Dynamic anti-collision method and system for unmanned ship formation

Abstract

The invention relates to a dynamic anti-collision method and system for an unmanned ship formation. According to the method, each unmanned ship collects environment information in real time and periodically broadcasts the state of the unmanned ship; on the basis of the information, a nonlinear track prediction model is used for judging the nearest intersection time and distance between ships, and if any one is smaller than a threshold value, conflict is judged. The priorities of the involved ships are scored after conflict, high-score ships keep course speed and low-score ships avoid when double ships conflict, an avoiding sequence is generated according to topological sorting when multiple ships conflict, and the lowest-priority ship firstly avoids; and after the avoidance is completed, calculating a shortest regression path by each ship through a dynamic surface control technology, and recovering a track regression main team. Compared with the prior art, the method has the advantages of high practicability, high-resolution perception capability, formation communication capability, dynamic anti-collision judgment capability and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned ship control, and in particular to a dynamic collision avoidance method and system for an unmanned ship formation. Background Art

[0002] With the rapid development of unmanned systems, fleets of multiple unmanned vessels are being widely used for tasks such as water sightseeing, mapping, and security. In complex, non-navigable waters such as inland rivers and lakes, the vessels in the fleet often have variable headings and face a high level of parallel processing, placing higher demands on collision avoidance capabilities.

[0003] The existing technology has the following defects:

[0004] 1. Rigid control methods and lack of individual dynamic judgment capabilities: Currently, most low-speed unmanned ship formation control methods still rely on centralized path scheduling or fixed avoidance rules (such as main channel priority and right-side avoidance). It is difficult to make personalized responses based on the current mission importance, energy level and dynamic environmental status of each ship, resulting in slow response and low coordination efficiency.

[0005] 2. Sensing methods are costly and low-precision, making them unsuitable for low-speed formation deployment: Traditional solutions often rely on single sensors such as AIS systems or medium- and long-range maritime radars. These devices are expensive and have low resolution, making it difficult to meet the requirements for fine-grained obstacle detection within 10 meters. They are not suitable for mass deployment in large numbers of low-speed unmanned ship formations and cannot accurately identify dynamic targets in the near field.

[0006] 3. The communication mechanism is difficult to support high-frequency status interaction: Existing fleet communications mostly use point-to-point methods or wireless broadcast methods with unstable bandwidth. It is difficult to achieve synchronous broadcasting of key status such as position, speed, and mission among multiple ships at the 100ms level, resulting in inconsistent information and delayed conflict response.

[0007] 4. Lack of adaptive avoidance scoring mechanism and strategy coordination logic: Most obstacle avoidance systems are unable to comprehensively consider factors such as mission level, battery status, and control stability for intelligent sorting. When multiple ship conflicts occur simultaneously, the lack of a graph-structured avoidance mechanism will lead to decision-making confusion or stalemate.

[0008] 5. Imperfect path reconstruction and return mechanisms: Traditional low-speed vessels mostly "avoid obstacles and then detour", lacking a trajectory generation algorithm with continuous curvature and smooth posture. This makes it difficult to naturally return to the formation after avoidance, resulting in an imbalance in the formation.

[0009] 6. Lack of fault-tolerance mechanism and susceptibility to communication interruptions: Existing systems often fall into a blind control state when communication is disconnected. They lack a state prediction and compensation mechanism based on historical trajectories and cached data, and lack robustness guarantees.

[0010] Currently, low-speed surface unmanned vessel systems generally rely on the centralized control and wide-range radar sensing methods of high-speed unmanned boats, making them difficult to adapt to the micro-scale conflict response requirements between densely packed individuals in a formation. This is particularly true in near-field mission collaboration (such as tourist cruises and surveying formations), where low-speed vessels have weak inertia and strong control capabilities but lack refined scheduling logic, leading to mission delays and wasted resources.

[0011] Therefore, there is an urgent need for an unmanned ship collaborative collision avoidance method with high-resolution perception capability, formation communication capability and dynamic collision avoidance judgment capability to meet the dual needs of practicality and deployability. Summary of the Invention

[0012] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a dynamic collision avoidance method and system for unmanned ship formation.

