Ocean Engineering 182 (2019) 162–173

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Coordinated path planning for an unmanned aerial-aquatic vehicle (UAAV)

and an autonomous underwater vehicle (AUV) in an underwater target
strike mission
Yu Wu
College of Aerospace Engineering, Chongqing University, Chongqing, 400044, China

A R T I C L E I N F O A B S T R A C T

Keywords: Unmanned system has become more and more popular as it can adapt to diverse environments and has pro­
Unmanned system spective applications. Especially, the coordination among heterogeneous vehicles is capable of completing
Unmanned aerial-aquatic vehicle complicated tasks, which is often beyond the ability of homogeneous vehicles. In this paper, the underwater
Autonomous underwater vehicle
target strike mission is concentrated, and the mission is completed by the coordination between a UAAV and an
Coordinated path planning
Particle swarm optimization algorithm
AUV. UAAV and AUV are deployed in this mission because UAAV has strong search ability in air and can
communicate with AUV directly after it dives into water. Firstly, to decompose the problem, the mission is
divided into two phases, i.e., single flying of UAAV and underwater coordination between UAAV and AUV. In the
coordinated path planning model, the motion of vehicles, the constraints in different media and the optimization
index in each phase are all formulated into mathematical forms. Based on the particle swarm optimization (PSO)
algorithm, the collocation points are used to determine the locations of control variables. Those points can
reduce the computation load and improve the solution quality, and they are distributed by height and moment
according to the forms of constraints in each phase. Besides, the strategy of addressing infeasible solutions is
generated to guarantee the normal operation of PSO-based algorithm. Simulation results demonstrate that the
proposed two-phase coordinated path planning method can generate coordinated paths, and the obtained results
is very close to the optimal solution in theory. Compared to the whole method, the two-phase method can better
deal with the complicated constraints in each phase.

1. Introduction heterogeneous system contains different types of vehicles, which enable

it to cope with the cooperative tasks crossing multiple areas, such as
Unmanned system is now becoming more common as it can complete air-sea surveillance, air-ground target track and communications relay
complicated missions without human intervention. Unmanned system is (Minaeian et al., 2016) (Murphy et al., 2008). In this paper, the un­
made up by a certain number of vehicles and can be applied in detection, derwater target strike mission is focused on, and a UAAV and an AUV are
surveillance and strike mission (Wang and Liu, 2012). The system is equipped to complete the mission cooperatively. UAAV is a new type of
usually consisted of unmanned aerial vehicles (UAVs), unmanned aircraft that can navigate both in the air and underwater, and the
ground vehicles (UGVs), unmanned surface vehicles (USVs) and introduction of UUAV extends the scope of mission and makes the task
autonomous underwater vehicles (AUVs). According to the types of completed more efficiently (Yang et al., 2015).
vehicles, the unmanned system can be classified into two categories, i.e., In existing literature, coordination in homogeneous system and
homogeneous system and heterogeneous system. In a homogeneous heterogeneous system are both studied. In homogeneous system, coor­
system, the vehicles are with the same type, and they execute specific dination among UAVs is the main concern, and the task assignment and
tasks in a single environment. For example, UAVs execute transportation coordinated path planning problem are widely studied. Task assignment
task in air (Quintero et al., 2016), and USVs conduct rescue task on the is a combinatorial optimization problem, and UAVs are usually assigned
water surface (Kim et al., 2017). However, as there are more and more more than one task during the whole process. In Ref. (Luitpold, 2018), a
complicated tasks crossing multiple areas, homogeneous system may be group of UAVs are required to arrive at the destinations simultaneously
incapable of finishing such tasks. Compared to homogeneous system, or sequentially with specified time delays, and the goal is to minimize

E-mail address: [email protected].

https://doi.org/10.1016/j.oceaneng.2019.04.062
Received 1 January 2019; Received in revised form 12 March 2019; Accepted 22 April 2019
Available online 3 May 2019
0029-8018/© 2019 Elsevier Ltd. All rights reserved.
Y. Wu Ocean Engineering 182 (2019) 162–173

