Received 12 January 2023, accepted 14 February 2023, date of publication 24 February 2023, date of current version 7 March 2023.

Digital Object Identifier 10.1109/ACCESS.2023.3246093

Ship Formation Algorithm Based on the

Leader–Follower Method
MINGYANG LI 1, KAI MENG 1, JIELU CHEN 2, AND HONGBO WANG 1, (Member, IEEE)
1 State
Key Laboratory on Integrated Optoelectronics, College of Electronic Science and Engineering, Jilin University, Changchun 130012, China
2 CSSC Marine Technology Company Ltd., Pudong, Shanghai 200136, China

Corresponding author: Hongbo Wang ([email protected])

This work was supported in part by the National Natural Science Foundation of China (NSFC) under Grant u1964202.

ABSTRACT In recent years, marine engineers have increasingly relied on multiship collaborative work
to complete complex tasks at sea and improve maritime transport efficiency. Ship formation significantly
contributes to the maintenance of stability during multiship-coordinated operations, but establishing the
formation structure while ensuring the safety of each ship remains challenging. Therefore, this study
proposes an improved leader–follower-based formation algorithm as a solution to this issue. First, the fast
marching square method was used for global static path planning. The leader–follower formation control
method then enabled the path tracking of the follower ship to the leader ship. Finally, an improved artificial
potential field method was used for local collision avoidance, and virtual obstacles were added to improve the
situation in which the ship probably entered a local minimum point. The experimental results revealed that the
proposed algorithm could plan a route with time, smoothness, and safety advantages for the formation leader
and quickly form a stable formation. Notably, the ship can quickly initiate collision avoidance operations
when it encounters an obstacle posing a collision risk.

INDEX TERMS Artificial potential field method, fast marching square method, formation of ships,
leader–follower method, local collision, path planning.

I. INTRODUCTION working efficiency than traditional single ships when con-

Most traditional ship motion research objects are single ships, fronting complex tasks [1], [2].
but single ships frequently have significant operational limi- Ship navigation relies heavily on path planning, which
tations. Owing to the complexity and diversification of nav- can be divided into global and local. Based on task require-
igation and marine operation tasks in recent years, multiple ments, global path planning uses advanced data such as
sea surface ships must collaborate to accomplish complex electronic charts and appropriate search algorithms to find
tasks that a single ship cannot, such as fleet cooperative oper- a feasible, barrier-free route in a large area [3], [4], [5],
ation, search and rescue, fleet roundup, and marine resource [6]. Local path-planning algorithms play an auxiliary role
exploration. With the rapid advancement and integration in global navigation systems. During global path navigation,
of communication and computer technology, the problem the ship detects and autonomously avoids unknown obstacles.
of the cooperative formation of surface ships has received After completing the avoidance, the ship returns to its global
increased attention from scholars around the globe. Notably, path [7], [8], [9], [10].
the study of ship formation systems is of great theoret- Currently, the most commonly used global path-planning
ical value and practical engineering significance because algorithms include Dijkstra [11], [12], [13], A∗ [14], [15],
such systems have greater fault tolerance, adaptability, and genetic [16], [17], [18], particle swarm optimization [19],
[20], [21], and ant colony algorithms [22]. Dijkstra and A∗
algorithms are used to solve the optimal path. The Dijkstra
The associate editor coordinating the review of this manuscript and algorithm was first proposed by Dijkstra [23] in 1959. The
approving it for publication was Emre Koyuncu . classic Dijkstra algorithm has a simple underlying principle,

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

VOLUME 11, 2023 For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ 21655
 M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

but its calculation process is overly complex and memory- the artificial potential field method to realize the local path
intensive, making it unsuitable for large-scale path planning. planning of the ship.
Moreover, the A∗ algorithm is a heuristic algorithm proposed With the maturation of the single ship control theory, Pro-
by Hart et al. [24]; this algorithm is the most effective method fessor Fossen and his team initiated research into multiship
for determining the shortest path in static networks, extend- cooperative formation control in 2002 [34]. Many important
ing Dijkstra’s algorithm. Although the A∗ algorithm has a theoretical and technical advances have been made in the
simple underlying principle and outperforms the Dijkstra study of formation control of surface ships, thanks to the
algorithm, it relies heavily on the heuristic function, resulting collaboration of international navigation and control circles.
in a massive amount of computation, and both algorithms Because maintaining formation stability is crucial for ship
require additional smoothing. Furthermore, James proposed formation, inspired by the behaviors of birds gathering, ant
a numerical method called fast marching to solve the Eikonal colonies, and bee swarms, researchers have proposed var-
equation’s boundary value problem [25]. This method gen- ious formation structures, such as leader–follower, virtual
erates potential fields by simulating the propagation of elec- structure [35], [36], [37], [38], [39], and behavior-based
tromagnetic waves as they travel from their source point to methods [40], [41], [42].
their destination. The potential field simulated by this method Among these formation methods, the formation control
has only one global minimum point, the starting point, and method based on leader-follower method has the advan-
its path is generated by the gradient descent method, which tages of simple principle and easy implementation, which
effectively avoids the problem of a local minimum. Based on is favored by scholars. The leader–follower method adopts
this, the enhanced fast marching square method is suitable for a centralized control structure. The basic concept is to des-
ship tracking due to its excellent smoothness and security. ignate one or more team members as leaders and the rest
The current local path planning algorithms mainly include as followers. The leader can track the desired trajectory
fuzzy logic algorithm [26], [27], artificial potential field and posture, and the follower can communicate with the
method [28], [29], [30], [31], velocity obstacle method [32], leader. Meanwhile, maintaining relative distance, azimuth,
[33] and so on. Guan et al. proposed a ship domain model and attitude consistency with the pilot is essential for achiev-
based on fuzzy logic aimed at providing early warning of ship ing formation control of a particular formation. Ding and
collision risk and a reasonable reference that can be used in Guo [43] studied the surface ship formation problem based
combination with the International Regulation for Preventing on the leader-follower method. By designating one ship in
Collisions at Sea [26]. Shang et al. designed a fuzzy con- a multiship system as the system’s leader and the other
troller to improve the problem of trajectory oscillation, which ships as its followers, the formation problem was viewed
outputs the environment danger factor to adjust the step size as the following and coordination problem of all following
of robot and enhance the trajectory smoothness of robot in ships to the position and direction of the system’s leader.
complex environment [27]. Yang studied the development Liu et al. used control theory to investigate the controllability
process of the obstacle avoidance system for autonomous of a leader–follower multiagent system in switched commu-
vehicles and propose an optimization scheme for the obstacle nication topology and presented the system’s controllability
avoidance algorithm [28]. Chen and Xu proposed a method of conditions in this setting [44]. Ajwad et al. studied the same
removing the shaking state based on a favorable path, and on system based on a continuo–discrete observer and demon-
this basis, visibility graph method is used to optimize the path strated the proposed algorithm’s stability using Lyapunov
of the AUV to avoid obstacles [31]. Wang et al. proposed a theory [45]. Wang et al. also studied the consistency of the
USV autonomous dynamic obstacle avoidance method based aforementioned system under a directed hostile communica-
on the enhanced velocity obstacle method in order to achieve tion topology [46].
path replanning [32]. This paper proposes an improved leader-follower forma-
However, in local path planning algorithms, ships often tion algorithm to establish the formation structure and ensure
work in a partially or completely unknown environment, the safety of the formation. The global static path planning
which requires continuous detection of the surrounding envi- is carried out by the fast marching square method, and it
ronment by sensors. According to the data detected by the is combined with the leader-following method for the first
sensors: the distance, size, shape, etc. between the obstacle time to realize the path tracking from the follower ship to the
and the ship itself, its position, speed, and direction are con- leader ship. Notably, the preceding studies did not consider
stantly adjusted. This obstacle avoidance method requires the the possible obstacles encountered by followers during navi-
performance of the ship’s own sensors to be very powerful, gation. Therefore, this study proposes an improved artificial
and the designed algorithm should be more and more accurate potential field method to solve this problem, enabling the
and practical. The artificial potential field method is a more follower to independently take collision avoidance behavior
mature and effective algorithm, which allows moving objects when collision risk occurs during the trajectory tracking of
to avoid obstacles while considering the motion performance the leader and to return to the team after a collision avoidance
of moving objects. It has the advantages of strong real-time operation.
performance, simple mathematical calculation, smooth path The remainder of this paper is structured as follows.
planning, and easy programming. Therefore, this paper uses Section II introduces the basic concepts and applications of