[0013] The purpose of the present invention can be achieved by the following technical solutions:

[0014] According to one aspect of the present invention, a dynamic collision avoidance method for an unmanned ship formation is provided. Figure 1 As shown, the specific steps include:

[0015] S1. Each unmanned vessel collects environmental information in real time and periodically broadcasts its own status information to other unmanned vessels in the queue;

[0016] S2, based on the environmental information and the state information of each unmanned vessel, using the nonlinear track prediction model to determine the closest intersection time and the closest intersection distance between the unmanned vessels;

[0017] S3: If either the closest intersection time or the closest intersection distance between the unmanned ships is less than the corresponding preset threshold, a conflict is determined to have occurred and the process jumps to S4; otherwise, S1 and S2 are executed in a loop.

[0018] S4. Prioritize all conflicting unmanned ships;

[0019] S5. If there is a two-vessel conflict, the one with the higher priority score maintains its current speed and course, and the one with the lower priority score performs avoidance. If there are multiple vessels, a topological sorting algorithm is used to generate an avoidance sequence number for each conflicting unmanned vessel, so that the one with the lowest priority gives way first, and the order is followed.

[0020] S6. After the avoidance is completed, each unmanned ship calculates the shortest return path based on the dynamic surface control technology, and performs trajectory recovery based on the shortest return path, so that all unmanned ships return to the main formation.

[0021] As a preferred technical solution, the environmental information in S1 includes the static or dynamic obstacle detection results within a preset range; the self-state information includes position, velocity vector, heading angle, remaining power percentage, task priority identifier and ship unique ID.

[0022] As a preferred technical solution, the specific calculation process of the closest intersection time in S2 is:

[0023] TCPA=-(P1-P2)·(V1-V2) / ||V1-V2|| 2 ;

[0024] Where TCPA is the closest intersection time; P1 is the position vector of own ship; P2 is the position vector of adjacent ship; V1 is the velocity vector of own ship; V2 is the position vector of adjacent ship.

[0025] As a preferred technical solution, the specific formula for priority scoring in S4 is:

[0026] Q=αW+βE+γC+δS;

[0027] Among them, Q is the priority score; W is the task weight, which is set to 0.9 for the main task and 0.3 for the auxiliary task, and is dynamically updated according to the scenario; E is the energy ratio, which indicates the current remaining power ratio, and the larger the value, the higher the stability; C is the control capability, which is the inverse of the servo response delay and represents the heading control accuracy; S is the safety margin, which is dynamically adjusted based on the success rate of 10 obstacle avoidance operations in historical task performance; α, β, γ, and δ are parameters trained through the deep reinforcement learning process and are dynamically optimized and updated based on the input policy network of historical task performance.

[0028] As a preferred technical solution, the avoidance methods in S5 specifically include: deceleration, changing course or waiting in place; among them, deceleration is to linearly reduce the output power of the thruster 43; changing the course is to the maximum steering angle of ±30°, and adjusting the path in combination with the obstacle avoidance vector; waiting in place is to maintain the position stability of the unmanned boat within a preset controllable range.

[0029] As a preferred technical solution, the specific process of generating avoidance sequence numbers for each conflicting unmanned ship based on the topological sorting algorithm in S5 includes: first constructing a set of conflicting ships, and then generating a directed graph G = (V, E) based on the set of conflicting ships, where V is the conflicting ship node and the edge E represents the "need to avoid" relationship; finally, generating an avoidance sequence number for each conflicting unmanned ship according to the directed graph.

[0030] As a preferred technical solution, the shortest regression path in S6 is optimized by fusion of Bezier curve interpolation and the target main path. At the same time, the shortest regression path is dynamically adjusted in combination with real-time state feedback to achieve a natural integration into the main formation.

[0031] According to another aspect of the present invention, a dynamic collision avoidance system for an unmanned ship formation is provided, the system comprising a sensing unit 1, a communication unit 3, a decision unit 2, and a control unit 4;

[0032] The perception unit 1 includes a lidar array 12, a binocular camera 13, and an inertial measurement module 11. This unit uses a multi-source time synchronization and calibration algorithm to obtain the detailed outlines, motion vectors, and behavioral characteristics of obstacles within a preset range. This unit then uses a graph-optimization-based fusion algorithm to output a fused positioning result.

