IEEE TRANSACTIONS ON ROBOTICS, VOL. 30, NO.

4, AUGUST 2014 845

Nonlinear RISE-Based Control of an Autonomous

Underwater Vehicle
Nicholas Fischer, Devin Hughes, Patrick Walters, Eric M. Schwartz, Senior Member, IEEE,
and Warren E. Dixon, Senior Member, IEEE

Abstract—This study focuses on the development of a nonlin- However, while these controllers provide suitable performance
ear control design for a fully-actuated autonomous underwater and stability results in theory and simulation, obtaining empiri-
vehicle (AUV) using a continuous robust integral of the sign of cal models of dynamic parameters (e.g., added mass, nonlinear
the error control structure to compensate for system uncertainties
and sufficiently smooth bounded exogenous disturbances. A Lya- Coriolis models, hydrodynamic damping forces, etc.) experi-
punov stability analysis is included to prove semiglobal asymptotic mentally are often inaccurate and extremely difficult to develop.
tracking. The resulting controller is experimentally validated on Motivated to provide robustness to model uncertainty, adap-
an AUV developed at the University of Florida in both controlled tive controllers for AUVs are developed in [6]–[8]. Efforts in [9]
and open-water environments. used traditional adaptation methods and switching theory to
Index Terms—Autonomous underwater vehicles (AUVs), marine compensate for unknown dynamics. In [10]–[16], fuzzy logic
robotics, nonlinear control, robust integral of the sign of the error or neural network (NN)-based methods are used to approximate
(RISE). uncertain dynamics; however, the presence of external distur-
bances and function approximation errors result in uniformly
I. INTRODUCTION ultimately bounded tracking results. An adaptive switching su-
pervisory control technique for uncertain underactuated AUVs
DVANCES in sensing and control capabilities are en-
A abling autonomous surface vehicles and autonomous un-
derwater vehicles (AUV) to become vital assets in search and
is developed in [17] yielding ultimately bounded tracking. Re-
sults in [14], [15], and [18] use a composite sliding mode con-
trol law to eliminate the steady state error of pure adaptive
recovery, exploration, surveillance, monitoring, and military ap- designs, in a similar manner to robust sliding mode control
plications [1]. Accurate and robust trajectory tracking is crucial results in [19]–[22]. Nonlinear observers to estimate unknown
to the performance of these vehicles and advancement of auton- hydrodynamic damping coefficients were coupled with a sliding
omy in the maritime environment. mode control law in [23] for diving and steering control of an
The dynamics of an AUV are time-varying, nonlinear, and AUV. However, while the sliding mode technique successfully
often include difficult to model effects, for example, hydrody- yields asymptotic tracking for uncertain nonlinear systems with
namic coefficients and external disturbances such as sea state additive disturbances, the resulting controller is discontinuous.
or ocean currents. Many results in the literature focus on AUV Discontinuous controllers suffer from limitations such as the
controllers that utilize exact knowledge of the dynamics [2]–[5]. demand for infinite bandwidth and chatter, motivating the need
for continuous methods that can yield asymptotic tracking in
Manuscript received April 23, 2013; revised September 9, 2013; accepted the presence of added disturbances and generalized uncertainty.
February 9, 2014. Date of publication March 14, 2014; date of current version Motivated by our previous work in [24] and preliminary ef-
August 4, 2014. This paper was recommended for publication by Associate Edi- forts in [25], a continuous robust integral of the sign of the
tor K. Kyriakopoulos and Editor W. K. Chung upon evaluation of the reviewers’
comments. error (RISE) control structure is used to compensate for un-
N. Fischer, D. Hughes, and P. Walters are with the Department of Mechanical certain, nonautonomous disturbances for a class of coupled,
and Aerospace Engineering, University of Florida, Gainesville, FL 32611 USA fully-actuated underwater vehicles. A Lyapunov-based stabil-
(e-mail: [email protected]; [email protected]; [email protected]).
E. M. Schwartz is with the Department of Electrical and Computer ity analysis is provided to show that the control method yields
Engineering, University of Florida, Gainesville, FL 32611 USA (e-mail: semiglobal asymptotic tracking. The resulting controller is ex-
[email protected]). perimentally validated on a 6-degree-of-freedom (DOF) AUV
W. E. Dixon is with the Department of Mechanical and Aerospace Engi-
neering, University of Florida, Gainesville, FL 32611 USA and also with the that has been developed at the University of Florida. Experi-
Department of Electrical and Computer Engineering, University of Florida, mental trials are conducted in a swimming pool to demonstrate
Gainesville, FL 32611 USA (e-mail: [email protected]). the performance of the controller. Additionally, an open-water
This paper has supplementary downloadable material available at
http://ieeexplore.ieee.org, provided by the author. The video illustrates exper- sea trial was completed in the Gulf of Mexico off the coast of
imental validation of work that focuses on the development of a nonlinear Panama City Beach, FL, USA, to illustrate the robustness of the
control design for a fully-actuated autonomous underwater vehicle (AUV) us- design in a real-world environment.
ing a continuous robust integral of the sign of the error (RISE) control structure
to compensate for system uncertainties and sufficiently smooth bounded exoge-
nous disturbances. The resulting controller is experimentally validated on an
II. VEHICLE CONFIGURATION AND MODELING
AUV developed at the University of Florida in both controlled and open-water The position and orientation of an AUV relative to an earth-
environments.
Color versions of one or more of the figures in this paper are available online fixed frame is given by the kinematic equation of motion [26]:
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TRO.2014.2305791 η̇ = J (η) ν (1)

