P. 1
Model-Based Output Feedback Control Of

Model-Based Output Feedback Control Of

|Views: 57|Likes:
Published by yamraj86

More info:

Published by: yamraj86 on Sep 14, 2010
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

09/14/2010

pdf

text

original

930

IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 16, NO. 5, SEPTEMBER 2008

Model-Based Output Feedback Control of Slender-Body Underactuated AUVs: Theory and Experiments
Jon E. Refsnes, Asgeir J. Sørensen, and Kristin Y. Pettersen, Senior Member, IEEE
Abstract—This paper presents the design and experimental results of a novel output feedback controller for slender-body underwater vehicles. The controller is derived using model-based design techniques. Two separate control plant models are employed: a 3-degree-of-freedom (DOF) current-induced vessel model accounting for the current loads acting on the vehicle and a 5-DOF model describing the vehicle dynamics. The main design objective behind this strategy is to incorporate the vehicle dynamics when estimating the current influence on the vehicle. Furthermore, the transit model is based on the notion of constant propeller revolution resulting in a partly linearized model, which subsequently leads to perspicuous and implementable controller and observer structures. The controller is derived using the observer backstepping technique, and the closed loop is proved to be asymptotically stable using Lyapunov and cascaded systems theory. The control objective is to track the desired pitch and heading angle generated by the line-of-sight guidance system while keeping constant forward thrust. Experimental results demonstrate successful performance of the proposed output feedback controller implemented on the Minesniper MkII AUV/ROV. Index Terms—Autonomous underwater vehicle (AUV), experimental results, nonlinear model-based output feedback control.

I. INTRODUCTION

F

OR underwater vehicles, moving with some forward speed, the dynamics are highly nonlinear and coupled. This presents control challenges that have led to considerable interest on nonlinear observer and controller design for underwater vehicles during the last decades. There are, however, relatively few reported results on model-based control (MBC) designs for underwater vehicles that include experimental tests. The main reason for this lies probably in the great difficulties in obtaining an accurate model of the vessels. Furthermore, unpredictable current loads and poor position measurements present challenges when employing MBC due to their potentially strong influence on the controller. For these reasons, in addition to
Manuscript received June 7, 2007; revised August 21, 2007. Manuscript received in final form January 7, 2007. Published July 30, 2008 (projected). Recommended by Associate Editor F. Caccavale. This work was supported by the Norwegian University of Science and Technology and Kongsberg ASA, Norway. J. E. Refsnes and A. J. Sørensen are with the Department of Marine Technology, Norwegian University of Science and Technology, N-7491 Trondheim, Norway (e-mail: jon.refsnes@marin.ntnu.no; asgeir.sorensen@ntnu.no). K. Y. Pettersen is with the Department of Engineering Cybernetics, Norwegian University of Science and Technology, N-7491 Trondheim, Norway (e-mail: kristin.y.pettersen@itk.ntnu.no). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TCST.2007.916347

issues related to the actual implementation, tuning, and debugging, nonmodel-based solutions are often preferred. There are, however, important advantages that can follow from employing MBC. Based on the model, one can predict the motion of the vehicle by using controller actuator inputs and available state measurements. Moreover, a model-based observer can provide estimates of unmeasured states in addition to filtering of noisy signals. This paper presents successful results of an MBC system of a slender-body underwater vehicle, demonstrating orientation tracking, estimation of unmeasured states, filtering, and dead reckoning. We will, in this paper, denote the model that is designed for the purpose of control design as the control plant model (CPM). A CPM is, according to [31], defined as a model that captures the main characteristics of the physical system. Unfortunately, poorly formulated CPMs that do not capture the important characteristics of the dynamic system may cause reduced performance and also stability problems. Hence, when deriving the CPM, emphasis should be placed on stability and robustness issues related to the system in addition to simplifying the model such that analysis is feasible. We will thus, in this paper, describe the development of the dynamic model and explain the most important hydrodynamic features of slender-body autonomous underwater vehicles (AUVs). The more complex process plant model (PPM) is a comprehensive description of the actual process and should be as detailed as needed. The main purpose of this model is to simulate the real plant and to test controllers and observers that are designed based on the corresponding CPM. A. Background There are some reported results on MBC of AUVs in the literature. In [16], a state feedback controller is proposed for tracking of the NPS ARIES AUV. The model is linearized about a constant forward velocity and decoupled into three separate systems: surge, horizontal steering (sway and yaw), and the diving system (heave and pitch). Sliding-mode controllers and observers [9] are proposed to solve the tracking problem. Experimental results, reported in [22], demonstrate successful controller performance. The NPS ARIES is an underactuated slender-body AUV intended for orientation tracking while maintaining some forward speed. This kind of streamlined AUVs should be distinguished from open box-framed vehicles. These are low-speed vehicles, usually fully actuated, and with hydrodynamic and stability properties that may vary significantly. In [30], a model-based positioning system is proposed for robotic vehicles where experimental evaluation