[0033] The communication unit 3 includes a 5G communication module 31, a LoRa wireless module 32 and a TDMA scheduler 33. Based on the 5G and LoRa dual-channel communication mechanism, the unit adopts a TDMA scheduling method to realize the status broadcast of the preset period within the formation;

[0034] The decision unit 2 includes a multi-factor scoring model 22, a conflict topology sorter 23 and a conflict detection unit 24. The unit is used for decision-making calculation operations based on the priority scoring model and the collision avoidance scheduling strategy. The calculation operations include local path reconstruction, conflict graph construction and topological sorting;

[0035] The control unit 4 integrates a propeller 43 and a vector servo 44, and also includes a curvature continuous interpolator 41 and a dynamic surface control module 42, supporting 0-5m / s propulsion speed adjustment and ±30° heading adjustment; and the unit is embedded with an inertia compensation algorithm, and has inertia compensation and attitude stabilization functions.

[0036] As a preferred technical solution, the communication unit 3 also includes a data storage module 34, which is used to store historical trajectory data; a communication disconnection fault tolerance mechanism is established based on the data storage module 34. When the status broadcast is lost, stable obstacle avoidance operation is achieved through the prediction and compensation of historical trajectory data in the data storage module 34.

[0037] As a preferred technical solution, the multi-factor scoring model 22 is embedded with a deep reinforcement learning optimizer 21, which is used to dynamically update the priority scoring function parameters based on the historical obstacle avoidance success rate and behavior feedback.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. In the present invention, through multi-dimensional information fusion (environment plus its own state) and hierarchical decision-making mechanism (first conflict detection, then priority scoring, and finally differentiated avoidance based on priority scoring), the micro-scale conflict response requirements between dense individuals in the formation are adapted, and dual threshold judgment (last intersection time plus distance) is used to avoid misjudgment of a single indicator and accurately identify potential conflicts; the topological sorting algorithm solves the scheduling problem when multiple ships conflict, ensures orderly avoidance, and realizes efficient collision avoidance of unmanned ship formations in complex waters. This method is highly practical and has high-resolution perception capabilities, formation communication capabilities, and dynamic collision avoidance judgment capabilities. It constructs a formation collision avoidance mechanism based on perception-communication fusion and improves overall redundancy capabilities.

[0040] 2. In the present invention, by giving a formula for the closest convergence time, the collision risk is accurately quantified, and the closest convergence time (TCPA) is calculated based on the relative position and velocity vector, providing a time dimension prediction basis for collision avoidance decision-making, which is more forward-looking than the traditional "distance priority" judgment.

[0041] 3. In the present invention, the priority score is composed of task weight, energy ratio, control capability and safety margin, and is combined with deep reinforcement learning to dynamically adjust parameters to make priority allocation more flexible and intelligent; a multi-factor scoring mechanism is adopted to adapt to complex formation relationships and realize flexible strategy adjustment.

[0042] 4. This invention utilizes a combination of deceleration, steering, and holding strategies to adapt to different scenarios: steering is prioritized in narrow waterways, while deceleration or holding is optional in open waters, enhancing collision avoidance flexibility. A ±30° steering angle limit prevents excessive maneuvers, and linear power regulation enables smooth deceleration, minimizing impact on the overall formation. Furthermore, mission-driven active avoidance logic is introduced to prevent interruptions to high-priority vessels.

[0043] 5. In the present invention, the topological sorting algorithm generates the optimal avoidance order by constructing a conflict directed graph (nodes are ships and edges are avoidance relationships), thereby solving the "chain reaction" problem when multiple ships cross and conflict.

[0044] 6. In the present invention, Bezier curve interpolation is combined with main path fusion to optimize the smooth regression trajectory, and combined with real-time state feedback dynamic adjustment, the unmanned ship can quickly and naturally return to the formation after collision avoidance, reducing interference with the overall mission.

[0045] 7. In the present invention, the anti-collision system includes a perception unit, a communication unit, a decision unit and a control unit. The perception unit provides high-precision environmental perception through multi-sensor fusion (lidar, camera and IMU), the communication unit 5G+LoRa dual-channel communication ensures real-time synchronization of status, and the control unit achieves precise control through vector servos and inertia compensation. The scheduling logic is refined through this system; and the system is highly deployable, the overall system structure is lightweight, the degree of modularity is high, and it has good portability and engineering implementation capabilities.

[0046] 8. In the present invention, historical trajectory data is stored and a communication disconnection fault-tolerant mechanism is established. Trajectory prediction and compensation are used when the signal is lost to ensure the continuity of collision avoidance operations. When the signal is briefly interrupted, the system calculates the position of the neighboring ship based on historical data to maintain the reliability of collision avoidance decisions and improve the robustness of the system.

