Group mixed type hunting control method based on space-time alternating layered leader structure

Abstract

The invention relates to a group mixed type hunting control method based on a space-time alternate layered leader structure, and belongs to the technical field of robot groups. The task execution efficiency and robustness in a space-time dynamic environment are optimized in combination with bionic principles such as a space-time alternate layered leader structure, informed leader and a hybrid hunting control strategy of biological population behaviors. According to the method, space-time alternate layered leader structure characteristics existing in biological group behaviors are applied to a control method of a robot group, the cooperative behaviors of the robot group are optimized, and a bionic group hybrid hunting control method is provided, so that more robust, efficient and flexible cooperation of the robot group in the dynamic hunting process is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot groups, and in particular relates to a group hybrid capture control method based on a time-space alternating hierarchical leadership structure. Background Art

[0002] With the rapid development of robot swarm technology, the demand for dynamic capture missions in heterogeneous dynamic environments (such as complex and unknown terrain, communication barriers, incomplete information, and uncertain target movement) is becoming increasingly urgent. However, existing control technologies still rely primarily on centralized global scheduling or distributed autonomous decision-making, making it difficult to achieve a dynamic balance between robustness and global optimization capabilities in scenarios with incomplete information, communication barriers, and complex terrain. Centralized control relies on a central node for global decision-making and task allocation. For example, the machine learning-based swarm control method in patent CN118642377B optimizes multi-robot collaboration through a centralized algorithm. However, this method has extremely high requirements for communication bandwidth and stability, lacks flexibility and robustness, and is prone to system crashes due to single point failures in dynamic capture environments with blocked communication. Distributed control emphasizes individual autonomous decision-making and collaborative interaction between robots. For example, the distributed multi-robot synchronous swarming control algorithm in patent CN109164826A achieves connectivity in the network topology during swarm evolution through local information exchange. While robust, these approaches suffer from limited adaptability and global coordination in complex environments with incomplete information, leading to efficiency bottlenecks in dynamic task allocation. While hybrid control approaches theoretically combine the advantages of distributed and centralized approaches, there is currently a lack of patented technologies for complex terrestrial environments, and research and application of biomimetic hybrid control methods is lacking.

[0003] Faced with a complex and ever-changing environment, in order to improve decision-making efficiency and respond to emergencies, biological groups (such as flocks of sheep) exhibit a unique spatiotemporal alternating hierarchical leadership structure in a complex natural environment: group members dynamically switch between the roles of leader and follower through random processes, forming a hierarchical decision-making mechanism in the time dimension, which not only ensures local autonomous response but also achieves global motion coordination. This bionic feature reveals the natural rationality of the hybrid roundup control strategy, that is, through the integration of centralized global optimization and distributed autonomous decision-making, taking into account both system robustness and task efficiency. However, the existing technology has insufficient bionic application of biological behavior, and its spatiotemporal dynamic adaptability is weak; when faced with an unknown environment, its coordination is poor, making it difficult to complete collaborative exploration. The patent field still lacks hybrid roundup control methods for unstructured scenarios such as complex and unknown terrain, dynamic roundups, incomplete information, and limited communication. Summary of the Invention

[0004] (1) Technical issues to be resolved

[0005] The technical problem to be solved by the present invention is: how to optimize the collaborative behavior of the robot group to achieve more robust, efficient and flexible collaboration of the robot group in the dynamic capture process.

[0006] (2) Technical solution

[0007] In order to solve the above technical problems, the present invention provides a group hybrid roundup control method based on a time-space alternating hierarchical leadership structure, comprising the following steps:

[0008] According to the characteristics of biological swarm hunting behavior and spatiotemporal alternating hierarchical leadership structure, the dynamic hunting process of the robot swarm is divided into different stages TP on the time scale, namely, searching TP1, tracking TP2 and hunting TP3. For these three stages, according to the needs of robot swarm collaboration, a hybrid hunting control strategy combining centralized and distributed methods is adopted in the spatial dimension to achieve hybrid hunting control of the hunting target.

[0009] (3) Beneficial effects

[0010] The present invention has the following advantages:

[0011] (1) This invention draws on the behavioral patterns of biological groups in nature. Based on the biological behavior and movement characteristics of the group, the group capture process is divided into multiple stages, including search, tracking, and capture. Each stage has its own specific group behavior control mode and task objectives, allowing the robot group to complete the capture task in a more orderly manner. This division method enables the robot group to more specifically allocate and adjust tasks, which will effectively improve the collaborative efficiency of the robot group and the success rate of target capture.

