Path planning method for USVs based on improved DWA and COLREGs

2.1. Kinematic model of USVs

Figure 1 presents a two-dimensional motion model for local path planning of USVs. For the responsive model, the heading is usually defined as the angle between the USV's heading (longitudinal axis direction) and the Earth's true north direction, with a value range of 0° 360°, 0° when it coincides with the true north direction, and increases in a clockwise direction. In the navigation coordinate system a, north (N) is the x-axis and east (E) is the y-axis. P is the position of the USV in the navigation coordinate system, represented by the coordinates of its center of gravity, and is also the origin of the hull coordinate system b. The hull coordinate system b takes the heading of the USV as the x-axis and the starboard direction of the USV as the y-axis. The motion of USVs is described using six degrees of freedom: surge, sway, yaw, heave, roll, and pitch ^[14]. To reduce the complexity of motion control for USVs, it is often sufficient to focus on the movements within the horizontal plane, which includes only the degrees of freedom for surge, sway, and yaw. This paper primarily investigates the motion of USVs on a two-dimensional horizontal plane, disregarding the forces of disturbance caused by environmental factors such as wind and ocean currents. Consequently, the relationships for the transformation of velocity vectors between the navigation coordinate system a and the body coordinate system b, as well as the USV's three-degree-of-freedom control model, are given as ^[15–18]:

Figure 1. Two-dimensional motion diagram for local path planning of USV. USV: Unmanned surface vehicle.

where $$ \begin{bmatrix}x&y&\psi\end{bmatrix} $$ represents the position and yaw angle of the USV in the navigation coordinate system a, and $$ \begin{bmatrix}u&v&r\end{bmatrix} $$ represents the surge velocity, sway velocity, and yaw rate in the body coordinate system b. As shown in Figure 1, assume the USV is moving forward, that is, the surge speed $$ u>0 $$. The direction of movement $$ \chi $$ is not equal to the heading angle $$ \psi $$, and the sailing speed is given by $$ \nu_t=\sqrt{u^2+\nu^2} $$ it is not merely the surge speed $$ u $$.

The dynamic model of the USV is established as follows^[19]:

where $$ m_{ii}, i = 1, 2, 3 $$ are positive constants that stand for ship inertia masses. $$ \tau_u $$ and $$ \tau_r $$ are control inputs. $$ \tau_{ud} $$, $$ \tau_{vd} $$ and $$ \tau_{rd} $$ stand for the time-varying unknown external disturbances, which are induced by ocean currents, winds and waves. The nonlinear hydrodynamic damping $$ h_i, i =u, v, r $$ are expressed as follows:

Here, $$ X(\cdot) $$, $$ Y(\cdot) $$ and $$ N(\cdot) $$ stand for the linear and quadratic hydrodynamic coefficients in the surge, sway and yaw motions, respectively.

Remark 1 It should be pointed out that there are only two control inputs $$ \tau_u $$ and $$ \tau_r $$ in model dynamics, while the control force in the sway channel is not provided. Therefore, the USV denoted by Equations (1) and (2) is underactuated. In trajectory tracking control issues of USVs, the difficulty lies in how to eliminate the cross-tracking error with uncontrollable sway dynamics. Compared to the fully actuated USV, the control problem for underactuated ones will be more challenging.

Remark 2 In engineering applications, the USV platform is often designed to be underactuated since the installation of the lateral thruster is not only costly, but also impairs hull hydrodynamic performance in high-speed scenarios. In general, most surface vessels are only equipped with propellers and rudders for surge and yaw motion. This means the contribution of this work will be meaningful for practice.

Remark 3 The vessels' hydrodynamic damping $$ h_i, i = u, v,r $$ are composed of serval hydrodynamic terms involving potential damping, wave drift damping, skin friction, and so on ^[20]. Thus, it poses many challenges to accurately acquire this kind of information in practice. In this paper, $$ h_i, i = u, v,r $$ are supposed to be unknown for controller designers.