21656 VOLUME 11, 2023

M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

Case 1: When T > max(Tx , Ty ), the Eikonal equation can

be written as
T − Tx 2 T − Ty 2 1
( ) +( ) = 2. (3)
1x 1y Fij
Assuming that the unit distance between the horizontal and
vertical discrete points is 1, i.e., 1x = 1y = 1, then the
Eikonal equation can be reduced to a quadratic equation of
one variable:
FIGURE 1. 2D discrete space. 1
(T − Tx )2 + (T − Ty )2 = 2 , (4)
Fij
the fast marching and fast marching square methods, reveal-
1 1
ing how the latter plans a collision-free path for the ship’s T 2 − (Ty + Tx )T + (Ty2 + Tx2 − 2 ) = 0. (5)
leader. Section III presents the basic knowledge of graph 2 Fij
theory and its application to ship formation. The consis- Thus, the solution of time T can be written as follows:
tency theory is adopted to solve the tracking problem of the r
follower to the leader and realize the effect of formation Tx + Ty + 22 − (Tx − Ty )2
Fij
keeping. Section IV presents an improved artificial potential T = . (6)
field method for solving the local collision avoidance prob- 2
lem, enabling followers to take collision avoidance behavior Because the other solution of time T is negative, it does not
independently. Section V presents the simulation and analysis satisfy the condition and is thus dropped.
of the proposed algorithm. Finally, Section VI concludes the Now, determine whether the solution satisfies the assump-
study. tions, i.e., if the following inequality is true:
r
II. FAST MARCHING METHOD AND FAST MARCHING Tx + Ty + 22 − (Tx − Ty )2
Fij
SQUARE METHOD T = > max(Tx , Ty ). (7)
A. FAST MARCHING METHOD 2
1) PROPOSAL OF THE FAST MARCHING METHOD Let Tx ≥ Ty , that is
In 1996, J. Sethian first proposed a numerical algorithm, the r
2
fast marching method (FMM), which can approximate the Tx + Ty + − (Tx − Ty )2
Fij2
viscosity solution of an equation, and used this algorithm to T = > Tx . (8)
iteratively solve the Eikonal equation to solve the interface 2
propagation problem. The solution to this inequality is
The function equation has the following form: 1
T = T2 + . (9)
|∇T (x, y)| W (x, y) = 1, (1) Fij

where (x, y) is the coordinates in the position and pose space Case 2: when T1 ≥ T > T2 , the Eikonal equation can be
of the calculated point, T (x, y) is the time when the interface written as follows:
reaches the point (x, y), and W (x, y) is the local propagation 1
(T − T2 )2 = 2 , (10)
speed of the interface at the point (x, y). Fij
Although the FMM used to solve the Eikonal equation
shares commonalities with the Dijkstra method, the Dijk- where T1 = max(Tx , Ty ), T2 = min(Tx , Ty ).
stra algorithm updates based on Euclidean distance between The solution to this equation is
nodes, whereas the FMM algorithm updates based on the 1
T = T2 + . (11)
approximate partial differential equation obtained from the Fij
simplified Eikonal equation.
Because the other solution of time T is negative, it does not
To use FMM, the space must be discretized. Fig. 1 shows
satisfy the condition and is thus dropped.
a discrete 2D space. The Eikonal equation can be written as
Now, determine whether the solution satisfies the assump-
follows:
tions, i.e., if the following inequality is true:
T − Tx T − Ty 1
max( , 0)2 + max( , 0)2 = 2 , (2) 1
1x 1y Fij T = T2 + ≤ T1 . (12)
Fij
where Tx = min(T(x−1x,y) , T(x+1x,y) ), Ty = min(T(x,y−1y) ,
The solution to this inequality is
T(x,y+1y) ).
Notably, the solution set for time T in this equation can be 1
T1 − T2 ≥ , (13)
obtained as follows: Fij

VOLUME 11, 2023 21657

 M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

FIGURE 2. Initializing the map.

FIGURE 3. The first iteration.
which is
1
Tx − Ty ≥ . (14)
Fij
Overall, the solution set of T can be written as follows:
 r
2
 T x + T y + − (Tx − Ty )2
Fij2


 1
, Tx − Ty <


T= 2 Fij

 1 1
 min(Tx , Ty ) + , .

