An Innovative Navigation Strategy for Autonomous Underwater Vehicles

Proceedings of the ASME 2015 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference IDETC/CIE 2015 August 2-5, 2015, Boston, Massachusetts, USA

DETC2015-46432

AN INNOVATIVE NAVIGATION STRATEGY FOR AUTONOMOUS UNDERWATER VEHICLES: AN UNSCENTED KALMAN FILTER BASED APPROACH

Benedetto Allotta Riccardo Costanzi Enrico Meli Alessandro Ridolfi

Luigi Chisci Claudio Fantacci

Department of Information Engineering University of Florence 50139, Florence, Italy Department of Industrial Engineering Email: [email protected] University of Florence 50139, Florence, Italy [email protected] Email: [email protected] [email protected] [email protected] [email protected]

ABSTRACT

Andrea Caiti Francesco Di Corato Davide Fenucci Centro Enrico Piaggio University of Pisa 50122, Pisa, Italy Email: [email protected] [email protected] [email protected]

sically adopted in the field of underwater robotics and the UKF is given. These navigation algorithms have been experimentally validated through the data related to some sea tests with the Typhoon class AUVs, designed and assembled by the Department of Industrial Engineering of the Florence University (DIEF) for exploration and surveillance of underwater archaeological sites in the framework of the THESAURUS and European ARROWS projects. The comparison results are significant as the two filtering strategies are based on the same process and sensors models. At this initial stage of the research activity, the navigation algorithms have been tested offline. The presented results rely on the experimental navigation data acquired during two different sea missions: in the first one, Typhoon AUV #1 navigated in a Remotely Operated Vehicle (ROV) mode near Livorno, Italy, during the final demo of THESAURUS project (held in August 2013); in the latter Typhoon AUV #2 autonomously navigated near La Spezia in the framework of the NATO CommsNet13 experiment, Italy (held in September 2013). The achieved results demonstrate the effectiveness of both navigation algorithms and the superior-

Developing reliable navigation strategies is mandatory in the field of Underwater Robotics and in particular for Autonomous Underwater Vehicles (AUVs) to ensure the correct achievement of a mission. Underwater navigation is still nowadays critical, e.g. due to lack of access to satellite navigation systems (e.g. the Global Positioning System, GPS): an AUV typically proceeds for long time intervals only relying on the measurements of its on-board sensors, without any communication with the outside environment. In this context, the filtering algorithm for the estimation of the AUV state is a key factor for the performance of the system; i.e. the filtering algorithm used to estimate the state of the AUV has to guarantee a satisfactory underwater navigation accuracy. In this paper, the authors present an underwater navigation system which exploits measurements from an Inertial Measurement Unit (IMU), Doppler Velocity Log (DVL) and a Pressure Sensor (PS) for the depth, and relies on either an Extended Kalman Filter (EKF) or an Unscented Kalman Filter (UKF) for state estimation. A comparison between the EKF approach, clas-

1

Copyright © 2015 by ASME

ity of the UKF without increasing the computational load. The algorithms are both affordable for online on-board AUV implementation and new tests at sea are planned for spring 2015.