[0047] 9. In the present invention, the priority scoring model continuously optimizes parameters through deep reinforcement learning and dynamically adjusts weights based on historical obstacle avoidance success rates and behavioral feedback, so that the system can continuously learn the best strategy in long-term operation, adapt to different water environments and task requirements, and improve its adaptive capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is the overall flow chart of the method of the present invention;

[0049] Figure 2 This is a flow chart of the scoring calculation and priority determination in the present invention;

[0050] Figure 3 Schematic diagram of the structure of the device in the present invention;

[0051] Figure 4 Schematic diagram of three-ship conflict scheduling in the embodiment;

[0052] In the figure, 1 is the perception unit; 11 is the inertial measurement module; 12 is the lidar array; 13 is the binocular camera; 2 is the decision unit; 21 is the deep reinforcement learning optimizer; 22 is the multi-factor scoring model; 23 is the conflict topology sorter; 24 is the conflict detection unit; 3 is the communication unit; 31 is the 5G communication module; 32 is the LoRa wireless module; 33 is the TDMA scheduler; 34 is the data storage module; 4 is the control unit; 41 is the curvature continuous interpolator; 42 is the dynamic surface control module; 43 is the thruster; 44 is the vector servo. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0054] Example 1

[0055] In this embodiment, a dynamic collision avoidance method for an unmanned vessel formation is provided, and the method steps include:

[0056] S1. Each unmanned vessel collects environmental information in real time and periodically broadcasts its own status information to other unmanned vessels in the queue;

[0057] S2, based on the environmental information and the state information of each unmanned vessel, using the nonlinear track prediction model to determine the closest intersection time and the closest intersection distance between the unmanned vessels;

[0058] S3: If either the closest intersection time or the closest intersection distance between the unmanned ships is less than the corresponding preset threshold, a conflict is determined to have occurred and the process jumps to S4; otherwise, S1 and S2 are executed in a loop.

[0059] S4. Prioritize all conflicting unmanned ships;

[0060] S5. If there is a two-vessel conflict, the one with the higher priority score maintains its current speed and course, and the one with the lower priority score performs avoidance. If there are multiple vessels, a topological sorting algorithm is used to generate an avoidance sequence number for each conflicting unmanned vessel, so that the one with the lowest priority gives way first, and the order is followed.

[0061] S6. After the avoidance is completed, each unmanned ship calculates the shortest return path based on the dynamic surface control technology, and performs trajectory recovery based on the shortest return path, so that all unmanned ships return to the main formation.

[0062] Environmental information in S1 includes detection results for static or dynamic obstacles within a preset range; vessel status information includes position, velocity vector, heading angle, remaining battery percentage, mission priority indicator, and vessel ID. This environmental information, encompassing both static and dynamic obstacles, is combined with other state parameters such as vessel battery level and mission priority to enable more comprehensive decision-making. For example, vessels with low battery levels are prioritized to ensure their return path, preventing them from running aground due to energy depletion. Vessels with high-priority missions are given priority passage, improving the overall efficiency of the fleet.

[0063] The specific calculation process of the nearest intersection time in S2 is:

[0064] TCPA=-(P1-P2)·(V1-V2) / ||V1-V2||2 ;

[0065] Where TCPA is the closest intersection time; P1 is the position vector of own ship; P2 is the position vector of adjacent ship; V1 is the velocity vector of own ship; V2 is the position vector of adjacent ship.

[0066] Priority scoring is composed of task weight, energy ratio, control capability, and safety margin, and is combined with deep reinforcement learning to dynamically adjust parameters, making priority allocation more flexible and intelligent. For example, ships with a high historical obstacle avoidance success rate will receive higher trust, and ships with low battery power will actively avoid obstacles, extending the overall endurance of the fleet. The specific scoring calculation and priority judgment process is as follows: Figure 2 As shown, the specific formula for priority scoring in S4 is:

[0067] Q=αW+βE+γC+δS;

[0068] Among them, Q is the priority score; W is the task weight, which is set to 0.9 for the main task and 0.3 for the auxiliary task, and is dynamically updated according to the scenario; E is the energy ratio, which indicates the current remaining power ratio, and the larger the value, the higher the stability; C is the control capability, which is the inverse of the servo response delay and represents the heading control accuracy; S is the safety margin, which is dynamically adjusted based on the success rate of 10 obstacle avoidance operations in historical task performance; α, β, γ, and δ are parameters trained through the deep reinforcement learning process and are dynamically optimized and updated based on the input policy network of historical task performance.