1063-6536/$25.00 © 2008 IEEE

Nevertheless. which has been thoroughly described and analyzed in the literature. A 3-degree-of-freedom (DOF) model in surge.g. The vehicle. Hence. according to [30]. a higher order model is preferred in order to avoid large jumps and oscillations in the current estimate. This approach was first introduced in [13]. The . The relatively small weight compared to the nominal speed implies that the dynamics are speed dominant and that the nonlinear characteristics of the hydrodynamics become decisive.g. the guidance kinematics algorithms are included in the controller derivation. However. This makes the system more tolerant to faults. This makes it possible to prove convergence to the desired path despite the lack of control actuators. The three-dimensional guidance system is based on the line-of-sight method. in which all require velocity feedback of some kind. The design and results of a guidance and control system for slenderbody underactuated AUVs measuring only position. to improve the performance. and heave is derived to serve as a foundation for current observer design. A drawback of this method is that the hydrodynamic properties of the vehicle are not properly accounted for when modeling the current loads. We then show that the unactuated states are bounded due to hydrodynamic damping by analyzing the inherent dynamics of the proposed controller. the controller provides good tracking results. this is a common approach for control plant modeling. and bounded signals. the ocean current has severe influence on the vehicle performance. Consequently. and it is a convenient tool which follows from using the backstepping method [20]. except [2]. the output feedback controller proposed in this paper may also improve the performance for slender-body vehicles with more sophisticated sensor suites since the proposed observer and controller may work independently of these velocity measurements. speed measurements can be obtained by using Doppler Velocity Log (DVL) [19] or by integrating accelerations measured by the IMU. The proposed model is completely decoupled. and thus contribute to increased reliability of the control system by providing analytical redundancy to the measurements. Moreover. depth. With this approach. Furthermore. or slowly varying bias. we will employ the modeling approach first introduced in [26] and more thoroughly described in [29]. This contributes to relatively simple solutions for the observer-controller design. this generation of the Minesniper does not carry any velocity or inertial measurement units (IMU). Furthermore. i. Successful experimental results of this observer concept can be found in [27] which reports the design of a three DOF current-induced vessel model coworking with a complete nonlinear six DOF vehicle model. The goal is to provide an estimate of the current velocity and thereby estimate the influence of the current loads on the vehicle. and all. we consider underactuated AUVs. the key hydrodynamic properties are taken into account when estimating the effect of the environmental disturbance. In [30] and [34]. the performance is greatly degraded when employing incorrect model parameters. [7] and [8]. This is a current-induced vessel model that can be interpreted as a third-order filter with constants obtained based on the vehicle parameters. and the current influence is difficult to predict even though measurements of both vehicle and water velocity are available. e. The AUV/ROV is a low-cost.g. this assumption is not realistic.e. In underwater applications. e. see. is descried by a three DOF horizontal model without nonlinear damping. The control objective is to track the desired pitch and heading angles while keeping constant forward thrust. a common approach is to model the disturbance as a constant. which can be contaminated by severe noise. Other reported methods involve using kinematics and filtering techniques to obtain an estimate of the current velocity. the reported controller is a linear PID controller. In this paper. For underwater vehicles. [34] and [14]. Successful tracking results of an MBC derived using the backstepping theory are presented in [2].. roll. this paper presents the results from experimental tests carried out on a full-scale vehicle in the ocean. sway. The position is measured by using a short base line acoustical measurement system. and [4]. dynamic positioning [21]. assume that the destabilizing Coriolis forces are dominated by the hydrodynamic damping in some sense. However. Therefore. In [34]. However. The work presented in this paper is motivated by the Minesniper MkII developed by Kongsberg ASA. since only the orientation. This paper adds to the results on output feedback control presented in [29] since we in this paper consider the case where velocity measurements are not available. This restriction in the instrumentation contributes to increased challenges for accurate tracking. Examples of this can be seen in. the hydrodynamic properties of a box shaped vehicle indicates that the damping is dominant.. time-varying.: MODEL-BASED OUTPUT FEEDBACK CONTROL OF SLENDER-BODY UNDERACTUATED AUVs 931 of the different controllers is performed on the JHRUROV vehicle. and depth. see. e. In [10] and [3]. IMUs are subject to drift in the derived velocity when integrating faulty acceleration measurements. are measured. We use a slightly different approach in this paper by considering the desired trajectories as external. torpedo-shaped underwater vehicle. it is not model-based since it does not incorporate the model dynamics in the controller. we propose an observer providing position and velocity estimates. Moreover. for slender-body vehicles with some forward speed. and orientation. Main Contribution and Paper Outline The main contribution of this paper is the following. and in addition. All these mentioned results have in common that the velocity is available for feedback.REFSNES et al. pitch. B.g. Therefore. and in particular the position. and that the hydrodynamic Coriolis forces are negligible.. due to cost reasons. the Coriolis forces are not explicitly included in the model. [5]. a vehicle property which often complicates the overall analysis.. since the estimated current velocity is explicitly used in calculation of the nonlinear hydrodynamic damping and Coriolis forces. velocity measurements are available for feedback. experimental study on tracking of the open frame vehicle ODIN is presented. and the hydrodynamics are dominated by linear and nonlinear damping. e. An output controller has not yet been tested with the observers in [27]. Comparing with low-speed applications for ships. The sensor suite also provides measurements of the heading.. an open-frame hovercraft. the DVL can only generate accurate velocity measurements provided that the distance to the seafloor is within a certain boundary. The paper concludes that fixed model-based controllers outperform the PD controller. In this paper. Although employed on a nonlinear model of the ODIN vehicle. [3].

mass. In this paper.g. we will consider AUVs in transit where the forward velocity is larger than the current velocity such that . e.. and damping matrices. This decoupling of the dynamics is a method that has shown to provide a successful basis for underwater vehicle control design. the guidance system. see. full-scale tional to basin tests of the Minesniper MkII have shown that (3) matches the actual thruster forces in a satisfactory manner. 1 depicts a simulation using the PPM (1) of the Minesniper MkII performing way-point tracking in a current. it is reasonable to determine a constant propeller revolution which the forward thruster is operating on. Note that is obtained analytically by solving (2) and (3) with respect to .e.e. see. Whitcomb and Yoerger [33] have presented a quasi-steady approach for mathematical modeling of thrust which has shown to match the thrusters on the Minesniper MkII well (3) where [rad/s] denotes the propeller revolution. Euler-angles. such that the relative velocity remains constant in steady state. Hence. An important objective of this work is to develop an observer-controller system that is easily implementable. This paper is organized as follows: A description of the mathematical modeling is given in Section II. note that the vehicle velocity is altering with the heading. Furthermore. see. Considering only steady-state. The right plot shows that the relative forward velocity is the same for all headings. The reason for this can be . respectively. only slightly varying when changing course direction. i.e. where is a known constant velocity. added mass Coriolis. and the observer and controller gains can be tuned separately. and heave. Due to the nonlinear coupling between the three DOF current-induced vessel model and the five DOF vehicle model. Moreover. Other common thruster models for underwater vehicles omit the term propor.g. respectively. VOL. These features are convenient for practical implementation. Neither high gain nor bounded controller feedbacks are required. where and denote the body-fixed vehicle and current velocity. the experimental results are shown. straight line motion. 16.g. The proposed vehicle CPM is semi-linearized. 5. it follows from (2) and (3) that the relative forward velocity is constant in steady-state . Furthermore. SEPTEMBER 2008 output feedback controller consists of a pair of coworking nonlinear Luenberger observers providing filtering and estimation of the position. provided that the observer converge to the actual state. The observer and controller design and analyses are given in Sections III and IV. To illustrate the scenario. nonlinear damping is included. The relative velocity is given by .. employing constant propeller revolution. The vectors and capture the restoring and control forces and moments.. [17]. gives the following steady state surge force equality: (2) where represents the linear and nonlinear hydrodynamic damping. since the approach in this paper is also valid when there is ocean currents present. Inaccuracies in the parameters are equivalent to the effect of a current which will be estimated by an observer. e. However. 4. This is mainly because obtaining a certain speed is of less importance compared to tracking of the orientation and depth. Steady-State Analysis of Surge Motion The motivation behind this section is to determine the constant about which to linearize the CPMs. More details regarding the thruster forces will be presented in Section V-A1. and finally. The key idea is to apply two coworking . the stability analysis becomes quite involved. a case study on the Minesniper MkII is presented in Section V. Ch. sway. and the vehicle and current velocities. and and are positive constants. e. respectively. The relative surge velocity is given by . However. although . Cascaded systems theory is employed to prove asymptotical stability of the closed loop system. This is valid independently of the vehicle orientation. Therefore. Fig. this verified by the following: It follows that means that the even though is dependent on the orientation of the vehicle relative to the current. [16]. In transit. which is time-varying. Some concluding remarks are given in Section VI. 13]. where is the current velocity vector containing the body-fixed current velocities in surge. The following subsections present the corresponding CPMs for the application considered in this paper. displaying all proofs and also elaborating on the guidance system and the actuator modeling. rigid-body Coriolis... Two Separate Systems The CPM presented in this section is based on the approach presented in [29]. a will be automatically compotential error in the calculated pensated for. chap. it follows from (2) and (3) that will vary accordingly.. describing the actuator modeling. II. MATHEMATICAL MODELING The CPMs proposed in this paper are based on the following six DOF dynamics presented as the PPM in ([12]. NO.3) and [31] (1a) (1b) and denote where the North-East-down (NED)-frame position/orientation and body-frame velocities in six DOF. (i. . respectively. We decouple the surge motion from the rest of the model and study the steady-state behavior. and are the frame transformation. respectively. the destabilizing Coriolis forces and moments are linearized about the relative forward speed when applying constant thruster revolution. B. but the resulting observers and controller are easily implementable. A. . Furthermore. the modeling approach proposed in this paper is an extension to the results in which constant forward velocity is assumed. Hence. [16] and [12. Experimental results show satisfactory performance of the proposed observers and controller.932 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. see Section V. This paper is an extension to the work presented in [28]. i.

