autonomous underwater robot, IEEE International Conference on Robotics and
Automation, 1055-1060. Pasadena, California.
[8] Fossen T.I. (1994) Guidance and Control of Ocean Vehicles. John Wiley & Sons.
[9] German C. R., Yoerger D. R., Jakuba M., Shank T. M., Langmuir C. H., Nakamura K.I.
(2008) Hydrothermal exploration with the autonomous benthic explorer. Deep Sea Research
Part I: Oceanographic Research Papers 55/2, 203-219.
[10] Malahoff A. (1987) Geology of the Summit of Loihi Submarine Volcano. Volcanism in
Hawaii: U.S. Geological Survey Professional Paper 1350, 133-144.
[11] Smith R.N. (2008a) Geometric control theory and its application to underwater vehicles.
Ph.D. dissertation. Ocean & Resources Engineering Department, University of Hawaii at
Manoa.
[12] Smith R.N., Chyba M., Wilkens G.R., Catone C. (2009) A Geometrical Approach to the
Motion Planning Problem for a Submerged Rigid Body. Int. J. of Control, 82/9, 1641 -1656.

Vous aimerez peut-être aussi