1552-3098 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
846 IEEE TRANSACTIONS ON ROBOTICS, VOL. 30, NO. 4, AUGUST 2014

unlikely due to metacentric restoring forces [26]. To handle sit-

uations when operation near θη = ±90 is required, additional
supplemental techniques can be used to modify the reference
coordinate system that is used to define the order of the Eu-
ler angles when the vehicle is nearing a singular orientation.
An arbitration algorithm must be used to correctly manage the
the transition periods when the coordinate system convention is
redefined.
Under assumptions that 1) the body-fixed frame coincides
with the center of mass of the AUV, 2) accelerations of a point
on the surface of the earth can be neglected (i.e., the reference
frame XYZ in Fig. 1 is considered to be inertial), and 3) added
Fig. 1. AUV reference frames and associated state vector directions. mass is constant (independent of wave frequency), the dynamic
motion of the AUV can be described by a body-fixed vector
representation as [26]
where ν ∈ R6 is a vector of linear and angular velocities with M ν̇ + C (ν) ν + D (ν) ν + g (η) + τd = τb (3)
coordinates in the body-fixed frame, η ∈ R6 is a vector of po-
sition and orientation with coordinates in the earth-fixed frame, where M ∈ R6×6 denotes inertia (including added mass), C :
and J : R6 → R6×6 is a Jacobian transformation matrix relat- R6 → R6×6 denotes Coriolis and centripetal effects (including
ing the two frames. The state vectors of the AUV from (1) are added mass), D : R6 → R6×6 denotes hydrodynamic damping
illustrated in Fig. 1 and are defined as [26] effects, g : R6 → R6 is the vector of hydrostatic (gravitational
and buoyancy) forces and moments, τd ∈ R6 is a vector of non-
η  [ xη yη zη φη θη ψη ]T linear disturbances (e.g., current, waves, tether forces, etc.), and
ν  [ uν vν wν pν qν rν ]T , τb ∈ R6 is a vector of external forces and moments about the
center of mass in the body-fixed frame. An earth-fixed repre-
where xη , yη , zη ∈ R represent the Cartesian position of the cen- sentation of the dynamics [26] can be generated by applying the
ter of mass, φη , θη , ψη ∈ R represent the orientation (roll, pitch, kinematic transformations in (1) to (3) to obtain
and yaw), uν , vν , wν ∈ R represent the surge, sway, and heave
velocities, and pν , qν , rν ∈ R represent angular velocities. The M̄ (η) η̈ + C̄ (η, η̇, ν) η̇ + D̄ (η, ν) η̇ + ḡ (η) + τ̄d = τn (4)
Jacobian which relates the state vectors in (1) is defined as where M̄  J −T M J −1 , C̄  J −T [C − M J −1 J]J ˙ −1 , D̄ 
  −T −1 −T −T
J DJ , ḡ  J g, τ̄d  J τd , and τn  J τb . The sub- −T
J1 (η) 03×3
J (η)  . (2) sequent development is based on the assumptions that η and ν
03×3 J2 (η)
are measurable (using sensors common to many AUVs as dis-
In (2), J1 : R6 → R3×3 and J2 : R6 → R3×3 are defined as cussed in [27]) and that M̄ , C̄, D̄, ḡ, and τ̄d are unknown
functions. 1
J1 (η) 
⎡ ⎤ Assumption 2: The disturbance term and its first two time
cψη cθη −sψη cφη + cψη sθη sφη sψη sφη + cψη cφη sθη derivatives are bounded, i.e., τd , τ̇d , τ̈d ∈ L∞ . 2
⎣sψη cθη cψη cφη + sφη sθη sψη −cψη sφη + sθη sψη cφη ⎦ Assumption 3: The desired trajectory ηd ∈ Rn is designed
−sθη cθη sφη cθη cφη (i)
such that ηd ∈ Rn , ∀ i = 0, 1, . . . , 4 exist and are bounded. 3
⎡ ⎤ Property 1: The inertia matrix M̄ is symmetric, positive def-
1 sφη tθη cφη tθη
⎢ ⎥ inite, and satisfies the following inequality ∀ ξ ∈ Rn , η ∈ R6 :
J2 (η)  ⎣ 0 cφη −sφη ⎦ ,
0 sφη /cθη cφη /cθη m ξ2 ≤ ξ T M̄ (η) ξ ≤ m̄ (η) ξ2

where s·, c·, t· denote sin (·), cos (·), and tan (·), respectively, where m ∈ R+ is a known constant, and m̄ : R6 → R+ is a
and 03×3 ∈ R3×3 represents a matrix of zeros. known function.
Assumption 1: The Jacobian and its inverse exist and are Property 2: The dynamics terms C̄, D̄, and ḡ are bounded
bounded by a known constant J¯ ∈ R+ such that supη J ≤ J¯ provided η and ν are bounded.
and supη J −1 ≤ J. ¯
The definition of J1 consists of a sequence of three rota- 1 Structural or measurable knowledge of the unknown terms M̄ , C̄ , D̄, ḡ

tions about each of the primary orientation axes. The order in allow the user to more appropriately define upper bounds such that the gain
which these rotations is completed is not arbitrary; however, conditions presented in Theorem 1 can be satisfied.
2 Many practical disturbance terms satisfy this assumption including wind
for guidance and control applications, it is common to use the disturbances, wave disturbances, ocean currents, etc. when simple bounding
xyz-convention in terms of Euler angles. Utilizing this conven- assumptions are imposed (e.g., bounded current velocity, bounded wind velocity,
tion, J2 is undefined for a pitch angle of θη = ±90◦ , which vio- etc.) [26], [28], [29].
3 Many guidance and navigation applications utilize smooth high-order dif-
lates Assumption 1. However, during routine “flight” operations ferentiable desired trajectories. Curve fitting or nonlinear filter methods can be
with underwater vehicles, the parameter regions θη = ±90◦ are used to generate sufficiently smooth time-varying trajectories.
FISCHER et al.: NONLINEAR RISE-BASED CONTROL OF AN AUTONOMOUS UNDERWATER VEHICLE 847