the total mission time. To deal with the problem, the algorithm for For the underwater target strike mission discussed in this paper, both
simultaneous arrival was presented firstly to calculate the shortest flight homogeneous and heterogeneous system can deal with this problem but
paths between all pairs of UAVs and targets. Then an optimal assignment there are still some weak points. When AUVs are deployed to execute
and a critical path are found by solving a linear bottleneck assignment this task, they lack of global search ability and it may cost a lot of time
problem, and the other paths are prolonged to the length of the critical searching the target (Li et al., 2017). As for the UAV-AUV system,
path by insertion. In Ref. (Ma et al., 2018), the coordinated motion communication from air to underwater (and from underwater to air) is
planning problem for a team of micro aerial vehicles with high density is forbidden because the GPS signal is easy to be detected, which makes
paid attention to. The problem was solved by the proposed decentralized AUV unsafe (Barbarino et al., 2011). To solve the above problems, a
motion planning method. In high-level planning, a prioritized algorithm UAAV-AUV system is developed. In this system, UAAV can search the
was combined, and in low-level trajectory generation and vehicle con­ underwater target in air and send the information to AUV after the
trol, the coordination strategy was based on barrier functions. Other UAAV has dived into water, thus both improving the work efficiency and
studies regarding homogeneous coordination are among UGVs or AUVs. ensuring the safety of vehicles.
The coordinated path planning problem for carrier aircraft taxiing on A new coordinated path planning model for a UAAV and an AUV is
flight deck is essentially a UGVs coordination problem, and a distributed proposed in this paper, and an effective particle swarm optimization
planning architecture with multiple levels is proposed in Ref. (Wu et al., (PSO)-based algorithm is also developed to solve the established model.
2018). Under this planning architecture, realistic constraints are The main contributions of this work can be concluded as follows:
modeled, and a distributed path planning algorithm based on the
asynchronous planning strategy is developed to avoid the conflict during 1. The coordinated path planning problem for a UAAV and an AUV in
taxi. As for AUVs, the underwater target tracking problem is focused on. an underwater target strike mission is described. The mission is
Considering the tracking control problem for unknown autonomous divided into two phases, i.e., single flying of UAAV and underwater
marine vehicles (AMVs) with uncertainties and disturbances, a novel coordination between UAAV and AUV. The superiority of the two-
data-driven adaptive tracking controller is designed based on the phase division is verified by comparing with treating the task as a
established equivalent data model of the AMV, and the stability analysis whole process.
for the closed-loop AMV system is also presented (Weng et al., 2018). 2. The coordinated path planning model is established. In this model,
The performance of coordination algorithm also has impact on the the maneuverability of UAAV in different media is taken into ac­
operation of AUVs system. In Ref. (Xing et al., 2017), the convergence count, and the constraints on communication, safe distance and co­
analysis on Multi-AUV system with leader-follower architecture is con­ ordinated strike are all included. Besides, the optimization indexes
ducted. A graph theory model based on second-order consensus protocol considering different goals of each phase are proposed.
is introduced. The contributions on enhancing the convergence rate are 3. The PSO-based algorithm is developed. In this algorithm, Chebyshev
discussed, especially for the influences of control gains, network topol­ polynomials are introduced to combine with PSO algorithm. The
ogy and topologies imposed on the proposed collaborative navigation locations of control variables optimized in PSO algorithm are
algorithm. determined by Chebyshev collocation points. Those points are
Coordination in heterogeneous system is often seen between UAV distributed according to the form of constraints in each phase, which
and other types of vehicle. There are strong complementarities between matches well with the established model. With this algorithm, the
UAV and UGV, and the advantages are summarized as follows (Chen solution which is very close to the theoretical optimal one can be
et al., 2015) (Minaeian et al., 2015). Firstly, UGV can be deployed to obtained.
accurately locate ground targets to make up for the weakness of sensors
located on UAV limited by operating airspeed and altitude. Secondly, 2. Description of the underwater target strike mission
UAV can be regarded as communication relay as it is less blocked by
obstacles than that among UGVs. Finally, UGV have larger payload The underwater target strike mission specified in this paper is
capability and longer voyage than UAV in a transportation task. In defined as follows: UAAV initially flies in air with high speed to search
Ref. (Li et al., 2016), a UAV is used to obtain a ground image from aerial for the underwater target quickly, and AUV keeps static before receiving
vision, which is beneficial for constructing the ground map and any command. Once UAAV has discovered the underwater target, it will
improving the recognition accuracy of obstacles. Then a hybrid path dive into water and communicate with AUV. The position information of
planning algorithm combining global path planning with local rolling underwater target will be sent to AUV. Then UAAV and AUV compute
optimization is proposed to optimize the planned path. The coordination the coordinated paths, head for the target and hit it simultaneously.
between UAV and USV is similar to that between UAV and UGV, and the According to the above description, the mission can be divided into two
UAV-USV system has been applied in sea rescue and marine pollution phases, as shown in Fig. 1.
clearance (Murphy et al., 2010). In Ref. (Ramirez et al., 2011), a coor­ UAAV carrying the position information of underwater target dives
dinated system for rescuing people at sea was proposed taking advan­ into water to communicate with AUV. In this phase, UAAV flies without
tage of the high speed and sensing capability of a searching UAV and the considering the position of underwater target. When UAAV is under­
computer power and load capability of a rescuing USV. In Ref. (Jung water and gets in touch with AUV, this phase ends.
et al., 2017), an algal bloom removal robotic system is proposed. AUV is
responsible for removing the algal bloom, and to improve the work ef­ 2. Underwater coordination between UAAV and AUV
ficiency, a UAV is utilized to detect algal blooms with an image-based
detection algorithm. UAV and AUV are deployed for underwater After the position information of underwater target has been sent to
exploration task. As they work in different media, they can not AUV, UAAV and AUV navigate coordinately to hit the target simulta­
communicate with each other once the AUV is tasked (Faria et al., neously. They hit the target from different directions because it can
2014). To deal with the constraints in communication, three strategies, i. maximize the damage and gain the possibility of success.
e., periodic coordination, shortest route coordination and probabilistic The advantages of dividing the mission into two phases can be
coordination are proposed in Ref. (Sujit and Saripalli, 2013), and the concluded as follows:
goal is to maximize the explored area while minimizing the idle time of
UAV and AUV. Comparisons are conducted to analyze the characteristic 1. In the whole mission, there are various constraints in different
of each strategy. Another typical heterogeneous system is composed of media. It is difficult to obtain feasible paths for vehicles all at once
AUV and USV, which is applied in disaster assessment and sea rescue satisfying all of constraints. Therefore, decentralization is a good way
(Bingham et al., 2008). to address the constraints separately.

163
Y. Wu Ocean Engineering 182 (2019) 162–173

UAAV

Phase 1: Single flying of UAAV

Air

Target
Phase 2: Underwater coordination between UAAV and AUV

AUV Water

Fig. 1. Two phases of the underwater target strike mission.

1. Single flying of UAAV

2. As UAAV navigates both in air and underwater, the goals of navi­ UAAV and AUV under the kinetic frame is presented as follows.
gating in different media may not the same. A unified optimization 8
x_ ¼ Vcosθcosφ
index can not reflect the real objective of each phase and has a bad >
>
>
> y_ ¼ Vcosθsinφ
>
influence on the efficiency of completing the mission. <
z_ ¼ Vsinθ
(1)
>
> θ_ ¼ q
>
3. Mathematical model of coordinated path planning for a UAAV >
>
: φ_ ¼ r
and an AUV V_ ¼ a

The underwater target strike mission has been divided into two where (x, y, z) and V are the position and velocity of vehicle. θ and φ are
phases. In this section, the motion and maneuverability of UAAV and the pitch angle and the yaw angle, and q and r are the angular rates
AUV in different media are depicted firstly. Then the constraints in each correspondingly. a is the acceleration of vehicle. Note that a subscript U
phase are set to ensure the success of communication, the safety of ve­ or A can be added to the above variables to denote the states of UAAV
hicles and the success of strike mission. Finally, optimization indexes and AUV respectively. In Eq. (1), the control variable is defined as u ¼
reflecting the demand of each phase are proposed to make the goals ½ q r a �T , and the state variable is x ¼ ½ x y z V θ φ �T . The
clear. motion of vehicle can be calculated according to Eq. (1) after the value
of control variable is determined.
3.1. Kinetic model and the maneuverability of vehicles Maneuverability is an important index to reflect the ability of
changing velocity, height and motion direction for a vehicle. The con­
In a path planning problem, the vehicle is usually treated as a par­ straints on the maneuverability of vehicle must be met, or the obtained
ticle, and a particle motion model (including the particle dynamic model path is unfeasible. Here the constraints on q, r and a are listed below.
and the particle kinetic model) is used. Even in some literature, only the 8 � a �
< qU � ��qUmax��
particle kinetic model is considered (Shen et al., 2016). This is because
r � �ra �� ; zU < 0 (2)
compared to control module, path planning module is an outer loop to : U �� Umax
aU � aa �
provide the command signal, which requires a fast computation speed. Umax

