You are on page 1of 6

Estimation Problems for Autonomous Underwater Vehicle

Under Finite Communication Constraints

Nanang Syahroni1), Young Bong Seo2) and Jae Weon Choi2)


1) Electronics Engineering Polytechnic Institute of Surabaya, email: nanang@eepis-its.edu
2) Pusan National University, Busan, Korea, email: ybseo@pusan.ac.kr, choijw@pusan.ac.kr

Abstract - In this paper, we investigate a linear constraint is considered in the paper. The coder-
optimal control system for design of a depth control estimator sequence algorithm is employed in order to
for autonomous underwater vehicle (AUV) under the transmit measurement data and to estimate the states
of the vehicles.
presence of a finite bandwidth of communication A coder-estimator sequence proposed in [9], and
network. The problem is addressed through the use of recursive coder-estimator sequence based on
remote control and estimation problem via a observation transmitted with finite communication
communication channel with a bandwidth limitation capacity for platoon merging also studied in [10]. In
and the additive noise. The coder-estimator sequence this paper addressing the problem for state estimation
for AUV under the situation where observation and
algorithm is employed in order to transmit control input are sent over communication
measurement data and to estimate the states of the constraints.
vehicles. The ocean depth measurement using In this paper we organized as follows: Section 2
frequency domain to estimate the ocean depth from presents AUV dynamic model. Section 3 gives the
acoustic backscatter returns is presented in detailed. controller algorithm. Section 4 gives recursive coder-
estimator sequence algorithm to estimate the states of
the vehicles. Section 5 gives depth estimation using
Keywords : autonomous underwater vehicle, remote frequency-domain approach. Section 6 presents
control, coder-estimator, communication constraints results from the simulations together with the
assumptions of problems solution. The last section
covers conclusions.
1. INTRODUCTION

In underwater application impose a severe 2. VEHICLE MODEL


restriction on power consumption and present
challenging problems in intelligent control. The vehicle to be controlled is the torpedo type
Navigation is an important requirement for any type AUV with free degree deflection fin to adjust a
of robot, three primary navigation methods for AUV vehicle depth, AUV Model is depicted in figure 1.
are dead-reckoning and inertial navigation [1,2], The simple’s form of equation of motion is
geophysical navigation [3,4,5], and acoustic obtained with body axes coincident with the
navigation [6,7]. principles axes of inertia, and the origin at the center
Acoustic modem technology can enable remote of mass center of gravity (CG), for this case the
control of AUV, but communication bandwidth and equation in the dimensionless form as in [11] are:
reliability are restricted by challenging acoustic
propagation conditions. Acoustic electromagnetic X  m[u  qw   r ]
energy propagates well in the ocean, and hence Y  m[  ru  pw]
acoustic transponders can be used as beacons to
Z  m[ w  pv  qu ]
guide the motion of an AUV without the need for (1)
resurfacing. Two types of system have been primarily K  I x p  ( I z  I y )qr
employed [6,7], long baseline (LBL) and ultra-short M  I y q  ( I x  I z )rp
baseline (USBL). Both systems employ external N  I z r  ( I y  I x ) pq
transducers or transducer arrays as aids to compute
position fixes by locating the intersection point of
spheres of appropriate radiuses from the beacons in The 6DOF components of the rigid body dynamic
the array, and integrate the raw time-of-flight equations of motion of the submerged vehicle are:
measurements into an appropriate Kalman filter [8].
A remote control and estimation problem via a
communication channel with a finite capacity
X  m[u   r  wq  xG (q 2  r 2 )  yG ( pq  r)  zG ( pr  q )] Dv  (m  Z w )( I y  M q )  (mxG  Z q )(mxG  M w )
zGB  zG  z B
Y  m[  ur  wq  xG ( pq  r)  yG ( p 2  r 2 )  zG (qr  p )]
(mxG  Z q )W
Z  m[ w  uq   p  xG ( pr  q )  yG (qr  p )  zG ( p 2  q 2 )] a21  
Dv
K  I x p  ( I z  I y )qr  I xy ( pr  q )  I yz (q 2  r 2 )  I xz ( pq  r)
(2) ( I y  M q ) Z w  (mxG  Z q ) M w
a22 
 m[ yG ( w  uq   p )  zG (  ur  wp )] Dv
( I y  M q )(m  Z q )  (mxG  Z q )( M q  mxG )
M  I y q  ( I x  I z ) pr  I xy (qr  p )  I yz ( pq  r)  I xz ( p 2  r 2 ) a23 
Dv
 m[ xG ( w  uq   p )  zG (u   r  wq )] (mxG  Z q )W
a31 
N  I z r  ( I y  I x ) pq  I xy ( p 2  q 2 )  I yz ( pr  q )  I xz (qr  p ) Dv
(m  Z w ) M w  (mxG  M w ) M 
 m[ xG (  ur  wp )  yG (u   r  wq )] a32 
Dv
(m  Z w )( Mq  mxG )  (mxG  M w )(m  Z q )
where, X, Y, and Z are surge, sway, and heave force; a33 
Dv
K, M, and N are roll, pitch, and yaw moment; p, q, ( I y  M q ) Z  (mxG  Z q ) M 
b1 
and r are roll, pitch, and yaw rate; u, v, and w are Dv
surge, sway, and heave velocity; x, y, and z are body ( I y  Z w ) M   (mxG  M q ) Z
b2 
fixed axes in positive forward, positive starboard, and Dv
positive down; Ix, Iy, and Iz is vehicle mass moment
of inertia around the x-axis, around the y-axis, and
around the z-axis; xG, yG, and zG are longitudinal 3. CONTROLLER DESIGN
position, athwart position, and vertical position of
center of gravity;  ,  , and  are roll, pitch, and Fig. 2 shows the considered systems, depth
yaw angle. control is completed if the AUV arrives at desired
We can further simply equations (2) by assuming depth subtracted by safety depth and vehicle size, in
that yG is small compared to the other terms. After order words:
several steps of linearization as in [12,13], vertical zv  rv  z s  Lv (5)
motion equations become:
  q
The controller based on linear quadratic (LQ)
servo regarding to the reference input. Consider state
( m  Z w ) w  ( mxG  Z q ) q  Z wUw  ( m  Z q )Uq  U 2 Z 
(3) space model:
(  M w  mxG ) w  ( I y  M q ) q  ( zGW  z B )  M wUw
 ( M q  mxG )Uq  M  U 2
x (t )  Ax(t )  Bu (t ) (6)
z  -U   w
with y (t )  C p x(t ) , and C p is [0m( n  m) I mm ] .
It can rewritten in the matrix form
LQ servo basically base on LQ optimal control
1 0 0 0    problem to find a control law u (t )  Gx(t ) to
  