1

Introduction Filtering algorithms for the estimation of the AUV state are a key factor to guarantee a satisfactory underwater navigation accuracy. In this paper, the authors present a navigation system specifically designed for AUVs and its experimental evaluation in different underwater missions. The developed system exploits inertial sensors (Inertial Measurement Unit - IMU), depth sensors, Doppler Velocity Log (DVL) and Global Positioning System (GPS) fixes and relies on Unscented Kalman Filter (UKF) ( [1], [2], [3]) for motion estimation. A comparison between the Extended Kalman Filter (EKF) approach, classically adopted in the field of underwater robotics and the UKF one is given. The navigation algorithm have been experimentally validated through the data related to some sea tests with the Typhoon class AUVs, designed and assembled by the Department of Industrial Engineering of the Florence University (DIEF) during the THESAURUS and European ARROWS projects. At this initial stage of the research activity, the navigation algorithms have been tested offline. The presented results rely on the experimental navigation data acquired during two different sea missions: in the first one Typhoon AUV #1 (TifOne in the following) navigated in a Remotely Operated Vehicle (ROV) mode near Livorno, Italy, during the final demonstration of THESAURUS project (held in August 2013); in the latter Typhoon AUV #2 (TifTu in the following) autonomously navigated near La Spezia (see Fig. 1, ( [4], [5]) in the framework of the NATO CommsNet13 experiment, Italy (held in September 2013). In the first scenario the experimental tests have been performed using the vehicle in remotely operated (ROV) mode and has consisted of two different phases: a surface navigation part (to exploit the GPS signal as ground truth for evaluating the navigation accuracy) and an underwater navigation part (wherein only GPS fixes during the surfacing are available). In the second scenario the navigation data are obtained thanks to the permanent testbed for underwater networking and communication purposes (LOON Littoral Ocean Observatory Network) of the Center for Maritime Research and Experimentation (CMRE, formerly NURC).

FIGURE 1.

TifTu AUV during the sea tests in La Spezia (Italy).

concerns the further developments of this research activities, the authors are going to test the proposed filtering algorithm online: according to this aim new tests at sea are already scheduled for spring 2015.

2

General architecture The typical test architecture is represented in the scheme in Fig. 2.

FIGURE 2.

Online testing of navigation systems.

The Typhoon AUV navigates in dead reckoning using the information coming from the available sensors. The physical quantities characterizing the Typhoon AUV dynamics (speed, orientation, depth and position) are measured by the set of on-board sensors. The measured quantities are then processed in dead reckoning to approximately estimate the vehicle dynamics. Starting from the estimated vehicle dynamics, the control is able to calculate the motor control signals ( [6]). Finally, the thrusts produced by the motors allow the vehicle to follow the desired dynamics.

A performance comparison between the proposed UKFbased navigation system and a standard EKF (Extended Kalman Filter) based system ( [6]) has been carried out. The achieved results demonstrate the effectiveness of both navigation algorithms and the superiority of the UKF (very suitable for AUV navigation and, up to now, still not used much in this field) in estimating the vehicle dynamic behaviour without increasing the computational load (affordable for online on-board AUV implementation). As

2


The sensor suite provides the real physical outputs to be used to test and compare the different navigation systems (i.e., the EKF-based and UKF-based ones). The preliminary validation and analysis of the navigation system presented by the authors in this work are performed offine. The adopted architecture for offine testing is depicted in Fig. 3.

FIGURE 3.

sions can be seen.

FIGURE 4. The Typhoon CAD design and its final built versions on the NATO Alliance RV (CommsNet13).

Offline testing of navigation systems.

The test of the navigation system relies on the data measured by the sensor suite (DVL, IMU, pressure sensor and GPS during periodic resurfacings). The measured quantities (speed, orientation, depth and position) are processed by both EKF and UKF algorithms in order to recursively generate estimates of the AUV state. Then, such estimates are compared to the USBL position fixes during the whole mission.

Typhoon propulsion system is composed of 6 actuators: two lateral thrusters, two vertical thrusters and two main rear propellers (both working in longitudinal direction). The propulsion system actively controls 5 Degrees Of Freedom (DOFs) of the vehicle (the only one left passive is the roll one).

3

3.1.1 Vehicle modelling To analyse the motion of the AUV, two different reference frames, shown in Fig. 5, are needed ( [9], [10]):

The Typhoon AUVs In this chapter, the main parts of the Typhoon AUV architecture for the testing of navigation systems are described.

• the body frame: reference frame with origin Ob placed in the center of mass of the body and axes lined up to the main inertia axes of the body itself. • the fixed frame: inertial reference frame, defined with the origin On placed on the surface and axes lined up to the ones of a NED (North-East-Down) frame.