Tx − Ty ≥


Fij Fij
(15)

2) DESCRIPTION OF THE FAST MARCHING METHOD

ALGORITHM
The algorithm description of FMM is shown in Algorithm 1.
• Step 1: The navigation environment is modeled and
FIGURE 4. Iterative process.
converted into a binary grid diagram. Separate the cells
that belong to the barrier and the cells that correspond to
the passable area. all feasible areas for the iteration result or until all points
• Step 2: FMM divides points on the map into three are set to frozen points or reach the starting point of the
groups: frozen, open, and unvisited points. Frozen points path-planning problem.
are points at which the arrival time can no longer be Each grid point’s value in this constructed potential field
changed. Unvisited points are unprocessed points. Open represents the arrival time of the interface. Notably, the inter-
points can be thought of as the interface between the face cannot propagate in the infeasible region; that is, the
frozen and unvisited areas of the map belonging to propagation speed of the interface in the infeasible region is 0,
the propagation surface. In the path-planning problem, and the arrival time approaches infinity. If the interface travels
to ensure that only one global minimum point exists, at the same speed throughout the map, the arrival time of the
the wave propagates from the target point. As shown in interface is proportional to the distance from the starting point
Fig. 2, in this step, all cells in the point set are initialized of the current grid data.
to an infinite value and set to unvisited points, except that
the propagation starting point is set to an arrival time B. FAST MARCHING SQUARE METHOD
of 0. The essence of FMM is to search for the interface’s shortest
• Step 3: In each iteration, the open point comprises four propagation path. The path length is prioritized over path
adjacent grids of a black sphere (marked blue), as shown safety and smoothness. If you drive along the path generated
in Fig. 3. The open point with the smallest T-value is by the FMM, only the shortest distance is guaranteed, and
then selected as the frozen point by solving the formula. safety cannot be ensured.
As shown in Fig. 4, the process continues until the end This study uses the fast marching square (FMS) method to
condition of interface propagation theoretically covers calculate a path with sufficient smoothness and safe distance,

21658 VOLUME 11, 2023

M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

Algorithm 1 Fast Marching Method Algorithm Description • Step 4: From the ship’s starting point, a gradient descent
Input: The velocity of a point in a 2-dimensional discrete is made across the entire arrival time map, moving
space Fij , toward its target point (the global minimum of the result-
Point set S, Discrete space step size 1h, ing map). By using FMS, a path that incorporates arrival
Output: Time T time, smoothness, and safety is obtained.
1: ∀p ∈ / S, T (p) = +∞
2: for p ∈ S do In this section, FMS solves the global path-planning problem
3: ps = Neighbor(p) and provides the formation leader with the planned path to
4: T (ps) = T (p) + 1h track. Then, the subsequent section will solve the problem of
5: Trial ← ps, Alive ← S how to realize the tracking of the follower to the leader in
6: end for order to maintain formation stability.
7: while Trial ̸ = ∅ do
8: m = min(Trial) III. LEADER–FOLLOWER SHIP FORMATION
9: Alive ← m A. STRUCTURE OF THE SHIP FORMATION
10: ms = Neighbor(m) With the development of ship clustering technology, which
11: for ∀p ∈ S, p ∈/ Alive do was inspired by natural behaviors, such as bird gathering, ant
12: {i, j} = p colonies, and bee swarms, researchers began exploring ways
13: Tx = min(T(i−1h,j) , T(i+1h,j) ) to control multiple smart ships simultaneously, resulting in
14: Ty = min(T(i,j−1h) , T(i,j+1h) ) the creation of ship formation technology. The implemen-
15: if Tx − Ty < 1h Fij then
tation of ship formation technology can increase efficiency
r
2 and performance while addressing the shortcomings of single
Tx +Ty + −(Tx −Ty )2
Fij2 ships, such as inadequate carrying capacity and insufficient
16: T = 2 > Tx
information processing capacity. Simultaneously, the stability
17: else
and safety of ship formation are improving thanks to advance-
18: T = min(Tx , Ty ) + F1ij
ments in computer science and technology, communication
19: end if technology, and navigation technology.
20: T (p) = min{T (p), T } Typical cluster collaborative control technologies include
21: Trail ← p leader–follower, virtual structure, and behavior-based
22: end for methods.
23: return T The virtual structure approach treats the entire system as
a rigid-body structure, with the individuals making up the
formation serving as the rigid body’s reference points with
addressing the path defects generated directly by FMM, fixed relative positions. When the entire body is in motion,
notably that the path is too close to the obstacle and is not each individual can achieve a fixed shape formation motion
smooth. by tracking the corresponding reference points on the rigid
FMS applies the basic FMM twice. body. Group behavior can be easily programmed using this
• Step 1: The environment is modeled similarly to when method, which also yields excellent results in the areas of
FMM is used. Convert the environment into a binary trajectory tracking and formation keeping.
grid. Separate the cells that belong to the barrier and the The fundamental concept underlying the behavioral
cells that correspond to the passable area. method is as follows: first, the agent’s expected basic behav-
• Step 2: By applying FMM for the first time, FMS iors are specified. In general, behaviors include collision
treats each obstacle-marked cell on the map as a wave avoidance, obstacle avoidance, target-directed driving, and
source and simultaneously expands multiple waves. The formation maintenance. When the sensor is stimulated by the
resultant value of each map cell represents the time external environment, it responds based on the sensor’s input
required for the wave to reach the nearest obstacle, information and outputs the response vector as the expected
which is proportional to the obstacle’s distance. The behavior response.
potential diagram obtained from this step is the velocity The benefits of the leader–follower method include a sim-
potential W (x). ple principle, easy implementation, and good stability. As a
• Step 3: Based on the potential diagram W (x), FMM result, this study uses a centralized control structure in which
is executed again from the target point. This time, the one ship acts as the formation leader and the others act as its
target point is treated as the only wave source to ensure a followers.
global minimum. The wave expands across the map until As a global path-planning algorithm, FMS described in
it reaches its initial point. For each cell in the feasible Section II can be used to plot the course of a ship formation’s
region, the wave expansion velocity is extracted from leader. In this section, we will determine how to maintain
the velocity potential diagram W (x) calculated in the the relative distance and azimuth between the follower and
previous step. leader.

VOLUME 11, 2023 21659

 M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

B. FOUNDATION OF GRAPH THEORY

In general, multiagent formation control has a large scale, and
each individual has a communication and control relationship
that can form a network structure. Therefore, the multiagent
system can be modeled naturally as a graph. The graph vertex
typically represents a single agent. The graph edges represent
the topological relationships between agents.
After the multiagent system is modeled as a graph, the
formation problem can be studied using the relevant knowl-
edge of graph theory. Graph theory methods make it simple
to design team shapes and formation control algorithms. FIGURE 5. Undirected connected graphs.

1) LAPLACIAN MATRIX
The Laplacian matrix, also known as the admittance matrix,
Kirchhoff matrix, or discrete Laplacian operator, is mainly
used in graph theory. As a matrix representation of a graph,
the Laplacian matrix can reflect the degree of graph con-
nectivity while describing the point–edge relationship in the
graph.
Given a figure G = {V , E}, as shown in Fig. 5, V =
{v1 , v2 , . . . , vn } represents the vertices in the graph, E ⊂ v×v
is the graph edge; the Laplacian matrix is defined as L =
D − A, where D stands for the degree matrix and A is the
adjacency matrix.
The degree matrix is a diagonal matrix whose diagonal
value is the sum of the total number of adjacent nodes that FIGURE 6. Leader–follower multiship formation system.
maintain a communication relationship with each vertex in
the graph. The degree matrix in this graph is
2) APPLICATION OF THE LAPLACIAN MATRIX IN SHIP
FORMATION
 
2

 2 
 The leader–follower multiship formation system treats each
 3 .
 ship in the system as a network node and the information
D=  (16) transmission link between ships as an edge connecting differ-
 3 