[0069] Among them, the avoidance methods in S5 specifically include: deceleration, changing course or waiting in place; among them, deceleration is to linearly reduce the output power of the thruster 43; changing course is to the maximum steering angle of ±30°, and adjusting the path in combination with the obstacle avoidance vector; waiting in place is to maintain the position stability of the unmanned boat within the preset controllable range.

[0070] The specific process of generating avoidance sequence numbers for each conflicting unmanned vessel based on the topological sorting algorithm in S5 includes: first, constructing a set of conflicting vessels, then generating a directed graph G = (V, E) based on the set of conflicting vessels, where V is the conflicting vessel node and the edge E represents the relationship that needs to be avoided; finally, generating an avoidance sequence number for each conflicting unmanned vessel based on the directed graph.

[0071] The shortest return path in S6 is optimized by fusion of Bezier curve interpolation and the target main path. At the same time, the shortest return path is dynamically adjusted in combination with real-time state feedback to achieve a natural integration into the main formation.

[0072] In summary, this method adapts to the micro-scale conflict response needs between dense individuals in the formation through multi-dimensional information fusion (environment plus its own status) and a hierarchical decision-making mechanism (first conflict detection, then priority scoring, and finally differentiated avoidance based on priority scores). Dual threshold judgment (last intersection time plus distance) avoids misjudgment of a single indicator and accurately identifies potential conflicts. The topological sorting algorithm solves the scheduling problem when multiple ships conflict, ensures orderly avoidance, and realizes efficient collision avoidance of unmanned ship formations in complex waters. This method is highly practical and has high-resolution perception capabilities, formation communication capabilities, and dynamic collision avoidance judgment capabilities. It constructs a formation collision avoidance mechanism based on perception-communication fusion and improves overall redundancy capabilities.

[0073] Example 2

[0074] In this embodiment, a dynamic collision avoidance system for low-speed unmanned ship formations is adopted to solve problems such as the delayed response of traditional formation control, insufficient accuracy of existing perception systems, and reliance on rigid rules, and to improve the collaborative obstacle avoidance capability and safety redundancy level of the formation in dynamic and complex inland water environments.

[0075] The overall structure of this system is as follows Figure 3 As shown, it includes a multi-source perception unit 1, an intra-formation state synchronization unit, namely a communication unit 3, a dynamic collision avoidance decision unit 2, and an obstacle avoidance path reconstruction unit, namely a control unit 4. The core technical solution is as follows:

[0076] (1) Multi-source environmental perception and state synchronization: Each unmanned vessel is equipped with a laser radar (for static and dynamic obstacle detection within 20 meters), a binocular camera 13 (for dynamic target tracking and recognition), and an IMU inertial navigation system (for heading and attitude compensation), forming a fusion multi-source perception system. Perception unit 1 samples environmental information at a 50ms cycle and combines a timing cache mechanism to perform dynamic target association and matching.

[0077] In terms of communication, the system uses a LoRa / 5G dual-channel communication mechanism to build a distributed state synchronization network within the formation. Each unmanned vessel periodically broadcasts its own status information every 100ms. The broadcast content includes location (GPS coordinates), velocity vector, heading angle, remaining battery percentage, mission priority identifier, and vessel unique ID. The broadcast is scheduled using TDMA time slots to avoid communication conflicts. All broadcast information is aligned with the timestamp and input into the Kalman filter, which integrates historical trajectories with multi-source perception data to ensure the consistency of the entire fleet. In the event of communication packet loss, the system retains the most recent 300 frames of status data for short-term prediction and has link break fault tolerance. The positioning error after fusion is controlled within ±0.2 meters.

[0078] (2) Dynamic collision avoidance decision model: Based on a nonlinear trajectory prediction model, the system integrates the position, speed, and heading of the own ship and adjacent ships to predict the sequence of ship positions within the next T seconds (T is generally 6 to 10 seconds). The following criteria are used to determine potential collisions:

[0079] TCPA (Time to Closest Point of Approach): Assume that the position vectors of own ship and neighboring ship are P1 and P2, and the velocity vectors are V1 and V2, respectively. Then:

[0080] TCPA=-(P1-P2)·(V1-V2) / ||V1-V2|| 2

[0081] If TCPA is less than 6 seconds, it is judged to enter the dangerous area;

[0082] DCPA (Distance at Closest Point of Approach): Calculate the predicted distance DCPA between the two ships at TCPA. If DCPA is less than 3 meters, the ship is considered a potential conflict target.