III. CONTROL DEVELOPMENT Filippov’s theory of differential inclusions [32]–[35], the ex-
istence of solutions can be established for υ̇F ∈ K [h1 ] (e2 ),
A. Error Systems
where h1 : R6 → R6 is defined as the right-hand side (RHS)
The objective is to design a controller that ensures the system of (10) and K [h1 ]  δ > 0 μS m =0 coh1 (B (e2 , δ) \ Sm ),
state tracks a sufficiently smooth desired trajectory. To quantify where μS m =0 denotes the intersection of all sets Sm (of
the control objective, a tracking error e1 ∈ R6 is defined as Lebesgue measure zero) of discontinuities, co denotes convex
closure, and B (e2 , δ)  {ς ∈ R| e2 − ς < δ} [36], [37].
e1  ηd − η. (5)
Remark 1: Typical control techniques that can achieve asymp-
Two auxiliary tracking errors e2 , r ∈ R6 are defined as totic convergence in the presence of a disturbance either uti-
lize discontinuous feedback or feedback with a discontinuous
e2  ė1 + α1 e1 (6) derivative. Continuously differentiable robust techniques such
r  ė2 + α2 e2 (7) as high-gain feedback can only achieve uniformly ultimately
bounded convergence. Discontinuous control techniques (such
where α1 , α2 ∈ R+ are constant gains. The auxiliary signal r as sliding mode or variable structure control) suffer from limita-
is introduced to facilitate the stability analysis and is not used tions such as demand for infinite bandwidth or chatter. Because
in the control design since the expression in (7) depends on the the controller in (9) utilizes the integral of a discontinuous signal,
unmeasurable state η̈. the implemented control law does not suffer from these restric-
tions, while still compensating for sufficiently smooth nonlinear
B. Open-Loop Error System disturbances and system uncertainties.
Premultiplying (7) by M̄ (η) and using (4)–(6), the open-loop
error system can be expressed as D. Closed-Loop Error System

M̄ (η) r = fd + S + τ̄d − τn (8) To facilitate the subsequent stability analysis, the controller
in (9) is substituted into (8) and the time derivative is determined
where the auxiliary function fd ∈ R6 is defined as as
fd  M̄ (ηd ) η̈d + C̄ (ηd , η̇d , νd ) η̇d + D̄ (ηd , νd ) η̇d + g (ηd ) 1 ˙
M̄ (η) ṙ = − M̄ (η, η̇) r + Ñ + Nd
2
the desired signal νd can be determined using the known kine-
− e2 − (ks + 1) r − βsgn (e2 ) (11)
matic relationship in (1) and the known signals ηd and η̇d , and
the auxiliary function S ∈ R6 is defined as where Ñ ∈ R6 and Nd ∈ R6 are defined as
S = M̄ (η) η̈ + C̄ (η, η̇, ν) η̇ + D̄ (η, ν) η̇ 1 ˙
Ñ  − M̄ (η, η̇) r + Ṡ + e2 (12)
2
+ ḡ (η) − fd + α2 ė1 + α2 e2 .
Nd  f˙d + τ̇d . (13)
Strategic grouping allows terms included in fd to be upper
bounded by constants since they are all functionally dependent Since C̄ is not skew symmetric in the earth-fixed representation
on bounded trajectories. The remaining terms in S are subse- of the dynamics in (4), a portion of the M̄ ˙ term is included
quently bounded by a state-dependent function for use in the in the closed-loop error system to help cancel cross terms in
analysis, as shown in the next section. the stability analysis, while the remainder of the term is placed
inside the Ñ term. Using (5)–(6) and the Mean Value Theorem,
C. Control Design4 the function Ñ in (12) can be upper bounded as [38, App. A]
From (8), the controller is designed using a RISE feedback Ñ ≤ ρ (z) z (14)
structure as [30], [31]
τn  (ks + 1) e2 − (ks + 1) e2 (0) + υF (9) where z ∈ R18 is defined as

where υF ∈ R6 is the Filippov solution to the following differ- z  [ eT1 eT2 r T ]T (15)
ential equation and ρ : R+ → R+ is a positive, globally invertible function.
υ̇F  (ks + 1) α2 e2 + βsgn (e2 ) , υF (0) = 0 (10) From Assumptions 2 and 3, the following inequalities can be
developed:
β, ks ∈ R are positive, constant control gains, and sgn(·)
is defined ∀ξ ∈ Rm = [ξ1 ξ2 · · · ξm ]T as sgn (ξ)  Nd  ≤ ζ1 , Ṅd ≤ ζ2
[sgn (ξ1 ) sgn (ξ2 ) · · · sgn (ξm )]T . The differential equa-
where ζ1 , ζ2 ∈ R+ are known constants.
tion given in (10) is continuous except when e2 = 0. Using
IV. STABILITY ANALYSIS
4 RISE feedback structure can be coupled with a feedforward term, i.e., NNs
[24], [25], model-based adaptive law [30] or known dynamics, for improved Theorem 1: The controller in (9)–(10) ensures that the states
performance. and controller are bounded and the tracking errors are regulated
848 IEEE TRANSACTIONS ON ROBOTICS, VOL. 30, NO. 4, AUGUST 2014

Fig. 2. AUV, “SubjuGator 7,” developed at the University of Florida.