[0012] (2) The present invention innovatively proposes a hybrid capture control method, which deeply integrates the advantages of centralized and distributed control, enhances the flexibility of dynamically adjusting the control strategy according to task requirements, overcomes the defects of the single-mode control method in the dynamic capture process, realizes the optimization of resource allocation and energy consumption, and improves the capture efficiency. Centralized control realizes the overall coordination and task allocation of the robot group, provides global information to ensure the orderly progress of the capture task; under distributed control, each robot can make autonomous decisions and actions based on the local information it perceives, giving individual autonomy and enhancing the flexibility and adaptability of the system. In the process of group capture, the hybrid capture control method realizes group collaboration through information sharing and local decision-making, which can more effectively and flexibly respond to communication obstruction and complex and changeable capture environment under complex terrain conditions, and will improve the reliability and fault tolerance of the system.

[0013] (3) The present invention combines the time-space alternating hierarchical leadership structure in group biobehavior with the dynamic process of robot group capture. By constructing a hierarchical leadership structure, different control methods are adopted in different movement time stages according to the needs of dynamic capture of the group's spatial distribution. The upper-level centralized control method generates the global capture path and task priority; the lower-level distributed control method perceives the environment in real time and dynamically switches the "temporary leader" role to trigger the adjustment of the local capture strategy. The present invention has made a breakthrough in solving the problem of task continuity and response hysteresis in dynamic target tracking in the scenario of communication interruption.

[0014] (4) The present invention adopts a leadership design based on informed strategy in group animal leadership behavior, which enables the leader to formulate a more accurate and reasonable capture strategy based on its own experience and the information and environmental perception of the informed followers, and flexibly adjust the robot's movement trajectory and speed according to actual conditions, giving full play to the leader's decision-making ability and the information advantages of group members, improving the comprehensiveness and accuracy of the robot group's decision-making, and ensuring the efficient and smooth completion of the capture task. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is the overall flow chart of the robot group roundup of the present invention;

[0016] Figure 2 This is a flow chart of the robot group search phase of the present invention;

[0017] Figure 3 Schematic diagram of the leader selection scheme in the tracking phase of the present invention;

[0018] Figure 4 This is a schematic diagram of the control decision principle of the robot group capture stage of the present invention;

[0019] Figure 5 Schematic diagram of the robot's perception area. DETAILED DESCRIPTION

[0020] In order to make the purpose, content and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below with reference to the accompanying drawings and examples.

[0021] This method combines biomimetic principles such as the spatiotemporal alternating hierarchical leadership structure, informed leadership, and hybrid roundup control strategies of biological swarm behavior to optimize task execution efficiency and robustness in spatiotemporal dynamic environments. This method applies the spatiotemporal alternating hierarchical leadership structure found in biological swarm behavior to the control of robotic swarms, optimizing their coordinated behavior and proposing a biomimetic hybrid roundup control method. This method achieves more robust, efficient, and flexible collaboration among robotic swarms during dynamic roundup processes.

[0022] The present invention provides a hybrid capture control method for a robot group based on a spatiotemporal alternating hierarchical leadership structure, comprising the following steps:

[0023] The latest research by Gómez Nava et al. has discovered a hierarchical leadership structure that alternates time and space. In this structure, the group exhibits a leader on a time scale at different stages of movement, guiding the group's movement. The experimental model also found that the group randomly alternates between the roles of leader and follower (intermittent collective strategies). Based on the characteristics of biological group hunting behavior and the hierarchical leadership structure that alternates time and space, the dynamic hunting process of the robot group is divided into different stages (TP) on a time scale, such as Figure 1 As shown in the figure, it includes three stages: searching (TP1), tracking (TP2), and capturing (TP3). For these three stages, a hybrid capture control strategy combining centralized and distributed capture control is adopted in the spatial dimension to achieve hybrid capture control of the capture target based on the needs of group collaboration.

[0024] In the search phase, the robot group uses a distributed control strategy to search and capture the target. The process is as follows: Figure 2 All robots receive external environmental information, obtain environmental parameters, and form a terrain model to autonomously search for the target. When a robot finds the target, the entire robot group enters the tracking phase. If no target is found, the search continues.