3.1

The vehicles Typhoon is a middle-sized AUV class able to reach a 300 [m] depth. The vehicles can carry suitable payload for the specific underwater mission to perform. Currently, two Typhoons AUVs are fully operative and already performed many missions at sea: the vehicles are called TifOne and TifTu. Typhoon class AUVs have a length of 3700 [mm], an external diameter of 350 [mm] and a weight of about 150 [kg] according to the carried payload (the vehicle can be considered an intermediate one compared to the smaller Remus 100 ( [7]) and the bigger Remus 600 ( [8])). Its autonomy is 8 − 10 [h] and the maximum reachable longitudinal speed is 6 [kn] (whereas the cruise speed is about 2 [kn]). The power needed by the propulsion on-board systems and payloads was approximately known (about 350 [W ]): considering a mission duration of about 8 − 10 [h], the needed energy was calculated in about 3 − 3.5 [kW h]. Li-Po (LithiumPolymer) batteries have been chosen. In Fig. 4, both the Typhoon CAD design and its final built ver-

FIGURE 5.

3

Body and fixed reference systems.


To describe the motion of the system, the following coordinate vectors have to be introduced: η=

η1 η2

=

Ob Φ

,

ν=

ν1 ν2

=

b O˙ ωb

mation algorithm; • STS DTM depth sensor: it is a digital pressure sensor used to measure the vehicle depth at a 10 [Hz] working frequency; DVL Teledyne Explorer: the sensor measures the linear speed of the vehicle, with respect to the seabed or with respect to the water column beneath the vehicle (working frequency of 7 [Hz]). Moreover, if it detects the seabed it is also able to measure the distance from it (like an altimeter). Due to its accuracy and reliability this sensor is quite expensive. • GPS: the global positioning system is included into the industrial PC-104. This sensor, working at 1 [Hz], is employed to get suitable fixes only during periodic and dedicated vehicle resurfacing; • USBL 18/34 by Evologics: this sensor is not exploited by the navigation systems and is only used as an external benchmark to evaluate the algorithm accuracy. The sensor works at a maximum working frequency of 0.2 [Hz]. Actually, the real working frequency can be quite lower depending on the quality of the acoustic channel.

! (1)

where η represents the vector of position η 1 and orientation η 2 (expressed in terms of Euler’s angles) in the fixed frame and ν includes the components of linear ν 1 and angular ν 2 velocities expressed in the body frame. Clearly, the previous physical quantities are linked together by the following kinematic equations:

η˙ 1 η˙ 2

J1 (η 2 ) 03x3 03x3 J2 (η 2 ) n Rb (Φ) 03x3 ν1 = ν2 03x3 T −1 (Φ) η˙ = J(η)ν,

=

ν1 ν2

= (2)

in which Rnb is the rotation matrix between the body and the fixed reference systems and T is the so called Euler’s matrix. On the other hand, the motion of the AUV is governed by the vehicle dynamic equations: M(ν)ν˙ +C(ν)ν + D(ν)ν + g(η) = τ(ν, u)

The sensor set available for the two existing versions of Typhoon class AUVs is summarized in table 1.

(3)

where M is the mass matrix, C is the centrifugal and Coriolis matrix, D is the drag matrix, g is the gravity and buoyancy vector and τ(ν, u) are the resultant forces and torques acting on the vehicle (u are the control signals of the vehicle motors, i. e. the moT tor rotation speeds). Introducing the state vector x = ν T η T , the system equations can be summarized as follows: x˙ = F(x, u) + v = M(ν)−1 τ(ν, u) −C(ν)ν − D(ν)ν − g(η) = +w J(η)ν

Sensor

TifOne

TifTu

Xsens MTi IMU

X

X

KVH DSP 1750 FOG

X

×

Teledyne Explorer DVL

X

×

STS DTM PS

X

X

Evologics S2CR 18/34 USBL

Localizable Modem

Localizing Transducer

PC-104 GPS

X

X

TABLE 1.