7. It is assumed that the vehicle is neutrally buoyant. Thus. In the NED-frame. we consider vehicles with control actuators in surge. These are included for stability purposes and will be determined in the upcoming section. since we consider underwater vehicles measuring only the position and orientation. Moreover. pitch. u and u . The bias relative velocity captures the external disturbances such as drag from an optical fibre connecting the vehicle to the surface vessel. respectively. we choose to employ a model-based observer for the current estimation. respectively. The velocity vector denotes the surge.4]. The mass and damping matrices yield Fig. the current is described by the vector . and denotes the Earth gravity. In this paper. sway. Following slender-body theory presented in [23. which is common for slender-body underwater vehicles in transit.: MODEL-BASED OUTPUT FEEDBACK CONTROL OF SLENDER-BODY UNDERACTUATED AUVs 933 and (6) is the vector containing the gravity and buoyancy moments. which are all positive. dynamic models: a five DOF vehicle model and a three DOF current induced vessel model (surge. Only the linear off-diagonal terms are included since these are dominating for slender-body vehicles [23. This gives the following expression for the . heave) which captures the main current loads on the vehicle. it is included to compensate for unmodeled vehicle and thruster dynamics. Ch. is the vehicle mass. pitch. Simulation with the Minesniper MkII performing way-point tracking. be defined as follows: (9) where and so forth are the linear and nonlinear damping coefficients. and are the distances between the buoyancy and gravity center in surge and heave direction. where is a diagonal matrix containing the positive time constants. and yaw velocity. self-stabilizing roll. and the fact that the length is much larger than the width. the vehicle dynamics are taken into account when modeling the current loads since the current velocity is used explicitly in the vehicle model. [12]. Ch. Furthermore.REFSNES et al. heave. Left: Horizontal position. the following CPM is proposed: (4a) (4b) (4c) denotes the NED-frame position and where orientation vector. (5) . by neglecting roll and applying a constant propeller set-point. Furthermore. It is modeled as a slowly varying process. Right: Surge velocities u. 7]. and since the current loads are strongly dependent on the vehicle dynamics. sway. When utilizing this method. The matrix transforms body-fixed vectors into NED-frame coordinates and is given by (7) (8) where denotes the mass elements included added mass. 1. respectively. and yaw resulting in the following control vector . and the vector contains the pitch and yaw angles. CPM 1: Vehicle Model: Key properties of a slender-body AUV are taken into account when deriving the CPM: port-starboard symmetry. The total damping is thus given by a linear and a nonlinear part according to . This vector is rotated into the body-frame as . where . and and are tunable scalar gains. we let the damping coefficients in (8).

Preliminaries The following assumption and properties yield throughout the paper. respectively. sway. The matrices are the top left submatrices of and in (4). [12. The Munk-moment is a hydrodynamic phenomenon which affects all geometric shapes in water except spheres. i. Hence. it follows that (11) where . and the control vector yields .g. Underwater vehicles with no control actuators in sway and heave are incapable of counteracting the forces induced by the current in these directions. In CPM 2. respectively. The Munk-moment is destabilizing in the sense that it tries to turn the vehicle perpendicular to the flow. are updated at low frequencies. respectively. For any positive .e. By comparing the nominal with the actual velocity . this yields the following property. where (skewsymmetric). the proposed CPM 2 of the main current loads on the vehicle is valid. but a model that serves as basis for an observer intended to estimate the dominant response of the vehicle due to the current. The Coriolis matrices are defined as follows: the current.g. 6)]. Furdenotes the Euclidian norm of a vector. there exists a constant . respectively. P. respectively. For most underwater vehicles.. A. where . it can be shown by using the assumption on constant current velocity in .e. i. see. Especially for slender-body vehicles with length much larger than the width causing .1: The pitch angle is limited by .e. Ch. P.. we propose two separate nonlinear Luenberger observers for the CPMs presented in Section II.. The following three DOF model is proposed: (12a) (12b) (12c) where represents the vehicle velocities in surge. which captures the slowly varying current forces. thermore.2: The magnitude and direction of the current is unknown such but upper bounded. including the Munk-moment. III. is measured. 16. VOL. i. the vector . [12].e. and heave. Here. In order to fully utilize the properties of the hydrodynamic damping. i. [32] and [11. In fact. A.. CPM 2: Current-Induced Vessel Model: This is a current induced vessel model that captures the slowly varying loads caused by the current. any nonzero velocities in sway and heave must originate from the bias . P.3: It follows from (8) that the nonlinear diagonal damping satisfies . the induced current velocity velocities in surge. we assume that the main current loads are captured in the linear motions: surge. SEPTEMBER 2008 where and denote the linear and nonlinear damping matrices. For box framed vehicles with . (12) is not a model of . sway. a current-induced vessel model. The main elements of the destabilizing overall Coriolis forces and moments shown in (1) are comprised in (10) and (11). However. and since and are diagonal matrices.. and heave may be obtained according to the following: (13) (10) Although the current velocity also affects the motions in pitch and yaw. See. the Munk-moment is decisive.e.. the resulting moments are often negligible. and heave. and denote definite and symmetric matrix the minimum and maximum eigenvalue of . 3] for further details regarding the coefficients in Coriolis matrices. exists a constant is diagonal. sway. respectively. This CPM includes the most dominant Munk-moments. especially the hori. respectively. NO. in a Lyapunov sense. e. The model may also be interpreted as a third-order filter with gains obtained from the vehicle parameters. that . Hence. Ch. Although the signals in . the Munk-moments in pitch and yaw are captured in and in .934 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY.. Hence. and denote the diagonal and off-diagonal part of the linear damping matrix. sway. i. The reason for this partition is the is not necessarily positive definite for slendermatrix body AUVs. rotation of the vehicle is contributing to neither stabilizing or destabilizing the system. In this model. To simplify Hence. Thus. Hence. and are the added mass coefficients in surge. e. this is realistic given the inherent restoring moments preventing the vehicle from large pitch angles. and heave. and in pitch and yaw. 5. the five DOF damping matrix in CPM 1 is separated into a diagonal and an off-diagonal part according to . that We use the following notation in this paper. Moreover. we conzontal position sider the error generated by the zero-order-hold in discrete time as bounded. in CPM 2. is considered continuous. the three DOF damping matrix . respectively. For more details on this. there matrix such that . and denotes the vehicle position in the NED-frame. OBSERVER DESIGN In the following section. it is clear that given . the NED-frame. The key task of this model is to function as a basis for observer design to obtain an estimate of the current velocity .1: Only the position and the Euler angles.