2.2. 2-D Ocean current environment model

Due to the rapid rotation of the Earth, the horizontal scale of ocean currents is significantly larger than the vertical scale of movement. Therefore, ocean currents can be considered as motion across multiple layers of two-dimensional horizontal planes. In this paper, the numerical equations for the ocean current model are represented by the superposition of two viscous Lamb vortices. The mathematical description is given as^[21]:

where $$ \vec{r}_0 $$ represents the position vector of the vortex center, while $$ \delta $$ and $$ \omega $$ denote the vortex radius and vortex intensity, respectively. $$ V_{x}(\vec{r}) $$ and $$ V_{y}(\vec{r}) $$ are the horizontal and vertical components of the vortex velocity, respectively. Equation (4) represents the physical model of the velocity field surrounding the USV. This study only considers the spatial variation of ocean currents, allowing us to assume that the ocean current motion in the corresponding region remains constant throughout the mission. In the paper, the current environmental parameters are set as $$ \delta = 1 $$, $$ \Gamma= 10 $$, and the vortex center vector $$ r_0 $$ is chosen from two points on the current plane. Thus, under the influence of ocean currents, the kinematic model of the USV can be given as:

2.3. Constraints of COLREGs

The COLREGs are traffic rules established by the International Maritime Organization to prevent collisions between vessels at sea. COLREGs delineate three types of encounters: head-on, cross, and overtaking. Specifically, the cross encounter includes two scenarios: crossing from the left (cross-left) and crossing from the right (cross-right). The regional division method adopted by COLREGs in this paper is illustrated in Figure 2. Article 8 of the COLREGs outlines the fundamental principles for collision avoidance during the design of local path planning, which should adhere to the following principles: Actions should be taken as early as possible; substantial evasive actions should be taken to ensure they are observable by visual or radar means and to ensure passing at a safe distance; when sufficient space is available, steering alone should be used, as it is the most effective means of avoiding a close-quarters situation and during evasion, avoiding a series of minor changes to speed or direction; continuous monitoring of the effectiveness of the collision avoidance actions should be maintained until the other vessel has past and is clear ^[22–24].

Figure 2. Regional division method of COLREGs. COLREGs: International Regulations for Preventing Collisions at Sea.

Articles 13, 14, and 15 of the COLREGs specify the judgment methods and action rules for vessels in overtaking, head-on, and cross-encounter situations ^{[25, 26]}, classifying the USV either as a give-way or stand-on vessel depending on the situation. Detailed action rules for the USV are as follows:

(1) Overtaking: When a USV is navigating behind a moving vessel within an angle range of [112.5°, 247.5°), and the USV is faster, it should take action to avoid the vessel by turning left or right.

(2) Head-on: A head-on encounter refers to two vessels sailing in opposite or nearly opposite directions, with the moving vessel within the bearing range of [0°, 10°) or [355°, 360°) from the USV. The USV should turn right, the oncoming vessel should turn left, and the USV should pass as far away as possible from the left side of the moving vessel.

(3) Cross-left: This encounter refers to a moving vessel being within the bearing range of [247.5°, 350°) from the USV. The USV is not the give-way vessel, so it should maintain its course.

(4) Cross-right: This type refers to a moving vessel within the bearing range of [10°, 112.5°) from the USV. The USV should turn right and navigate behind the moving vessel as much as possible, avoiding crossing in front of the other vessel.

The correct evasive routes for these four encounter scenarios are shown as dashed lines in Figure 3. The local path planning algorithm should also comply with these action rules.

Figure 3. Avoidance methods for the USV. USV: Unmanned surface vehicle.

3.1. Velocity search space

The velocity search space is a two-dimensional space of speeds derived from the constraints specific to the USV and the operational environment. The purpose of this search space is to identify the optimal movement strategy for the USV to avoid obstacles, achieve path planning, and complete tasks within the current environment. The limiting factors can be categorized into three types: intrinsic speed limits of the USV, acceleration limits, and obstacle constraints ^{[30, 31]}.

Intrinsic speed limits of the USV refer to the physically feasible range of speeds for the USV, including limitations on linear and angular velocities. These limits are determined by factors such as the power structure and physical design of USVs. The intrinsic speed limits of the USV are given as

where $$ u $$ represents the linear speed (surge velocity) of the USV, and $$ r $$ represents the angular velocity (yaw rate), $$ u_{min}=0 $$, $$ r_{min}=-r_{max} $$. The acceleration limits of a USV are defined to ensure that the motion paths selected within the dynamic window are physically feasible and stable. These limits are primarily influenced by the power and torque of the USV's drive motors, imposing a maximum acceleration limit ^[32]. Since the USV operates as a nonlinear system, its acceleration is dependent on its current state of motion, and thus, traditional computation methods are not applicable. The relationships between $$ u_{max} $$, $$ r_{max} $$, $$ u $$ and $$ r $$ are given as

where $$ l_r $$ represents the longitudinal distance from the center of gravity of the USV to the thruster, $$ X_u $$, $$ X_{|u|u} $$, $$ X_{uuu} $$, $$ N_r $$, $$ N_{|r|r} $$ and $$ N_{rrr} $$ denote hydrodynamic coefficient, $$ m $$ stands for the mass of the USV, $$ F_x $$ means the longitudinal force produced by the thruster, Fy points to the lateral force produced by the thruster, and $$ I_z $$ indicates the moment of inertia about the z-axis.