in the sense that Fig. 3. Time history of the position of the AUV with respect to the helical
trajectory.
e1  → 0 as t→∞

provided ks is selected sufficiently large based on the initial To localize the state of the vehicle, an indirect unscented
conditions of the states, and the remaining control gains are Kalman filter estimates the error in position, velocity, and ori-
selected based on the following sufficient conditions: entation generated by the inertial navigation system (INS),
1 1 1 which includes high-speed sensory inputs (205 Hz) from an
β > ζ1 + ζ2 , α1 > , α2 > . (16) analog device ADIS16405 9-DOF IMU (providing triaxis mag-
α2 2 2
netometer, accelerometer, and gyroscopic inputs). Input error
Proof: See the Appendix  signals for the Kalman filter are generated using low-speed
reference sensors: a Teledyne Explorer DVL (providing three-
V. EXPERIMENTAL RESULTS axis velocity and height over bottom), a 14-channel GPS re-
ceiver (while surfaced), a 10-bar pressure sensor, and a filtered
A. Vehicle Configuration
tilt/magnetometer/gravity-based estimation of attitude. Mission
The validation experiments are completed using an AUV commands are executed by interpreting information from on-
(known as “SubjuGator 7”) that has been developed at the Uni- board optical sensors and are then converted to sufficiently
versity of Florida. The AUV, shown in Fig. 2, is a hybrid AUV smooth desired vehicle trajectories that are based on the dy-
that is designed with emphasis on modularity and fault toler- namic capabilities of the vehicle. By computing smooth desired
ance. The vehicle has physical dimensions of 1.3 m × 0.5 m trajectories from waypoints in real time, the controller avoids
× 0.5 m and is configured with eight bidirectional thrusters in large steps in error (potentially producing large actuation efforts
a redundant configuration with four heave thrusters, two sway and posing risks of actuator saturation). State estimation and
thrusters, and two primary surge thrusters, allowing for maneu- control loops for the vehicle are sampled at 50 Hz.
vering in 6 DOF. The relationship between the force/moment
acting on the vehicle and the control input of each individual
B. Controlled Environment Study
thruster can be described by a thruster mapping algorithm, such
as the one described in [39]. 1) Experimental Setup: The first set of experiments were
The vehicle utilizes specialized independent pressure vessels performed in a swimming pool at the University of Florida un-
to house core computing functions. Motor controllers, network- der the influence of parametric system uncertainties and small
ing capabilities, and platform-specific processing are located unknown disturbance effects such as pump currents and tether
in a central pressure vessel. This vessel also houses process- forces. For this enclosed space, the vehicle is commanded to
ing of external sensors (e.g., cameras, sonars, etc.) utilizing a track a helical desired trajectory beginning from a depth of
2.13-GHz Quad-core Xeon processor. Navigation and control approximately 0.25 m, traveling to a depth of 2.25 m. Possible
capabilities are located in a separate T-shaped navigation pres- kinematic singularities are avoided by choosing a desired trajec-
sure vessel at the front of the vehicle. The navigation vessel tory that remains sufficiently far from pitch angles of ±90◦ , and
includes: vehicle-independent navigation sensors (e.g., inertial the vehicle is designed such that metacentric restoring forces
measurement unit (IMU), Doppler velocity log (DVL), depth help regulate the pitch and roll of the vehicle.
sensor, temperature sensor, and GPS receiver) and the process- 2) Results: Fig. 3 depicts the time history of the vehicle in
ing capability to unify the data sources. On-board localization, the inertial frame. The circle denotes the starting coordinate and
navigation, and control are computed in the navigation vessel the square denotes the goal coordinate. Tracking errors for the
on a 720-MHz OMAP processor. inertial position and orientation are shown in Fig. 4. The errors
FISCHER et al.: NONLINEAR RISE-BASED CONTROL OF AN AUTONOMOUS UNDERWATER VEHICLE 849

Fig. 4. Tracking errors for the inertial position (top) and attitude (bottom) of Fig. 5. Control efforts commanded about the center of mass of the vehicle.
the vehicle.

TABLE I
RMS TRACKING ERRORS FROM THE CONTROLLED ENVIRONMENT AND
OPEN-WATER STUDIES

are also numerically represented in Table I. The control forces

and moments about the center of mass of the vehicle are shown
in Fig. 5.
Results can be viewed in the video accompanying this
paper at the following url: http://ncr.mae.ufl.edu/index.php?
id=research/sub_RISE. Fig. 6. Time history of the position of the AUV with respect to a linear search
pattern for the RISE controller.
C. Open-Water Study
1) Experimental Setup: An open-water sea trial was com- sufficiently far from pitch angles of ±90◦ , and the vehicle is de-
pleted in the Gulf of Mexico off the coast of Panama City signed such that metacentric restoring forces help regulate the
Beach, FL, USA, to analyze and compare the effectiveness of pitch and roll of the vehicle.
the controller in a real-world environment. The study was com- 2) Results: Fig. 6 illustrates the RISE controller in three-
pleted in shallow water (approximately 5-m deep), with wave dimensional space as a function of time as compared with the
heights of approximately 0.5–1.25 m and a measured current commanded desired trajectory. Fig. 7 illustrates the position
of 0.08 m/s. Gulf currents and surface effects are considered errors in meters and the attitude errors in degrees for the RISE
unknown disturbances to the system, and all coefficients of the controller. The errors are also numerically represented in Table I.
AUV’s dynamic model (i.e., inertia, hydrodynamic forces, etc.) Fig. 8 illustrates the control efforts provided for each controller.
are uncertain. The experiment considers a continuous linear Because of the shallow water trajectory, wave interactions
search pattern that is converted in real time to smooth C 4 tra- and surface effects are apparent in the depth positioning of the
jectories; the pattern consists of segments 2 and 10 m in length. AUV as illustrated in Fig. 6. However, the AUV still maintains
Beginning at the surface, the vehicle submerges to a depth of accurate tracking of the trajectory despite these disturbances.
2 m and follows the linear segments. Possible kinematic singu- The experimental results demonstrate that the control strat-
larities are avoided by choosing a desired trajectory that remains egy provides robustness to both parametric uncertainty and
850 IEEE TRANSACTIONS ON ROBOTICS, VOL. 30, NO. 4, AUGUST 2014

formance is well within the desired performance level to operate

the vehicle. With kinematic modifications, the RISE-based feed-
back control law can also be applied to slender-bodied AUVs.