 2  ent nodes. Thus, a multiship system and its communication
2 network can be modeled as a diagram. However, unlike the
undirected graph introduced in Section III, Part B(1), due
The adjacency matrix is a symmetric matrix whose value to the existence of information flow in formation, the ship
on the main diagonal is 0. If a communication relationship formation system is constructed as a directed graph, as shown
exists between points vi and vj , then aij = aji = 1; otherwise, in Fig. 6, where the arrow direction represents the direction
aij = aji = 0. The adjacency matrix is of information transmission.
  Given the directed graph G = {V , E}, V =
0 1 1 0 0 0
1 0 1 0 0 0 {v1 , v2 , . . . , vn } represents the vertices in the graph, i.e., N

1 1 0 1 0 0
 ships in the system, E ⊂ v × v is the graph edge.
A=  . (17) In this digraph, the adjacency matrix A is transformed into
0 0 1 0 1 1

0 0 0 1 0 1 a weight matrix; if vertex v is the starting point of edge E,
then aij = 1, otherwise aij = 0. The degree matrix D
0 0 0 1 1 0
is transformed into an indegree matrix, with the diagonal
Thus, the Laplacian matrix can be written as elements representing the sum of the nodes’ intake degree.
  Thus, the weight matrix of the system can be written as
2 −1 −1 0 0 0 follows:
 −1
 2 1 0 0 0 

3 0 0 
 −1 −1  
−1 . 0 0 0 0 0
L =D−A=  0
 0 −1 3 −1 −1 
 1 0 0 0 0
1 1 0 0 0.
 0  
0 0 −1 2 −1  A=  (19)
0 0 0 −1 −1 2 0 1 0 0 0
(18) 0 0 1 1 0

21660 VOLUME 11, 2023

M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

The indegree matrix is as follows:

 
0
 1 
.
 
D=  2  (20)
 1 
2
The Laplacian matrix can be written as follows:
 
0 0 0 0 0
 −1 1 0 0 0
.
 
L = D − A =  −1 −1
 2 0 0 (21)
 0 −1 0 1 0
0 0 −1 −1 2
Define x = [x1 , x2 , . . . x5 ]T as the current coordinate
value of each ship and η as the expected relative positions FIGURE 7. Schematic diagram of the artificial potential field method.

of adjacent two points.

Thus, for each edge on the digraph, we can obtain the A. ARTIFICIAL POTENTIAL FIELD
following:
Fig. 7 is a schematic representation of the artificial potential
The error of the expected distance from point xj to point xi
field method.
is as follows:
Assuming that the real-time coordinate position of the ship
eij = xi − xj − η. (22) is X = (x, y) and the position of the target point is Xg =
(xg , yg ), then the gravitational field function is
The sum of the errors of the expected distance between
point xj and all other points is as follows: 1
Ua = k(X − Xg )2 , (24)
X 2
ej = eij = Ax − Dx − kη = −Lx − kη. (23)
where the coefficient k is the gain coefficient of the gravi-
i
tational field, and the appropriate constant is selected by the
To enable the follower to track the leader’s trajectory, experiment.
we must reduce the value of ej to 0, thereby affecting the Notably, the size of the gravitational field is positively
speed of each follower by ej and achieving formation stability correlated with the distance between the ship and the target
by eliminating the error. point.
The following relation can be derived from the relationship
IV. LOCAL COLLISION AVOIDANCE BASED ON AN
between gravitation and gravitational field:
IMPROVED ARTIFICIAL POTENTIAL FIELD METHOD
In Section III, we used the leader–follower method to track Fa = −grad(Ua ) = −k(X − Xg ). (25)
the leader’s path, allowing the ship team to reach the end In the artificial potential field method, the repulsive force
point stably. The analysis in the preceding section reveals, is inversely proportional to the square of the distance of the
however, that the follower in the team follows only the leader obstacle. That is, as the distance to the obstacle decreases, the
mechanically. When the follower is too close to an obstacle repulsion intensifies and vice versa. However, the influence of
on the tracked path, the collision risk is likely to materialize. obstacles on ships is not infinite and only has an effect within
Therefore, each follower must independently achieve local a certain range, as indicated by the black dotted line area in
collision avoidance in order to deal with this situation. This Fig. 7. The repulsive force becomes 0 outside this range. The
section focuses on the independent collision avoidance of mathematical expression of the repulsive force field is
each follower when encountering collision risk, as well as on 1 1
the maintenance of collision avoidance within the formation.  β( − 1 )2 ρ ≤ ρ0
The artificial potential field method proposed by Ur = 2 ρ ρ0 (26)
ρ > ρ0 ,

KHATIB [47] establishes a constraint relationship among 0
ships, target positions, and obstacles through a virtual force where the coefficient β is the gain coefficient of the repulsive
field. When sailing in formation, the follower ship naturally force field, ρ is the ship–obstacle distance, ρ0 is the influence
receives the gravitational force brought about by the preset radius of the repulsive force field, and the repulsive force
target position of the formation. However, if it senses that beyond this range is zero.
the distance to the obstacle is too close, it will enter collision The repulsive force can be expressed as follows:
avoidance mode and experience extra repulsive force from
 β( − 1 ) 1 ∂ρ ρ ≤ ρ0
 1
the obstacle. The follower ship will move under the combined
influence of gravity and repulsive force until it can avoid the Fr = ρ ρ0 ρ 2 ∂X (27)
ρ > ρ0 .

obstacle and reach a safe area. 0

VOLUME 11, 2023 21661

 M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

FIGURE 8. Local minimum case in the artificial potential field method.

FIGURE 9. Adding virtual obstacle.