[0083] Once a conflict is determined, the system assigns a priority score to all conflicting vessels:

[0084] Q=αW+βE+γC+δS

[0085] in:

[0086] W (task weight): 0.9 for main tasks and 0.3 for auxiliary tasks, dynamically updated according to the scenario;

[0087] ●E (Energy Ratio): The current remaining power ratio. A larger value indicates higher stability.

[0088] C (control capability): The inverse of the servo response delay, representing the heading control accuracy;

[0089] S (Safety Margin): Dynamically adjusted based on the success rate of the past 10 obstacle avoidance maneuvers;

[0090] α, β, γ, and δ are parameters trained through the deep reinforcement learning process. The system inputs historical task performance into the strategy network for dynamic optimization and update.

[0091] The scoring results are used for subsequent avoidance strategy allocation to ensure that the decision is explainable and robust.

[0092] (3) Collaborative avoidance strategy: Once the conflict detection unit 24 identifies a potential risk, the system immediately triggers the avoidance strategy judgment.

[0093] In a two-vessel collision scenario, the system calculates a priority score, Q, for each vessel. The vessel with the lower Q score will perform an evasive maneuver, while the vessel with the higher Q score will maintain its current speed and heading. Avoidance methods include: deceleration (linearly reducing thruster 43 output power), changing course (maximum steering angle ±30°, combined with obstacle avoidance vectors for path adjustment), or holding position (maintaining a stable position within a controllable range).

[0094] In scenarios where three or more vessels face a potential collision risk, the system constructs a set of conflicting vessels and generates a directed graph G = (V, E), where V represents the conflicting vessel node and E represents the "give way to" relationship. Using a topological sorting algorithm, the system generates an avoidance sequence number for each vessel, with the lowest priority vessel giving way first, and so on.

[0095] After completing the avoidance maneuver, the vessel activates the trajectory recovery unit, which calculates the shortest return path based on dynamic surface control technology. This path is integrated with the target main path through Bezier curve interpolation, ensuring path continuity and smooth curvature, and avoiding sudden movements that could disrupt the formation. Simultaneously, the recovery path is adjusted based on real-time status feedback, enabling a natural integration into the main formation without external human intervention.

[0096] (4) System structure description: This anti-collision system is highly modular in design. Its structure consists of four core functional units: perception unit 1, communication unit 3, decision unit 2 and control unit 4. It is integrated on the unmanned hull platform and has scalability and easy maintenance.

[0097] ① Perception Unit 1: Utilizing a multi-sensor fusion configuration, it comprises a 360° surround-view LiDAR array 12 (with a point cloud accuracy of 0.05m and a refresh rate of 20Hz), a binocular camera 13 (supporting depth estimation and dynamic tracking, with a resolution ≥1080p), and an IMU (inertial measurement unit 11). Through multi-source time synchronization and calibration algorithms, it acquires the detailed outlines of obstacles, motion vectors, and behavioral characteristics of neighboring vessels within a 10m range. Combined with a graph-optimization-based fusion algorithm, the system outputs fused positioning results within 0.2 seconds, with an error within ±0.2m.

[0098] ② Communication Unit 3: Equipped with an industrial-grade 5G communication module 31 and a LoRa wireless module 32, it achieves redundant communication for both long and short distances. Status information is broadcast within the formation using TDMA time-sharing scheduling, with a 100ms broadcast cycle. Broadcast content includes: own ship ID, timestamp, latitude and longitude coordinates, velocity vector, heading angle, mission status, remaining energy, etc. The system supports link loss prediction. During disconnections, it estimates the status of neighboring ships based on cached historical trajectories and a dynamic inertial model to fill in information gaps.

[0099] ③ Decision Unit 2: Based on a low-power embedded AI inference module, it features a reinforcement learning-driven priority scoring model and collision avoidance scheduling strategy. It supports computational operations such as local path reconstruction, conflict graph construction, and topological sorting, with a control latency of less than 50ms, meeting real-time response requirements.

[0100] ④ Control Unit 4: Integrates a multi-channel propulsion system (PWM-controlled motor) with a high-response vector servo 44, supporting propulsion speed adjustment from 0 to 5 m / s and ±30° heading adjustment. The system incorporates an inertia compensation algorithm (based on IMU angular velocity feedback) to achieve stable attitude control and interference rejection, improving steering control accuracy and response robustness.