Based on the current motion state of the USV and its acceleration, the range of velocity values $$ v_d $$ can be determined by

The schematic diagram of velocity values is given as Figure 5.

Figure 5. The schematic diagram of velocity values.

The process of velocity search and generation of potential trajectory sets is one of the most critical processes in DWA. Only after determining the set of velocities can trajectory sets be generated and evaluated using an evaluation function, with the optimal trajectory ultimately being selected. In this algorithm, FISA is employed for velocity search within the velocity search space. Compared to traditional search methods, the use of FISA has refined the velocity search mechanism of DWA. The improved algorithm significantly reduces the likelihood of DWA getting trapped in local optima, thus alleviating issues where the USV is unable to navigate out of complex environments with multiple obstacles.

The FISA is an intelligent optimization algorithm that operates on the principle of random interactions between the best and worst solutions obtained during the optimization process, as well as among candidate solutions. It requires only common control parameters such as population size and the number of iterations, without needing any algorithm-specific control parameters. FISA relies on a difference vector obtained by subtracting the position of the worst individual from the best individual in the current iteration. This ensures that the set of solutions always progresses towards better outcomes. Compared to traditional optimization algorithms, FISA improves the optimization of the shift function while maintaining the simplicity and lack of control parameters characteristic of traditional algorithms, pushing the dataset towards superior solutions ^[33]. The flowchart of FISA is shown in the Figure 6, and the process is given as

Figure 6. Flowchart of FISA. FISA:

Where $$ X_i^{Iter} $$ represents the position of the $$ i-th $$ solution in the current iteration Iter; $$ j $$ (ranging from 1 to D) denotes the $$ j-th $$ dimension of each solution; $$ X_{best}^{Iter} $$ and $$ X_{worst}^{Iter} $$ sequentially represent the positions of the members with the highest and lowest performance in the population during the current iteration; $$ r_1 $$ and $$ r_2 $$ are two randomly selected values between $$ 0 $$ and $$ 1 $$ for each of the D dimensions; $$ X_{k}^{Iter} $$ represents the position of the $$ k-th $$ solution chosen indiscriminately; and $$ f(\cdot) $$ denotes the numerical output of the function currently being optimized for the corresponding solution, which includes cost computation. Using Equation (10), the position of the $$ i-th $$ solution for the next iteration is calculated.

In fact, in FISA, each member moves away from the average position of individuals with poorer fitness within the population and closer to the average position of individuals with better fitness than the respective member. Then, Equation (10) is used to update the position of each member. In Equation (9), the values of $$ MX_\text{best}^\text{Iter} $$ and $$ MX_\text{worst}^\text{Iter} $$ in each iteration are given as

where $$ B_i $$ and $$ W_i $$ represent the subsets of population members whose fitness values are respectively better and worse than that of the $$ i-th $$ member in iteration $$ Iter $$. The function $$ length(\cdot ) $$ indicates the number of individuals in each subset.

In the application of this integrated algorithm, the current motion state's linear and angular velocity limits of the USV are input into FISA. By continuously iterating the values of $$ MX_{best}^{Iter} $$ and $$ MX_{worst}^{Iter} $$ simultaneously calculating costs, the set of velocities $$ X_{best}^{Iter} $$ that minimizes cost is selected. The DWA then selects the optimal trajectory from this set of velocities using the evaluation function, which becomes the final operational trajectory for the USV.

3.2. Obstacle constraints

Local planning necessitates dynamic and real-time obstacle avoidance capabilities. For the safety of USVs, it is essential to ensure that the vessel does not collide with any obstacles. The USV must maintain a sufficient distance at any given moment to allow for stopping under maximum deceleration ^[34]. The constraint that the USV does not collide with surrounding obstacles at any moment is given as

where $$ \varphi(u,r) $$ is the distance evaluation function. If there are no obstacles around the USV or if the nearest distance to any obstacle is greater than a set threshold, then the value of distance evaluation function is set to a large constant. When the distance to an obstacle is less than the set threshold, the distance evaluation function is defined as the closest distance from the simulated trajectory at the current speed to the obstacle.