 m  Z w  (mxG  Z q ) 
0
0  mxG  M w  ( I y  M q )
0   w 
0   q 
 
minimize J  [ xT (t )Qx(t )  u T (t ) Ru (t )]dt , where
0
   G  R 1 BT P and P is solution of Riccati equation.
0 0 0 1   z 

0 0
1
0     0  A control gain G consists of G  [G y Gr ] , where

 0 Z wV  m  Zq V   
0   w  Z V 2 

Gy  R mm and Gr  R m( n  m ) , control input became:
  z W  z B  M V  
 G B w  
M q  mxG V 0   q   M  V 2 
 U 1 0   z   0  u (t )  G y y (t )  Gr xr (t )  G y r (t ) (7)
0  

Substitute it into state space model we have


Finally, we have: differential equation of close-loop system:

   0 0 1 0     0  x (t )  [ Ar  BG y C p  BG y D p ]x(t )  BG y r (t ) (8)


    
 w    a21VGB a22V a23V 0   w b1V 2  (4)
 
 q   a31 zGB a32V a33V 0   q  b2V 2 
      
 z   V 1 0 0   z   0 
where:
the kth state at time 4in , where   1 / R , and R
bits per second is data rate of the communication
channel.
Denote by M j  ( j / 24 n )(U i  Di ) , we have:

xˆi , k ( ci ,k  j )  Di , k  M j , k
 i 1, k ( ci ,k  j )  max( Di , k  M j 1, k  Vk , Di , k  Vk )
(15)
 i 1, k ( ci ,k  j )  min( Di , k  M j 1, k  Vk , Di , k  Vk )

Finally, using the estimated value in (15) and


from (7) we can obtain the control input:
ui , k  Gi , y yi , k  Gi , r xˆi , k  Gi , y ( rv|i ,v  z s  Lv ) (16)
Fig. : AUV Model

4. CODER ESTIMATOR 5. DEPTH ESTIMATION

A coder-estimator sequence is a sequence of The acoustic echo sounding methods provide


depth data along a path directly beneath the UAV.
triples {( hi , ti , Xˆ i )}i  0 , where {( hi , ti )}i  0 a coder
 