Employing cascaded systems where theory [24]. and are positive definite where and diagonal observer gain matrices. e. we define the following perturbation vector (21) since is proportional to the current estimation error. which growth is unknown. According to (13). Subtracting (14) from (12) and letting and .4: The mass matrix and its submatrices are positive and definite and symmetric.g.3 gives the following observer error dynamics (20a) (20b) (20c) and . See. where is the mass matrix transformed into the NED-frame. which by (NED and body-frame). . the observer error dynamics become (16a) (16b) (16c) where refers to in P. P. Current Observer The following Luenberger observer is proposed. (19a) (19b) (19c) . The error dynamics (16) will be proven uniformly globally exponentially stable (UGES) in the upcoming section. [12. This asvelocity is bounded according to sumption will be lifted when the overall closed loop system.. the damping matrix (8) gives that there such that . including the controller error dynamics. Subtracting (19) from (18) and using P.g. exists a constant P.3. e.5: The rotation matrix is orthogonal and satisfies . C. we derive a nonlinear Luenberger observer for the vehicle dynamics. The following property yields: such P. where . This is a common method in output feedback controller design. and are positive definite and diagonal observer gain matrices. Furthermore. are analyzed.. Vehicle Observer In this section. the estimated current velocity is derived as (15) and Furthermore. The suthe NED-frame relative velocity yields perscript marks the NED-frame matrices. under A. Furthermore.2: There exists a constant . Note that the gray box indicates that this is implemented in the control system (14a) (14b) (14c) where . This results in the following nominal observer error dynamics: (22a) (22b) (22c) Notice that also evolves linearly with the estimated velocity . there exist constants such that yields denotes the NED-frame velocities. and is on the line segment joining and . Hence. Ch. To avoid technicalities using two frames . see.6: There exist sufficiently large constants that the Coriolis matrices are upper bounded according to We propose the following Luenberger observer by copying the dynamics in (18) and adding correction terms B. we define the error vectors .: MODEL-BASED OUTPUT FEEDBACK CONTROL OF SLENDER-BODY UNDERACTUATED AUVs 935 the notation we define the function employing the mean value theorem gives .REFSNES et al. CPM 1 (4) is rewritten in NED-frame coordinates as follows: (18a) (18b) (18c) where . we apply the following assumption: such that the relative A.1. Hence. 3.3] for details regarding the matrix transformations. Let (17) denote the overall current estimation error state.

Proof: Consider the following positive definite and radially . the origin of the nominal observer . gives that there exists a constant . unbounded Lyapunov function candidate with respect to time gives Differentiating is the minimum eigenvalue of Using P.6. it follows that if (24) is satisfied.2. We thus arrive at the following upper bound on the Lyapunov function derivative: (26) Using Lyapunov theory [18]. of system (22) respectively. In this case. since of system (23). since the current velocity is upper bounded. as claimed such that in P. Thus. these representations will either have one sinerties of gularity or two equilibrium points. is not contributing to alleviate the condition on . which is dominant for most states. Proposition 1: The origin of the nominal observer error dynamics (23) is globally exponentially stable (GES) if the following condition is satisfied: (24) . Euler parameters. To circumas a general timevent this problem.936 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. or similar. Consequently. However. In order to fully exploit the dissipative property of the hydrodynamic damping. Notice. VOL. we can consider varying signal using forward completeness as in [21]. error dynamics (23) is GES. we have that . the vector consists of the following functions: Notice that system (23) is autonomous since all the error vari. The matrix is Hurwitz. there exists a constant such that . We thus lows that the origin defined in (17) can be bounded have that the error variable by (27) where and are positive constants. however. Due to the topological prop. ables are concatenated in the error vector we apply autonomous Lyapunov stability theory. it folof system (16) is UGES. Let the error vectors be defined as . 5. Hence. The nominal observer error dynamics (16) and (22) can then be written in compact form as the following: where . We asexists for all . SEPTEMBER 2008 [6]. can be upper bounded as follows: (25) (23) which gives that for all .3 as follows: Moreover. 16.2. the result clearly indicates that the observer feedback gains must be sufficiently high to dominate the destabilizing Coriolis forces and the bias loads. Under A. it follows that is GES. NO. This sume that the time-varying vector is trivial since in the following proofs is part of the tracking error which automatically lifts this assumption. something which precludes Recall that where and where The nonlinear and the diagonal linear damping matrices are collected in using P. Eulerframe of Rodrigues parameters. we will analyze the current estimation error and the vehicle observer error dynamics using one Lyapunov function. whereas the system (16) includes the time-varying vector current error dynamics only involve . Note here that the error vector includes the current estimation error. that the right-hand side of the . which only consists of error variables and thus corresponding with cascaded systems theory methods. Remark 2: Note that the globalness is given with respect to the chosen coordinate frame. It is not topologically possible to using any coordinate obtain results that are global in like the Euler angles. and . and of (16) are GES and UGES. Remark 1: The Lyapunov analysis provides only sufficient conditions for stability which often leads to conservative results. and hence. Recall that the current estimation error dynamics (16) is a stand-alone system independent on the vehicle observer error dynamics (22). Moreover. there exist positive definite and symmetric matrices and such that . and . the demand on (24) may be unnecessarily strict since the nonlinear damping. Moreover.

the time variable is included for the reference trajectories since these are external time-varying signals. Step 1: The nonlinear damping introduces undesirable coupling terms that complicate the stability analysis. Furthermore. Controller Design Based on the observers designed in the prior section. Inserting for in (32) into (31) yields (33a) (33b) where contains the stabilizing functions for the actuated states: pitch and yaw. we define two tracking error vectors as follows: (30) The reason for this will become clear in the upcoming stability analysis. It will become clear that due to hydrodynamic damping in all degrees of freedom. . Note that we have used in the -function in order to avoid nonlinear coupling terms involving the estimation error . 7)]. A. This concept is based on the fact that given a nonzero forward speed and some orientation of the vehicle. This results in the following expression for the . the fact that all states. are subjects to linear hydrodynamic damping. we choose the stabilizing function to evolve according to (34) where is a positive diagonal controller gain matrix. and is the velocity tracking error vector. and it is a convenient feature of the backstepping procedure. Then. any point in the global frame can be reached. Hence. we analyze the inherent dynamics of the controller that arise since there are no controls in sway and heave. IV. see. computing the corresponding error dynamics by and with respect to time and using (29a) differentiating and that give (31a) (31b) where Here. the controller error dynamics are nonautonomous. Since the vehicle is underactuated. OUTPUT FEEDBACK CONTROL In this section. and are the continuous reference trajectories for the angular velocities. The second error vector is defined as (32) where is a vector of stabilizing functions that we will choose. Thus. where contains the smooth and continuously differentiable reference trajectories. 1) Design a control system such that the control objective stated in (28) is guaranteed. way-points or a path can be obtained if the control objective (28) is met. and where we have multiplied with the positive and diagonal controller gain matrix in order to increase the design flexibility in the controller. Therefore. [7]. it is desirable to back-step into the following observer dynamics: (29a) (29b) (29c) In this section.REFSNES et al.: MODEL-BASED OUTPUT FEEDBACK CONTROL OF SLENDER-BODY UNDERACTUATED AUVs 937 global results on . including the unactuated states sway and heave. global tracking of. the Euler angle symbol is omitted when it is used in a transformation matrix for notational simplicity. we utilize the observer backstepping method. However. Moreover. since the pitch and heading angles are measured. In order to render (33) stable differential equations. The key idea behind the approach shown in this paper is to split the total tracking task in two. For more details. enables a separate study of the tracking error dynamics and the unactuated states. the controller design and analysis become more involved. we design a nonlinear controller utilizing the observer backstepping technique [20. in which they prove global convergence to the desired track/path using a line-of-sight-based method in the guidance system. The control objective is defined as tracking the desired orientation as follows: (28) as . 2) By carefully designing the guidance system. for instance. Second. since the control vector naturally evolves in the bodyframe. for instance. This approach was first introduced in [13]. the estimated position and velocity are available for feedback. the state space defined in . and the results in this this paper does not include paper are thus only global in the chosen coordinate frame. Ch. the velocities of the unactuated states converge to a bounded set. The method is shortly described by first designing the control vector considering it as an arbitrary vector in .and -dynamics (35a) (35b) where the stabilizing function is rewritten as to fit into (35b). Therefore.