Traditional DWA uses the above methods to restrict velocities, preventing collisions between mobile robots and obstacles. However, when applying DWA for obstacle avoidance in USVs, the USV itself is also considered an obstacle, necessitating improvements to obstacle constraints. The COLREGs prescribe the division of responsibilities and avoidance strategies when two vessels meet, and based on COLREGs, priorities among multiple USVs are assigned and specific avoidance strategies are designated ^[35].

When assessing the motion states of other USVs, given that USVs are typically smaller and often operate in relatively open and calm waters, this paper determines the priorities and establishes avoidance strategies based on the distance and coordinate relationships between USVs. Additionally, when multiple USVs are operating together, the relationships between them are continuously changing. Therefore, it is crucial to determine whether a previous avoidance action has ended in order to improve the smoothness of the travel trajectory.

In response to multiple USVs encountering each other, this paper draws from the COLREGs to establish the following strategies for division of responsibilities when two vessels meet. The specific rules are as follows:

(1) Head-On: The vessel on the starboard (right) side generally has the right of way, while the vessel on the port (left) side should take evasive action. Head-on encounters are divided into two scenarios: USVa traveling from south to north and USVb traveling from north to south: In this case, USVa has the right of way and USVb should yield. USVa traveling from west to east and USVb traveling from east to west: Here, USVa has the right of way and USVb should yield; (2) Overtaking: If USVa is faster than USVb and is behind it with a relative angle of -10° to 10°, USVa needs to overtake USVb from either the left or right side; (3) Crossing Situations follow these rules: Cross-right: If USVa is traveling from south to north and USVb from west to east, USVa has the right of way and USVb should yield. Cross-left: If USVa is traveling from south to north and USVb from east to west, USVb has the right of way and USVa should yield.

When two USVs reach a judgment distance within a predictive range and one USV has a higher priority, its navigation is not affected by the other USV, and the traditional approach can still be used for trajectory selection using the evaluation function. The USV with lower priority needs to take evasive action, thus requiring optimization of the distance evaluation function, as given in

where $$ d_{ij} $$ represents the distance between the USV and another USV, and $$ R_d $$ is the judgment distance. When the USV has a higher priority, $$ k=1 $$; when it has a lower priority, $$ k=2 $$.

If the minimum value of $$ d_{ij} $$ is less than the obstacle avoidance distance of the USV, the lower-priority USV should stop and wait for the higher-priority USV to pass before resuming its actions.

3.3. Evaluation function

After determining the velocity range constrained by the USV's capabilities, some simulated velocity trajectories may be feasible, while others may not meet the required standards. Therefore, it is necessary to evaluate and select the best trajectories from the multiple sets of sampled trajectories. By evaluating the trajectories using a standard evaluation function and comparing the scores, the optimal trajectory is selected to determine the USV's speed. The trajectory evaluation function is given as

where $$ \mu(u,r) $$ is the heading angle evaluation function, which is used to assess the error between the direction of the trajectory's endpoint at the current sampled speed and the line connecting to the target point, denoted as $$ \Delta\theta $$. Since a larger value of the evaluation function typically indicates a better result, $$ \pi-\Delta\theta $$ is used in the evaluation, as expressed by $$ \mu(u,r)=\pi-\Delta\theta:\delta(u,r) $$. $$ \delta(u,r) $$ is the speed evaluation function, where a larger value indicates a faster speed along the planned trajectory. This function is represented by the magnitude of the current linear speed. However, since the USV operates on the water surface and has a slower speed response, the planned speed needs to gradually change while ensuring obstacle avoidance. Thus, it is expressed as $$ \delta(u,r)=||u|-\Delta u| $$. $$ \alpha\beta\gamma $$ represents the coefficient of the evaluation function. Since local path planning requires data collection from multiple sensors, and continuous information gathering is not always possible, this can result in significant differences in the evaluation outcomes. Therefore, normalization (smoothing) is necessary ^[36], where $$ \sigma $$ represents the normalization process. The normalization procedure is given as

where $$ i $$ represents the $$ i-th $$ simulated trajectory, and $$ n $$ represents the total number of sampled trajectories under the given constraints. Using the above formula, a path that avoids obstacles and moves quickly toward the target point can be obtained, enabling the USV to achieve optimal local path planning.