[0025] In the tracking phase, the robot group uses a centralized strategy to track the target. There are two modes for selecting the leader (either one can be selected), including selecting the leader based on visual information (leader_visual) and selecting the leader based on distance information (leader_distance). The specific process is as follows: Figure 3 shown.

[0026] (1) The process of selecting the leader leader_distance based on distance information is as follows: at a certain moment, after the robot finds the target, it synchronizes the target position information to all robots, calculates the distance xi_target between each robot and the target, calculates the minimum value min(xi_target) of the distance between all robots and the target, and defines the robot closest to the target in the robot group as the leader leader_distance.

[0027] (2) The process of selecting the leader leader_visual based on visual information is as follows: at a certain moment, record the time point ti_target when the robot detects the encirclement target, calculate the minimum value min(ti_target) of the time points detected by all robots, and define the robot in the robot group that first detects the encirclement target in the visual information as the leader leader_visual.

[0028] After determining the leader (leader_visual or leader_distance), the leader commands the robot swarm to begin tracking based on the direction of the target. During the tracking process, the leader calculates the distance xi_target between itself and the target. When xi_target is less than or equal to the preset tracking distance threshold_distance, the leader initiates the tracking command and the robot swarm enters the tracking phase. Otherwise, the tracking phase continues.

[0029] In the capture phase, the robot group adopts a hybrid capture control strategy that combines centralized and distributed control to capture the capture target. In this capture phase, the robot group is evenly divided into a groups (the number of members in each group = the total number of robots / the number of groups a), and each group of robots carries out the capture process according to the distributed control strategy. There is no central node in the distributed control strategy. All individuals are autonomous and have certain decision-making capabilities. They can respond quickly to environmental changes without relying on central control. In this capture phase, the leader adopts an informed leadership strategy, that is, if there is an informed follower in each group of members of the robot group (a robot that can perceive local perception information (environment, direction, target, etc.) is an informed follower), the local perception information perceived by the individual can be transmitted to the leader. The leader will combine the local perception information transmitted by the informed follower to make a global decision on the direction and a real-time judgment of the capture status. The control decision diagram is shown in the figure below. Figure 4 When judging the capture status, the leader needs to determine whether the capture target has been successfully captured based on the degree of closure of the encirclement. If the capture is successful, the robot group enters the next round of capture target search phase. If not, the capture continues.

[0030] In the above three stages, the robot needs to calculate the expected direction. The specific calculation methods for each stage are introduced below.

[0031] Let the serial number of a robot in the robot group be i, and n represents the number of robots in the robot group, i = 1, 2, 3...n. At each time point t, the position vector corresponding to a robot i is x i (t), the direction vector is v i (t), the speed maintained is s iIn each time step, the robot is based on the position of the neighboring robots in the three sensing areas (attraction area Az, repulsion area Rz and direction area Oz) (e.g. Figure 5 As shown) to calculate the expected direction (desired direction) d i (t). The group interaction network describes the topological structure and dynamic interaction rules of information transmission between individuals in a robot group. The social connection matrix is a mathematical abstract representation of the group interaction network, which usually quantifies the connection strength or interaction weight between individual robots in matrix form. Therefore, the connection strength between two robots is predefined, S ij (t) is the group interaction network, represented by the social connection matrix between individuals, S ij The value of (t) represents the relative influence of robot j on robot i. If robot i can be influenced by robot j, then S ij (t)≠0, otherwise S ij (t) = 0. In different robot group motion stages, the structure of the group interaction network can be fixed or set to change over time.

[0032]

[0033] Among them, for any i, j, S ij When all (t) are greater than 0, the robot group is in the search phase, and the group interaction network is a fully connected network; when S ij When the (t) part is greater than 0, the robot group is in the tracking or encirclement stage, and the group interaction network is a partially connected network, as shown in formula (1).

[0034] To prevent collisions between individual robots, an attraction zone Az, a repulsion zone Rz, and a direction zone Oz are defined for each robot. Within the attraction zone Az, individual robots align their orientation with that of adjacent robots, preventing the robot group from dispersing. Within the repulsion zone Rz, individual robots actively move away from adjacent robots to avoid collisions. Within the direction zone Oz, individual robots align their orientation with that of the group.