The ocean depth can be measured from the time
is sequence and Xˆ i is an estimator based on delay between transmit and reflections from bottom
of the ocean. Since the backscatter bottom return is
{c0 ,..., ci }. much weaker than the transmitted sound, the most
critical performance factor in depth measurement
The sampled-data representation is:
accuracy is the resolution between transmitted sound
xi 1, k  f ( xi , k , ui ,k , t )  Gk wi , k , for k  1,..., 4 (9) and bottom reflections.
For the purpose of enhancing the precision of
and depth estimation, a number of signal processing
techniques have been proposed in recent years. In
yi 1,k  xi 1,k  vi 1, k , for k  1,..., 4 (10) this paper, we are addressing the depth measurement
problem in the frequency domain by making use of
Where xi ,k the kth element of the state x is at some of the high resolution signal subspace.
Classical high resolution frequency domain base
time it , t is the sampling interval, wi , k is the
method is used to estimate the ocean depth from
measured output, and vi , k is the measurement noise. backscatter noisy returns.

A pdf for x0 , wi , and vi for all i have interval Consider a transmitted signal f (t ) that gets
[ X 0 , X 0 ],[W , W ], and [-V ,V ] respectively. returned from the surface and bottom of the ocean. If
T1 and T2 represent the two round trip delays from
Let consider coder-estimator sequence algorithm the transmitter to the ocean surface and its bottom
representation: and back to the receiver, respectively, then the
received signal can be written as:
 0    X 0  (11)
  
 0   X 0 
r (t )  a1 f (t  T1 )  a2 f (t  T2 )  n(t ) (17)
for i  0 and k  1,...4,
where a1 and a2 represent the attenuation
We define: coefficients to and from the top and bottom surfaces
of the ocean, respectively, and n(t ) represent the
U i ,k  max fk ( X k , ui , k , 4n )  GkWk  Vk , (12) noise present.
X k [ i ,k ,  i ,k ]
Given the received waveform r (t ) over some
Di , k  min fk ( X k , ui , k , 4n )  GkWk  Vk , (13)
X k [ i ,k ,  i ,k ] duration, the problem is to fit a model as in (16) and
And use it to obtain the two delays T1 and T2 . The ocean
depth is given by c(T2  T1 ) / 2r , where c is the
hi , k ( yi , k )  j if Di , k  M j 1, k  yi , k  Di , k  M j , k (14)
velocity of propagation of sound in the ocean and r is
where hi , k ( yi , k ) is a coding function from a the refractive index of sea water. The above
wideband time-delay estimation problem has a nice
source ( yi ,k ) defined on the probability space to interpretation in the frequency domain. Consider the
the set (ci ,k ) of finite-length string of symbols for Fourier transform of (16), which assumes the form:
R ( )  a1F ( )e  jT1  a2 F ( )e  jT 2  N ( ) (18)  1 
 jTk 
e 
In the (17) represents sinusoidal components that sk   , k  1, , K . (26)
 
modulate the common signal F ( ) . The problem of  
obtaining the unknowns T1 and T2 in (17) is similar  e jnTk 
to sinusoidal frequency estimation in the time domain.
To estimate T1 and T2 , two high resolution sk represent the delay vector associated with
schemes are described, first for the noiseless and then unknown delay Tk . Equation (23) show that rank H
in noisy situation. If N ( )  0 , (17) can be expressed = rang A = K, as illustrated in [15]. Then H is
as: singular and n – K + 1 eigenvalues of H are zero. The
eigenvectors associated with the distinct eigenvalues
K of H are orthogonal. In particular, let
R ( )
F ( )
 a e
k 1
k
 jTk
(19) (u1 , u2 , , uK ) represent eigenvectors associated with
the K nonzero eigenvalue of H, and then
(u1 , u2 , , uK ) span a K-dimensional subspace.
where K=2 in our problem. If we let Tk  Tk and
Let 1 , 2 , , K represent the nonzero
define:
eigevalues of H, and K 1 , K  2 , , n 1 the zero
R ( ) eigenvalues of H.
ri 
F ( )
|  i
, i  0,1, 2, (20)
Hui  i ui , i  1, , K (27)

then, we get:
using (23) and (27), we obtain
K
ri  a e
k 1
k
 jTk
, i  0,1, 2, (21) S * AS H ui  i ui (28)

it can rewritten as:


The structure present in (21) can be used to
estimate the unknowns Tk , k  1, , K as describe in S * i  0, i  1, , K (29)
[15].
Consider the symmetric Hankel matrix: where i  AS H ui / i
Equation (28) can be expressed as:
 r0 r1  rn 
r r2  rn 1  U  [u1 , u2 , , u K ]
H 1 (22)
      S *[1 ,  2 , ,  K ] (30)
 
 rn rn 1  rn  2   S *
If U1 and U2 represent the first and the last n rows
The above Hankel matrix allows the direct of U, respectively, then proceeding as in [15] we
evaluation of the unknown delays Tk , k  1, , K , have:
even when the attenuation constants
ak , k  1, , K are complex. U1  U 2  S1* ( I   B* ) (31)
By direct calculation, it is easy to verify that the
above H can be expressed as: And, consequently, the K generalized eigenvalues
of matrix pair (U1,U2) are given by:
H  S * AS H (23)
Where  k  e jTk , k  1, , K . (32)
Alternatively, the expression
S  ( s1 , s2 , , sK ), (24)  I  (U 2H U 2 ) 1U 2H U1 can be used to obtain the
 a1  above unknowns.
 a2 
A  (25)
   6. SIMULATION
 
 aK 
And The vehicle to be controlled is the torpedo type
AUV with free degree deflection fin to adjust a
vehicle depth. The controller design based on linear
quadratic (LQ) servo, and the depth control is
completed when AUV arrives at current depth where
the desired depth is subtracted by safety depth and
vehicle size.
For simulation, we suppose to control pitch angle
θ of AUV to near zero and depth z nearby -5 m for
first 30 times counter duration, then change depth z
to -14 m until 70 times duration, and etc., we assume
AUV reach 5o angle of dive when pitch angle
deviates to 5o from zero, and the AUV reaches the
desired depth with 0.05m of deviation.
We refer to the physical parameter of NPS AUV1
in [14], the state space equation:

 0 0 1.0000 0   0  Fig. 2: Remote Controlled System


 0.0177 -0.0342 -0.4538 
0  0.1334 
A ,B ,
 -0.0317 -0.0342 -2.0702 0  7.9401  0
    Desired Depth
 -9.1440 1.0000 0 0  0 
-2 Actual Depth

C   0 0 0 1 , D   0
-4

-6

We assume there is no noise disturbance by channel


z (m)

-8
and that measurements are directly observed by
-10
sensor. Measurement noise of pitch angle, heave rate,
pitch rate, and depth are mutually independent zero -12

mean white Gaussian noise with standard deviation


-14
of 0.2o, 0.5m/s, 0.3m/s, and 0.15m.
For LQ Servo control, obtained gain from -16
0 5 10 15 20 25 30 35 40 45 50
R 1 BT P where P is solution of Riccati equation, we time (sec)