Inserting the modified control vector (42) .3: There exists a constant velocity is bounded according to . which claims bounded bation vector velocities of the vehicle.2. and not the estimated state . will be considered in Section IV-C. The DP model in [21] is nonlinear only because of the rotation matrix between the NED and the body-frame. we define the following new error vectors: (40) by using (32) and (34). This is an advantage since the horizontal position provided by the acoustical measurement system can be contaminated by severe noise. The fact that the second and third element are zero. we lift A. however. the controller (37) can be rewritten as follows: (42) Here. In order to avoid circularity in the following stability analysis. independently of the contents of vector . Hence. the error between the desired state and the actual state . where the origin is proven GES in the Proof of Proposition 1. which is the natural result of the observer backstepping technique used for the controller design in Section IV-A. consists only on error variables and can be treated as a stand-alone perturbation. Note that the proof of Proposition 1 still holds when replacing A.3. Hence. In order to rewrite the control vector (37) so that it consists of the actual states and not the estimated states. This is. Controller Analysis In [21]. provided that (37) is satisfied. and that . and is a diagonal controller gain matrix. we need to define new controller error vectors. The complete observer system. by employing cascaded systems theory [25]. The paper presents. The error system (39b) represents the nominal observer error dynamics depicted in (23). Recall that the perturis derived under A. It is shown that the separation principle holds for the nonlinear case in the sense that the controller and observer can be tuned separately. Notice that we have used feedback gain and not in (37) in order to keep the level of measurement noise in the controller to a minimum. 16.2 with A. We want to formulate the controller error dynamics and observer estimation error as a cascaded system. If controller terms proportional to are included. note that the conand that troller is considered as a general vector in are left undecided.e. where also will be determined. NO. Remark 3: The proposed controller (37) is not directly dependent on the estimation error . a convenient method for deriving observers and controllers for nonlinear systems. we have that the perturbation vector to the nominal observer error dynamics (23) can be upper bounded according to (37) where . Another favorable result of the following analysis is that the controller and do not need to meet high gain criteria in gains order to ensure stability. Furthermore.. in(21) will be analyzed in the cluding the perturbation vector proof of the final theorem of this paper. B. In order to achieve this.3. 5. scattering and thereby degradation of the controller is likely to occur. we use that and resulting in (41) Using (41). VOL.938 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. such that the desired A. a solution for output feedback control of dynamic positioning (DP) of ships is reported. trivial since the Lyapunov analysis is valid. Under A. i. we have collected the terms involving the estimation error into the perturbation vector as follows: (38) where we have used (41).2 and replace it with the following assumption on bounded desired velocities. SEPTEMBER 2008 Step 2: Proceeding with the -dynamics With the new error states. we will show that the controller error can be written in a cascade with the nominal obdynamics as follows: server dynamics (36) (39a) We choose the control vector according to (39b) where the perturbation is to be defined. This paper is thus an extension to that result since the CPM models in this paper are coupled and involve nonlinear damping.

Furthermore. and if (24) is satisfied. and A. This completes the proof. The perturbation can be shown to be upper bounded by the tracking error state and the estimation error as follows: (47) where (44) and . denoted as in (39). We thus include the current estimation error in the controller stability analysis and propose the following radially unbounded Lyapunov function candidate where we have used that . and where the perturbation vector . i. is uniformly globally asymptotically stable (UGAS) under A. and (43) is UGAS under A.1. A. Proposition 2: The origin and of the cascaded system (39). The following theorem establishes uniform global asymptotic stability of the overall output feedback controller. Proof: Following cascaded systems theory arguments each of the functions in (39) are analyzed separately in three steps. is proven GES in the Proof of Proposition 1.3. 3) This step involves determining the growth of the perturba. it follows that the rewritten tracking error state (43) can be viewed as a cascade with the nominal observer error dynamics (23). Let the error vector denote the complete error state excluding the current estimation error. it follows that the origin and of the cascaded system (39) is UGAS [24. we want to establish the stability properties of the unperturbed system shown in (43). under A. and if condition (24) is satisfied. Lemma 2]. where . and is some constant.4. the current estimation error dynamics (48b) are proven UGES in Proposition 1. Consequently. The origin of the unperturbed system in (48a) is UGAS given Proposition 1 and 2. we have only considered the nominal observer error dynamics (22). Theorem 1: The origin and of the cascaded system (16).. Using P. which will be proven to hold in the Appendix.e. and therefore also the origin of the unperturbed system (39a) is UGAS.3. it follows that condition (46) holds. A.2. the origin Lyapunov function (44) satisfies (46) where . P. In this case. the linear growth restriction on in the perturbation is satisfied.5 gives tion vector (43) where . and thus is UGAS. Similarly as in Section III. 2) At this step. while we only consider the tracking error . we want to exploit the dissipative property of the hydrodynamic damping. Up to this point.REFSNES et al. and P. The following assumption. Theorem 2. is applied. based on the three prior steps.4.: MODEL-BASED OUTPUT FEEDBACK CONTROL OF SLENDER-BODY UNDERACTUATED AUVs 939 into the actual vehicle dynamics (4) results in the following -dynamics: where we have inserted for the current estimation error dynamics (16). which are rewritten in the body-frame.8].4: The time-varying signals exist for all . Differentiating where (44) with respect to time and inserting for the tracking dynamics (43) without the perturbation gives Thus. We thus consider as general time-varying signals using forward completeness [21]. . terms involving the estimation error becomes Having established this. Following cascaded systems (45) . and is a class function satisfying . (20). described in (43) and (23).2. Proof: We write the overall system including the current error dynamics in the following compact form: (48a) (48b) . Acis UGAS if the cording to [1. is given by Notice here that the nominal observer error dynamics (23) includes the time-varying signals . and from (35a) and (41) the perturbation vector for the tracking error capturing the . A. as depicted in (39). 1) The origin of the nominal observer error dynamics (23). since is proven UGES and satisfying (27).6.