Finally, through the superposition of repulsion and gravity,

the resultant force can be obtained as follows: The improved repulsive force expression changes the force
n state by adding virtual obstacles. When determining the posi-
X
F = Fa + Fr , (28) tion where the ship’s force balance enters the local minimum
1 point, introducing virtual obstacles will alter the repulsive
force at the position of the point and then change the overall
where n is the total number of obstacles that can affect the force so that the force is no longer 0, and the local minimum
ship’s force at this time. problem can be improved. Fr and Fr′ represent the repulsive
force of the obstacle and virtual obstacle on the robot, respec-
B. THE DEFECTS AND IMPROVEMENT OF THE ARTIFICIAL tively. Combined with the gravity formula of the target point,
POTENTIAL FIELD METHOD the resultant force can be expressed as follows:
When the ship, obstacle, and target point are in the same n m
straight line, and the obstacle is in the middle position, the X X
F = Fa + Fr + Fr′ . (31)
ship will be in a critical state. At this time, the ship’s force
1 1
is balanced, and stagnation or oscillation may occur. In this
state, the ship is said to have fallen into the local minimum When the follower tracks the leader’s trajectory navigation,
point (Fig. 8). if obstacles exist within the follower’s perception range, the
When the local minimum is detected, a virtual obstacle improved artificial potential field method proposed in this
is added by evaluating the obstacle distribution to assist the section can be used for local collision avoidance operations.
ship in escaping the local minimum point. Virtual obstacles Once the collision danger is removed, the consistency theory
are usually set in a direction perpendicular to the forward is reapplied to track the leader’s trajectory and return to the
direction to provide virtual forces (Fig. 9). By adding virtual team. Fig. 10 demonstrates the flow chart.
obstacles at this time, the resultant force on the ship will be • Step 1: Input the parameters of each ship, such as
altered, and the virtual obstacles will provide the ship with current position, sailing speed and direction.
an additional escape force to evade the local minimum point. • Step 2: Determine whether the ship has entered the
In addition, the presence of repulsive force prevents the ship influence range of the obstacle, if so, calculate repul-
from reentering the minimum point. sive and gravitational forces, otherwise continue to sail.
The virtual repulsive force field is as follows: • Step 3: When the ship is already affected by the obsta-
1 cle, whether the resultant force is 0 is judged, if it is 0,
 β ′ ( 1 − 1 )2 ρ ′ ≤ ρ ′ add virtual obstacles and then calculate the direction
Ur = 2 ρ ′
′ ρ0′ 0
(29) of motion, if not, directly calculate the direction of
ρ > ρ0′ ,
′

0 motion. This step is repeated until it leaves the influ-
ence range of the obstacle.
where β ′ is the virtual repulsion potential field constant • Step 4: When the ship is out of collision risk, it contin-
greater than zero, ρ ′ is the distance between the ship and ues to sail according to the original trajectory.
the virtual obstacle, ρ0′ is the influence distance of the virtual
obstacle on the ship.
V. SIMULATION EXPERIMENT
Then, the virtual repulsive force is as follows:
This section focuses on two MATLAB simulations conducted
 β ′ ( 1 − 1 ) 1 ∂ρ ρ ′ ≤ ρ ′
 ′ to validate the proposed algorithm’s effectiveness. In the first

Fr′ = ρ′ ρ0′ ρ ′2 ∂X 0
(30) simulation, the USV group path-planning algorithm based on
FMS and the leader–follower method is validated in a static
ρ > ρ0 .
′ ′

0

environment, whereas the local collision avoidance problem

21662 VOLUME 11, 2023

M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

FIGURE 11. Initial map.

FIGURE 10. Flow chart of improved artificial potential field method.

FIGURE 12. Potential field diagram.

solved by the improved artificial potential field method is
validated in the second simulation.

A. FORMATION PATH PLANNING BASED ON FMS

1) ALGORITHM SIMULATION OF FMM
In this section, we will first verify the path-planning problem
based on FMM. Figures are used to demonstrate the steps of
the FMM method. In Fig. 11, MATLAB is used to generate an
initial map with a size of 200 × 200 pixels, where black areas
represent obstacles, such as land or islands, and white areas
represent free space. FMM-based USV group path planning
is based on this map. Fig. 12 depicts the potential field FIGURE 13. Path planning based on FMM.

diagram obtained by first constructing a binary grid diagram,

followed by iterating the end point on the figure’s propagation
surface. Finally, after the above space is established, the
gradient descent method is used to iterate continuously from
the starting point (marked as a black point in the figure)
along the direction of the fastest gradient descent in the
generated potential field, and a path with the shortest distance
to the end point (marked as a red point in the figure) is
obtained (Fig. 13).
Fig. 13 demonstrates that the FMM-obtained path is not
smooth and is too close to the obstacle. This is because the
essence of FMM is to search for the shortest propagation path FIGURE 14. Velocity potential field diagram.
of the interface, which prioritizes the length of the path over
its safety and smoothness. route planning is as follows: by applying the initial FMM to
iterate over all units regarded as obstacles on the propaga-
2) ALGORITHM SIMULATION OF FMS tion surface of the figure, the velocity potential diagram is
The simulation environment uses the same binary grid dia- obtained, as shown in Fig. 14, where each grid has a value
gram as FMM. As shown in Fig. 11, cells belonging to obsta- representing its distance from the nearest obstacle, ranging
cles are marked with black, and cells corresponding to the from 0 to 1, and a low value indicates that the current position
passable area are marked with white. The principle of FMS may be too close to the obstacle, indicating that the USV

VOLUME 11, 2023 21663

 M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

FIGURE 15. Path planning based on FMS.

TABLE 1. Ship’s initial information.

FIGURE 16. Initial position.

should sail in areas with high values. Subsequently, based

on the potential diagram W (x), FMM is reapplied to iterate
from the end point on the propagation surface of the diagram,
and the potential field diagram is obtained, as shown in the
figure. Finally, gradient descent is performed on the entire
arrival time map to obtain the optimal path in terms of arrival
time, smoothness, and security (Fig. 15).
Notably, the track obtained by path planning using the FMS
method is superior in terms of time, security, and smoothness
FIGURE 17. Trajectory of formation motion.
and can be used for the leader’s path planning (Fig. 15).

3) FORMATION SIMULATION BASED ON THE

LEADER–FOLLOWER METHOD
The first simulation experiment is conducted to verify the
stability of the leader–follower method.
The simulation of the USV group path-planning algorithm
based on the FMS method includes four followers and a
leader. Table 1 lists the initial information of the leader and
each follower. As shown in Fig. 16, the leader is represented
by a black circle, and the followers are represented by circles
of other colors.
The leader sails following the preplanned path, as stipu-
lated in the simulation, to drive in a straight line with uniform
speed toward the target point (5, 5). The rest of the followers
follow the leader in a preset formation (the two wings are at
60◦ , and the distance between the two adjacent ships on the FIGURE 18. Position error curve between the leader and each follower.
same side is 0.2 km). The error distance affects the follower’s
speed, but to prevent the follower from moving too quickly in 12 min, all followers reach the preset position and can main-
the simulation, a maximum speed limit is set for the follower tain a stable structure with the leader after that.
in this experiment. Finally, employing the simulation experiment based on
The trajectory of the ship formation and relative error FMS in the last section, the leader–follower method is used
curves of the leader and each follower are shown in Fig. 17 to define the formation to move forward to the end with the
and Fig. 18. The relative error curve shows that, at about angle between the two wings set at 60◦ , enabling the follower

21664 VOLUME 11, 2023

M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

TABLE 2. Ship’s initial information.

FIGURE 19. Route of a ship formation.

FIGURE 21. Trajectory of formation motion.

FIGURE 20. Initial position of ship and obstacle.

to maintain the same navigation with the leader. The path

curve in this case is shown in Fig. 19. Notably, the method
can achieve formation stability and consistency.

