Multi-usv cooperative hunting method based on topology perception and evaluation of herding algorithm

Abstract

This invention relates to the field of unmanned surface vessel (USV) capture technology, and discloses a multi-USV collaborative capture method based on topology perception and evaluation using the shepherding algorithm. The method includes: (1) topological modeling of the target group, dividing it into a main group and dynamic subgroups; (2) constructing an evaluation mechanism based on subgroup structural characteristics, establishing intervention priorities and resource requirements; (3) generating collection and driving tasks according to priorities and resource requirements, and allocating collaborative tasks to achieve parallel collection of divergent subgroups and continuous advancement of the main group towards the target area; (4) converting USVs performing collection tasks to driving tasks based on dynamically updated group structure and evaluation results, achieving dynamic resource reallocation. This invention achieves stable capture in scenarios with dynamic changes in multiple subgroups, reducing the number of USVs captured while ensuring effectiveness, and effectively improving capture efficiency and stability in complex environments.

Description

Technical Field

[0001] This invention relates to a multi-USV (unmanned surface vessel) cooperative encirclement method based on topological perception and group assessment using a shepherding algorithm, belonging to the field of multi-agent cooperative control and unmanned surface vessel encirclement technology. Background Technology

[0002] Unmanned surface vehicles (USVs) have been widely used in fields such as maritime security, target surveillance, and countermeasures due to their high maneuverability, low cost, and ability to perform tasks in complex marine environments. Among these applications, multi-USV cooperative encirclement missions, as a typical multi-agent cooperative control problem, have significant research and engineering value.

[0003] Current research on USV capture tasks mainly includes heuristic methods, distributed control methods, and methods based on deep reinforcement learning. Some methods achieve target herding and capture through rule design, while others achieve cooperative behavior through learning strategies. However, existing methods still have the following problems:

[0004] First, most methods are designed for single targets or single group structures. When the target group undergoes dynamic changes such as splitting or merging, it is difficult to accurately characterize the group structure, which can easily lead to the failure of the encirclement strategy.

[0005] Second, existing methods rely heavily on local information or simple rules for decision-making, lacking a unified model of the overall structure of the population. In the case of multiple subgroups, it is difficult to determine the intervention priority, which can easily lead to unstable target selection and frequent switching.

[0006] Third, in the process of multi-USV coordinated encirclement, there is a lack of a unified task allocation and resource scheduling mechanism, which can easily lead to multiple USVs repeatedly acting on the same target or some targets being unattended, thereby reducing the overall encirclement efficiency.

[0007] Fourth, existing methods typically do not consider the stability of task progress during the collaborative process. When the target state changes continuously, the USV task allocation is frequently updated, which can easily cause system jitter and oscillation, affecting the stability and reliability of collaborative capture.

[0008] In addition, although applying the shepherding method to the USV encirclement task can achieve basic driving behavior, it is usually based on the assumption that the target group is a single whole. It is difficult to be effectively applied in scenarios with multiple subgroups that are dynamically evolving and highly mobile, and it has problems such as insufficient structural expression ability and imperfect cooperation mechanism.

[0009] Therefore, for the problem of coordinated capture of multiple USVs in complex, dynamic, and highly mobile group environments, there is an urgent need for a capture method that can perform group structure modeling, unified evaluation and decision-making, and achieve stable coordinated scheduling. Summary of the Invention

[0010] To address the problems of existing multi-USV cooperative trapping methods, such as difficulty in accurately modeling the target population structure under dynamic changes, unclear intervention priorities, and insufficient stability of cooperative scheduling, this invention proposes a multi-USV cooperative trapping method based on the shepherding algorithm for topology perception and evaluation, in order to improve the trapping efficiency and system stability in multi-subgroup dynamic evolution scenarios.

[0011] The multi-USV cooperative trapping method based on the shepherding algorithm for topology awareness and evaluation in this invention includes the following steps:

[0012] (1) Perform topological modeling of the target group:

[0013] A connectivity topology graph is constructed based on the perceptual relationships between target individuals. By identifying and correcting the weak connections in the topology graph, the dynamic division of the main group and subgroups is achieved. At the same time, the structural feature parameters of each subgroup are extracted, including subgroup size, subgroup center location and spatial distribution characteristics.

[0014] (2) Construct a population criticality assessment mechanism based on the aforementioned subgroup structural characteristics:

[0015] A comprehensive assessment of the task progress risk, group dispersion and size of each subgroup is conducted to obtain the intervention priority and corresponding resource requirements of each subgroup, thereby achieving unified decision-making on the selection of intervention targets and resource allocation for each subgroup.

[0016] (3) Generate capture tasks based on the priority and resource requirements of each subgroup, and coordinate the task allocation of multiple USVs based on the utility function, so that some USVs perform subgroup capture tasks and the rest of USVs perform main group drive-off tasks, thereby realizing the parallel capture of discrete subgroups and the continuous advancement of the main group to the target area.

[0017] (4) The number of USVs performing the gathering and driving tasks is dynamically adjusted according to the changes in the target group status. When the number of subgroups decreases or the subgroups are successfully gathered, the USVs performing the gathering task are switched to the driving task to achieve dynamic redistribution of collaborative resources.