is bounded. it follows that the origin of the error dynamics (16). which subsequently enhance the complexity of the stability analysis. whereas the damping and thruster coefficients were calculated based on basin tests . instead of tracking . Hence. Consequently. It is clear from the analysis in Section IV that the -dynamics are UGAS provided that the control vector satisfies (37). This approach. Then. NO.. This is the fundamental idea behind MBC in this paper. This leads to (53) is inserted into since is proven where bounded in Section III-B. notice that consists solely of bounded and converging signals . we have that The vector captures the off-diagonal terms in the mass. -dynamics by rewriting We proceed by analyzing the (51) into compact form and collecting all the bounded and converging signals into the vector function (52) The linear growth restriction on in the perturbation term is satisfied. Hence. and the following upper bound on (53) is obtained: (51) where the bounded and converging variables are concatenated in the function . For pitch and heave.. Hence. This implies that the system is ISS from to [18.940 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. we add to the function . Recall from Section II that this captures the unmodeled dynamics. damping and Coriolis matrices ( and ). is not employed in this paper since only leads to negative semidefinite Lyapunov function derivatives. Recalling that the control vector yields where and (50) This results in a dynamic constraint in the controller for the unactuated states. it remains to show bounded growth on the perturbation .e. converges to the bounded set . and must satisfy the differential equation (37) that arises since there are no controls in sway or heave . CASE STUDY: THE MINESNIPER MKII The parameters of the Minesniper MkII (see Fig. Obviously.6]. the obvious choice is to design the stabilizing function rendering the tracking error dynamics stable since we can assign and arbitrarily. 16. It thus follows from [24] that the origin of the system (48) is UGAS. and (43) are UGAS. not feasible for and . VOL. the velocities in sway and heave. we consider the fact that the vehicle is underactuated. Consequently. which clearly manifests the statement above. Finally. sway and heave since we cannot assign control force in these directions. Clearly.e. there exists a constant that satisfies . are bounded and converge to the same set. C. in this sense. This completes the proof. From (37). Moreover. Proof: This can be proven by applying for instance the Lyapunov function candidate . Underactuated In this section. Therefore. 2) are shown in Table I. however.e. the proposed CPM (4) needs to resemble the real world to a certain extent in order to obtain a stable MBC for real systems. Proposition 3: The —subsystem is input-to-state stable (ISS) from to . It will become clear that the -dynamics converge exponentially to a bounded set due to the linear hydrodynamic damping in sway and heave. (20). it follows that as . however. due to the dissipative hydrodynamic damping. All the signals in are shown to be bounded or converging to zero except the bias term . we achieve UGES of of the unforced system. It follows from (40) that it can be bounded vector and hence satisfying linearly by (49) where [12]. which differentiating with respect to time along the solutions of gives . Lemma 4. i. 5. This is. i. it follows naturally to declare the effect of the unmodeled dynamics as bounded and hence. V. as implies that as . Instead. since and as . Another common approach in marine applications is to define the bias as constant.. i. SEPTEMBER 2008 theory. Furthermore. The added mass coefficients were obtained by using the computer program WAMIT. the -dynamics track the actual velocities .

Recall from (3) that the . Minesniper MkII. Notice also that the contrally buoyant. Movement of the pitch mass leads to the following . (typically 0. 3. Based on the linear approximation and are positive constants. Norway. Recall that the vehicle is neu. we solve the equation mass actuator with respect to . where the constants thruster force is are defined as [33] (54) (57) . Furis the ambient water velocity in thermore. Pitch control by mass movement.4) is denoted as steady-state where the wake fraction number [12]. to obtain the expression for the distance of the pitch . and and denote the propeller diameter and the thrust deduction coefficient. to generate the pitch moment given this pitch angle. which contains the trol moment obtained in (37) involves . The corresponding revolutions are then obtained by (56a) (56b) where is the constant revolution providing surge velocity. 2) Pitch Actuator: The Minesniper is equipped with a moveable mass for pitch control. 3 describes the system of pitch control by mass movement. Norway. Courtesy of KDA. Fig. expression for the center of gravity: is the distance from the center to the mass. 2. Control Forces and Moments This section describes the mapping from the control moments derived in Section IV-A to the control action of the actuators on the vehicle. where is the water density. respectively. it follows from Fig.REFSNES et al.1–0. expression for the static pitch angle given a certain Therefore. 1) Thrusters: On the Minesniper MkII.: MODEL-BASED OUTPUT FEEDBACK CONTROL OF SLENDER-BODY UNDERACTUATED AUVs 941 Fig. and where denotes the weight of the actuator. is the distance from the propeller to the center of the and body. It is a common phenomenon that underwater vehicles can have difficulties obtaining initial pitch angle when surfacing. Then. Table II presents the sensor suite of the vehicle. the heading control is obtained by two horizontal thrusters located on each side of the hull at the center of the vehicle. see (6). 3 that the pitch mass distance is at Kongsberg Defence and Aerospace at Stjørdal. A. TABLE I MINESNIPER MKII CPM CONSTANTS Fig. hence . The equations for forward thruster force and yaw moment can then be described as (55) TABLE II SENSOR PROPERTIES where and are port and starboard propeller revolutions. This is avoided by this type of control action since the static pitch angle varies with the location of the pitch mass.

Fig. We start by defining the total track length be. see Fig. where heading yields Guidance is provided by two decoupled line-of-sight systems for heading and pitch. The cross-track error . The distances by where the if then or where Care must be taken when determining the size of prior to a run. and the forward speed. 2) Pitch: The LOS method is modified for pitch control according to Fig. 5. The LOS-point on the track (see Fig.e. respectively. Line-of-Sight Guidance System which . 1) Heading: In order to avoid potentially large cross-track errors. respectively.942 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. [22]. 4. the aiming point is moved from the next way-point to a point on the track line by setting the length of the aiming vector to . 4) is given by In order to ensure correct desired heading at all times. Kinematics of the horizontal LOS guidance system. SEPTEMBER 2008 Fig. i. The angle or if then Activate next way-point. The most usual case is if the vehicle has penetrated the or if the vehicle has passed a way-point watch radius way-point by a distance larger than . VOL. 16. where is saturated such that inherently includes the limit B. 5. This corresponds well with tests carried out on the Minesniper revealing that when was too small. It is shown in [15] that for way-point maneuvering of ships. the guidance system automatically switches to way-point tracking according to the following: ordered pairs and are the present and the previous and are given way-points. must be larger than some constant which is dependent on the vehicle dynamics. The next way-point is activated if one of two conditions is met. The pitch angle between two way-points is Note that ranges from 0% to 100% of between these two vectors is then defined by (58) where may now be defined as . 5. Kinematics of the vertical LOS guidance system. excessive control action and an increase in the cross-track error occurred. actuators. 4. tween way-point and where and sign if otherwise. . NO. the desired .. Based on the . cross-track-error and the length of the aiming vector.