B. LOCAL COLLISION AVOIDANCE SIMULATION BASED

ON THE IMPROVED ARTIFICIAL POTENTIAL
FIELD METHOD
When using the leader–follower method to track the path
of the follower to the leader, the collision risk is likely to FIGURE 22. The position error curve between the leader and each
occur if a situation exists in which the follower is too close follower.
to the obstacle on the tracked path. This section simulates the
solution to the local collision avoidance problem. The rest of the followers follow the leader in a preset forma-
tion (the two wings are at 60◦ , and the distance between the
1) ALGORITHM SIMULATION OF THE IMPROVED ARTIFICIAL two adjacent ships on the same side is 0.2 km).
POTENTIAL FIELD METHOD As shown in Fig. 21–24, the follower tracks the leader’s
The initial information of the leader and each follower is path for the first 2 min and 30 s. In 2.7 min, follower-3 senses
listed in Table 2. As shown in Fig. 20, the leader is rep- an obstacle and enters the collision avoidance state. The col-
resented by a black circle, and followers are represented by lision avoidance operation is performed using the improved
circles of other colors. artificial potential field method, and the collision avoidance is
The preset obstacles are represented by a cross in Fig. 20. completed in 3.2 min. After the collision avoidance operation
The coordinates of the obstacle points are (0.4, 0.8), (−0.5, is complete, the trajectory points are retracked. The same
1) and (3.1, 3.4). The collision radius of the obstacles is 50 m, follower-2 senses an obstacle at 4.6 min and completes the
and the influence distance is 200 m. collision avoidance operation at 5.2 min.
The leader sails following the preplanned path, as stipu- Follower-4 appears to be in the same straight line as the
lated in this experiment, to move toward the target point (5, 5). obstacle and the path tracking point in the subsequent sailing

VOLUME 11, 2023 21665

 M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

TABLE 3. Ship’s initial information.

FIGURE 23. Error analysis of follower-3.

FIGURE 26. Initial position of ship and obstacle.

FIGURE 24. Error analysis of follower-2.

FIGURE 27. Trajectory of formation motion.

FIGURE 25. Improved artificial potential field method for collision 2) ALGORITHM SIMULATION OF THE IMPROVED ARTIFICIAL
avoidance. POTENTIAL FIELD METHOD IN MULTI-OBSTACLE
ENVIRONMENT
process. At about 12 min, the force balances and falls into the The initial information of the leader and each follower is
position of the local minimum point. At this time, a virtual listed in Table 3. As shown in Fig. 26, the leader is rep-
obstacle is introduced in the left direction of the object (the resented by a black circle, and followers are represented by
red cross in Fig. 25). This virtual obstacle will provide a force circles of other colors.
perpendicular to the direction of travel, so that the resultant The preset obstacles are represented by a cross in Fig. 26.
force of the follower is not zero, and the added virtual obstacle The leader sails following the preplanned path, as stip-
prevents the ship from reentering the local minimum point. ulated in this experiment, to move toward the target point
Finally, at about 14 min, all followers reach the preset (9, 9). The rest of the followers follow the leader in a preset
position and maintain a stable structure with the leader after formation (the two wings are at 60◦ , and the distance between
that. the two adjacent ships on the same side is 0.5 km).