Sensors sets for TifOne and TifTu

(4) 3.2.1 Sensors modelling Hereafter, the measurement equations modelling the sensor behaviour will be derived by taking into account the main features of the employed sensors and the main noise sources affecting the measurements:

where w is the process noise. 3.2

The sensors In the following the set of on-board sensors employed during the navigation by the Typhoon AUV is briefly described:

• the linear speed of the vehicle ν 1 measured by the DVL Teledyne Explorer:

• IMU Xsens MTi: the device consists of a 3D gyroscope, a 3D accelerometer and a 3D magnetometer providing measurements at a maximum working frequency of 100 [Hz]. The device estimates the orientation of the vehicle in a 3D space in an accurate way by means of an embedded attitude esti-

ν meas = ν 1 + bν1 + δ ν1 1

(5)

is the linear velocity, bv1 is the bias error and where ν meas 1 δ v1 is the measurement noise;

4


meas and η are the measured and, respectively, true where η1z 1z depth and δz is the measurement noise.

• vehicle orientation η 2 provided by the IMU Xsens MTi (including 3D gyroscope, 3D accelerometer and 3D magnetometer) through the attitude estimation filter starting from the following measures:

Introducing the measurement vector z = T measT measT measT η1 η2 ν1 , the measurement equations can be summarized as follows:

• Vehicle angular velocity ν 2 measured by the 3D gyroscope:



= ν 2 + δ ν2 ν meas 2

 I3x3 O3x3 O3x3 O3x3 z = h(x) + v, h(x) =  O3x3 O3x3 I3x3 O3x3  x, O3x3 O3x3 O3x3 I3x3 T v = δ Tν1 δx δy δz δ Tη2 .

(6)

and ν 2 denote the measured and, respecwhere ν meas 2 tively, true angular velocity and δ ν2 is the measurement noise; • Vehicle linear acceleration measured by the 3D accelerometer: ameas = a + δ a

At each time step, the dimension of the measurement vector z can be different, depending on the available measurements: the sensors have different working frequencies and may not be always available (see for instance the GPS). Despite the different working frequencies of the various involved sensors, the correction step of the filter runs recursively at a desired constant rate (10 [Hz] for the implemented filters). In order to manage the variable size of the measurement vector, the matrix describing the h(x) function in equation 12 is resized accordingly by eliminating the rows corresponding to the missing measurements for the particular instant.

(7)

where ameas and a denote the measured and, respectively, true linear acceleration and δ a is the measurement noise; • Magnetic field M measured by the 3D magnetometer: M meas = M + δ M

(8) 4

where M meas and M denote the measured and, respectively, true magnetic field and δ M is the measurement noise.

(9)

xk = f (xk−1 , uk−1 ) + wk−1 = = xk−1 + ∆t F(xk−1 , uk−1 ) + wk−1

(14)

zk = h(xk ) + vk . The standard navigation system used to estimate the vehicle dynamics is based on the classical EKF estimator (tested offline using the data provided by the Typhoon AUV) ( [11], [12], [13], [5], [6]).

(10)

meas , η meas and η , η are the measured and, rewhere η1x 1x 1y 1y spectively, true positions and δx , δy are the measurement noises. • vehicle depth η1z measured by the STS DTM depth sensor: meas η1z = η1z + δz

(13)

has been discretized by the Euler method with sampling interval ∆t = 0.01s taking into account the different working frequencies of the employed sensors:

where δ η2 is a further measurement noise. • vehicle position η1x and η1y measured by the GPS sensor: meas = η + δ , η meas = η + δ η1x x y 1x 1y 1y

EKF-based navigation system The complete system describing the AUV motion x˙ = F(x, u) + w, z = h(x) + v

By suitably processing the above measurements, the attitude estimation filter is able to estimate the vehicle orientation η 2 that will be used in the subsequent developments of this paper as a virtual measurement η meas modelled as: 2 η meas = η 2 + δ η2 2

(12)

5

UKF-based navigation system The UKF is based on the Unscented Transform (UT), a derivative-free technique capable of providing a more accurate

(11)