Nevertheless. Norway. Moreover. .: MODEL-BASED OUTPUT FEEDBACK CONTROL OF SLENDER-BODY UNDERACTUATED AUVs 943 Fig. respectively. Right: Measured (red) and estimated (blue) depth and orientation of the Minesniper MkII. according to Fig. D. This caused the vehicle to slightly miss the to third and the fourth way-point. 6 shows large variations of the heading. and controller were implemented as functions written in C code with 20-Hz update rate. From Fig.. The observers. A saturation is included . Sea Trials—Results and Discussion The sea trials of Minesniper MkII were performed in Trondheimsfjorden nearby Stjørdal. The reference trajectories are then given by (60) (61) is the relative damping ratio. Fig. 6. which demonstrates the performance of the controller. guidance. However. and the guidance functions are collected in . the reference system . The way-point watch radius was set m. the heading shown in the bottom right plot of Fig. 5. C. dead-reckoning can be seen between the second and third way-point. the observer provides satisfactory position estimated despite measurement drop outs. we experienced increased noise in the acoustical measurements mainly due to the topography of the seabed blocking the view of the acoustical receivers. 6 shows the measured and the estimated position and orientation of the Minesniper MkII. (62) (59) Hence. Clearly. i. Toward the end of the run. the desired pitch angle is given by . The observer provides satisfactory estimates for the entire run with small deviations and little noise.e. we utilize a second-order filter cascaded with a low-pass filter [12]. The bay area is quite shallow and narrow which limited the location of the way-points. in order to keep gain was chosen relatively low. The distance between the next way-point and the LOS point is given by to ensure that the desired on the final reference pitch angle pitch angle is feasible at all times. where are determined by setting the pitch weight in where its maximum and minimum position. the observer seemed to cope with measurement drop outs and the noise in a satisfactory manner. Reference Trajectories In order to obtain smooth and continuously differentiable acceleration. is the where natural frequency. and position reference trajectories. A 5-m radius of acceptance around each way-point is included.REFSNES et al. 6. velocity. Left: Measured (red dots) and estimated (blue line) horizontal position of the Minesniper MkII performing way-point (green squares) tracking.

This is carried out by including the reference system algorithms presented in Section V-C in the overall stability analysis and thereby show exists all . 7 presents the tracking results in pitch and heading and the actuator action. SEPTEMBER 2008 Fig. Part of the vehicle CPM was linearized about the relative surge velocity. Center: The tracking error in pitch and heading. An advantage of the employed method is that it does not require any high gain nor bounded feedback controller gains.. . The main reason for this was that given the center location of the thrusters. Furthermore. i. VOL. thruster revolution. The tracking performance is satisfactory with relatively small deflections of the actuators. and position of the pitch weight. the pitch controller calms the motion relatively fast despite limited rate in the pitch mass actuator. Fig. 16.944 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. Therefore. The current observer gains were set equal to the upper left matrices of the vehicle observer gains. This was because the forward acceleration generated a pitch moment. The CPM system consisted of two separate models: a five DOF vehicle model and a three DOF current induced vessel model accounting for the main current loads. This is mainly due to and uncertainty of the damping coefficients which. Several should be set small runs indicated that the controller gain large for optimal tracking results. 8 showing satisfactory observer performance. Tracking results. However. The controller gains were found by trial and error since there exists. we prove that A. emphasis was placed on keeping the yaw motion within relatively restrictive boundaries. several sea trials revealed that erratic usage of the thrusters could cause intractable yaw motion. NO. 5. no formal methods for tuning gains in a backstepping controller at the present time. The nonlinear Luenberger observers and the controller. z and [z Right: The controller action. to the author’s best knowledge. some error occurred in the beginning of the run.e.z ] . Table III shows the controller/observer gains used in this run. togethers with a . the estimates indicate a slight current from South-East. the observer and controller gains can be tuned separately. We were unable to measure the actual current. The estimation results are presented in Fig. Left: The desired trajectories provided by the reference generator. Proof: Consider the following Lyapunov function: (63) . aggravated the performance. were proven UGAS using Lyapunov and cascaded system theory. However. In pitch. APPENDIX In this section. the yaw rate small. CONCLUSION AND FURTHER WORK An output feedback controller was proposed for slender-body underactuated underwater vehicles. 7. Further work involves obtaining more accurate vehicle parameters and optimal tuning of the controller feedback gains in attempt to optimize the performance of the vehicle. which was designed using the observer backstepping technique. Experimental sea trials on the Minesniper MkII were presented showing satisfactory observer and tracking performance.4 holds by showing forward completeness of the closed loop system. that the global position Proposition 4: The time-varying vector exists for all . large VI.

M. J. Silvestre. Control. 2006. 4421–4426. 2003. where is a sufficiently large and can be linearly upper constant. be continued for all ACKNOWLEDGMENT The authors would like to thank their colleagues at CeSOS for valuable discussions and contributions to the testing of the Minesniper MkII. Maui. it can be shown that . T. Kokotovic. pp.” in Proc. . Based on these observations.4 since is in. “Position tracking for a nonlinear underactuated hovercraft: Controller design and experimental results. 2002. 41st IEEE Conf.REFSNES et al. [4] F. Pascoal. Decision Control. [5] P. Decision Control. vol. where inequality that is a constant. P. V. [2] A. REFERENCES ´ [1] O. NV. and P. F. M. P. pp. 7. TABLE III CONTROLLER. Aguiar and A. 8. 51–56. A. San Diego. The right column describes ~ ^ ^  v ^ the estimated current velocity  (t) and the (t)-dynamics versus [^(t). Arcak. 42nd IEEE Conf. Hespanha. pp. CA. and J. Raimondi. 649–658. Las Vegas. “Dynamic positioning and way-point tracking of underactuated AUVs in the presence of ocean currents. 3858–3863. Decision Control. 47. M. Oliveira. no. AND LOS GAINS 2) The overall observer-controller error dynamics given in (16). Alonge. w (t)]. the vector exists and can . Decision Control. bounded by signals captured in . Estimation results.” Int. P. L. 2001. FL. Ippolito. Orlando. Moreover. it can be cluded in shown by inserting the error dynamics and using Schwartz satisfies . and (43) can now be analyzed using the Lyapunov function (63) without applying A. pp. Hence. 40th IEEE Conf. Thus. Differentiating with respect to time gives (64) Before proceeding we note the following: 1) Based on the guidance algorithms presented in Section V-B. pp.” in Proc.” in Proc. Cremean. “Trajectory tracking of underactuated underwater vehicles. and P. 45th IEEE Conf. C.4 holds. The left column shows the estimation error (t). [3] A.” in Proc. Aguiar. Aamo. D. I. “A quaternion sensor based controller for homing of underactuated AUVs. 2001. The center column presents the estimated velocities  (t). and F. 2105–2110.: MODEL-BASED OUTPUT FEEDBACK CONTROL OF SLENDER-BODY UNDERACTUATED AUVs 945 Fig. OBSERVER. Batista. “Global output tracking control of a class of Euler-Lagrange systems with monotonic nonlinearities in the velocities. HI. (20). Fossen.