[0101] Unlike existing passive avoidance systems based on AIS, radar and fixed rules, this solution uses a dynamic formation collision avoidance system based on close-range composite perception, autonomous scoring and active decision-making. It is particularly suitable for non-standard inland waterway operation environments with frequent task switching and dynamic path adjustment, and can significantly improve collaborative efficiency and system safety.

[0102] Avoidance maneuvers include deceleration, waiting in place, and detour path reconstruction. The task weight W in the scoring model supports dynamic configuration at the task level. The system has a disconnection fault tolerance mechanism, relying on historical trajectories and scoring caches to perform local obstacle avoidance in the event of temporary communication interruptions.

[0103] Typical applications of this solution in actual scenarios include Figure 4 As shown, take three low-speed unmanned boats in a lake sightseeing cruise mission as an example:

[0104] Ship A performs the main mission, i.e., cruise and shoot along the waterway, and has the highest mission weight.

[0105] W = 0.9;

[0106] Ship B performs boundary area mapping mission, with weight W = 0.6;

[0107] ●Ship C is returning empty-handed, with the lowest weight W = 0.3.

[0108] Three ships sailed into the confluence of a lake from different directions. Based on information such as position, speed, and heading, the system predicted that they would enter the intersection point at the same time within the next 5 seconds. The DCPA (estimated minimum distance) was 3 meters, reaching the avoidance judgment threshold set by the system and triggering the conflict detection mechanism.

[0109] The system first receives the state parameters broadcast by the three ships and uses the scoring function Q = αW + βE + γC + δS to calculate the passage priority Q values ​​of the three ships (the calculation process is omitted here):

[0110] Ship A: Q A =α×0.9+...

[0111] Ship B: Q B =α×0.6+...

[0112] Ship C: Q C =α×0.3+...

[0113] The final priority is Q A >Q B >Q C .

[0114] According to the coordinated avoidance strategy of this scheme:

[0115] Vessel A is the lead mission vessel and will proceed straight through the intersection area on its original course without making any evasive moves.

[0116] Ship B determines that it cannot pass directly through the channel, so it slows down and gives way, waiting for Ship A to pass before moving sideways;

[0117] Ship C is a low-priority empty ship. It adopts the waiting and avoiding strategy by turning and slowing down to avoid the high-priority task ship in front and delay its passage through the intersection area.

[0118] The above operations ensure task priority, path safety and high efficiency of system scheduling, and avoid the phenomena of "mutual waiting" and "total stop".

[0119] This solution uses a multi-factor fusion scoring mechanism to achieve dynamic avoidance priority decisions in local conflicts. Combined with path reconstruction and trajectory regression mechanisms, it ensures the overall mission continuity and navigation safety of the formation. It is suitable for a variety of low-speed unmanned vessel formation missions, including lake sightseeing, urban security, and hydrographic mapping.

[0120] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A dynamic collision avoidance method for an unmanned vessel formation, characterized in that: The method steps include: S1. Each unmanned vessel collects environmental information in real time and periodically broadcasts its own status information to other unmanned vessels in the queue; S2, based on the environmental information and the state information of each unmanned vessel, using the nonlinear track prediction model to determine the closest intersection time and the closest intersection distance between the unmanned vessels; S3: If either the closest intersection time or the closest intersection distance between the unmanned ships is less than the corresponding preset threshold, a conflict is determined to have occurred and the process jumps to S4; otherwise, S1 and S2 are executed in a loop. S4. Prioritize all conflicting unmanned ships; S5. If there is a two-vessel conflict, the one with the higher priority score maintains its current speed and course, and the one with the lower priority score performs avoidance. If there are multiple vessels, a topological sorting algorithm is used to generate an avoidance sequence number for each conflicting unmanned vessel, so that the one with the lowest priority gives way first, and the order is followed. S6. After the avoidance is completed, each unmanned ship calculates the shortest return path based on the dynamic surface control technology, and performs trajectory recovery based on the shortest return path, so that all unmanned ships return to the main formation.

2. A dynamic collision avoidance method for unmanned ship formation according to claim 1, characterized in that: The environmental information in S1 includes the detection results of static or dynamic obstacles within a preset range; the self-state information includes position, velocity vector, heading angle, remaining power percentage, task priority identifier and ship unique ID.