5


statistical characterization of a Random Variable (RV) undergoing a nonlinear transformation ( [1], [2]). In particular, the UT is a deterministic technique suited to provide mean and covariance of a given RV subjected to a nonlinear function given a minimal set of its samples. The pseudo-code of the UT is reported in Fig. 6. Taking into account mean x and associated covariance matrix P of a generic RV as well as a transformation function g(·), the UT proceeds as follows:

α, β and κ. Moment matching properties and performance improvements are discussed in ( [1], [2]) resorting to specific values of α, β and κ. It is common practice to set these three parameters as constants computing the weights at the beginning of the estimation process. The UT can be applied in the KF recursion allowing to adopt a nonlinear recursive estimator known as UKF ( [1]). The pseudo-code of the UKF is shown in Fig. 8.

• generates 2nx + 1 samples X ∈ Rnx ×(2nx +1) , the so called σ -points, starting from the mean x with deviation given by the matrix square root Σ of P; • propagates the σ -points through the transformation function g(·) resulting in G ∈ Rnx ×(2nx +1) ; • calculates the new transformed mean xˆ and associated covariance matrix Pgg as well as the cross-covariance matrix Pxg of the initial and transformed σ -points.

FIGURE 8.

The Unscented Kalman Filter algorithm.

The main advantages of the UKF approach are the following:

FIGURE 6.

• it does not require the calculation of the Jacobians. Therefore the UKF algorithm is very suitable for highly nonlinear problems and represents a good trade-off between accuracy and numerical efficiency; • being derivative free, it can cope with functions with jumps and discontinuities; • it is capable of capturing higher order moments of nonlinear transformations better then the Taylor series based approximation ( [1]).

The Unscented Transformation algorithm.

6

Validation of the navigation system In this chapter, the navigation system described in section 2 will be validated offline for both classical EKF-based and UKFbased implementations throughout experimental data collected in two different mission scenarios. 6.1 FIGURE 7.

The experimental campaigns 6.1.1 TifOne at Livorno - THESAURUS Project The considered experimental data were collected during suitable sea tests performed in Livorno (Italy) in August 2013 (see Fig. 9). Such tests have been carried out in cooperation with Vigili del Fuoco di Livorno and Nucleo Operativo Subacqueo della Soprintendenza Archeologica della Toscana.

The Unscented Transformation weights algorithm.

It is worth pointing out that the weights c, wm and Wc are calculated exploiting the algorithm in Fig. 7, given three parameters

6


was defined by three waypoints respectively called Janus1, M2 (position of the second one of the four LOON modems) and Typhoon1. The length of the path side was about 190 [m]) and the mission was performed at the depth of about 4.5 [m] (see Fig. 10). The TifTu AUV navigated underwater for a total of 1150 [s] at a speed of 0.6 [m/s].

FIGURE 9.

The Typhoon AUV during the sea tests in Livorno (Italy).

Throughout the experiments, the Typhoon AUV TifOne operated in ROV mode equipped with an IMU, a pressure sensor for the depth, a DVL and a GPS for a total of two missions: 1. WATER SURFACE MISSION : the AUV is guided on the water surface for a total of 1299 [s] covering a distance of 1012 [m]. The collected GPS fixes will be used only for initializing the filter and as a benchmark for comparing the two different navigation system implementations. The aim of the mission is to understand the position drift error generated during the filtering process. The drift error is due to the fact that the available measurements provided by the exploited sensors (IMU, pressure sensor for the depth and DVL) do not directly provide the position of the AUV. 2. U NDERWATER MISSION : the AUV is guided underwater at a depth of 2 [m] for a total of 2540 [s]. The only available GPS fixes are the starting and ending point of the AUV path that are collected since both points are on the water surface. The starting point will be used to initialize the filters, while the ending point will be used to compare the estimation error of the arrival. The aim of the mission is twofold: 1) understand whether or not the filters are capable of reconstructing the shape of the trajectory given by the pilot; 2) understand which filter provides the estimated ending point closest to the GPS fix and, consequently, the smallest drift error.

FIGURE 10. Layout of the autonomous mission: triangle-shaped path with vertices placed in the waypoints Janus1, M2 and Typhoon1.