. “Observer design for underwater vehicles with angle and position measurement. vol. [14] T. no. [8] M. 466–476. In 1999. Decision Control. I. Sørensen received the M. 4. J. and A. in the Department of Engineering Cybernetics. 26. Berghuis and H.” IEEE Trans. Aug. and ships. Loría. both from the Norwegian University of Science and Technology (NTNU). navigation and control of ships. [34] S. Practice. [16] A. 627–634. MA. IFAC Manoeuvring Control of Marine Craft (MCMC). A.” in Proc. Spain. no. Stjørdal. Refsnes. Fjellstad. [29] J. Trondheim. 152–160.: Cambridge Univ. U. “Global output feedback universal controller for stabilization and tracking of underactuated ODIN: A spherical underwater vehicle.” IEEE J. Her research interests include nonlinear control of mechanical systems with applications to robotics. 2004. I. vol. Trondheim. Tech. Fossen. K. 15th IEEE Mediterranean Conf.. Traiantafyllou and F. pp. Y. Control. “Output feedback control of an AUV with experimental results. Skjetne. Loría. 602–608. Girona.” IEEE Trans. E. and H.” Marine Cybernetics. From 1993–2002. pp. L. degrees in electrical engineering from the Norwegian University of Science and Technology (NTNU). J. A. Princeton University. [32] M. Seville. no. “A 6 DOF nonlinear observer for AUVs with experimental results. 2005. pp. P. and A. 43rd IEEE Conf. he has held the position of Professor of Marine Cybernetics at the Department of Marine Technology.” in Proc. 12. no. respectively. C. Eng. respectively. Fossen and O. 1995. She holds several board positions. (CAMS95). vol. Y. 695–703. Refsnes received the M. Control Automation (MED). vol. Decision Control. “Model-based dynamic positioning of underwater robotic vehicles: Theory and experiments. pp.Sc. 5360–5367. pp. Christi. [31] A. Smallwood and L. 1999.D. J. Panteley. in 1992 and 1996. Healey and D. 66–74. 33.” IEEE J. S. Nonlinear and Adaptive Control Design. IFAC Workshop Motion Control. Princeton. Sørensen. Cambridge. pp. “A passivity approach to controller-observer design for robots. and Ph. Trondheim. Control Lett. Rep. 21. “Structural issues in the design and operation of marine control systems. Contr.K. pp.” in Proc. control. Refsnes.. 80.. 4. Jan.Sc. 8. Eng. Ocean. 2003. 3. Cambridge. Athens. Greece. 6. he was employed in the ABB Group in various positions as a Research Scientist. vol.” IEEE Trans. Healey.” in Proc. 4. “Robust adaptive control of underwater vehicles. Nijmeijer. Pettersen. Kinsey and L. France. Available: http://www. M. 12. “Command. 2000. “Marine control systems: Guidance. 2. 15th IEEE Mediterranean Conf. L. A. J. 2006. Spain.com [13] T. Oct. 169–186. Loría. Whitcomb and D. NJ.. Control Automation (MED).” in Proc. E. [19] J. I. Kristin Ytterstad Pettersen (SM’04) received the M. 1. 2001. Miles. Dec. J. and K. no.” Int.” IEEE J. 2002. pp. [18] H. Athens. T. Spain. Papoulias. 40. Nijmeijer. 1993. [17] A. Eng. AUVs. [15] E. Sørensen. 2005. [21] A. Healey. Oct. Technol.Sc.. 24. “Maneuvering and control of marine vehicles. 1998. Rev. and Technical Manager in the Business Area Automation Marine and Turbochargers. 2004. Panteley and A. 327–339. Norway. Pettersen. pp. 2004. pp. “Principles of guidance-based path following in 2D and 3D. Panteley. Trondheim. no. vol. vol. Kokotovic. I. Pettersen received the IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY Outstanding Paper Award in 2006. 2. Breivik and T. Jul. M. Portugal. Aug. Asgeir J. Y. pp. F. NTNU. Control Syst. 5.marinecybernetics. and in 2002 Professor. IFAC Manoeuvring and Control Marine Craft (MCMC). Z. Lefeber. The project is in cooperation with Kongsberg Defence and Aerospace. Eng. Kanellakopoulos. no. N.” in Proc. M. Pan. vol. 2007. 244–249. 3. [28] J. 332–343. Eng. [23] J. she was a Visiting Fellow in the Department of Mechanical and Aerospace Engineering. Nonlinear Systems. SEPTEMBER 2008 [6] H. Ocean.” Syst. Y. and P. “Development. J. 3rd ed. Pettersen. She has published more than 70 conference and journal papers. Y. [27] J. Yuh. Control Eng. 2002 [Online]. Sea Loads on Ships and Ocean Structures. D.D. His research interests are underwater vehicles. “Output feedback control of underwater vehicles with current estimation. “Preliminary field experience with the DVLNAV integrated navigation system for oceanographic sumbersibles. 1–7..” in Proc.” Automatica. S. Jul. and J. “Adaptive sliding mode control of autonomous underwater vehicles in the dive plane. where he is currently working toward the Ph. 221–226. degree in engineering cybernetics. [12] T. Decision Control.” J. pp. S. vol. Cambridge. Dr. Børhaug and K. D. pp.” IEEE J. Ocean. Cody. 1990. Khalil. degree in marine technology . Ocean. “Line-of-sight path following of underactuated marine craft. Atlantis. vol. S. E. observer design. 2007. 481–494. She became a Associate Professor in 1996. 3rd IFAC Workshop Control Applicat. and H. A. Grenoble. 7. 1136–1150. vol. Marco and A. Faltinsen. Ocean. no. Norway. E. J. J. Norway. Pettersen. 2005. vol. Refsnes. pp. 1977.” in Proc. pp. K. I. B. comparison and preliminary experimental validation of nonlinear dynamic thruster models.” Annu. 1998. J. NJ: Prentice-Hall. 18. [30] D. Press. Jon E. “On global uniform asymptotic stability of nonlinear time-varying non autonomous systems in cascade. Lienard. and R. Ocean. no. 1990. pp. [33] L.946 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. 354–361. 740–753. 2007. 117–124. “Toward an improved understanding of thruster dynamics for underwater vechicles.. modeling and hydrodynamics. and K.” in Proc. 29. Fossen. pp.” Maneuvering and control of marine vehicles. Healey. Bahamas. no. MA: MIT Press. [10] K.” IEEE J. satellites. no. Norway. “Multivariable sliding-mode control for autonomous diving and steering of unmanned underwater vehicles.. Greece. “Exponential tracking control of a mobile car using a cascaded approach. “A separation principle for dynamic positioning of ships: Theoretical and experimental results. in 1988 and 1993.. rigs and underwater vehicles. Sørensen. 1. Sørensen. and E. Project Manager. New York: Wiley. 44th IEEE Conf. Whitcomb. Breivik. Seville. “Experimental study on advanced underwater robot control. Fredriksen and K. NO. 4. NTNU. Whitcomb. Pettersen. A. Sørensen. 1993. Hover. E. 2004. 1–8. Marine Syst. 1995. 1995. Eng. [26] J. degree in marine technology and the Ph. Department Manager.” J. 20. Jiang. Do. Upper Saddle River. no. [9] R. in 2003. A. 9. pp. [22] D. “-exponential way-point manoeuvering of ships. [11] O. Y. Oct. Yoerger. 131–138. 16. Marine Hydrodynamics. I. Fossen. Robotics Autom. Newman. vol. pp. no. Pettersen. Fossen. pp. and control. Paradise Island. V. E. 125–149. Trondeim. 29. ´ ´ [20] M. Since 1999. Krstic. Robotics. J. Lisboa. pp. and K. Refsnes. J. 15. 44th IEEE Conf. vol. pp. 1541–1549. 2002. [24] E. [25] E. [7] E. . Brown. Nijmeijer. “Cross-track control for underactuated autonomous vehicles.. VOL. and navigation: Experimental results with the NPS ARIES AUV.D. Zhao and J. 2005. degree in marine technology from the Norwegian University of Science and Technology (NTNU). Rock.

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->