For a rigid body, the motion model is a set of nonlinear differential 8 � w �

< qU � ��qUmax��
equations with variable coefficients. It will cost much time to solve such
r � �rw �� ; zU � 0 (3)
complicated equations, and it is the duty of control system to deal with : U �� Umax
aU � awUmax �
the problem. In a coordinated path planning problem, only the kinetic
model of vehicles are taken into account in existing studies to save more 8
< qA � jqAmax j
time for computing on coordination (Cui et al., 2017) (Zhu et al., 2013). r � jrAmax j (4)
Besides, UAAV is a new type of vehicle, and a completed dynamic model : A
aA � jaAmax j
for a submersible UAAV has not been reported so far, and only wind and
water tunnel testing are conducted to obtain the dynamic characteristics Eqs. (2) and (3) are the constraints considering the maneuverability
(Siddall et al., 2017). In view of the above reasons, the particle kinetic of UAAV in different media, and Eq. (4) contains the constraints of
models of UAAV and AUV are used in this study. In the ground frame, control variables for AUV.
the sea level is defined as the plane XOY, and the axis OZ is vertically
downward to obey the right hand rule. The particle kinetic model of

164
Y. Wu Ocean Engineering 182 (2019) 162–173

3.2. Constraints in each phase of strike mission qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðxU ðtÞ xA ðtÞÞ2 þ ðyU ðtÞ yA ðtÞÞ2 þ ðzU ðtÞ zA ðtÞÞ2 � dsafe ; tf 1 � t � tf 2
In the coordinated path planning model, the state of UAAV and AUV (13)
must obey specific constraints to ensure the safety of vehicles and the 8� � ��
success of strike mission. Those constraints can be concluded as follows. < �θU tf 2 θA tf 2 � � θmin
or �� (14)
: �� �
1. Constrains for single flying of UAAV φU tf 2 φA tf 2 � � φmin

Eq. (8) means the navigation time of UAAV and AUV must be the
When UAAV flies, its velocity is limited within a certain range ac­
same to hit the target simultaneously, and tf2 is the ending moment of
cording to different media because of its flight performance, which is
phase 2. Eqs. (9) and (10) guarantee that UAAV and AUV are always
determined by vehicle design and flight environment. Especially, the
underwater in this phase, and the velocity of AUV is within a certain
velocity near the water surface must be small enough to alleviate the
range according to its navigation performance. The constraints of sub­
disturbance, because a big impact may have a bad influence on the
merged debris and underwater terrain are expressed in Eqs. (11) and
motion stability of UAAV (Lock et al., 2010). The constraints on velocity
(12) respectively, and note that the positive direction of axis Oz is
of UAAV in different media and near the water surface can be presented
downward. Eq. (13) denotes that the distance between UAAV and AUV
in Eqs. (5) and (6).
must be greater than a minimal value (expressed as dsafe ) to avoid
�
0 � VU � V aUmax ; zU < 0 collision. At last, UAAV and AUV must hit the target from different di­
(5)
0 � VU � V wUmax ; zU � 0 rections to make the greatest damage. This constraint is formulated into
Eq. (14).
VU � VUcri ; jzU j � zUcri (6)
3.3. Optimization indexes of coordinated path planning
where V aUmax and V wUmax and the maximum velocity in air and water
respectively, and VUcri and zUcri are the maximum permitting velocity The constraints of coordinated path planning problem are all
near the water surface and the critical distance from UAAV to the water expressed in their mathematical forms in chapter 3.2. The number and
surface. complexity of constraints will have an influence on the solution quality,
As the communication between UAAV and AUV is made underwater i.e., the value of optimization index. The mathematical model with
to ensure the safety, UAAV must be underwater and within the fewer constraints has a higher probability to result in good solutions.
communication range of AUV at the end of this phase. This constraint Besides, complicated constraints, which are hard to be satisfied, will
can be expressed in Eq. (7). lead to the decrease of feasible solutions. As stated earlier, there are two
( z t >0
� phases in the underwater target strike mission. In phase 1, only UAAV is
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiUffiffiffiffiffiffifffi1ffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffi in motion, and the goal is to reach the water and communicate with AUV
� ��2 � � 2 � � 2
xU tf 1 xA tf 1 þ yU tf 1 yA tf 1 þ zU tf 1 zA tf 1 � dcom within the shortest time. In phase 2, UAAV and AUV navigate cooper­
atively to complete the strike mission, and they must hit the target
(7)
simultaneously with the minimum terminal position error. The goal of
where tf1 is the ending moment of phase 1 and dcom is the maximum each phase can be expressed in Eqs. (15) and (16) respectively.
underwater communication distance. J1 ¼ tf 1 (15)

2. Constraints for underwater coordination qffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffi

2 2 2
J2 ¼ maxð xU tf 2 xtar þ yU tf 2 ytar þ zU tf 2 ztar ;
After UAAV dives into water and sends the information regarding the rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffiffi �
target position to AUV, they go ahead for hitting the target coopera­ �2 2 2
xA ðtf 2 Þ xtar þ ðyA ðtf 2 Þ ytar þ ðzA ðtf 2 Þ ztar (16)
tively. In this phase, as the vehicles navigate underwater, the influence
of underwater terrain and submerged debris must be taken into account
to ensure the navigation safety. The boundary of space that the sub­ where (xtar , ytar , ztar ) is the position of underwater target. Sometimes it
merged debris occupies is defined as fi ðxd ;yd ;zd Þ ¼ 0 ði ¼ 1;…;PÞ, where can not be measured accurately, and an effective method is needed to
P is the number of debris, fi is the function describing the boundary of predict the real position of target, which can reduce the value of opti­
debris and ðxd ; yd ; zd Þ is a point on the boundary of debris. Besides, the mization index in Eq. (16). The above separate optimization indexes
height of underwater mountain can be denoted as zm ¼ hðx;yÞ, where zm reflect the real goal in each phase and avoid the inappropriate
is the height corresponding to a point (x, y) in xOy plane and h is the description of demand by using a unified optimization index.
function describing the underwater terrain. With the above descriptions
on submerged debris and underwater terrain, the following constraints 4. Coordinated path planning based on PSO algorithm
must be met to realize the mission goal.
To obtain the paths of UAAV and AUV, a coordinated path planning
tU2 ¼ tA2 ¼ tf 2 (8)
algorithm must be developed considering the characteristic of the
� established model. Firstly, the distributions of control variables in each
zU ðtÞ > 0
; t � t � tf 2 (9) phase are determined to make the constraints being satisfied easier and
zA ðtÞ > 0 f 1
improve the solution quality. Then PSO algorithm is applied to deter­
mine the values of control variables at Chebyshev points, and the
0 � VA � VAmax (10)
strategy of dealing with the solutions violating the constraints is
� generated. Finally, the procedures of coordinated path planning are
fi ðxu ðtÞ; yu ðtÞ; zu ðtÞÞ > 0
; t � t � tf 2 (11) presented to provide a clear explanation on solving this problem.
fi ðxA ðtÞ; yA ðtÞ; zA ðtÞÞ > 0 f 1
�
hðxU ðtÞ; yU ðtÞÞ > zU ðtÞ 4.1. Operations on the control variables
; t � t � tf 2 (12)
hðxA ðtÞ; yA ðtÞÞ > zA ðtÞ f 1
According to the kinetic model of vehicle, u ¼ ½ q r a �T is the