use simple weighting matrices Q = CTC, and R = 0.1. Fig. 3: Depth of AUV
The obtained gains are applied for remote controller
system including noise.
The regular square linear system (22) can be 7. CONCLUDING REMARKS
solved quicker with very simple recursive method
using Levinson-Durbin [16]. The two delay Coder-Estimator sequence is employed to
T1 and T2 as well as their strengths are estimated in transmit the measurement and to estimate the state
above method. T1 and T2 correspond to the time for feedback control. We believe that the employed
delays to and from target or bottom of the ocean. coder-estimator sequence algorithm for a
Extraction of target range R is computed by communication network is suitable for underwater
measuring the time delay t as summation from two environment in order to deal with limitation of
communication bandwidth.
delays a pulse to travel the two-way path between the
Linear predictive method has been formulated an
acoustic sound transmitter and the target T1 and T2 .
applied in the frequency domain to estimate the
Where R  ct / 2 is in meters, t is in seconds, c is depth of ocean by utilizing on Hankel matrix
speed of acoustic sound in m/s, and the factor 1/2 is structure decomposition.
needed to account for two way time delay. Although the simulation with estimation problem
Fig 3 shows the actual depth of AUV presented in here is quite simple, it serves the illustrations of the
solid line with the desired depth presented in solid proposed algorithm of communicating feedback
dashed line. The time response of the differential control with finite bandwidth and estimate the
equation of close-loop system is computed by measurement, even in small and noisy data.
classical numerical iteration Runga-Kutta 4th order. The major portion of this paper was devoted to
As we predicted that the actual depth fluctuated with all state and measurement estimation with simple,
maximum distance of 0.75m from desired depth, the inexpensive and effective for communication
value of measurement depth reach almost 1.3m from network with finite capacity. Future research should
desired depth. be directed at acquiring a better understanding of
the problems in implemented system modeling and
developing appropriate methodologies to deal with
these problems.
REFERENCES [10] . J. W. Choi, T. H. Fang, S. Kwong, and Y. H.
Kim, “Remote-Controlled Platoon Merging
[1] . E. S. Maloney, Dutton's Navigation and via Coder-Estimator Sequence Algorithm for
Piloting, Naval Institute Press, Annapolis, a Communi-cation Network”, IEEE
1985. Transaction on Industrial Electronics, vol.50,
[2] . Cox and G. Wilfong, Autonomous Robot no.1,pp. 30-36, February 2003.
Vehicles, Springer-Verlag, 1990. [11] . Sname, The Society of Naval Architects and
[3] . C. Tyren, “Magnetic Anomalies as a Marine Engineers, Nomenclature for Treating
Reference for Ground-Speed and Map- the Motion of a Submerged Body Through a
Matching Navigation”, The Journal of Fluid, Technical Research Bulletin, no.1-5,
Navigation, vol. 35, pp. 242-254, May 1982. USA, 1950.
[4] . E. Geyer, P. Creamer, J. D'Appolito, and R. [12] . R. Cristi, F. A. Papoulias, and A. J. Healey,
Gains, “Characteristics and Capabilities of “Adaptive Sliding Mode Control of
Navigation Systems for Unmanned Autonomous Underwater Vehicles in the Dive
Untethered Submersibles”, Proceeding of Plane”, IEEE Journal of Oceanic Engineering,
International Symposium on Unmanned vol. 15, pp. 152-160, Piscataway, NJ, USA,
Untethered Submersible Technology, pp. 320- 1999.
347, 1987. [13] . J. S. Riedel, “Pitchfork Bifurcations and Dive
[5] . S. T. Tuohy, J. J. Leonard, J. G. Bellingham, N. Plane Reversal of Submarines at Low Speeds”,
M. Patrikalakis, and C. Chryssostomidis, Engineer’s Thesis, Naval Postgraduate School,
“Map Based Navigation for Autonomous Monterey, California, USA, 1993.
Underwater Vehicles”, International Journal [14] . J. Healey, P. A. Papoulias, and R. Cristi,
of Offshore and Polar Engineering, vol. 6, pp. “Design and Experimental Verification of a
9-18, March 1996. Model Based Compensator for Rapid AUV
[6] . D. B. Heckman and R. C. Abbott. “An Depth Control”, Proceeding of the
Acoustic Navigation Technique”, IEEE International Symposium on Unmanned
Oceans, pp. 591-595, 1973. Untethered Submersible Technology, pp.
[7] . M. Hunt, W. Marquet, D. Moller, K. Peal, W. 458-474, Washington DC, USA, 1989.
Smith, and R. Spindel, “An Acoustic [15] . S. R. Pillai and A. Antoniou, “A Frequency-
Navigation System”, Technical Report WHOI- Domain Approach to Shallow-Water Depth
74-6, Woods Hole Oceanographic Institution, Measurement”, IEEE Transactions on
1974. Geoscience And Remote Sensing,” vol. 35, no.
[8] . J. Vaganay, J. G. Bellingham, and J. J. 3, pp. 540-545, May 1997.
Leonard, “Outlier Rejection for Autonomous [16] . J. Makhoul, “Linear Prediction: A Tutorial
Acoustic Navigation”, Proceeding of IEEE Review,” Proceedings of the IEEE, Vol. 63,
International Conference Robotics and No. 4, pp. 561–580, April 1975.
Automation, pp. 2174-2181, April 1996. [17] . R. Roy, A. Paulraj, and K. Kailath, “ESPRIT –
[9] . W. S. Wong and R. W. Brockett, “System with A Subspace Rotation Approach to Estimation
Finite Communication Bandwidth Constraint of Parameters of Cissoids in Noise”, IEEE
– Part I: State estimation problem”, IEEE Transactions on Accoustic, Speech, Signal
Transaction on Automatic Control, vol. 42, pp. Processing, vol. ASSP-34, pp. 1340-1342,
1294-1299, September 1997. October 1986.

You might also like