[0018] The specific process of step (1) is as follows:

[0019] Construct a connectivity topology graph based on the perceptual relationships between target individuals, and represent the set of target individuals as a set of nodes. When any two target individuals and The Euclidean distance between them satisfies At that time, establish undirected edges between nodes. Thus, an undirected graph is obtained. ,in, Indicates the distance between target individuals. Indicates the preset sensing range. Represents the set of edges;

[0020] In the undirected graph Connectivity analysis is performed to obtain all connected components as the subgroup partitioning results, and the connectivity degree of each node is calculated. Used to characterize the adjacency relationship of nodes;

[0021] Based on this, a stability analysis is performed on the connectivity relationships in the topology graph, and edges that increase the number of connected components after deletion are identified as the bridge edge set. It is used to characterize weak connections between subgroups; and for undirected edges Define stability weights:

[0022] ;

[0023] In the formula: Representing an edge Stability weights, and These are the weighting coefficients;

[0024] When performing stability analysis on the connectivity relationships in the topology graph, it is clarified that the connectivity topology graph is a perceptual connectivity graph consisting of target individuals as nodes and perceptual relationships between individuals as edges, and the edges correspond to stability weights; for those satisfying the condition of belonging to a preset set of bridge edges... Or the stability weights satisfy Edges are deleted or have their weights reduced, where A preset threshold is used to obtain the corrected topology map. ;

[0025] Based on the revised topology graph Re-partition the connected components, determine the structure of the main group and subgroups, and calculate the center position of each subgroup. and subgroup size ,in Describes the k-th subgroup. Indicates the node position. Indicates the center position of the subgroup. This indicates the subgroup size, used for subsequent group assessment and multi-USV collaborative capture decisions.

[0026] The group criticality assessment mechanism in step (2) is:

[0027] Based on the subgroup partitioning results obtained in step (1), assume that the target population is divided into a set of subgroups. And extract the structural feature parameters of each subgroup, including the subgroup size. subgroup center position And the spatial distribution range of the subgroup;

[0028] Based on this, for each subgroup Constructing key assessment indicators for groups The key evaluation index is used to characterize the intervention priority of the subgroup in the encirclement task. This index is determined by the task progress risk, the degree of group dispersion, and the size weight, and is obtained through a weighted combination function.

[0029] ;

[0030] In the formula: This represents the key evaluation metric for the k-th subgroup. This represents the risk factor for mission progress, used to characterize the degree of deviation of the subgroup from the target region. Indicates the degree of dispersion of a subgroup, used to characterize the spatial dispersion of individuals within the subgroup. Indicates the size of the subgroup. , , These are the weighting coefficients;

[0031] Based on the key evaluation indicators of each subgroup Sort all subgroups to obtain a subgroup intervention priority sequence;

[0032] At the same time, for each subgroup Based on its topological instability requirements, escape strength requirements, and operation space constraints, a resource requirement function is constructed to obtain the corresponding resource requirements. The resource requirement function is used to characterize the number of USVs required to perform the subgroup capture task; the resource requirement function satisfies:

[0033] ;

[0034] In the formula, This represents the topological instability demand term of the k-th subgroup, used to characterize the fragility of the connections within the subgroup. This represents the escape intensity of the k-th subgroup, used to characterize the number of USVs required to capture all individuals in the subgroup due to their angular deviation relative to the target point. This represents the operation space-constrained requirement of the k-th subgroup, used to characterize the requirement for the number of USVs to be captured due to obstacles, boundaries, or traversable space limitations.

[0035] The aforementioned subgroup priority sequence and corresponding resource requirements serve as the basis for subsequent multi-USV collaborative task allocation decisions.

[0036] The process of collaboratively allocating tasks among multiple USVs for the encirclement operation based on group multi-factor evaluation and utility function in step (3) is as follows:

[0037] Based on the subgroup priority and resource requirements obtained in step (2), a utility function is constructed between the capture USV and the subgroup to characterize the suitability of the i-th capture USV for performing the k-th subgroup capture task. The utility function is expressed as:

[0038] ;

[0039] In the formula: This represents the utility value between the i-th USV and the k-th subgroup. Indicates the key evaluation indicators for subgroups. This represents the resource requirement of the k-th subgroup. This represents the distance between the i-th USV used for encirclement and the center of the subgroup;

[0040] The utility function is used to comprehensively characterize the impact of subgroup priority, resource requirements, and spatial costs on task allocation results, and through resource requirements... Constrain the number of USVs to be captured and assigned to the corresponding subgroup;

[0041] Based on the utility function, all capture USVs are matched with subgroups, so that each capture USV prioritizes selecting the subgroup with the larger utility value to perform the capture task, and the allocation result satisfies the resource requirement constraints of each subgroup, thereby realizing the collaborative task allocation of multiple USVs to multiple subgroups.

[0042] The dynamic adjustment process in step (4) is as follows:

[0043] During the collaborative capture of multiple USVs, the number of USVs to be captured is dynamically adjusted based on the number of subgroups, subgroup size, and spatial distribution of the subgroups; let the total number of USVs captured in the system be... The number of USVs performing the subgroup convergence task is The number of USVs performing the main group expulsion mission was ,satisfy ;

[0044] When the number of subgroups increases, the size of subgroups becomes larger, or the dispersion of subgroups increases, the capture USVs should be prioritized for subgroup convergence tasks. Increase the number of USVs used to perform subgroup merging tasks when the number of subgroups decreases, subgroups are gradually merged, or the main group structure becomes stable. The released USVs will be redistributed to the main swarm drive-off mission, making... The corresponding increase enables coordinated resource scheduling for subgroup recovery and main group advancement;

[0045] Meanwhile, to avoid frequent task assignment switching caused by changes in the target group's state, after each task assignment, the task assignment results for each USV to be captured remain unchanged within a preset time range, so that each USV continues to execute the currently assigned gathering or driving task; after the time range ends, task assignment and resource adjustment are redistributed based on the updated subgroup structure and the group criticality assessment results, thereby reducing system oscillations and improving the stability of the multi-USV collaborative capture process.