165
Y. Wu Ocean Engineering 182 (2019) 162–173

control variable in each phase. Actually, it will cost much time if the 4.2. Generation of feasible coordinated paths integrating PSO algorithm
value of control variable at every sampling moment is optimized, which
will reduce the efficiency of algorithm. To balance the computation load PSO is a well-known swarm based optimization algorithm and has
and efficiency, collocation points are set in each phase of mission. The been widely applied in path planning. In this problem, the formulas of
values of control variables at sampling points can be calculated by linear updating solutions in PSO algorithm are used, and the strategy of
interpolation. Considering the characteristic of each phase, the distri­ dealing with unfeasible solutions is designed. In many previous studies,
bution modes of control variable are not the same. each particle in PSO algorithm is a collection of discrete points, which
presents a possible path of vehicle. In this problem, as the kinetic model
1. Distribution of collocation points in phase 1 is considered, each particle represents a set of control variables, and the
corresponding path can be obtained by solving the kinetic equations. For
The height between the initial height of UAAV (denoted as zU ðt0 Þ) each particle, the dimensions equal to the number of collocation points
and the water surface is discretized as a certain number of Chebyshev (denoted as D1 and D2 in phase 1 and 2 respectively). The scale of par­
points, and they are the locations where the control variables are opti­ ticle swarm, i.e., the number of initial solutions is set as N in each phase.
mized. The locations of Chebyshev points are calculated by the The formulas of updating solutions are presented below.
following formula (Fahroo and Ross, 2000).
� � Δui ðtÞ ¼ ω ⋅ Δui ðt 1Þ þ c1 ⋅ rand1 ⋅ ðuib ðtÞ ui ðtÞÞ þ c2 ⋅ rand2 ⋅ ugb ðtÞ
ðn mÞ ⋅ π �
zm ¼ cos ; m ¼ 0; 1; 2; …; n (17) ui ðtÞ
n
(20)
where the number of collocation points is nþ1, and the two endpoints, i.
ui ðt þ 1Þ ¼ ui ðtÞ þ Δui ðtÞ (21)
e., zU ðt0 Þ and 0 are also included. zm is the height of the mth collocation
point. The Chebyshev points are dense at both ends and sparse in the where ui ðtÞ is the control vector of the ith particle in the tth iteration, and
middle. This feature is just suitable for meeting the terminal constraints uib ðtÞ and ugb ðtÞ are the control vectors corresponding to the individual
easily that there are more control variables to be operated at the end of best solution and the global best solution. ω is the inertia weight, c1 and
phase. As zm 2 ½ 1; 1� in Eq. (17), a conversion is needed to make the c2 are acceleration constraints, and rand1 and rand2 are random number
value of zm fall into the interval [zU ðt0 Þ, 0]. obeyed uniform distribution in interval [0,1]. One times of iteration is
0 zU ðt0 Þ 0 þ zU ðt0 Þ finished after the solutions are updated according to the formulas in Eqs.
hm ¼ ⋅ zm þ (18) (20) and (21). Note that if any control variable in ui ðt þ 1Þ goes out of
2 2
range, it will be set to the closest threshold to ensure the normal oper­
where hm (m ¼ 0, 1, …, n) are the real height of collocation points, and ation of algorithm. For each ui ðt þ 1Þ, it is put into the kinetic model of
the values of control variables at other heights (denoted as ho) are ob­ vehicle to obtain the states of UAAV and AUV, and the duration of phase
tained by linear interpolation, as expressed in Eq. (19). 1 and the terminal state of phase 2 can be used to calculate the fitness
8
qUhmþ1 qUhm value according to Eq. (15) or Eq. (16) respectively. If the state of vehicle
>
> violates any constraint, the fitness value of ui ðt þ 1Þ will be set to a large
> qUho ¼ h
> hm
⋅ ðho hm Þ þ qUhm
>
value to denote that the solution is infeasible and must be abandoned.
> mþ1
>
< rUh rUhm
rUho ¼ mþ1 ⋅ ðho hm Þ þ rUhm ; hm < h0 < hmþ1 (19)
>
> hmþ1 hm
>
>
> aUhmþ1 aUhm 4.3. Procedures of solving the coordinated path planning problem
>
>
: aUho ¼ ⋅ ðho hm Þ þ aUhm
hmþ1 hm
The procedures of PSO-based algorithm solving the coordinated path
As it is presented in the path planning model, phase 1 will end when planning problem for UAAV and AUV are shown in Fig. 2.
UAAV is underwater and gets in touch with the static AUV. The ending In Fig. 2, t1 and t2 are the current iteration number in phase 1 and 2
moment (also the optimization index of phase 1, J1 ¼ tf1 ) is determined respectively, and Iter is the maximum iteration times. The steps in phase
by the terminal state of UAAV. 1 and 2 are roughly the same. Compared to phase 1, there are more
control variables in phase 2 (the control variables are qU , rU and aU in
2. Distribution of collocation points in phase 2 phase 1 and are qU , rU , aU , qU , rU , aU and tf 2 in phase 2), and the con­
straints in this phase are more complicated. The final optimal coordi­
In this phase, the navigation time for UAAV and AUV is tf2 . tf2 is free nated paths will be outputted when the maximum iterations is reached,
and is also regarded as a control variable to be optimized. As the initial and the corresponding states of UAAV and AUV can be obtained by
positions of UAAV and AUV are different, it is inappropriate to discretize solving the kinetic model. Note that when using the PSO algorithm in
the collocation points by height. Considering that the navigation time phase 1 and phase 2, there are two random numbers in Eq. (20), which
for UAAV and AUV is the same, it is better to distribute the Chebyshev will cause some uncertainties. For example, a greater value of rand1
points according to different moments, which makes it convenient to makes the new solution refer more to the individual best solution, while
know the states of vehicles at any moment. The formula of generating the global best solution is paid more attention when rand2 is larger.
the collocation points and calculating the values of control variables are
the same with those from Eq. (17) to Eq. (19). 5. Simulation studies
By comparison, the distribution modes of collocation points are
different in phase 1 and phase 2, and both modes are reasonable ac­ To investigate the rationality of the established model for the un­
cording to the characteristic of each phase. In phase 1, it is easier to derwater strike mission of UAAV and AUV and the validity of the pro­
judge the height of UAAV by discretizing the collocation points by posed PSO-based coordinated path planning algorithm, simulations
height. This is important because many constraints in phase 1 are set by under different situations are conducted. In the first scenario, a static
different heights, such as the constraints in Eqs. (2), (3), (5) and (6). In underwater target is considered, and the strike mission begins after the
phase 2, both UAAV and AUV are underwater, and the constraints in this target is detected by UAAV. The paths of vehicles in each phase are
phase are set by different moments. In conclusion, to make the coordi­ generated by the proposed two-phase method, and the optimality of
nated path planning algorithm more effective, the distribution mode of solution is analyzed. Based on the setting in the first scenario, a moving
collocation points should be in accordance with the form of constraints. underwater target with known path is taken into account in the second
scenario. Two methods, i.e., the two-phase method and the whole