3. The dynamic collision avoidance method for unmanned ship formation according to claim 1, characterized in that: The specific calculation process of the latest intersection time in S2 is: TCPA=-(P1-P2)·(V1-V2) / ||V1-V2|| 2 ; Where TCPA is the closest intersection time; P1 is the position vector of own ship; P2 is the position vector of adjacent ship; V1 is the velocity vector of own ship; V2 is the position vector of adjacent ship.

4. The dynamic collision avoidance method for unmanned ship formation according to claim 1, characterized in that: The specific formula for the priority score in S4 is: Q=αW+βE+γC+δS; Among them, Q is the priority score; W is the task weight, which is set to 0.9 for the main task and 0.3 for the auxiliary task, and is dynamically updated according to the scenario; E is the energy ratio, which indicates the current remaining power ratio, and the larger the value, the higher the stability; C is the control capability, which is the inverse of the servo response delay and represents the heading control accuracy; S is the safety margin, which is dynamically adjusted based on the success rate of 10 obstacle avoidance operations in historical task performance; α, β, γ, and δ are parameters trained through the deep reinforcement learning process and are dynamically optimized and updated based on the input policy network of historical task performance.

5. The dynamic collision avoidance method for unmanned ship formation according to claim 1, characterized in that: The avoidance methods in S5 specifically include: deceleration, changing course or waiting in place; wherein, the deceleration is to linearly reduce the output power of the propeller; the changing course is to adjust the path by a maximum steering angle of ±30° combined with the obstacle avoidance vector; the waiting in place is to maintain the position stability of the unmanned vessel within a preset controllable range.

6. The dynamic collision avoidance method for unmanned ship formation according to claim 1, characterized in that: The specific process of generating an avoidance sequence number for each conflicting unmanned vessel based on the topological sorting algorithm in S5 includes: first, constructing a set of conflicting vessels, then generating a directed graph G = (V, E) based on the set of conflicting vessels, where V is a conflicting vessel node and the edge E represents a "need to avoid" relationship; finally, generating an avoidance sequence number for each conflicting unmanned vessel based on the directed graph.

7. The dynamic collision avoidance method for unmanned ship formation according to claim 1, characterized in that: The shortest regression path in S6 is optimized by fusion of Bezier curve interpolation and the target main path. At the same time, the shortest regression path is dynamically adjusted in combination with real-time state feedback to achieve a natural integration into the main formation.

8. A dynamic collision avoidance system for an unmanned ship formation, characterized in that: The system applies a dynamic collision avoidance method for an unmanned ship formation as described in any one of claims 1 to 7; the system comprises a sensing unit (1), a communication unit (3), a decision unit (2) and a control unit (4); The perception unit (1) includes a laser radar array (12), a binocular camera (13) and an inertial measurement module (11); the unit obtains the fine outline of obstacles, motion vectors and behavioral characteristics of neighboring ships within a preset range through a multi-source time synchronization and calibration algorithm, and the unit cooperates with a fusion algorithm based on graph optimization to output a fusion positioning result; The communication unit (3) includes a 5G communication module (31), a LoRa wireless module (32) and a TDMA scheduler (33). The unit is based on a 5G and LoRa dual-channel communication mechanism and adopts a TDMA scheduling method to realize status broadcasting of a preset period within the formation; The decision unit (2) includes a multi-factor scoring model (22), a conflict topology sorter (23) and a conflict detection unit (24), which is used for decision-making calculation operations based on the priority scoring model and the collision avoidance scheduling strategy, and the calculation operations include local path reconstruction, conflict graph construction and topological sorting; The control unit (4) integrates a propeller (43) and a vector servo (44), and also includes a curvature continuous interpolator (41) and a dynamic surface control module (42), supporting propulsion speed adjustment of 0-5m / s and ±30° heading adjustment; and the unit is embedded with an inertia compensation algorithm and has inertia compensation and attitude stabilization functions.

9. The dynamic collision avoidance system for unmanned ship formation according to claim 1, characterized in that: The communication unit (3) further includes a data storage module (34) for storing historical trajectory data; a communication disconnection fault tolerance mechanism is established based on the data storage module (34); when the status broadcast is lost, stable obstacle avoidance operation is achieved through prediction and compensation of the historical trajectory data in the data storage module (34).

10. The dynamic collision avoidance system for unmanned ship formation according to claim 9, characterized in that: The multi-factor scoring model (22) is embedded with a deep reinforcement learning optimizer (21) for dynamically updating the priority scoring function parameters based on historical obstacle avoidance success rates and behavioral feedback.

Images