[0046] This invention is based on the shepherding mechanism, combined with topology perception, group assessment and cooperative task allocation. By modeling the topology of the target group and combining it with group criticality assessment, it achieves unified decision-making on the intervention priority and resource requirements of subgroups. Furthermore, based on the cooperative task allocation and dynamic task scheduling mechanism, it achieves stable cooperation among multiple USVs, thereby improving the capture efficiency and system stability in multi-subgroup dynamic evolution scenarios.

[0047] This invention constructs a topology-aware population structure modeling method and combines it with a multi-factor population evaluation and multi-USV collaborative task allocation mechanism to achieve orderly intervention and rational resource scheduling of subgroups. It can effectively improve the encirclement efficiency in the dynamic evolution scenario of multiple subgroups. Compared with existing methods, this invention can reduce the number of USVs participating in the encirclement while ensuring the success rate of the encirclement, and realize the collaborative encirclement of multiple target adversarial USVs with fewer encircled USVs, thereby improving the overall resource utilization efficiency of the system.

[0048] Experimental results show that the method proposed in this invention performs well in multi-target adversarial USV capture missions and can stably achieve the capture of targets in complex environments. It is an efficient method with engineering application value. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the sheepdog encirclement mechanism in the shepherding algorithm.

[0050] Figure 2 This is a schematic diagram of an improved encirclement and capture mission model.

[0051] Figure 3 This is a schematic diagram illustrating the matching and allocation of the USV and subgroups based on the utility function.

[0052] Figure 4 It is a map showing the trajectory of multiple USVs in the encirclement and capture operation.

[0053] Figure 5 This is a schematic diagram showing the changes in distance between all USVs and the target area. Detailed Implementation

[0054] This invention proposes a multi-USV cooperative encirclement method based on topology awareness and swarm assessment. Building upon the drive-and-gathering principles of the shepherding algorithm, it addresses the issues of swarm splitting, merging, and dynamic discrepancies that easily occur during the escape process of multi-target high-performance adversarial USVs. The method improves upon the swarm structure description, intervention target selection, and multi-encirclement USV cooperative scheduling process in the traditional shepherding algorithm. First, a topology graph based on connectivity is constructed to model the target swarm structure and identify subgroups. Then, based on subgroup size, dispersion, and task progress status, a multi-factor swarm assessment is performed on each subgroup to determine the intervention priority and resource requirements of different subgroups. Finally, based on the assessment results, cooperative tasks are allocated among the multiple encirclement USVs, with some USVs performing subgroup gathering tasks and the remaining USVs performing main swarm drive-and-gathering tasks. A task maintenance mechanism within a preset time range ensures stable encirclement scheduling and task progress, thereby achieving the cooperative encirclement of multiple adversarial USVs with a smaller number of encirclement USVs. Specific steps include:

[0055] (1) Perform topological modeling of the target group:

[0056] To address the issue of traditional shepherding target groups easily splitting and merging during movement, a group modeling method based on topological connectivity is constructed. By establishing a perception-driven graph and identifying and correcting weak connections, dynamic partitioning of subgroups is achieved.

[0057] An undirected graph is constructed based on the perceptual relationship between target individuals. The target individuals are used as nodes. When the distance between any two target individuals meets a preset perceptual threshold, a connection edge is established between them. The main group and at least one subgroup are determined based on the connected components of the undirected graph.

[0058] Based on this, the undirected graph is structurally corrected. By identifying bridge edges and low-stability connections in the undirected graph, connection edges that meet preset stability conditions are deleted or their weights are adjusted to obtain a corrected topology graph. Based on the corrected topology graph, the structural features of each subgroup are extracted, including the subgroup center position and subgroup size.

[0059] The specific process includes:

[0060] Construct a connectivity graph based on perceptual distance, representing the set of target individuals as a set of nodes. When any two nodes and Euclidean distance between At that time, establish undirected edges between nodes. To obtain an undirected graph ,in For the preset sensing range; in the undirected graph Connectivity analysis is performed to obtain all connected components as the subgroup partitioning results, and the degree of each node is calculated. and edge connection relationships;

[0061] Based on this, the undirected graph is structurally modified by identifying edges that increase the number of connected components after deletion and grouping them into a bridge edge set. It is used to characterize weak connections between subgroups; and for undirected edges Define stability weights The weights are related to the distance between nodes and the node connectivity, satisfying the following:

[0062] ;

[0063] in, and These are the weighting coefficients;

[0064] When performing structural correction on the undirected graph, it is clarified that the connectivity graph includes a perceptual relation connectivity graph consisting of target individuals as nodes and perceptual relationships between individuals as edges, and each edge has a corresponding stability weight; for edges that satisfy the condition that belong to a preset set of bridge edges... Or the stability weights satisfy Edges are deleted or have their weights reduced, where A preset threshold is used to obtain the corrected topology map. Based on the corrected topology graph The connected components are re-partitioned to determine the subgroup structure, and the center position and size of each subgroup are calculated for subsequent group evaluation and task allocation.