166
Y. Wu Ocean Engineering 182 (2019) 162–173

Initial state information of

UAAV and AUV

Initialize the control variable ui

Input the current state of UAAV
and the corresponding fitness
and AUV
value, t1=1

Initialize the control variable ui

Update ui , uib and ugb according
and the corresponding fitness
to the rules in PSO algorithm
value, t2=1

t1=t1+1 Update ui , uib and ugb according

to the rules in PSO algorithm

Is t1=Iter? No
t2=t2+1
Yes

Output the optimal control

Is t2=Iter? No
variables at collocation points

Yes
Calculate the control variables Output the optimal control
at sampling moments variables for UAAV and
AUV at collocation points

Obtain the state of UAAV at

Calculate the control variables Obtain the state of UAAV and
sampling moments
at sampling moments AUV at sampling moments
Phase 1 Phase 2

Fig. 2. Flow chart of coordinated path planning algorithm.

method (in this method the task is treated as a single process, which is
often used in UAVs coordination problems), are both applied to solve the
problem, and comparison is made. A moving target with unknown path
is considered in the third scenario, and Kalman filtering is used to pre­
dict its position. In this scenario, two methods are also used to obtain the
results, and comparison is conducted again. Note that in these simula­
tions, the time interval between two adjacent sampling moments is set to
Δt ¼ 1s, and the initial positions of UAAV and AUV are set as ( 2000,
500, 200) and ( 1000, 100, 20) respectively. All the results are
obtained by running the programs on a desktop with Intel Core i7-3370
3.40 GHz. The parameters in PSO algorithm are referred to Ref. (Shi and
Eberhart, 1998). In both phases, the scale of the swarm is set to 100, the
number of collocation points is 20 and the maximum iteration time is
100. Other initial conditions regarding the state of vehicles and con­
straints are presented in Table 1.

5.1. Coordinated path planning for hitting a static underwater target

In the first phase, AUV is static underwater and there is no coordi­ Fig. 3. 3-D paths for UAAV and AUV hitting the static underwater target.
nation between UAAV and AUV. The static target is defined as a sphere

Table 1
Settings of initial states and the constraints.
Item θU ðt0 Þ φU ðt0 Þ VU ðt0 Þ θA ðt0 Þ φA ðt0 Þ VA ðt0 Þ qaUmax

Value 0deg 0deg 100 m/s 0deg 0deg 0 m/s 0.1 rad/s
Item raUmax aaUmax qwUmax rwUmax awUmax qAmax rAmax
Value 0.2 rad/s 10m/s2 0.1 rad/s 0.2 rad/s 10m/s2 0.1 rad/s 0.2 rad/s
Item aAmax VaUmax VwUmax VAmax VUcri zUcri dcom
Value 10m/s2 100 m/s 100 m/s 50 m/s 30 m/s 10m 50m
Item dsafe θmin φmin
Value 10m 30deg 30deg

167
Y. Wu Ocean Engineering 182 (2019) 162–173

with the radius of 5m, and the position of mass center is set to (0, 0, 30). Compared to the optimization index in phase 2, it is more likely for the
After UAAV has communicated with AUV, they go head for the target limited number of discrete values to be stable with fewer times of iter­
cooperatively. The 3-D coordinated paths for UAAV and AUV in each ations. Besides, the accuracy of solution can be judged by comparing
phase is presented in Fig. 3. with the optimal solution, and the optimal solution can be estimated by
In Fig. 3, after the target is detected, UAAV dives into water and hits the following method. According to the constraint in Eq. (13), the dis­
the underwater target with AUV. To make a better understanding of the tance between UAAV and AUV must be greater than dsafe (dsafe is set to
results, the states and control variables of UAAV and AUV in each phase 10m in this scenario). As there is a rule that any two sides of a triangle is
are presented from Fig. 4, Fig. 5 and Fig. 6. greater than the third one, the distance between the vehicle and the
It can be seen from Fig. 4 that it costs 18s for UAAV to finish the first target must be greater than 5m for UAAV or AUV. Therefore, the theo­
phase (tf1 ¼ 18s). It dives into water at the end of this phase (zU ðtf 1 Þ ¼ retical optimal solution is greater than 5, which is very close to the final
2:46m), and the final distance between UAAV and AUV is 46.08m, fitness value 5.02 obtained with the PSO-based algorithm.
which meets the constraint in Eq. (7). Besides, when UAAV is diving into The above results demonstrate that the PSO-based algorithm is
water, the velocity drops to 27.32 m/s and satisfies the constraint in Eqs. effective in generating the optimal coordinated paths for UAAV and
(5) and (6). In the second phase, UAAV and AUV are always underwater, AUV, and the solution can meet all of the constraints proposed in the
and the distance between UAAV and AUV (the minimum value is path planning model.
10.002m and appears at the ending moment) is always great than dsafe ,
which ensures the safety of vehicles. The velocities of UAAV and AUV 5.2. Coordinated path planning considering the moving underwater target
are smaller than the maximum values, and the values of control vari­
ables are within their ranges in both phases. To check the position error In the first scenario, the static underwater target is considered. While
and the direction of reaching the target, the terminal states of UAAV and in some cases, the target will move, which makes the ideal terminal
AUV are listed in Table 2. positions of UAAV and AUV change. In this scenario, the target moves
The symbol “-” means that there is no ideal value for this state, and with the velocity of (Vx , Vy , Vz )¼(5, 4, 3) (m/s), and other settings are
the item “difference” means the gap between UAAV and AUV in the the same with those in the first scenario. Two methods are used to obtain
corresponding state. In Table 2, both UAAV and AUV can hit the target, the coordinated paths for UAAV and AUV, and the 3-D paths are shown
and they hit the target from different direction, i.e., the constraint in Eq. in Fig. 8.
(14) is satisfied. The above results demonstrate that UAAV and AUV can In Fig. 8, with the proposed two-phase method, UAAV and AUV can
hit the underwater target simultaneously, and their paths can meet all of cooperatively hit the target again, but UAAV and AUV deviate from the
the constraints proposed in the mathematical model. target greatly with the whole method. To further investigate the reasons
To further verify the validity of PSO-based algorithm, the conver­ for the failure of the whole method, the detailed information regarding
gence curves for each phase are presented, as shown in Fig. 7. the states of UAAV and AUV are presented in Fig. 9 and Fig. 10.
The fitness values in the two phases have different scales, so they are In Figs. 9 and 10, in both methods, the vehicles can satisfy all of the
presented separately. In phase 1, the fitness value decreases from 106 to constraints in the mathematical model, and the terminal states of ve­
18 and becomes convergent after about 40 times of iterations. In phase hicles with different methods are listed in Table 3.
2, the fitness value can converge within about 50 times of iteration, and The symbol “-” and the item “difference” have the same meanings
final fitness value is 5.02. Note that the convergence rate of PSO algo­ with those in Table 2. With the proposed two-phase method, UAAV and
rithm in phase 2 is slower than that in phase 1. This is because the AUV can hit the target cooperatively, while the vehicles can not finish
motion duration of UAAV is the optimization index in phase 1, which this task under the whole method. The convergence curves of fitness
only can be selected from some discrete values in this problem. value are shown in Fig. 11.
In Fig. 11, only the fitness value of phase 2 is presented in the two-