After sampling a set of $$ (u,r) $$, trajectory simulation is conducted based on the USV's kinematic model. Once all trajectories are generated, the trajectory with the best evaluation in the set is selected. After choosing the optimal trajectory from the set, the DWA algorithm is used to continue predicting the trajectory at the next time step until the final target point is reached.

4.1. Case1: static obstacles

In this case, the planned map is a rectangular area of 4 m by 14 m, with multiple static obstacles of 0.3 m radius added to simulate small coral reefs and buoys that a USV might encounter at sea. The environment model also includes the influence of ocean currents on the USV. In this simulation, there is only one USV, with an initial pose of [0, 0, 90] and a target point of [0, 10]. To facilitate comparison of the improved algorithm with the traditional DWA, the same experimental environment was set up and the results were compared. The simulated trajectory indicates that the improved algorithm can avoid obstacles and reach the target point, whereas the traditional DWA falls into a local optimum during obstacle avoidance, colliding with obstacles and failing to reach the target. Furthermore, the improved algorithm generates a shorter and smoother path compared to the traditional DWA. The simulation time is 21.67 s, and the simulated trajectory is shown in Figure 7. Since the traditional DWA algorithm failed to reach the target point in the previous simulation, the positions of the simulated obstacles were changed for another experiment. In this case, the traditional DWA algorithm was able to avoid obstacles and reach the target point, with the simulated trajectory shown in Figure 8. Comparing the trajectories, it can be seen that the path of the traditional DWA algorithm is longer, while the improved algorithm generates a shorter and smoother path, making it easier for real-world trajectory tracking. A comparison of the USV running times during the simulations of the improved and traditional algorithms is shown in Figure 9. Based on 20 simulations, the average running time for the improved algorithm is around 20 s, whereas for the traditional DWA, it is around 35 s, indicating a reduction in running time. The comparison of running distances is shown in Figure 10. When using the improved algorithm, the average running distance for the USV is about 10 m, considering the straight-line distance between the target point and the initial position is 9.5 m ($$ d < 0.5 $$ m is considered reaching the target). The traditional DWA's average running distance is 11.3 m, demonstrating a significant improvement.

Figure 7. Simulated trajectories of the USV. USV: Unmanned surface vehicle.

Figure 8. Simulated trajectories of the USV. USV: Unmanned surface vehicle.

Figure 9. Average travel time of the USV. USV: Unmanned surface vehicle.

Figure 10. Average travel distance of the USV. USV: Unmanned surface vehicle.

4.2. Case2: dynamic obstacles

In this case, the planned area is a 140 m by 140 m square region. The map includes two static obstacles and five dynamic obstacles, simulating larger obstacles that the USV might encounter at sea, such as large islands and bridge piers of sea-crossing bridges. The coordinates of the static obstacles are [40, 40] and [60, 60], while the dynamic obstacles start from random positions within the map area and move in random directions. The influence of ocean currents on the USV movement is also included in the environment model. This simulation involves four USVs: USV1, USV2, USV3 and USV4, with initial poses of [0, 0, 90], [100, 0, 90], [80, 100, -90], and [20, 100, -90], respectively, and target points of [100, 100], [0, 100], [20, 0], and [80, 0].

In the first step of the simulation, USV1's movement is normal, adhering to COLREGs and featuring dynamic obstacle avoidance with the improved algorithm. USV2, USV3 and USV4 exhibit abnormal movements, as they use only the traditional obstacle avoidance algorithm, with their simulated trajectories shown in Figure 11. Since only USV1 has the integrated obstacle avoidance functionality, it can reach the target point quickly and safely. At the beginning of the simulation, there is a collision risk between USV1 and USV4, but they use an improved algorithm and the motion rules comply with COLREGs. According to the rule, USV1 has higher priority and will pass in front of USV4. During the simulation, USV3 and USV4 face collision risks, having reached the minimum safe distance($$ d < 0.5 $$ m). A comparison of USV running times for the improved and traditional algorithms is shown in Figure 12. Based on 20 simulations, USV1's travel time is shorter than that of USV2. A comparison of travel distances is shown in Figure 13. After 20 simulations, USV1's travel distance is shorter than that of USV2. In the second step of the simulation, the experiment was conducted to verify that the improved algorithm allows the USV's motion trajectory to comply with COLREGs while maintaining dynamic obstacle avoidance capabilities. USV1, USV2, USV3, and USV4 all demonstrated normal movement, utilizing the improved algorithm's obstacle avoidance functionality. The motion trajectories are shown in Figure 14, indicating that all USVs were able to safely reach their target points while following obstacle avoidance rules. The USV running times are shown in Figure 15, and the comparison of running distances is illustrated in Figure 16. Since the start positions and target points of USV1 and USV2 have the same distance of 141.42 m, and those of USV3 and USV4 have the same distance of 128.06 m, the running distance is greater than the straight-line distance due to the need for obstacle avoidance maneuvers. Based on 20 simulation results, USV1 and USV2 have similar results, as do USV3 and USV4, verifying the stability of the improved algorithm.