[0065] When the target individual is resisting the USV, we consider its outward divergent motion characteristics during the escape process. We remove the group cohesion behavior term from its motion model and introduce a marine environmental disturbance term to describe the impact of external disturbances on its motion.

[0066] (2) Conduct multi-factor population assessment on the subgroups:

[0067] Based on the structural characteristics of each subgroup, a group evaluation index is constructed to comprehensively evaluate the task progress risk, group dispersion and size of each subgroup, and obtain the subgroup priority sequence and corresponding resource requirements.

[0068] The specific process includes:

[0069] Based on the corrected topology graph The target group is divided into subgroups to obtain a set of subgroups. And for each subgroup Extract its structural feature parameters, including subgroup size. subgroup center position And the spatial distribution characteristics of the subgroups;

[0070] Based on this, for each subgroup Constructing key assessment indicators for groups The indicators are used to comprehensively characterize the priority of encirclement and intervention for this subgroup, wherein the key assessment indicators The key evaluation indicators consist of a task advancement risk term, a group dispersion risk term, and a size weight term. The task advancement risk term characterizes the degree of deviation of the subgroup from the target region. The group dispersion risk term characterizes the spatial dispersion of individuals within the subgroup. The size weight term characterizes the number of target individuals in the subgroup. This is a weighted combination function of the above factors;

[0071] Based on the key evaluation indicators of each subgroup Sort all subgroups to obtain a subgroup intervention priority sequence;

[0072] At the same time, for each subgroup Based on its topological instability requirements, escape strength requirements, and operation space constraints, a resource requirement function is constructed to obtain the corresponding resource requirements. The resource requirements This is used to characterize the number of USVs required to execute the multi-encirclement USV task of this subgroup, and serves as a resource constraint for utility function allocation in step (3); and the subgroup priority sequence and corresponding resource requirements are also used. This serves as a unified decision-making basis for subsequent capture mission generation and multi-USV collaborative allocation.

[0073] (3) Generate and coordinate the allocation of capture tasks:

[0074] Under the shepherding drive-away mechanism, gathering and driving-away tasks are generated based on the subgroup priority and resource requirements. Based on the utility function, multiple USVs are assigned tasks, so that some USVs perform subgroup gathering tasks and the remaining USVs perform driving-away tasks from the main group to the target area.

[0075] Under the shepherding expulsion mechanism, key evaluation indicators for each subgroup are used. and corresponding resource requirements As a basis for task allocation, a utility function is constructed between USV and subgroups. The utility function is used to characterize the suitability of the i-th USV for performing the k-th subgroup encirclement task, wherein the utility function satisfies:

[0076] ;

[0077] In the formula: This represents the key evaluation metric for the k-th subgroup. This indicates the resource requirements of the subgroup. The utility function is used to characterize the spatial cost of the i-th USV performing the k-th subgroup task; the utility function varies with... Increase and increase, with Increase and decrease, and combine Assignment to subgroup The number of USVs captured is constrained;

[0078] According to the utility function All USVs involved in the capture operation are matched and assigned to subgroups, ensuring that each USV prioritizes selecting a subgroup with a higher utility value to perform the capture mission, and that the USV is assigned to the subgroup... The number of USVs captured and their resource requirements Matching; when the number of USVs captured is insufficient to meet the resource needs of all subgroups, priority is given to subgroups with higher key evaluation indicators, thereby achieving collaborative task allocation of multiple USVs to multiple subgroups.

[0079] (4) Perform dynamic coordinated encirclement and task updates:

[0080] Based on the dynamically updated group structure and its key evaluation results, the number of USVs performing the gathering and driving tasks is redistributed and adjusted. The task allocation of USVs remains unchanged for a preset time. After the time expires, the task allocation is uniformly updated based on the updated group structure and evaluation results to reduce system oscillations caused by frequent task switching.

[0081] Let the total number of USVs captured in the system be... The number of USVs that perform the subgroup convergence task is denoted as The number of USVs that performed the main group expulsion task is denoted as . ,satisfy Based on the current number of subgroups and the spatial distribution of the subgroups, the following... and Dynamic adjustments are made, specifically increasing the number of USVs performing the convergence task when the number of subgroups increases or the dispersion of subgroups increases. When the number of subgroups decreases or the main group structure stabilizes, reduce the number of USVs performing the swarming task. The number of USVs carrying out the expulsion mission will be increased accordingly. ;

[0082] At the same time, set the task retention time. During the specified time period, the task allocation results for each USV remain unchanged. After the time period ends, the task allocation is uniformly updated based on the updated target group topology and group criticality assessment results, thereby reducing system oscillations caused by frequent task switching and improving the stability of the multi-USV cooperative encirclement process.

[0083] The multi-USV cooperative trapping method proposed in this invention will be described in detail below with reference to the accompanying drawings.

[0084] I. Shepherding algorithm and its improvements.

[0085] 1. Shepherding algorithm.

[0086] like Figure 1 As shown, the shepherding algorithm guides and controls the target group by simulating the behavior of sheepdogs herding sheep. In this model, the target individuals and the individuals being herded have different behavioral characteristics, and their motion states are determined by their respective force models.