APPENDIX
PROOF OF THEOREM 1
Proof: Let y ∈ R 19
be defined as
√ T
y  [ zT P] . (17)

In (17), the auxiliary function P ∈ R is defined as the Filippov

solution to the following differential equation:

Ṗ = −rT (Nd − βsgn (e2 ))

n
P (t0 ) = β |e2 i (t0 )| − e2 (t0 )T Nd (t0 ) (18)
i=1

Fig. 7. Inertial position errors (top) and attitude errors (bottom) for the RISE where the subscript i = 1, 2, . . . , n denotes the ith element of
controller.
the vector, and β is chosen according to the sufficient conditions
in (16). Similar to the development in (10), existence of solutions
for P can be established using Filippov’s theory of differential
inclusions for Ṗ ∈ K [h2 ] (e2 , r, t), where h2 : R12 × R+ →
R+ is defined as h2  −rT (Nd − βsgn (e2 )) and K [h2 ] 
δ>0 μS m =0 coh2 (B (e2 , δ) \ Sm , r, t) as in (10). Integrating
(18) by parts and provided the sufficient conditions in (16) are
satisfied, P ≥ 0 (see [24] for details).
 √ 
Let D  y ∈ R19 | y ≤ ρ−1 2 λ1 ks be an  open and
connected set where λ1 = min α1 − 12 , α2 − 12 , 1 , and VL :
D × [0, ∞) → R be continuously differentiable in y, locally
Lipschitz in t, regular and defined as
1 T 1 1
VL (y, t) = e e1 + eT2 e2 + rT M̄ (η (t)) r + P (19)
2 1 2 2
which satisfies the following inequalities:

U1 (y) ≤ VL (y, t) ≤ U2 (y) (20)

where U1 : R19 → R and U2 : R19 → R are positive defi-

nite functions defined as U1  12 min {1, m} y2 and U2 

Fig. 8. Control effort forces (top) and moments (bottom) about the center of max 12 m̄ (η) , 1 y2 .
mass for the RISE controller. Under Filippov’s framework, strong stability of the closed-
loop system ẏ = h3 (y, t) can be established, where h3 : R19 ×
R+ → R19 denotes the RHS of the closed-loop error signals.
measurement noise. The use of industry-standard navigation The time derivative of (19) exists almost everywhere (a.e.), i.e.,
a.e.
sensors (MEMs-based IMU and DVL) illustrate the utility of for almost all t ∈ R+ , and V̇L (y (t) , t) ∈ Ṽ˙ L (y (t) , t) where
the control method on real-world systems. 
Ṽ˙ L =
T
ξ T K [ ėT1 ėT2 ṙT 21 P − 2 Ṗ 1 ]
1

ξ ∈∂ V L (y ,t)
VI. CONCLUSION
where ∂VL is the generalized gradient of VL [36], [37], [40].
A continuous robust controller has been developed and ex- Since VL is continuously differentiable with respect to y
perimentally validated for an AUV. The control scheme com-
Ṽ˙ L ⊂ ∇VLT K[ ėT1 1 − 12 T
pensates for complete model uncertainty yielding semiglobal ėT2 ṙT 1] (21)
2P Ṗ
asymptotic tracking. The control design is implemented on an
AUV in both controlled and open-water environments to il- where ∇VL  [ eT1 eT2 rT M̄ 2P 2 12 rT M̄
1 ˙ r]T .
lustrate the performance of the controller on a physical system. Using the calculus for K [·] from [37], substituting (5), (6),
Despite larger orientation errors in the open-water study, the per- (9), (11), and (18), and canceling similar terms, the expression
FISCHER et al.: NONLINEAR RISE-BASED CONTROL OF AN AUTONOMOUS UNDERWATER VEHICLE 851