21666 VOLUME 11, 2023

M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

[2] J. Tang, G. Liu, and Q. Pan, ‘‘A review on representative swarm intelligence
algorithms for solving optimization problems: Applications and trends,’’
IEEE/CAA J. Autom. Sinica, vol. 8, no. 10, pp. 1627–1643, Oct. 2021, doi:
10.1109/JAS.2021.1004129.
[3] C. Pan, Z. Zhang, W. Sun, J. Shi, and H. Wang, ‘‘Development of ship
weather routing system with higher accuracy using SPSS and an improved
genetic algorithm,’’ J. Mar. Sci. Technol., vol. 26, no. 4, pp. 1324–1339,
Feb. 2021, doi: 10.1007/s00773-021-00807-z.
[4] P. Zhou, Z. Zhou, Y. Wang, and H. Wang, ‘‘Ship weather rout-
ing based on hybrid genetic algorithm under complicated sea condi-
tions,’’ J. Ocean Univ. China, vol. 22, no. 1, pp. 28–42, Feb. 2023,
doi: 10.1007/s11802-023-5002-1.
[5] W. Zhao, Y. Wang, Z. Zhang, and H. Wang, ‘‘Multicriteria ship route
planning method based on improved particle swarm optimization–genetic
algorithm,’’ J. Mar. Sci. Eng., vol. 9, no. 4, p. 357, Mar. 2021, doi:
10.3390/jmse9040357.
[6] W. Zhao, H. Wang, J. Geng, W. Hu, Z. Zhang, and G. Zhang, ‘‘Multi-
objective weather routing algorithm for ships based on hybrid particle
swarm optimization,’’ J. Ocean Univ. China, vol. 21, no. 1, pp. 28–38,
Feb. 2022, doi: 10.1007/s11802-022-4709-8.
FIGURE 28. The position error curve between the leader and each [7] J. Li, H. Wang, W. Zhao, and Y. Xue, ‘‘Ship’s trajectory planning based
follower.
on improved multiobjective algorithm for collision avoidance,’’ J. Adv.
Transp., vol. 2019, pp. 1–12, Apr. 2019, doi: 10.1155/2019/4068783.
[8] J. Li, H. Wang, Z. Guan, and C. Pan, ‘‘Distributed multi-objective algo-
As shown in Figures 27-28, the follower enters the multi- rithm for preventing multi-ship collisions at sea,’’ J. Navigat., vol. 73, no. 5,
obstacle area at about 7.5min and enters the collision avoid- pp. 971–990, Sep. 2020, doi: 10.1017/S0373463320000053.
ance state. Finally regains the stability of the formation at [9] G. Zhang, Y. Wang, J. Liu, W. Cai, and H. Wang, ‘‘Collision-avoidance
decision system for inland ships based on velocity obstacle algo-
about 17min. rithms,’’ J. Mar. Sci. Eng., vol. 10, no. 6, p. 814, Jun. 2022, doi:
10.3390/jmse10060814.
VI. CONCLUSION [10] Y. Wang, Y. Zhang, H. Zhao, and H. Wang, ‘‘Assessment method based
on AIS data combining the velocity obstacle method and Pareto selection
The FMS and leader–follower method are combined in this for the collision risk of inland ships,’’ J. Mar. Sci. Eng., vol. 10, no. 11,
study. The FMS is used to plot a collision-free and smooth p. 1723, Nov. 2022, doi: 10.3390/jmse10111723.
course for the ship’s sailing reference, after which the con- [11] M. Saad, A. I. Salameh, S. Abdallah, A. El-Moursy, and C.-T. Cheng,
‘‘A composite metric routing approach for energy-efficient shortest path
sistency theory is used to make the follower track the leader planning on natural terrains,’’ Appl. Sci., vol. 11, no. 15, p. 6939, Jul. 2021,
and to realize the effect of formation keeping. In addition, doi: 10.3390/app11156939.
an improved artificial potential field method is proposed to [12] W. Zhang, D. Xia, G. Chang, Y. Hu, Y. Huo, F. Feng, Y. Li, and H. Li,
‘‘APFD: An effective approach to taxi route recommendation with mobile
solve the possible obstacles encountered by the follower dur- trajectory big data,’’ Frontiers Inf. Technol. Electron. Eng., vol. 23, no. 10,
ing navigation, enabling the follower to take collision avoid- pp. 1494–1510, Sep. 2022, doi: 10.1631/FITEE.2100530.
ance behavior independently. This method can effectively [13] H. Niu, Z. Ji, A. Savvaris, and A. Tsourdos, ‘‘Energy efficient path
planning for unmanned surface vehicle in spatially-temporally variant
prevent the ship from entering the local minimum point posi- environment,’’ Ocean Eng., vol. 196, Jan. 2020, Art. no. 106766, doi:
tion. Meanwhile, after the independent collision avoidance 10.1016/j.oceaneng.2019.106766.
is completed, the consistency theory is reapplied to return [14] X. Chen, N. Bose, M. Brito, F. Khan, G. Millar, C. Bulger, and T. Zou,
the ship to the team. The method and algorithm presented in ‘‘Risk-based path planning for autonomous underwater vehicles in an oil
spill environment,’’ Ocean Eng., vol. 266, Dec. 2022, Art. no. 113077, doi:
this paper are verified through computer simulation, and the 10.1016/j.oceaneng.2022.113077.
results demonstrate the method’s feasibility. [15] X. Tong, S. Yu, G. Liu, X. Niu, C. Xia, J. Chen, Z. Yang, and
This study proposes the following ideas for the next stage Y. Sun, ‘‘A hybrid formation path planning based on A* and multi-target
improved artificial potential field algorithm in the 2D random environ-
of research: When the FMS method is used to perform the ments,’’ Adv. Eng. Informat., vol. 54, Oct. 2022, Art. no. 101755, doi:
final step of gradient descent, the selected path is mainly 10.1016/j.aei.2022.101755.
considered to travel in the highest area of the potential [16] L. Wang, Z. Zhang, Q. Zhu, and S. Ma, ‘‘Ship route planning based
on double-cycling genetic algorithm considering ship maneuverabil-
field diagram obtained through FMM operation, ensuring the ity constraint,’’ IEEE Access, vol. 8, pp. 190746–190759, 2020, doi:
path’s safety. However, in some cases, selecting this area is 10.1109/ACCESS.2020.3031739.
unnecessary, as the navigation’s safety may be satisfied in the [17] W. Lu and M. Wu, ‘‘Optimization of vehicle automatic navigation
path based on remote sensing and GIS,’’ Optik, vol. 270, Nov. 2022,
area above a certain value, suggesting that the selection of a Art. no. 169952, doi: 10.1016/j.ijleo.2022.169952.
slightly lower area can not only ensure safety but also reduce [18] H.-B. Wang, X.-G. Li, P.-F. Li, E. I. Veremey, and M. V. Sotnikova,
the distance sailed. Future research will modify the algorithm ‘‘Application of real-coded genetic algorithm in ship weather rout-
presented in this study to set a fitness function that balances ing,’’ J. Navigat., vol. 71, no. 4, pp. 989–1010, Jul. 2018, doi:
10.1017/S0373463318000048.
the speed and safety of the planned path using FMS. [19] E. Krell, S. A. King, and L. R. Garcia Carrillo, ‘‘Autonomous surface
vehicle energy-efficient and reward-based path planning using particle
swarm optimization and visibility graphs,’’ Appl. Ocean Res., vol. 122,
REFERENCES May 2022, Art. no. 103125, doi: 10.1016/j.apor.2022.103125.
[1] J. Tang, X. Chen, X. Zhu, and F. Zhu, ‘‘Dynamic reallocation model of [20] J. Zhong, B. Li, S. Li, F. Yang, P. Li, and Y. Cui, ‘‘Particle swarm opti-
multiple unmanned aerial vehicle tasks in emergent adjustment scenarios,’’ mization with orientation angle-based grouping for practical unmanned
IEEE Trans. Aerosp. Electron. Syst., early access, Aug. 1, 2022, doi: surface vehicle path planning,’’ Appl. Ocean Res., vol. 111, Jun. 2021,
10.1109/TAES.2022.3195478. Art. no. 102658, doi: 10.1016/j.apor.2021.102658.
VOLUME 11, 2023 21667
 M. Li et al.: Ship Formation Algorithm Based on the Leader–Follower Method