[0087] In the sheep model, the movement of an individual is influenced by a combination of factors, including inertia, group cohesion, inter-individual repulsion, avoidance by predators, and random disturbances. The combined force can be expressed as:

[0088] ;

[0089] In the formula: This represents the net force acting on an individual sheep at time t. Inertial force represents the influence of a sheep's previous state of motion on its current motion. It represents group cohesion and is used to characterize the tendency of individuals to gravitate towards the center of the group. This represents the repulsive force between target individuals, used to prevent collisions between sheep. This indicates the sheep's repulsive force against the sheepdog. Indicates random perturbation force. These represent the weighting coefficients for the stress term.

[0090] In the sheepdog model, the target group is regulated by executing driving and gathering behaviors, and their combined force can be represented as:

[0091] ;

[0092] In the formula: This represents the net force acting on the sheepdog at time t. This refers to driving force, used to propel a target toward a predetermined target area. This represents a gathering force, used to bring stray target individuals back to the vicinity of the main group. Indicates random perturbation force. Represents inertial force. This indicates the collision avoidance ability between sheepdogs. This indicates the collision avoidance capability between the sheepdog and the target individual. These represent the weighting coefficients for the stress term.

[0093] 2. Improve the shepherding algorithm.

[0094] To apply the shepherding algorithm to multi-USV cooperative trapping tasks and adapt it to the escape characteristics of USVs in the marine environment, the original shepherding model needs corresponding improvements. Traditional shepherding algorithms typically assume that target individuals exhibit agglomerative behavior, meaning they tend to move towards the group center and are driven away by a single global centroid. Figure 1 As shown. However, in actual adversarial scenarios, adversarial USVs tend to disperse and escape in order to break free from encirclement, making it difficult for the original model to accurately describe the target's behavior; at the same time, the original model does not consider the disturbance of the marine environment and the cooperative constraints between multiple USVs, making it difficult to directly apply to actual encirclement tasks.

[0095] To address the aforementioned issues, this invention improves the target individual model and the encirclement individual model in the shepherding algorithm.

[0096] Because the sheep model is used to guide the individual target model of the adversarial USV, it should not gather towards the center but rather escape to the surrounding areas. Therefore, the group cohesion term in the traditional shepherding algorithm is removed, so that the adversarial USV will no longer gather towards the group center during its movement. At the same time, an environmental disturbance term is introduced to describe the impact of ocean disturbance on its movement. The improved force model can be expressed as:

[0097] ;

[0098] In the formula: This represents the net force acting on USV at time t. Indicates the disturbance force of the marine environment. These represent the weighting coefficients for the stress term.

[0099] In the USV encirclement model guided by sheepdogs, an environmental disturbance term is also introduced to describe the impact of ocean disturbances on its motion. When the USVs perform the subswarming task, the gathering force in equation (2) Dominant role, driving force The weight is temporarily zero; conversely, when the target group structure is stable and enters the overall advancement phase, the driving force... It plays a dominant role. Therefore, the combined force for capturing the USV at different mission phases can be dynamically adjusted according to the mission status, which can be uniformly expressed as:

[0100] ;

[0101] In the formula: This represents the net force acting on the USV at time t during the capture. This represents the weight of the USV during the expulsion task at time t. This represents the weight of the USV during the convergence task at time t. This represents the net force acting on the sheepdog at time t. This refers to the herd drive force, used to propel a target group toward a predetermined target area. This represents the group's cohesive force, used to bring stray target subgroups back to the vicinity of the main group. Indicates random perturbation force. Represents inertial force. This indicates the collision avoidance ability between sheepdogs. This indicates the collision avoidance capability between the sheepdog and the target individual. These represent the weighting coefficients for the stress term. When the USVs are being pursued and the main swarm is being driven away, When the USV is being captured and its subgroup is being gathered, The weights are determined jointly by the topological structure of the target group, the multi-factor evaluation results of the subgroups, and the task allocation results.

[0102] Based on the improved model described above, the capture of USVs no longer employs the traditional shepherding rule that only targets a single outlier. Instead, it uses a subgroup structure modeling and collaborative task allocation mechanism to gather and drive away subgroups. When target individuals disperse, subsequent group structure modeling and collaborative task allocation mechanisms are used to intervene in and recover the subgroups, thereby completing the collaborative capture of multiple adversarial USVs.

[0103] 3. An improved encirclement task model was obtained.

[0104] Based on the improved shepherding algorithm described above, to achieve the cooperative encirclement task of multiple USVs against multiple adversarial USVs, the encirclement task model of this invention is constructed, as follows: Figure 2 As shown.

[0105] In the encirclement task model, the encircling USV is the encircling party, and the opposing USV is the escaping party. Let the number of encircling USVs be m, and the number of opposing USVs be n, where... Furthermore, in practical applications, m < n is allowed to enable the cooperative capture of multiple adversarial USVs with fewer USVs captured.

[0106] The USVs being captured move according to the improved force model and are guided to the target area by driving behavior. The USVs also move according to the improved force model and perform escape behavior. During the capture process, the USVs are not captured by the traditional individual capture rules, but by the capture task based on subgroups. Subgroups are intervened and recovered through subsequent group structure modeling and collaborative task allocation mechanisms.

[0107] When the USV being surrounded detects an enemy USV, it starts from its initial position. The vehicle starts moving towards the target USV. When it enters the effective range R of the target USV, it adjusts to the driving position along a trajectory with the target USV as the center and R as the radius. The driving position is located on the side away from the target area in the direction of the line connecting the target area and the anti-USV, so that the anti-USV is located between the encirclement USV and the target area, thereby achieving the driving of the anti-USV toward the target area;

[0108] Let G be the center of the target region, and let the radius of the target region be... When all opposing USVs enter a region centered on G with a radius of... When the USV is within the designated area, it is determined that the adversary USV has been successfully captured.