in (21) becomes REFERENCES

Ṽ˙ L ⊂ eT1 e2 − α1 eT1 e1 − α2 eT2 e2 + rT Ñ + rT Nd [1] G. Griffiths, in Technology and Applications of Autonomous Underwater
Vehicles, G. Griffiths, Ed. Boca Raton, FL, USA: CRC, 2003.
− (ks + 1) rT r − rT βK [sgn (e2 )] [2] M. Santhakumar and T. Asokan, “Coupled, non-linear control system
design for autonomous underwater vehicle (AUV),” in Proc. Int. Conf.
− rT (Nd − βK [sgn (e2 )]) (22) Control, Autom. Robot., Vis., 2008, pp. 2309–2313.
[3] S. Liu, D. Wang, and E. Poh, “Non-linear output feedback tracking control
where K [sgn(e2 )] = SGN (e2 ) [37] such that SGN (e2 i ) = 1 for AUVs in shallow wave disturbance condition,” Int. J. Control, vol. 81,
pp. 1806–1823, 2008.
if e2 i > 0, [−1, 1] if e2 i = 0, and −1 if e2 i < 0.5 Utilizing the [4] Z. Liu, “Reinforcement adaptive fuzzy control of wing rock phenomena,”
fact that the set in (22) reduces to a scalar equality since the IET Control Theory Appl., vol. 152, pp. 615–620, 2005.
RHS is continuous a.e., i.e., the RHS is continuous except for [5] M. Breivik and T. Fossen, “A unified control concept for autonomous
underwater vehicles,” presented at the Amer. Control Conf., Minneapolis,
the Lebesgue negligible set of times when rT βK [sgn (e2 )] − MN, USA, 2006.
rT βK [sgn (e2 )] = 06 [36], [44], an upper bound for V̇L is given [6] K. Do and J. Pan, “Robust and adaptive path following for underactuated
as autonomous underwater vehicles,” in Proc. Amer. Control Conf., 2003,
vol. 3, pp. 1994–1999.
a.e.
[7] M. Jordan and J. Bustamante, “A speed-gradient adaptive control with
V̇L ≤ −α1 e1 2 + e1  e2  − α2 e2 2 state/disturbance observer for autonomous subaquatic vehicles,” in Proc.
IEEE Conf. Decision Control, 2006, pp. 2008–2013.
+ ρ (z) r z − (ks + 1) r2 . (23) [8] R. P. Kumar, A. Dasgupta, and C. S. Kumar, “A new tracking controller
design for underwater vehicles using quadratic stabilization,” J. Dyn. Syst.
Utilizing Young’s Inequality, the expression in (23) can be Meas. Control, vol. 130, no. 2, pp. 024502-1–024502-6, 2008.
reduced to [9] L. Lapierre and B. Jouvencel, “Robust nonlinear path-following control
a.e. of an AUV,” IEEE J. Ocean. Eng., vol. 33, no. 2, pp. 89–102, Apr. 2008.
V̇L ≤ −λ1 z2 − ks r2 + ρ (z) r z (24) [10] K. Venugopal, R. Sudhakar, and A. Pandya, “On-line learning control
of autonomous underwater vehicles using feedforward neural networks,”
where z was defined in (15). Provided the sufficient conditions IEEE J. Ocean. Eng., vol. 17, no. 4, pp. 308–319, Oct. 1992.
in (16) are satisfied, and by completing the squares for r, the [11] M. Polycarpou, “Stable adaptive neural control scheme for nonlinear sys-
tems,” IEEE Trans. Autom. Control, vol. 41, no. 3, pp. 447–451, Mar.
expression in (24) can be upper bounded as 1996.
[12] J.-H. Li, P.-M. Lee, and Sang-Jeong Lee, “Neural net based nonlinear
a.e. ρ2 (z) z2
V̇L ≤ −λ1 z2 + ≤ −U (y) (25) adaptive control for autonomous underwater vehicles,” in Proc. IEEE Int.
4ks Conf. Robot. Autom., 2002, vol. 2, pp. 1075–1080.
[13] Y. Zhao, J. Farrell, and M. Polycarpou, “Localized adaptive bounds for
where U : R19 → R is positive definite function defined as U  online based control,” in Proc. Amer. Control Conf., 2004, vol. 1, pp. 590–
595.
c z2 , for some positive constant c ∈ R. [14] M. B. Cheng and C. C. Tsai, “Hybrid robust tracking control for a mobile
The inequalities in (20) and (25) can be used to show that VL ∈ manipulator via sliding-mode neural network,” in Proc. IEEE Int. Conf.
L∞ , thus, e1 , e2 , r, P ∈ L∞ . Given that e1 , e2 ∈ L∞ , standard Mechatron., Taipei, Taiwan, 2005, pp. 537–542.
[15] E. Sebastian and M. A. Sotelo, “Adaptive fuzzy sliding mode controller
linear analysis can be used to show that ė1 , ė2 ∈ L∞ from (6) for the kinematic variables of an underwater vehicle,” J. Intell. Robot.
and Assumption 1. Since e1 , e2 , r ∈ L∞ and ηd is sufficiently Syst., vol. 49, no. 2, pp. 189–215, 2007.