Height of vehicle
200
z [m]

-200
0 10 20 30 40 50 60
t [s] UAAV (in phase 1)
Path in xOy plane UAAV (in phase 2)
500
AUV
Target
y [m]

Obstacles
0

-500
-2000 -1500 -1000 -500 0 500
x [m]
Distance between UAAV and AUV
1500
In phase 1
1000 In phase 2
d [m]

500

0
0 10 20 30 40 50 60
t [s]

Fig. 4. The positions of UAAV and AUV and their distance.

168
Y. Wu Ocean Engineering 182 (2019) 162–173

Pitch angle and yaw angle

100 Pitch angle of UAAV
Yaw angle of UAAV
Pitch angle of AUV
50
Yaw angle of AUV

Angle [deg]
0

-50

-100
0 10 20 30 40 50 60
t [s]
Velocity of vehicles
100
UAAV
80 AUV

60
V [m/s]

40

20

0
0 10 20 30 40 50 60
t [s]
Fig. 5. The directions and velocities of UAAV and AUV.

Pitch rate of UAAV 150

20
Rate of pitch angle and yaw angle Yaw rate of UAAV Cost value in phase 1
Fitness value

Pitch rate of AUV

Rate of angle [deg/s]

10
Yaw rate of AUV 100

0
50
-10
0
-20
0 10 20 30 40 50 60 0 20 40 60 80 100
t [s] Times of iteration
Acceleration of vehicles 600
10
UAAV Cost value in phase 2
AUV
Fitness value
Acceleration [m/s ]
2

5
400
0

-5
200

-10 0
0 10 20 30 40 50 60
t [s] 0 20 40 60 80 100
Times of iteration
Fig. 6. Angular rates and accelerations of UAAV and AUV.
Fig. 7. Change of fitness value over the time of iteration.

Table 2
Details of terminal states for UAAV and AUV. method, the fitness value keeps to be 1000 for about 40 times of itera­
tion, and the final fitness value is 531.16. As there are too many con­
State x(m) y(m) z(m) θ(deg) φ(deg) V(m/s)
straints to be satisfied simultaneously in the whole method, all of spare
UAAV 4.37 0.39 32.44 5.11 66.47 33.22 solutions in PSO algorithm are not feasible before 40 times of iteration.
AUV 4.07 0.17 27.06 13.53 22.86 23.81
Ideal 5~5 5~5 25–35 – – –
The fitness value is set to a larger value (1000) to express the infeasi­
Difference – – – 18.64 89.33 – bility of the corresponding solution, and the final fitness value is far from
satisfaction. Compared to the whole method, the proposed two-phase
method decomposes the complicated constraints into two parts, i.e.,
phase method. This is because the optimization index of phase 1 is to constraints in the air and underwater, and each part of constraints are
dive into water and communicate with AUV, and the state of underwater dealt with respectively with the PSO-based algorithm, which makes it
target will not have an influence on the results in phase 1. For this easier to obtain satisfactory solution. The final fitness value is 5.007 in
reason, only the convergence curve in phase 2 is drawn. With the whole the two-phase method, which is very close to the optimal solution again.

169
Y. Wu Ocean Engineering 182 (2019) 162–173

Fig. 8. Paths of UAAV and AUV and motion of moving target.

Distance between UAAV and AUV

2000
The two-phase method
d [m]

1000 The whole method

0
0 10 20 30 40 50 60
t [s]
Pitch angle of vehicles
Angle [deg]

50
0
-50 UAAV (two-phase method)
0 10 20 30 40 50
AUV (two-phase method) 60
t [s] UAAV (whole method)
Yaw angle of vehicles AUV (whole method)
Angle [deg]

100
0
-100
0 10 20 30 40 50 60
t [s]
Velocity of vehicles
100
V [m/s]

50
0
0 10 20 30 40 50 60
t [s]
Fig. 9. The distances, directions and velocities of UAAV and AUV with different methods.

The above results demonstrate the superiority of the proposed two- controller and a model predictive control (MPC)-based fault tolerant
phase method solving the coordinated path planning problem. controller are developed to apply in a class of Takagi–Sugeno fuzzy
systems and air-breathing hypersonic vehicles respectively. In the third
scenario, the motion of underwater target is unknown in advance, and
5.3. Coordinated path planning based on predicting the position of Kalman filter approach is used to predict the position of target inte­
underwater target grating the motion equations of target. The target moves with the ve­
locity of (Vx , Vy , Vz )¼(3, 4, 5) (m/s), but only its current position can be
Sometimes the noise and sensor failure may affect the precision of measured with some error. With those settings, the two-phase method
measurement, which will result in some error from the real position of and the whole method are used again to generate paths for UAAV and
target. In Ref (Karimi et al., 2014). and (Selvaraj et al., 2017), actuator AUV, and the fitness values are shown in Fig. 12.
saturation with time-varying delay and external disturbances are With the whole method, the fitness value always keeps to be 1000
considered in the design of controller, and a fault-tolerant sliding mode

170
Y. Wu Ocean Engineering 182 (2019) 162–173

Pitch rate of vehicles

10

Pitch rate [deg/s]

0

UAAV (two-phase method)

-10
0 10 20 30 40 AUV (two-phase
50 method) 60
UAAV (whole method)
t [s]
AUV (whole method)
Yaw rate of vehicles
20
Yaw rate [deg/s]

-20
0 10 20 30 40 50 60
t [s]
Acceleration of vehicles
Acceleration [m/s ]

10
2

-10
0 10 20 30 40 50 60
t [s]

Fig. 10. Angular rates and accelerations of UAAV and AUV with different methods.