[0109] To avoid collisions between USVs during the capture process, the effective range R should be greater than the minimum safe distance between the USVs during the capture, i.e. , where L represents the length of the encirclement USV;

[0110] During the multi-target encirclement process, by continuously driving away multiple adversarial USVs and combining them with the subsequent multi-USV collaborative task allocation mechanism, a limited number of encirclement USVs can be scheduled and switched between different subgroups, achieving orderly encirclement of multiple adversarial USVs. This reduces the number of USVs required to be encircled while ensuring the encirclement effect, thereby improving the overall resource utilization efficiency of the system.

[0111] II. Collaborative task allocation based on group evaluation and utility function.

[0112] In a multi-USV encirclement mission, let the number of USVs to be encircled be m and the number of USVs to be countered be n. In the actual encirclement process, encircling USVs not only requires selecting the target, but also requires coordinated scheduling based on the structural state of the target group. Otherwise, it is easy for multiple USVs to act on the same target or for some targets to be left unattended.

[0113] To address the aforementioned issues, this invention, based on the subgrouping results obtained in step (2) and the group criticality evaluation index (see step (2) in the invention content section), performs collaborative task allocation for multiple USVs. Assume the target group is divided into a set of subgroups. For each subgroup Its key evaluation indicators have been obtained. and resource requirements .

[0114] In the method of this invention, a utility function is constructed between the capture USV and the subgroup to characterize the suitability of the i-th capture USV for performing the capture task of the k-th subgroup. The utility function is defined as follows:

[0115] ;

[0116] In the formula: Indicates the i-th USV and its subgroup. The utility values ​​between Indicates the key evaluation indicators for subgroups. Subgroup resource requirements, This represents the distance between the captured USV and the center of the subgroup.

[0117] In the specific task allocation process, the utility function varies with the key evaluation index of the subgroup. The utility function increases with increasing distance to ensure that the key subgroup has priority in obtaining capture resources; the utility function increases with increasing distance. The resource requirements decrease as the number of USVs increases, in order to reduce the maneuver costs associated with long-range missions; Used to constrain assignment to subgroups The number of USVs to be captured is determined so that subgroups with larger subgroup sizes, higher dispersion, or higher risk levels receive a corresponding number of USVs to be captured.

[0118] Based on the utility function, such as Figure 3 As shown, all USVs involved in the capture are matched and assigned to subgroups, ensuring that each USV prioritizes selecting a subgroup with a higher utility value to perform the capture mission; when a subgroup has already met its resource requirements... When the number of USVs captured matches the target number, the remaining captured USVs are no longer allocated to that subgroup but are redirected to other subgroups or the main group for herding tasks. When the number of captured USVs is insufficient to meet the resource needs of all subgroups, the task is determined according to the subgroup's criticality assessment indicators. The priority order is used to allocate resources, thereby achieving multi-target collaborative encirclement under limited encirclement resources.

[0119] Furthermore, during the encirclement process, the task allocation for encircling USVs is dynamically adjusted as the subgroup structure changes, enabling the encircling USVs to switch and coordinate between different subgroups. This allows for the continuous encirclement of multiple adversary USVs even when the number of encircling USVs is less than the number of adversary USVs.

[0120] III. Encirclement and Capture Implementation Examples

[0121] In the multi-USV encirclement experiment, three encircling USVs and six adversary USVs were deployed. The initial positions of encircling USV0, USV1, and USV2 were (180, 60), (130, 60), and (80, 60), respectively. The initial positions of adversary USV0, USV1, USV2, USV3, USV4, and USV5 were (220, 350), (200, 310), (-90, 350), (150, 310), (-50, 310), and (140, 350), respectively. The target area was set at (0, 90) with a radius R of 60m. Encirclement was considered successful when all adversary USVs entered within 60m of the target area. The initial positions show that the six adversary USVs were dispersed above the target area, forming two local subgroups on the left and right sides, respectively. This arrangement is suitable for verifying the cooperative capability of the method of this invention in a multi-subgroup dynamic encirclement scenario.

[0122] To enhance the experimental representation of the algorithm's characteristics, the adversarial USV was designed to exhibit escape behavior and random perturbations during its movement. This caused the target group to dynamically split and merge during the encirclement process, resulting in a multi-subgroup structure. The maximum speed of the encircling USV was slightly higher than that of the adversarial USV, ensuring that it had the ability to drive away but not completely suppress the target.

[0123] Experiments using a shepherding-based encirclement algorithm with topology sensing and dynamic task allocation yielded the following encirclement trajectories for multiple USVs: Figure 4 As shown. By Figure 4 It can be seen that in the initial stage of the experiment, the adversarial USVs were located above the target area and exhibited two local subgroups, one on the left and one on the right. After the experiment began, the adversarial USVs continued to spread outward due to escape behavior and random perturbations, with some adversarial USVs maneuvering to the upper left and upper right, further expanding the spatial distribution of the target group. At this point, the USV capture operation did not pursue individual adversarial USVs one by one, but instead identified local subgroups based on changes in the topology of the target group and prioritized intervention in subgroups with a spreading trend.