smooth based on Assumption 3, (5) and (6) can be used to show [16] L.-J. Zhang, X. Qi, and Y.-J. Pang, “Adaptive output feedback control
based on DRFNN for AUV,” IEEE J. Ocean. Eng., vol. 36, no. 9–10, pp.
that η, ν ∈ L∞ . Property 2, Assumption 2 and (8) can be used 716–722, Jul. 2009.
to show that τn ∈ L∞ . Let SD ⊂ D denote the set defined as [17] A. Aguiar and J. Hespanha, “Trajectory-tracking and path-following of un-
   2  deractuated autonomous vehicles with parametric modeling uncertainty,”
1
SD  y ∈ D | U2 (y) < min {1, m} ρ−1 2 λ1 ks . IEEE Trans. Autom. Control, vol. 52, no. 8, pp. 1362–1379, Aug. 2007.
2 [18] M. Corradini and G. Orlando, “A discrete adaptive variable-structure con-
troller for MIMO systems, and its application to an underwater ROV,”
The region of attraction in SD can be made arbitrarily large IEEE Trans. Control Syst. Technol., vol. 5, no. 3, pp. 349–359, May 1997.
to include any initial conditions by increasing the control gain [19] D. Yoerger and J. Slotine, “Robust trajectory control of underwater vehi-
cles,” IEEE J. Ocean. Eng., vol. 10, no. 4, pp. 462–470, Oct. 1985.
ks . From (25), [45, Corollary 1] can be invoked to show that [20] K. Goheen and E. Jefferys, “Multivariable self-tuning autopilots for au-
c z (t)2 → 0 as t → ∞ ∀y (0) ∈ SD . Based on the defini- tonomous and remotely operated underwater vehicles,” IEEE J. Ocean.
tion of z in (15),e1 (t) → 0 as t → ∞ ∀y (0) ∈ SD .  Eng., vol. 15, no. 3, pp. 144 –151, Jul. 1990.
[21] P.-M. Lee, S.-W. Hong, Y.-K. Lim, C.-M. Lee, B.-H. Jeon, and J.-W. Park,
“Discrete-time quasi-sliding mode control of an autonomous underwater
5 The sgn (·) function can alternatively be defined as sgn (0) = 0; however, vehicle,” IEEE J. Ocean. Eng., vol. 24, no. 3, pp. 388–395, Jul. 1999.
this restriction lacks robustness with respect to measurement noise. As described [22] G. Indiveri, M. Pino, M. Aicardi, and G. Casalino, “Nonlinear time-
in results such as [41]–[43], Filippov’s notion of a solution for discontinuous invariant feedback control of an underactuated marine vehicle along a
differential equations is appropriate to capture the possible closed-loop system straight course,” in Proc. IFAC Conf. Manoeuvr. Control Marine Craft,
behavior in the presence of arbitrarily small measurement noise. By utilizing the 2000, pp. 221–226.
set valued map SGN (·) in the analysis, we account for the possibility that when [23] J. Kim, K. Kim, H. Choi, W. Seong, and K.-Y. Lee, “Estimation of hy-
the true state satisfies x = 0, sgn (x) (of the measured state) falls within the set drodynamic coefficients for an AUV using nonlinear observers,” IEEE J.
[−1, 1]. Therefore, the presented analysis is more robust to measurement noise Ocean. Eng., vol. 27, no. 4, pp. 830–840, Oct. 2002.
than an analysis that depends on sgn (0) to be defined as a known singleton. [24] P. M. Patre, W. MacKunis, K. Kaiser, and W. E. Dixon, “Asymptotic
6 The set of times Λ  {t ∈ [0, ∞) : r(t) T βK [sgn(e (t))] − r(t) T βK
2
tracking for uncertain dynamic systems via a multilayer neural network
[sgn(e2 (t))] = 0} ⊂ R+ is equivalent to the set of times {t : e2 (t) = feedforward and RISE feedback control structure,” IEEE Trans. Automat.
0 ∧ r(t) = 0}. From (7), this set can also be represented by Control, vol. 53, no. 9, pp. 2180–2185, Oct. 2008.
{t : e2 (t) = 0 ∧ ė2 (t) = 0}. Provided e2 (t) is continuously differentiable, [25] N. Fischer, S. Bhasin, and W. E. Dixon, “Nonlinear control of an au-
it can be shown that the set of time instances {t : e2 (t) = 0 ∧ ė2 (t) = 0} is tonomous underwater vehicle: A RISE-based approach,” in Proc. Am.
isolated, and thus, measure zero. This implies that the set Λ is measure zero. Control Conf., San Francisco, CA, USA, 2011, pp. 3972–3977.
852 IEEE TRANSACTIONS ON ROBOTICS, VOL. 30, NO. 4, AUGUST 2014