Figure 11. Simulated trajectories of the USV. USV: Unmanned surface vehicle.

Figure 12. Average travel time of the USV. USV: Unmanned surface vehicle.

Figure 13. Average travel distance of the USV. USV: Unmanned surface vehicle.

Figure 14. Simulated trajectories of the USV. USV: Unmanned surface vehicle.

Figure 15. Average travel time of the USV. USV: Unmanned surface vehicle.

Figure 16. Average travel distance of the USV. USV: Unmanned surface vehicle.

In the third step of the simulation, the experiment aimed to verify that the improved algorithm also has dynamic obstacle avoidance capabilities in an irregular environment. In this scenario, USV1 and USV4 demonstrated normal operation, using the improved algorithm and having trajectories that comply with COLREGs. USV2 and USV3, however, exhibited abnormal movement, using only the traditional algorithm. The trajectories are shown in Figure 17. During the simulation, there is a collision risk between USV1 and USV4, but they use an improved algorithm and their motion rules comply with COLREGs. According to the rule, USV1 has higher priority and will pass in front of USV4. At the same time, USV1 and USV2 also have collision risks, and changing the direction of motion of USV1 according to the rules avoids collisions. USV2, after performing an obstacle avoidance maneuver, got trapped in a local optimum, failing to move towards the target point and ultimately hitting the map's edge, unable to reach the target point, while the other three USVs successfully reached their target points. Compared to USV2, USV1 was able to promptly adjust its direction after obstacle avoidance and quickly reach the target point without hitting the irregular map edges. The USV running times are illustrated in Figure 18, and the comparison of running distances is shown in Figure 19. Based on 20 simulation runs (with all USVs reaching their target points), USV1 and USV4 had shorter running times compared with USV2 and USV3, and their running speeds were faster. Throughout the experiment, USV1 and USV4 consistently reached their target points without any issues. The comparison shows a significant improvement in the performance of the improved algorithm over the traditional algorithm.

Figure 17. Simulated trajectories of the USV. USV: Unmanned surface vehicle.

Figure 18. Average travel time of the USV. USV: Unmanned surface vehicle.

Figure 19. Average travel distance of the USV. USV: Unmanned surface vehicle.

Based on the two types of simulation experiments, it can be concluded that the USV, using the improved algorithm, can safely and accurately perform obstacle avoidance maneuvers when dealing with both dense static obstacles and complex dynamic obstacles, ultimately navigating the USV to the target point. Additionally, when faced with irregular maps, the USV is able to adjust its course in a timely manner. Compared to the traditional algorithm, the improved algorithm shows significant improvements in both average running time and average running distance.

Authors' contributions

Made substantial contributions to the conception and design of the study and performed the analysis of the results: Liu S, Wang X

Carries out algorithm design and improvement and conducted theoretical analysis: Wu Y, Li Q, Yan J, Levin E

Availability of data and materials

Not applicable.

Financial support and sponsorship

This work was supported by the Shandong Provincial Natural Science Foundation (Grant No. ZR2024QF114) and the major innovation project for the science education industry integration pilot project of Qilu University of Technology (Shandong Academy of Sciences) (Grant No. 2023JBZ03).

Conflicts of interest

Wu Y is the Guest Editor Assistant of the Special Issue of "Intelligent Control and Decision-making of Unmanned Ships" of the journal Intelligence & Robotics, while the other authors have declared that they have no conflicts of interest.

Ethical approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Copyright

© The Author(s) 2024.