[0124] In the early stages of the encirclement process, USV0 primarily maneuvers towards the outer right subgroup, restricting the opposing USVs on the right; USV2 moves towards the left subgroup, converging the scattered targets on the left; and USV1, positioned in the central area, drives and connects the subgroups, thus forming a coordinated encirclement pattern combining local convergence and overall deflection. Figure 4 The trajectory changes show that the three USVs engaged in the pursuit acted in different spatial areas during the pursuit, thus avoiding the problem of multiple USVs pursuing the same target at the same time, which would result in other targets being left unattended.

[0125] Before the encirclement process reached approximately 1000 steps, the opposing USVs were still in a strong escape phase, with the target trajectory exhibiting multiple turns and localized diffusion. As the encirclement progressed, the encircling USVs gradually completed the confinement and recovery of the subgroup structure, causing the left and right subgroups to converge towards the central area. Subsequently, the encircling USVs readjusted their task allocation based on the updated group structure. The encircling USVs that were originally tasked with subgroup convergence gradually shifted to the task of driving the main group, forming a advancing structure towards the target area together with other encircling USVs, causing the overall target group to gradually move towards the target area.

[0126] Depend on Figure 5 It can be seen that during the encirclement process, the distance between all USVs and the center of the target area showed obvious phased changes. In the initial stage of the experiment, the average distance between the combat USVs and the center of the target area was about 280m, which then gradually increased and remained above 300m, indicating that the combat USVs had a clear escape trend in the early stage; at the same time, the average distance between the encirclement USVs and the center of the target area increased rapidly from about 130m to about 400m, indicating that the encirclement USVs were actively maneuvering towards the periphery of the target group in order to establish a position for gathering local subgroups and driving away the main group.

[0127] After approximately 1000 steps in the encirclement process, the average distance between the combatant USVs and the center of the target area began to gradually decrease, indicating that the encircling USVs had completed the initial adjustment of the subgroup structure and gradually established an effective driving direction. After approximately 2000 steps, the rate of decrease in the average distance of the combatant USVs accelerated significantly, and the target group as a whole was continuously pushed towards the target area. Finally, between approximately 2800 and 3100 steps, the combatant USVs as a whole entered the vicinity of the target area's radius, and the encirclement mission was completed.

[0128] Further comparing with traditional shepherding methods, under the same initial conditions, without topological subgroup identification and group criticality assessment, the USVs being captured tend to be chased according to local distance or single-target rules, leading to a lack of unified coordination between left and right subgroups. Some adversarial USVs may continue to escape outwards, thus prolonging the capture time or even causing capture failure. In this embodiment, the system can identify local subgroups based on the dynamic topological structure of the target group and dynamically adjust the task allocation for capturing USVs according to the subgroup state. This allows a limited number of USVs to switch between subgroup convergence and main group driving, thereby maintaining the continuity and stability of the capture process.

[0129] In summary, the encirclement method proposed in this invention, in a mission involving 3 encircling USVs against 6 opposing USVs, can identify the local splitting state of the target group through topological sensing and achieve coordinated control of subgroup recovery and main group expulsion by combining group criticality assessment and dynamic task allocation. Experimental results show that even when the number of encircling USVs is less than the number of opposing USVs, this invention can still stably complete multi-target encirclement tasks, demonstrating the advantage of a small number of encircling USVs in the coordinated encirclement of multiple opposing USVs.

Claims

1. A multi-USV cooperative trapping method based on topology sensing and evaluation using the shepherding algorithm, characterized in that, Includes the following steps: (1) Perform topological modeling of the target group: A connectivity topology graph is constructed based on the perceptual relationships between target individuals. By identifying and correcting the weak connections in the topology graph, the dynamic division of the main group and subgroups is achieved. At the same time, the structural feature parameters of each subgroup are extracted, including subgroup size, subgroup center location and spatial distribution characteristics. (2) Construct a population criticality assessment mechanism based on the aforementioned subgroup structural characteristics: A comprehensive assessment of the task progress risk, group dispersion and size of each subgroup is conducted to obtain the intervention priority and corresponding resource requirements of each subgroup, thereby achieving unified decision-making on the selection of intervention targets and resource allocation for each subgroup. (3) Generate capture tasks based on the priority and resource requirements of each subgroup, and coordinate the task allocation of multiple USVs based on the utility function, so that some USVs perform subgroup capture tasks and the rest of USVs perform main group drive-off tasks, thereby realizing the parallel capture of discrete subgroups and the continuous advancement of the main group to the target area. (4) The number of USVs performing the gathering and driving tasks is dynamically adjusted according to the changes in the target group status. When the number of subgroups decreases or the subgroups are successfully gathered, the USVs performing the gathering task are switched to the driving task to achieve dynamic redistribution of collaborative resources.