[26] T. I. Fossen, Handbook of Marine Craft Hydrodynamics and Motion Con- Devin Hughes received the M.S. degree in me-
trol. New York, NY, USA: Wiley, 2011. chanical engineering in 2009 from the University of
[27] J. C. Kinsey, R. M. Eustice, and L. L. Whitcomb, “A survey of underwater Florida, Gainesville, FL, USA, where he is currently
vehicle navigation: Recent advances and new challenges,” presented at the working toward the Ph.D. degree.
IFAC Conf. Manoeuvr. Control Marine Craft, Lisbon, Portugal, 2006. His research interests include real-time path plan-
[28] M.-C. Fang, P. Chang, and J.-H. Luo, “Wave effects on ascending and ning for underwater vehicles and control of auto-
descending motions of the autonomous underater vehicle,” Ocean Eng., mated machinery.
vol. 33, pp. 1972–1999, 2006.
[29] L. Moreira and C. G. Soares, “H 2 and H ∞ designs for diving and course
control of an autonomous underwater vehicle in presence of waves,” IEEE
J. Ocean. Eng., vol. 33, no. 2, pp. 69–88, Apr. 2008.
[30] P. M. Patre, W. Mackunis, C. Makkar, and W. E. Dixon, “Asymptotic Patrick Walters received the M.S. degree in me-
tracking for systems with structured and unstructured uncertainties,” IEEE chanical engineering from the University of Florida,
Trans. Control Syst. Technol., vol. 16, no. 2, pp. 373–379, Mar. 2008. Gainesville, FL, USA, in 2012, where he is currently
[31] B. Xian, D. M. Dawson, M. S. de Queiroz, and J. Chen, “A continuous working toward the Ph.D. degree with the Nonlinear
asymptotic tracking control strategy for uncertain nonlinear systems,” Controls and Robotics Group.
IEEE Trans. Autom. Control, vol. 49, no. 7, pp. 1206–1211, Jul. 2004. His research interests include reinforcement
[32] A. Filippov, “Differential equations with discontinuous right-hand side,” learning-based feedback control, approximate dy-
Amer. Math. Soc. Transl., vol. 42, no. 2, pp. 199–231, 1964. namic programming, and robust control of uncertain
[33] A. F. Filippov, Differential Equations with Discontinuous Right-hand nonlinear systems with a focus on the application of
Sides. Norwell, MA, USA: Kluwer, 1988. underwater vehicles. He also develops autonomous
[34] G. V. Smirnov, Introduction to the Theory of Differential Inclusions. underwater vehicles with the Machine Intelligence
Providence, RI, USA: Amer. Math. Soc., 2002. Laboratory, University of Florida.
[35] J. P. Aubin and H. Frankowska, Set-Valued Analysis. Cambridge, MA,
USA: Birkhäuser, 2008. Eric M. Schwartz (SM’13) received the Ph.D. de-
[36] D. Shevitz and B. Paden, “Lyapunov stability theory of nonsmooth sys- gree in 1995 from the Department of Electrical and
tems,” IEEE Trans. Autom. Control, vol. 39, no. 9, pp. 1910–1914, Sep. Computer Engineering, University of Florida (UF),
1994. Gainesville, FL, USA.
[37] B. Paden and S. Sastry, “A calculus for computing Filippov’s differen- He is a Master Lecturer with the Electrical and
tial inclusion with application to the variable structure control of robot Computer Engineering Department, UF, and the As-
manipulators,” IEEE Trans. Circuits Syst., vol. 34, no. 1, pp. 73–82, Jan. sociate Director of the Machine Intelligence Labora-
1987. tory (MIL). He is also the Director of several MIL
[38] M. de Queiroz, J. Hu, D. Dawson, T. Burg, and S. Donepudi, “Adaptive robot teams including several competition teams. His
position/force control of robot manipulators without velocity measure- robot submarines have competed for 16 years in the
ments: Theory and experimentation,” IEEE Trans. Syst., Man, Cybern., AUVSI Foundation and ONR’s International Robo-
vol. 27-B, no. 5, pp. 796–809, Sep. 1997. Sub Competitions, with his SubjuGator teams winning three times and placing
[39] A. Hanai, H. T. Choi, S. K. Choi, and J. Yuh, “Experimental study on in the top 3 in 10 of the 16 years of this worldwide competition. His robot
fine motion control of underwater robots,” Adv. Robot., vol. 18, no. 10, boat team, PropaGator, participated for the first time and won the 2013 AUVSI
pp. 963–978, 2004. Foundation and ONR’s International RoboBoat Competition. His MIL teams
[40] F. H. Clarke, Optimization and Nonsmooth Analysis. Philadelphia, PA, have also competed in several national lawnmower championships, winning in
USA: SIAM, 1990. the static division once.
[41] H. Hermes, “Discontinuous vector fields and feedback control,” in Differ- Dr. Schwartz has been the Advisor for the IEEE Student Branch and their
ential Equations and Dynamical Systems. New York, NY, USA: Academic, robot team since 2001. He has also been the IEEE Gainesville Section Treasurer
1967. since 2001. He won the 2002–2003 UF Teacher of the Year award. In 2012–
[42] J.-M. Coron and L. Rosier, “A relation between continuous time-varying 2013, he won the University of Florida IEEE and HKN 2012–2013 Teacher of
and discontinuous feedback stabilization,” J. Math. Syst. Estimation Contr the Year award.
ol, vol. 4, no. 1, pp. 67–84, 1994.
[43] R. Goebel, R. Sanfelice, and A. Teel, “Hybrid dynamical systems,” IEEE Warren E. Dixon (M’94–SM’05) received the Ph.D.
Control Syst. Mag.,, vol. 29, no. 2, pp. 28–93, Apr. 2009. degree in 2000 from the Department of Electrical and
[44] R. Leine and N. van de Wouw, “Non-smooth dynamical systems,” in Sta- Computer Engineering, Clemson University, Clem-
bility and Convergence of Mechanical Systems with Unilateral Constraints son, SC, USA.
(Lecture Notes in Applied and Computational Mechanics). Berlin, Ger- In 2004, he joined the Mechanical and Aerospace
many: Springer-Verlag, 2008, vol. 36, pp. 59–77. Engineering Department Faculty, University of
[45] N. Fischer, R. Kamalapurkar, and W. E. Dixon, “LaSalle–Yoshizawa Florida, Gainesville, FL, USA, where he is currently
corollaries for nonsmooth systems,” IEEE Trans. Automat. Control, a Full Professor and a Charles Taylor Faculty Fel-
vol. 58, no. 9, pp. 2333–2338, Sep. 2013. low (2012–2015) and where he holds a University of
Florida Research Foundation Professorship (2012–
2015). He has authored or coauthored three books,
an edited collection, nine book chapters, and more than 250 refereed journal
and conference papers. His main research interests include the development
and application of Lyapunov-based control techniques for uncertain nonlinear
systems.
Dr. Dixon served as an Appointed Member of the IEEE Control Systems
Nicholas Fischer received the Ph.D. degree in 2012 Society Board of Governors (BoG) in 2008 and now serves as the Director of
from the Department of Mechanical and Aerospace Operations for the Executive Committee of the BoG. He is currently or formerly
Engineering, University of Florida, Gainesville, FL, an Associate Editor of the American Society of Mechanical Engineers (ASME)
USA. Automatica; the IEEE TRANSACTIONS ON SYSTEMS MAN AND CYBERNETICS-
He is currently a Systems Engineer with Bluefin PART B: CYBERNETICS; and the International Journal of Robust and Nonlinear
Robotics where his primary role includes simulation Control. He received the 2011 ASME Dynamics Systems and Control Division
modeling, control architecture design, platform au- Outstanding Young Investigator Award, the 2009 American Automatic Control
tonomy, and sensor integration of autonomous un- Council O. Hugo Schuck Award, the 2006 IEEE Robotics and Automation Soci-
derwater vehicles. His research interests include the ety Early Academic Career Award, the National Science Foundation CAREER
development of Lyapunov-based control methods for Award (2006–2011), the 2004 U.S. Department of Energy Outstanding Mentor
uncertain nonlinear systems and their application to Award, and the 2001 Oak Ridge National Laboratory Early Career Award for
unmanned systems. Engineering Achievement.