[0035] The first area is the exclusion zone Rz, which maintains the robot's "individual space". If a robot j appears in the Rz area of robot i, at each time step Δt (representing the movement direction at the next moment), robot i calculates the expected direction d based on the relationship between it and the adjacent robots in the three perception areas. i (t+Δt|TP), in different stages TP, corresponds to the position, speed and other values of the corresponding stage. The specific calculation formula is based on the classic couzin aggregation motion model and introduces the group interaction network s ij (t) and noise ε i(t) is used to reflect the changes in the connection characteristics between groups at different movement stages and the influence of noise during the movement process. Therefore, the expected direction of the next step of robot i is defined as the following formula (2):

[0036]

[0037] Among them, ε i (t) is the random noise of robot i, ε i (t)~N(0,σ 2 ), adding random noise in the direction to simulate the impact of noise on the direction caused by complex terrain can improve the robustness of the system.

[0038] This expected direction ensures that no collision will occur when the particles converge within t+Δt. To ensure that all robots do not collide, robot j should move close to other robots and enter their attraction zone Az, with its direction facing the direction zone Oz.

[0039] When using the distributed control strategy to search for the target in the search phase, it is necessary to calculate the expected direction and adjust the direction of the next moment according to the calculated expected direction. In the search phase, the regional attraction intensity α1 of the attraction zone Az and the movement direction of the direction zone Oz will affect the expected direction of the robot i. At this time, the expected direction of the next step is defined as follows:

[0040]

[0041] Where α1 is the regional attraction strength of the attraction zone Az in the search phase, α1∈(0,1).

[0042] In the tracking phase, a centralized control strategy is used to track the target. The robot group is organized in a structured leader manner. The leader determines the movement direction of the robot group and defines k i is the level of individuals in the robot group, where k i =1 represents the leader, k i =2 represents the follower, and the leader's selection plan is as follows Figure 3 In addition, based on whether the leader is fixed during the tracking phase and whether it changes dynamically over time, the leader's fixedness is divided into the following three cases:

[0043] (1) In this stage, the leader is fixed, then k i fixed;

[0044] (2) In this stage, the leader is randomly fixed, then k i (t) changes over time. At time t, a robot is randomly selected from the group and defined as the leader;

[0045] (3) In this stage, leaders change along the time and space dimensions, and the fixed methods are different at different times.

[0046] In the tracking phase, the regional attraction intensity α2 of the attraction zone Az and the movement direction of the direction zone Oz will affect the expected direction of robot i. In the tracking phase, enhancing the group aggregation can increase the attraction intensity. α2>α1, at this time, the leader robot i (k i =1) The expected direction of the next step is defined as shown in the following formula (4):

[0047]

[0048] Normalization:

[0049]

[0050] Since the individual's movement direction needs to be adjusted according to the leader's direction during the tracking phase, the leader's direction is dominant at this time. In a robot group, except for the leader, the expected direction of other robots must not only consider the situation of adjacent robots within the sensing range, but also the leader's movement direction. At this time, the preference intensity for the leader's movement direction is W. The larger the value of W, the greater the influence of the leader's movement direction on the next movement direction of robot i. At this time, the follower robot i (k i =2) The expected direction of the next step is defined as shown in the following formula (6):

[0051]

[0052] Where W is the follower robot's preference for the leader's movement direction, W>1. k＝1 (t) represents the speed of the leader robot at time t;

[0053] In the capture phase, a hybrid capture control strategy is used to capture the target. The robot group is organized according to the informed leadership method and consists of a leader and followers. The group's movement direction is dominated by the leader and is affected by the real-time local perception information of other robots. The leader fixation method in the capture phase is the same as that in the tracking phase. There are three leader fixation methods (leader fixed, leader randomly fixed, and leader changes with time and space dimensions). Further strengthening the group aggregation in the capture phase can increase the attraction strength. That is, the regional attraction strength of the attraction zone Az in the capture phase is set to α3, and α3>α2. At this time, the expected direction of robot i in the next step is defined as follows:

[0054]

[0055] Normalizing formula (7) yields formula (8):

[0056]

[0057] In order to simulate the informed leader model, for some followers, there is a preference direction g based on “knowledge” i , the leader will receive the preferred directions of the informed followers and integrate them to assist in decision-making and achieve more efficient decision-making. Since the expected direction is jointly affected by the directions of the leader and the informed followers, the expected directions of the leader and the informed followers are weighted according to the weight w. Among them, w determines the degree of participation of the informed leader information in the direction decision. The smaller w is, the higher the degree of participation is. In this roundup phase, the leader robot i(k i =1, the expected direction of the leader robot's next step is shown in the following formula (9):

[0058]