2. The multi-USV cooperative trapping method based on the shepherding algorithm for topology sensing and evaluation according to claim 1, characterized in that, The specific process of step (1) is as follows: Construct a connectivity topology graph based on the perceptual relationships between target individuals, and represent the set of target individuals as a set of nodes. When any two target individuals and The Euclidean distance between them satisfies At that time, establish undirected edges between nodes. Thus, an undirected graph is obtained. ,in, Indicates the distance between target individuals. Indicates the preset sensing range. Represents the set of edges; In the undirected graph Connectivity analysis is performed to obtain all connected components as the subgroup partitioning results, and the connectivity degree of each node is calculated. Used to characterize the adjacency relationship of nodes; Based on this, a stability analysis is performed on the connectivity relationships in the topology graph, and edges that increase the number of connected components after deletion are identified as the bridge edge set. It is used to characterize weak connections between subgroups; and for undirected edges Define stability weights: ； In the formula: Representing an edge Stability weights, and These are the weighting coefficients; When performing stability analysis on the connectivity relationships in the topology graph, it is clarified that the connectivity topology graph is a perceptual connectivity graph consisting of target individuals as nodes and perceptual relationships between individuals as edges, and the edges correspond to stability weights; for those satisfying the condition of belonging to a preset set of bridge edges... Or the stability weights satisfy Edges are deleted or have their weights reduced, where A preset threshold is used to obtain the corrected topology map. ; Based on the revised topology graph Re-partition the connected components, determine the structure of the main group and subgroups, and calculate the center position of each subgroup. and subgroup size ,in Describes the k-th subgroup. Indicates the node position. Indicates the center position of the subgroup. This indicates the subgroup size, used for subsequent group assessment and multi-USV collaborative capture decisions.

3. The multi-USV cooperative trapping method based on the shepherding algorithm for topology sensing and evaluation according to claim 1, characterized in that, The group criticality assessment mechanism in step (2) is: Based on the subgroup partitioning results obtained in step (1), assume that the target population is divided into a set of subgroups. And extract the structural feature parameters of each subgroup, including the subgroup size. subgroup center position And the spatial distribution range of the subgroup; Based on this, for each subgroup Constructing key assessment indicators for groups The key evaluation index is used to characterize the intervention priority of the subgroup in the encirclement task. This index is determined by the task progress risk, the degree of group dispersion, and the size weight, and is obtained through a weighted combination function. ； In the formula: This represents the key evaluation metric for the k-th subgroup. This represents the risk factor for mission progress, used to characterize the degree of deviation of the subgroup from the target region. Indicates the degree of dispersion of a subgroup, used to characterize the spatial dispersion of individuals within the subgroup. Indicates the size of the subgroup. , , These are the weighting coefficients; Based on the key evaluation indicators of each subgroup Sort all subgroups to obtain a subgroup intervention priority sequence; At the same time, for each subgroup Based on its topological instability requirements, escape strength requirements, and operation space constraints, a resource requirement function is constructed to obtain the corresponding resource requirements. The resource requirement function is used to characterize the number of USVs required to perform the subgroup capture task; the resource requirement function satisfies: ； In the formula, This represents the topological instability demand term of the k-th subgroup, used to characterize the fragility of the connections within the subgroup. This represents the escape intensity of the k-th subgroup, used to characterize the number of USVs required to capture all individuals in the subgroup due to their angular deviation relative to the target point. This represents the operation space-constrained requirement of the k-th subgroup, used to characterize the requirement for the number of USVs to be captured due to obstacles, boundaries, or traversable space limitations.

4. The multi-USV cooperative trapping method based on the shepherding algorithm for topology sensing and evaluation according to claim 1, characterized in that, The process of collaboratively allocating tasks among multiple USVs for the encirclement operation based on group multi-factor evaluation and utility function in step (3) is as follows: Based on the subgroup priority and resource requirements obtained in step (2), a utility function is constructed between the capture USV and the subgroup to characterize the suitability of the i-th capture USV for performing the k-th subgroup capture task. The utility function is expressed as: ； In the formula: This represents the utility value between the i-th USV and the k-th subgroup. Indicates the key evaluation indicators for subgroups. This represents the resource requirement of the k-th subgroup. This represents the distance between the i-th USV used for encirclement and the center of the subgroup; The utility function is used to comprehensively characterize the impact of subgroup priority, resource requirements, and spatial costs on task allocation results, and through resource requirements... Constrain the number of USVs to be captured and assigned to the corresponding subgroup; Based on the utility function, all capture USVs are matched with subgroups, so that each capture USV prioritizes selecting the subgroup with the larger utility value to perform the capture task, and the allocation result satisfies the resource requirement constraints of each subgroup, thereby realizing the collaborative task allocation of multiple USVs to multiple subgroups.

5. The multi-USV cooperative trapping method based on the shepherding algorithm for topology sensing and evaluation according to claim 1, characterized in that, The dynamic adjustment process in step (4) is as follows: During the collaborative capture of multiple USVs, the number of USVs to be captured is dynamically adjusted based on the number of subgroups, subgroup size, and spatial distribution of the subgroups; let the total number of USVs captured in the system be... The number of USVs performing the subgroup convergence task is The number of USVs performing the main group expulsion mission was ,satisfy ; When the number of subgroups increases, the size of subgroups becomes larger, or the dispersion of subgroups increases, the capture USVs should be prioritized for subgroup convergence tasks. Increase the number of USVs used to perform subgroup merging tasks when the number of subgroups decreases, subgroups are gradually merged, or the main group structure becomes stable. The released USVs will be redistributed to the main swarm drive-off mission, making... The corresponding increase enables coordinated resource scheduling for subgroup recovery and main group advancement; Meanwhile, to avoid frequent task assignment switching caused by changes in the target group's state, after each task assignment, the task assignment results for each USV to be captured remain unchanged within a preset time range, so that each USV continues to execute the currently assigned gathering or driving task; after the time range ends, task assignment and resource adjustment are redistributed based on the updated subgroup structure and the group criticality assessment results, thereby reducing system oscillations and improving the stability of the multi-USV collaborative capture process.

Images