Table 3
Terminal states of UAAV and AUV with different methods.
Method State x(m) y(m) z(m) θ(deg) φ(deg) V(m/s)

The two-phase method UAAV 199.53 157.14 148.79 5.87 81.94 41.98
AUV 190.35 155.28 145.29 49.36 14.69 26.57
Ideal 190–200 151–161 142–152 – – –
Difference – – – 55.23 96.63 –
The whole method UAAV 411.98 17.48 197.93 15.41 14.45 43.57
AUV 390.91 290.78 140.32 14.42 20.75 38.42
Ideal 85–95 67–77 79–89 – – –
Difference – – – 0.99 6.30 –

conversion point between phase 1 and phase 2 in the two-phase method.

Cost value
1000 This statistics results just demonstrate the rationality of dividing the
The two-phase method (in phase 2)
The whole method whole process into two phases, thus reducing the probability of infea­
800 sible solutions significantly. Other constraints, such as the small velocity
of UAAV near the water surface, the distance between UAAV and AUV
Fitness value

600 and different directions to damage the target also have an influence on
the solution quality to some extent.
400 To verify the effectiveness of the Kalman filter in predicting the
location of moving target, the final predicted position of target and the
200
terminal state of vehicles with the two-phase method are listed in
Table 5.
0
0 20 40 60 80 100 There is no information on the direction of target as only the position
Times of iteration of target is concerned. The position of target can be predicted with high
accuracy by the Kalman filter, which ensures that UAAV and AUV can
Fig. 11. The best fitness value of each iteration with the two methods.
hit the target precisely.
In summary, in all of three scenarios, UAAV and AUV can finish the
during the iterative process, which means none of solutions in PSO-
underwater target strike mission cooperatively, and coordinated paths
based algorithm can meet all the constraints. The fitness value be­
satisfying various constraints can be generated by the proposed two-
comes convergent after about 50 times of iteration with the two-phase
phase method. The PSO-based algorithm has fast convergence rate,
method, and the final fitness value is 5.003. As there is no feasible so­
and the obtained results are very close to the theoretical optimal solu­
lution with the whole method, only the 3-D paths generated by the two-
tion. Compare to the whole method, the two-phase method can better
phase method are given in Fig. 13.
address the constrains in different media and decrease the probability of
To further investigate how the solutions with the whole method fail
infeasible solutions, which provide good conditions to result in high-
to meet the constraints, an analysis is conducted among the 100 solu­
quality solutions.
tions after 100 times of iteration. The detailed information on violation
of constraints is concluded in Table 4.
In Table 4, nearly all the solutions violate the constraint that UAAV
must be underwater and get in touch with the static AUV, which is the

171
Y. Wu Ocean Engineering 182 (2019) 162–173

Cost value
1000
The two-phase method (in phase 2)
The whole method
800

Fitness value
600

400

200

0
0 10 20 30 40 50 60 70 80 90 100
Times of iteration
Fig. 12. Convergence curves in the case of moving target with unknown motion.

Fig. 13. 3-D paths of vehicles and with the two-phase method.

consideration. Compared to the coordination among UAVs, USVs or

Table 4
AUVs, the coordination between UAAV and AUV has advantages in
Violation of constraints in two-phase method.
search and communication in the underwater target strike mission.
Number of Eqs (6) (7) (13) (14) The mission is divided into two phases, and the complicated con­
Rate of violation 64% 92% 31% 48% straints in different media can be decomposed into each phase, which
reduces the difficulty of solving this problem. When establishing the
mathematical model of the coordinated path planning problem, the ki­
Table 5 netic model of vehicles is considered, and the constraints are classified
Terminal states of UAAV, AUV and the target with the two-phase method. into two categories, that is, regarding the vehicles and regarding the
strike mission. Besides, considering different goals in each phase, the
State x(m) y(m) z(m) θ(deg) φ(deg) V(m/s)
optimization index aims to reduce the motion duration of UAAV and
UAAV 103.00 137.69 204.78 45.65 0.63 26.68 minimize the terminal position error respectively.
AUV 101.33 133.52 195.83 11.52 62.51 41.35
Target (predicted) 102.20 135.6 200.30 – – –
To solve the established coordinated path planning model, a PSO-
Target (actual) 102 136 200 – – 7.07 based algorithm is developed to generate the optimal paths in each
phase. Collocation points are set to reduce the computation load of
search and improve the solution quality. According to the form of con­
6. Conclusion straints in each phase, the collocation points are distributed by height
and moment in the two phases. Also the strategy of dealing with
The coordinated path planning problem for a UAAV and an AUV is infeasible solutions is formulated, which ensures the normal operation
studied in this paper. Literature investigation shows that the unmanned of the whole algorithm.
system containing heterogeneous vehicles is not given sufficient In the simulation studies, both the static and moving underwater