6.2

The mission 6.2.1 TifOne at Livorno - THESAURUS Project The results for the two mission (the surface one and the underwater one) described in section 6.1.1 are proposed. 1. S URFAFCE MISSION : in this paragraph, the EKF-based and the UKF-based navigation systems will be compared during a water surface mission. Fig. 11 shows the estimated Cartesian position coordinates η1x , η1y provided by both navigation system meas , η meas is also implementations. The GPS measurements η1x 1y provided for the sake of comparison. 12 the position estimation errors

Fig.

shows

EKF

UKF

meas meas

η 1 − η 1 and η 1 − η 1 with respect to the time and to the travelled distance. The results highlight quite similar performance for both filters, however the error of the UKF presents a flat behaviour in the final part, i.e. starting from time 1000 [s] (or after having travelled for about 750 [m]), while the EKF error keeps growing. It is also worth noticing that the shape of the estimated trajectories looks like the one provided by the GPS and that major errors occur when the AUV is turning. 2. U NDERWATER MISSION : In this section, the EKF-based and UKF-based navigation systems will be compared during an underwater mission. Fig. 13 illustrates the estimated trajectories provided by both EKF and UKF, along with the GPS fix representing the ending point. It is worth pointing out that the trajectory given by the pilot to the AUV involves quite a lot of turning points which, as noticed

6.1.2 TifTu at CommsNet13 The CommsNet13 experiment took place in September 2013 in the La Spezia Gulf, North Tyrrhenian Sea, with the support of NATO Research Vessel (NRV) Alliance (see Fig. 1). In the area the CMRE has a permanent testbed for underwater networking and communication purposes (LOON - Littoral Ocean Observatory Network). One Typhoon class vehicle by UNIFI, in particular TifTu, took part to this experimentation at sea. Throughout the experiments, the TifTu autonomously travelled in dead reckoning and was equipped with the IMU, the pressure sensor for the depth, the GPS sensor (used during the periodic and dedicated resurfacings: one resurfacing every 2.5 minutes) and the USBL unit (used only as ground truth). TifTu travelled along a triangle-shaped path on the LOON area. The reference path for the underwater mission

7


250 Or i gi n: N 43.52226◦ , E 10.27952◦

200

GPS Endi ng Poi nt : N 43.52340◦ , E 10.28003◦

100 Origin: N 43.52226 ◦ E 10.27952 ◦ m ea s meas η 1x , η 1y 50

St ar t i ng Poi nt EK F EK F η1x , η1y

EK F Endi ng Poi nt : N 43.52491◦ , E 10.27899◦

150

Nor t h di r ect i on [m ]

North direction [m]

200

EKF EKF η 1x , η 1y U KF U KF η 1x , η 1y Starting Point

0

150

UK F UK F η1x , η1y

UK F Endi ng Poi nt : N 43.52458◦ , E 10.27893◦

100

50

0

−50

0

50

100 150 East direction [m]

200

250

300

−50

FIGURE 11. AUV trajectory estimated by the classical EKF-based and UKF-based navigation system along with the GPS fixes.

0

Position Estimation Error [m]

100

150 200 East di r ect i on [m ]

250

300

350

FIGURE 13. AUV trajectory estimated by the classical EKF-based and UKF-based navigation system.

50 45

50

EKF UKF

40 35

duration of the mission and with respect to the estimated covered distance. The classical EKF-based navigation systems shows poorer performance compared to the UKF-based navigation system in terms of ending point estimation error and for both GRs. Therefore, the employment of UKF filter turns out to be quite promising to provide a better accuracy during complex navigation and cooperation tasks and a good trade-off between performance and computational load/implementation complexity.