[21] A. Serani, G. Fasano, G. Liuzzi, S. Lucidi, U. Iemma, E. F. [41] G. Sun, R. Zhou, B. Di, and H. Yu, ‘‘A physicochemically inspired
Campana, F. Stern, and M. Diez, ‘‘Ship hydrodynamic optimization approach to flocking control of multiagent system,’’ Nonlinear Dyn.,
by local hybridization of deterministic derivative-free global algo- vol. 102, pp. 2627–2648, Nov. 2020, doi: 10.1007/s11071-020-06062-y.
rithms,’’ Appl. Ocean Res., vol. 59, pp. 115–128, Sep. 2016, doi: [42] M. Fernando, ‘‘Online flocking control of UAVs with mean-field approx-
10.1016/j.apor.2016.04.006. imation,’’ in Proc. IEEE Int. Conf. Robot. Autom. (ICRA), May 2021,
[22] G. Zhang, H. Wang, W. Zhao, Z. Guan, and P. Li, ‘‘Application of improved pp. 8977–8983, doi: 10.1109/ICRA48506.2021.9560899.
multi-objective ant colony optimization algorithm in ship weather rout- [43] L. Ding and G. Guo, ‘‘Formation control for ship fleet based on backstep-
ing,’’ J. Ocean Univ. China, vol. 20, no. 1, pp. 45–55, Jan. 2021, doi: ping,’’ Control Decis., vol. 27, no. 2, pp. 299–303, 2012.
10.1007/s11802-021-4436-6. [44] B. Liu, T. Chu, L. Wang, and G. Xie, ‘‘Controllability of a leader–follower
[23] E. W. Dijkstra, ‘‘A note on two problems in connexion with dynamic network with switching topology,’’ IEEE Trans. Autom. Control,
graphs,’’ Numer. Math., vol. 1, no. 1, pp. 269–271, Dec. 1959, doi: vol. 53, no. 4, pp. 1009–1013, May 2008, doi: 10.1109/TAC.2008.919548.
10.1007/BF01386390. [45] S. A. Ajwad, T. Ménard, E. Moulay, M. Defoort, and P. Coirault, ‘‘Observer
[24] P. E. Hart, N. J. Nilsson, and B. Raphael, ‘‘Correction to a formal basis for based leader-following consensus of second-order multi-agent systems
the heuristic determination of minimum cost paths,’’ ACM SIGART Bull., with nonuniform sampled position data,’’ J. Franklin Inst., vol. 356, no. 16,
no. 37, pp. 28–29, Dec. 1972, doi: 10.1145/1056777.1056779. pp. 10031–10057, Nov. 2019, doi: 10.1016/j.jfranklin.2019.09.025.
[25] A. S. James, Level Set Methods and Fast Marching Methods. New York, [46] Q. Wang, K. Liu, X. Wang, L. Wu, and J. Lv, ‘‘Leader-following consensus
NY, USA: Cambridge Univ. Press, 1996. of multi-agent systems under antagonistic networks,’’ Neurocomputing,
[26] Z. Guan, Y. Wang, Z. Zhou, and H. Wang, ‘‘Research on early warning vol. 413, pp. 339–347, Nov. 2020, doi: 10.1016/j.neucom.2020.07.006.
of ship danger based on composition fuzzy inference,’’ J. Mar. Sci. Eng., [47] O. Khatib, ‘‘Real-time obstacle avoidance for manipulators and mobile
vol. 8, no. 12, p. 1002, Dec. 2020, doi: 10.3390/jmse8121002. robots,’’ Int. J. Robot. Res., vol. 5, no. 1, pp. 90–98, Mar. 1986, doi:
[27] M. Shang, M. Chu, and M. Grethler, ‘‘Path planning based on 10.1177/027836498600500106.
artificial potential field and fuzzy control,’’ in Proc. IEEE Int.
Conf. Power Electron., Comput. Appl., Mar. 2021, pp. 304–308, doi:
10.1109/ICPECA51329.2021.9362519.
[28] X. Yang, ‘‘Development of obstacle avoidance for autonomous vehicles MINGYANG LI was born in Siping, China,
and an optimization scheme for the artificial potential field method,’’ in in 1998. He received the bachelor’s degree in elec-
Proc. 2nd Int. Conf. Comput. Data Sci. (CDS), Jan. 2021, pp. 12–18, doi: tronic information science and technology from
10.1109/CDS52072.2021.00010. Jilin University, Jilin, in 2020, where he is cur-
[29] Z. Xin, X. Rongwu, and C. Guo, ‘‘AUV path planning in dynamic environ- rently pursuing the M.D. degree in circuits and
ment based on improved artificial potential field method based on visibility
systems. His research interest includes ship forma-
graph,’’ J. Phys., Conf., vol. 2383, no. 1, Dec. 2022, Art. no. 012090, doi:
tion algorithm.
10.1088/1742-6596/2383/1/012090.
[30] L. Liu, B. Wang, and H. Xu, ‘‘Research on path-planning algo-
rithm integrating optimization A-star algorithm and artificial potential
field method,’’ Electronics, vol. 11, no. 22, p. 3660, Nov. 2022, doi:
10.3390/electronics11223660.
[31] C. Zheyi and X. Bing, ‘‘AGV path planning based on improved
artificial potential field method,’’ in Proc. IEEE Int. Conf. Power KAI MENG was born in Xuzhou, China, in 1998.
Electron., Comput. Appl. (ICPECA), Jan. 2021, pp. 32–37, doi: He received the bachelor’s degree in electronic and
10.1109/ICPECA51329.2021.9362519. information engineering from Yangzhou Univer-
[32] J. Wang, R. Wang, D. Lu, H. Zhou, and T. Tao, ‘‘USV dynamic sity, Jiangsu, in 2021. He is currently pursuing
accurate obstacle avoidance based on improved velocity obstacle the M.D. degree in circuits and systems with Jilin
method,’’ Electronics, vol. 11, no. 17, p. 2720, Aug. 2022, University. His research interests include localiza-
doi: 10.3390/electronics11172720. tion and navigation.
[33] T. Xu, S. Zhang, Z. Jiang, and Z. Liu, ‘‘Collision avoidance of high-
speed obstacles for mobile robots via maximum-speed aware velocity
obstacle method,’’ IEEE Access, vol. 8, pp. 138493–138507, 2020, doi:
10.1109/ACCESS.2020.3012513.
[34] R. Skjetne, S. Moi, and T. I. Fossen, ‘‘Nonlinear formation control of
marine craft,’’ in Proc. 41st IEEE Conf. Decis. Control., Dec. 2002,
pp. 1699–1704, doi: 10.1109/CDC.2002.1184765. JIELU CHEN was born in Nantong, China,
[35] I.-A.-F. Ihle, M. Arcak, and T. I. Fossen, ‘‘Passivity-based designs for in 1988. She received the master’s degree in
synchronized path-following,’’ Automatica, vol. 43, no. 9, pp. 1508–1518, computer science and engineering from Shanghai
Sep. 2007, doi: 10.1016/j.automatica.2007.02.018. Jiao Tong University, in 2020. Her research
[36] R. Ghabcheloo, A. P. Aguiar, A. Pascoal, C. Silvestre, I. Kaminer, and interest includes navigation automation based on
J. Hespanha, ‘‘Coordinated path-following control of multiple underac- auto-pilot.
tuated autonomous vehicles in the presence of communication failures,’’
in Proc. 45th IEEE Conf. Decis. Control, Dec. 2006, pp. 4345–4350, doi:
10.1109/CDC.2006.376989.
[37] M. Breivik, V. E. Hovstein, and T. I. Fossen, ‘‘Straight-line target tracking
for unmanned surface vehicles,’’ Model., Identificat. Control, Norwegian
Res. Bull., vol. 29, no. 4, pp. 131–149, 2008, doi: 10.4173/MIC.2008.4.2.
[38] R. Cui, S. Sam Ge, B. Voon Ee How, and Y. Sang Choo, ‘‘Leader–
follower formation control of underactuated autonomous underwater vehi- HONGBO WANG (Member, IEEE) was born
cles,’’ Ocean Eng., vol. 37, nos. 17–18, pp. 1491–1502, Dec. 2010, doi: in Changchun, China, in 1969. She received the
10.1016/j.oceaneng.2010.07.006. Ph.D. degree from Saint Petersburg State Uni-
[39] I.-A.-F. Ihle, J. Jouffroy, and T. I. Fossen, ‘‘Formation control of marine versity, Russia. She is currently with the Col-
surface craft: A Lagrangian approach,’’ IEEE J. Ocean. Eng., vol. 31, no. 4, lege of Electronic Science and Engineering, Jilin
pp. 922–934, Oct. 2006, doi: 10.1109/JOE.2006.880426. University. Her research interests include ship
[40] F. Huang, P. Wu, and X. Li, ‘‘Distributed flocking control of quad-rotor motion control, weather routing, and ship collision
UAVs with obstacle avoidance under the parallel-triggered scheme,’’ Int. avoidance.
J. Control, Autom. Syst., vol. 19, no. 3, pp. 1375–1383, Jan. 2021, doi:
10.1007/s12555-019-0315-y.

21668 VOLUME 11, 2023