172
Y. Wu Ocean Engineering 182 (2019) 162–173

target are considered, and the superiority of the two-phase method are Luitpold, B., 2018. Coordinated target assignment and uav path planning with timing
constraints. J. Intell. Robot. Syst. https://doi.org/10.1007/s10846-018-0910-9.
verified by comparing with the theoretical optimal solution and the
Ma, X., Jiao, Z., Wang, Z., Panagou, D., 2018. 3-D decentralized prioritized motion
whole method. The two-phase method can better address the compli­ planning and coordination for high-density operations of micro aerial vehicles. IEEE
cated constraints and result in high quality solution that is very close to Trans. Control Syst. Technol. 26 (3), 939–953.
the theoretical optimal solution. Minaeian, S., Liu, J., Son, Y.J., 2015. Vision-based target detection and localization via a
team of cooperative uav and ugvs. IEEE Transactions on Systems, Man, and
In the future, more complicated motion model of vehicle will be Cybernetics: Systems 46 (7), 1005–1016.
established, and an improved Kalman Filter (Zhao et al., 2019) (Jiang Minaeian, S., Liu, J., Son, Y.J., 2016. Vision-based target detection and localization via a
et al., 2018) is also expected to better address the noise and failure of team of cooperative UAV and UGVs. IEEE Transactions on systems, man, and
cybernetics: Systems 46 (7), 1005–1016.
sensors. Besides, other applications of the UUAV-AUV system need to be Murphy, R.R., Steimle, E., Griffin, C., Cullins, C., Hall, M., Pratt, K., 2008. Cooperative
discovered to expand its military and civil use. use of unmanned sea surface and micro aerial vehicles at hurricane wilma. J. Field
Robot. 25 (3), 164–180.
Murphy, R.R., Steimle, E., Griffin, C., Cullins, C., Hall, M., Pratt, K., 2010. Cooperative
References use of unmanned sea surface and micro aerial vehicles at hurricane wilma. J. Field
Robot. 25 (3), 164–180.
Barbarino, S., Bilgen, O., Ajaj, R.M., Friswell, M.I., Inman, D.J., 2011. A review of Quintero, S.A.P., Ludkovski, M., Hespanha, J.P., 2016. Stochastic optimal coordination
morphing aircraft. J. Intell. Mater. Syst. Struct. 22 (9), 823–877. of small uavs for target tracking using regression-based dynamic programming.
Bingham, B.S., Prechtl, E.F., Wilson, R.A., 2008. Seakeeping System Trades for J. Intell. Robot. Syst. 82 (1), 135–162.
Coordinated Air-Surface-Underwater Operations. Oceans. IEEE. Ramirez, F.F., Benitez, D.S., Portas, E.B., Orozco, J.A.L., 2011. Coordinated sea rescue
Chen, J., Zhang, X., Xin, B., Fang, H., 2015. Coordination between unmanned aerial and system based on unmanned air vehicles and surface vessels. Oceans. IEEE.
ground vehicles: a taxonomy and optimization perspective. IEEE Transactions on Selvaraj, P., Kaviarasan, B., Sakthivel, R., Karimi, H.R., 2017. Fault-tolerant smc for
Cybernetics 46 (4), 959–972. takagi–sugeno fuzzy systems with time-varying delay and actuator saturation. IET
Cui, R., Li, Y., Yan, W., 2017. Mutual information-based multi-auv path planning for Control Theory & Appl. 11 (8), 1112–1123.
scalar field sampling using multidimensional RRT*. IEEE Transactions on Systems, Shen, C., Shi, Y., Buckham, B., 2016. Integrated path planning and tracking control of an
Man, and Cybernetics: Systems 46 (7), 993–1004. AUV: a unified receding horizon optimization approach. IEEE ASME Trans.
Fahroo, F., Ross, I.M., 2000. Direct trajectory optimization by a Chebyshev Mechatron. 22 (3), 1163–1173.
pseudospectral method. J. Guid. Control Dyn. 25 (1), 160–166. Shi, Y., Eberhart, R., 1998. Parameter selection in particle swarm optimization. In:
Faria, M., Pinto, Jos�e, Py, Fr�ed�eric, Fortuna, Jo~
ao, Dias, H., Martins, R., et al., 2014. Evolutionary Programming VII. Springer Berlin/Heidelberg, pp. 591–600.
Coordinating UAVs and AUVs for oceanographic field experiments:challenges and Siddall, R., Ortega, A.A., Kova�c, M., 2017. Wind and water tunnel testing of a morphing
lessons learned. In: IEEE International Conference on Robotics & Automation, aquatic micro air vehicle. Interface Focus 7 (1), 20160085.
pp. 6606–6611. Sujit, P.B., Saripalli, S., 2013. An empirical evaluation of co-ordination strategies for an
Jiang, K., Zhang, H., Karimi, H.R., Lin, J., Song, L., 2018. Simultaneous input and state auv and uav. J. Intell. Robot. Syst. 70 (1–4), 373–384.
estimation for integrated motor-transmission systems in a controller area network Wang, Y.C., Liu, J.G., 2012. Evaluation methods for the autonomy of unmanned systems.
environment via an adaptive unscented Kalman filter. IEEE Transactions on Systems, Chin. Sci. Bull. 57 (26), 3409–3418.
Man, and Cybernetics: Systems. https://doi.org/10.1109/TSMC.2018.2795340. Weng, Y., Wang, N., Qin, H., Karimi, H.R., Qi, W., 2018. Data-driven adaptive tracking
Jung, S., Cho, H., Kim, D., Kim, K., Han, J.I., Myung, H., 2017. Development of algal control of unknown autonomous marine vehicles. IEEE Access 6, 55723–55730.
bloom removal system using unmanned aerial vehicle and surface vehicle. IEEE Wu, Y., Wang, Y., Qu, X., Sun, L., 2018. Exploring mission planning method for a team of
Access 5, 22166–22176. carrier aircraft launching. Chin. J. Aeronaut. https://doi.org/10.1016/j.
Karimi, H.R., Hu, X., Guo, Y., Wu, L., 2014. Model predictive control-based non-linear cja.2018.08.012.
fault tolerant control for air-breathing hypersonic vehicles. IET Control Theory & Xing, W., Zhao, Y., Karimi, H.R., 2017. Convergence analysis on multi-AUV systems with
Appl. 8 (13), 1147–1153. leader-follower architecture. IEEE Access 5, 853–868.
Kim, Y.S., Lee, H., Kim, J., 2017. Coordinated weathervaning control of two surface Yang, X., Wang, T., Liang, J., Yao, G., Liu, M., 2015. Survey on the novel hybrid
vessels in a tandem configuration. Ocean Eng. 130, 142–155. aquatic–aerial amphibious aircraft: aquatic unmanned aerial vehicle (aquauav).
Li, J., Deng, G., Luo, C., Lin, Q., Yan, Q., Ming, Z., 2016. A hybrid path planning method Prog. Aero. Sci. 74, 131–151.
in unmanned air/ground vehicle (uav/ugv) cooperative systems. IEEE Trans. Veh. Zhao, D., Ding, S.X., Karimi, H.R., Li, Y., Wang, Y., 2019. On robust Kalman filter for two-
Technol. 65 (12), 9585–9596. dimensional uncertain linear discrete time-varying systems: a least squares method.
Li, Y., Ma, T., Chen, P., Jiang, Y., Wang, R., Zhang, Q., 2017. Autonomous underwater Automatica 99, 203–212.
vehicle optimal path planning method for seabed terrain matching navigation. Zhu, D., Huang, H., Yang, S.X., 2013. Dynamic task assignment and path planning of
Ocean Eng. 133, 107–115. multi-auv system based on an improved self-organizing map and velocity synthesis
Lock, R.J., Vaidyanathan, R., Burgess, S.C., Loveless, J., 2010. Development of a method in three-dimensional underwater workspace. IEEE Transactions on
biologically inspired multi-modal wing model for aerial-aquatic robotic vehicles Cybernetics 43 (2), 504–514.
through empirical and numerical modelling of the common guillemot, uria aalge.
Bioinspiration Biomimetics 5 (4), 1–15.

173