30 25 20 15 10 5 0 0

200

400

600 Time [s]

800

1000

1200

50

Position Estimation Error [m]

45

EKF UKF

6.2.2 TifTu at CommsNet13 In this subsection, the EKF-based and the UKF-based navigation systems will be compared during an underwater mission. Fig. 14 and Fig. 15 show the estimated Cartesian position coorEKF , η EKF and η UKF , η UKF provided by both navigadinates η1x 1y 1x 1y GPS , η GPS (during tion system implementations. The GPS fixes η1x 1y USBL USBL are also prothe resurfacings) and the USBL fixes η1x , η1y vided for the sake of comparison. USBL , At the same time, Tab. 3 summarizes the USBL fixes η1x

USBL (used as reference), the estimation error EKF − η USBL η1y

η 1

1

and the estimation error η UKF − η USBL affecting η EKF

affect1 1 1

40 35 30 25 20 15 10 5 0 0

100

200

300

400

500 600 Travelled Distance [m]

700

800

900

1000

FIGURE 12. Position estimation error as a function of time (up) and of the travelled distance (down).

ing η UKF . 1 The classical EKF-based navigation system, representing the standard for the underwater navigation, shows poorer performance compared to the UKF-based one during the whole underwater mission. Therefore, the employment of UKF filter turns out to be quite promising to provide a better accuracy during complex navigation and cooperation tasks

during the water surface mission described in the previous subsection, are the main sources of error.

Table 2 summarizes

the ending point estimation errors

EKF meas and UKF − η meas the estimated covered − η

η 1

η 1

, 1 1 distances and the error Growing Rates (GR) with respect to the

8


TABLE 2.

Performance summary for both classical EKF-based and UKF-based navigation system.

Ending point error

Estimated covered distance

GR over time

Distance GR

EKF

187.7 [m]

2065 [m]

0.074 [m/s]

9.1%

UKF

158.7 [m]

1953 [m]

0.062 [m/s]

8.1%

USBL , η USBL ) and errors (in [m]) affecting TABLE 3. USBL fixes (η1x 1y EKF UKF η1 and η 1 .

FIGURE 14. AUV trajectory estimated by the EKF-based navigation system along with the GPS and USBL fixes (origin coordinates: 44.03042984 o N, 9.81893253o E).

USBL fix

USBL η1x

USBL η1y

Fix N o 1

44.031985 o N

Fix

No2

44.031820

oN

Fix

No3

44.032042

oN

Fix

No4

44.032093

oN

EKF Er.

UKF Er.

9.829877 o E

22.3

11.1

9.829461

oE

32.4

17.0

9.829189

oE

7.8

7.7

9.828454

oE

10.4

13.6

Fix N o 5

44.032076 o N

9.828269 o E

20.7

17.3

Fix N o 6

44.032063 o N

9.828184 o E

18.9

20.8

Fix

No7

44.032069

oN

9.828319

oE

14.8

10.6

Fix

No8

44.031664

oN

9.828586

oE

26.4

12.4

Fix

No9

44.031357

oN

9.828791

oE

40.3

18.1

Fix N o 10

44.031175 o N

9.828890 o E

4.0

3.8

Fix N o 11

44.030865 o N

9.829867 o E

6.6

4.0

7

Conclusions In this work, a new navigation system for AUVs is presented by the authors. The innovative strategy is based on a sensor set including an Inertial Measurement Unit (IMU), a Pressure Sensor (PS) for the depth and a GPS (employed only during periodic and dedicated resurfacings). To estimate the system state during the AUV navigation, an Unscented Kalman Filter (UKF) approach has been used with the aim of overcoming the limitations of standard Extended Kalman Fiter (EKF) based strategies. Tha authors tested offline both the EKF and the new UKF navigation systems by means of experimental sea tests performed in Livorno (Italy) and in La Spezia (Italy) with the Typhoon AUV. The considered vehicle navigated in dead reckoning in the framework of the NATO CommsNet13 experiment. The comparison with experimental data showed a good agreement in terms of localization accuracy for both EKF and UKF; more particularly the UKF turned out to be more accurate than the UKF while the computational loads are similar. As regards the further developments, the UKF-based navigation system will be implemented on-board and tested directly online on the Typhoon AUV. Thanks to the online tests, the new navigation system will be investigated in different critical scenar-

FIGURE 15. AUV trajectory estimated by the UKF-based navigation system along with the GPS and USBL fixes (origin coordinates: 44.03042984 o N, 9.81893253o E).

and a good trade-off between performance and computational load/implementation complexity.

9


ios. Subsequently, the model of the AUV employed inside the navigation system will be further improved and more advanced control techniques (including nonlinear and robust control) will be considered.

[8]

ACKNOWLEDGMENT This work has been partially supported by the Italian THESAURUS project (funded by PAR FAS Regione Toscana, Linea di Azione 1.1.a.3) and by the European ARROWS project (this project has received funding from the European Unions Seventh Framework Programme for Research technological development and demonstration, under grant agreement no. 308724). The authors would like to thank the NATO Scientific and Technological Center for Maritime Research and Experimentation (CMRE) (La Spezia, Italy) for providing facilities and support during all the research activity. A special thanks also goes to the Comune di San Miniato (Pisa, Italy) and the rowing club Canottieri San Miniato for their precious help during the experimental campaigns in Tuscany. Finally, the authors would like to thank the colleagues of the MDM Lab of the University of Florence for the theoretical and experimental work during the whole research projects and Alessio Cuccoli who worked on this topic during his Master Thesis.

[9] [10]

[11]

[12] [13]

Proceedings of the IEEE OCEANS 2013, San Diego, CA, USA, 2013. R.P. Stokey, A. Roup, C. von Alt, B. Allen, N. Forrester, T. Austin, R. Goldsborough, M. Purcell, F. Jaffre, G. Packard, and A. Kukulya. Development of the remus 600 autonomous underwater vehicle. In Proc. of the IEEE OCEANS 2005, Washington D.C.,USA, 2005. T.I. Fossen. Guidance and Control of Ocean Vehicles. John Wiley and Sons, London, UK, 1994. R. Furferi, L. Governi, Y. Volpe, and M. Carfagni. Design and assessment of a machine vision system for automatic vehicle wheel alignment. International Journal of Advanced Robotic Systems, 10, 2013. Y. Bar-Shalom, X. Rong Li, and T. Kirubarajan. Estimation with Applications to Tracking and Navigation. Wiley and Sons, New York, NY, USA, 2001. G. Evensen. Data Assimilation. Springer Verlag, Heidelberg, Germany, 2009. A. H. Sayed. Adaptive Filters. Wiley and Sons, Hoboken, NJ, USA, 2009.

REFERENCES [1] S. Julier and J. Uhlmann. Unscented filtering and nonlinear estimaion. Proc. of the IEEE, 92:401–422, 2004. [2] E. A. Wan and R. Merwe. The Unscented Kalman Flter. S. Haykin Edition, New York, NY, USA, 2001. [3] B. Ristic, S. Arulampalam, and N. Gordon. Beyond the Kalman Filter. Artech House Publishers, Boston, MA, USA, 2004. [4] B. Allotta, F. Bartolini, R. Costanzi, N. Monni, L. Pugi, and A. Ridolfi. Preliminary design and fast prototyping of an autonomous underwater vehicle propulsion system. Journal of Engineering for the Maritime Environment, DOI 10.1177/1475090213514040, 2014. [5] B. Allotta, R. Costanzi, E. Meli, L. Pugi, A. Ridolfi, and G. Vettori. Cooperative localization of a team of auvs by a tetrahedral configuration. Robotics and Autonomous Systems, 62:1228–1237, 2014. [6] B. Allotta, R. Costanzi, N. Monni, L. Pugi, A. Ridolfi, and G. Vettori. Design and simulation of an autonomous underwater vehicle. In Proceedings of the ECCOMAS 2012 Congress, Vienna, Austria, 2012. [7] G.E. Packard, A. Kukulya, T. Austin, M. Dennett, R. Littlefield, G. Packard, M. Purcell, R. Stokey, and G. Skomal. Continuous autonomous tracking and imaging of white sharks and basking sharks using a remus-100 auv. In

10