[0059] Where w is the leader’s preference strength for the direction of the informed follower, w∈(0,1). w=0 means the leader is completely

[0060] Relying on the preferences of informed followers, w = 1 means that the leader completely relies on its own preferences and is not affected by the preferences of informed followers. w is between 0 and 1, which means that the leader partially combines the preferences of informed followers to make decisions. At this time, the follower robot i(k i =2) The expected direction of the next step is defined as shown in the following formula (10), where W has the same meaning as in formula (6);

[0061]

[0062] Judgment criteria for successful capture: The closure of the encirclement is used as an indicator to measure the integrity of the encirclement in the robot group capture task. The higher the closure, the more complete the encirclement and the lower the possibility of the captured target escaping. The encirclement closure formula (11) is shown below. Let Q be the encirclement closure, LC be the actual encirclement circumference, and LC be the circumference of the ideal completely closed encirclement. The closure Q can be expressed as:

[0063]

[0064] Formula (12) assumes that the circumference of the ideal encirclement is known, which can be the circumference of a perfect circle around the target. Assuming R is the radius of the encirclement, the calculation of LC is shown in Formula (12):

[0065] LC=2πR (12)

[0066] A threshold Q_threshold of the closed boundary of the robot group is set. The leader judges that when all robots form a closed boundary, that is, when the closure degree Q is greater than or equal to the defined closed boundary threshold Q_threshold (Q≥Q_threshold), the target is considered to be completely surrounded inside, and the capture is successful.

[0067] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A group hybrid roundup control method based on a time-space alternating hierarchical leadership structure, characterized in that: The following steps are involved: According to the characteristics of biological swarm hunting behavior and spatiotemporal alternating hierarchical leadership structure, the dynamic hunting process of the robot swarm is divided into different stages TP on the time scale, namely, searching TP1, tracking TP2 and hunting TP3. For these three stages, according to the needs of robot swarm collaboration, a hybrid hunting control strategy combining centralized and distributed methods is adopted in the spatial dimension to achieve hybrid hunting control of the hunting target.

2. The method according to claim 1, wherein During the search phase, the robot swarm uses a distributed control strategy to search for the target. All robots in the robot swarm receive external environmental information, obtain environmental parameters, and form a terrain model to autonomously search for the target. When a robot finds the target, all robots in the robot swarm enter the tracking phase. If no target is found, all robots continue to search.

3. The method according to claim 2, wherein In the tracking phase, the robot group adopts a centralized strategy to track the target. In particular, one of the following two modes is used to select the leader: leader_visual based on visual information and leader_distance based on distance information; The process of selecting a leader based on distance information is as follows: at a certain moment, after a robot finds a target, it synchronizes the target location information to all robots, calculates the distance xi_target between each robot and the target, calculates the minimum value min(xi_target) of the distances between all robots and the target, and defines the robot closest to the target in the robot group as the leader leader_distance; The process of selecting the leader leader_visual based on visual information is as follows: at a certain moment, record the time point ti_target when the robot detects the encirclement target, calculate the minimum value min(ti_target) of the time points detected by all robots, and define the robot that first detects the encirclement target in the visual information in the robot group as the leader leader_visual; After the leader is determined, it commands other robots to start tracking according to the direction of the target. During the tracking process, the leader calculates the distance xi_target between itself and the target. When xi_target is less than or equal to the preset tracking distance threshold_distance, the leader initiates the tracking command and the robot group enters the tracking phase. Otherwise, the tracking continues.

4. The method according to claim 3, wherein In the capture phase, the robot swarm adopts a hybrid capture control strategy that combines centralized and distributed control to capture the capture target. The robot swarm is evenly divided into a groups, and each group of robots carries out the capture process according to the distributed control strategy. When adopting the distributed control strategy, all individuals are autonomous, have certain decision-making capabilities, and can respond autonomously to environmental changes. The leader of this capture phase adopts an informed leadership strategy, that is, if there is an informed follower in each group of members of the robot swarm, the local perception information perceived by the individual can be transmitted to the leader. The leader combines the local perception information transmitted by the informed follower to make a global decision on the direction and real-time judgment of the capture status. The informed follower is a robot that can perceive local perception information. When judging the capture status, the leader determines whether the capture target is successfully captured based on the closure of the encirclement. If the capture is successful, the robot swarm enters the next round of capture target search phase. If not, the capture continues.

5. The method according to claim 4, wherein Let the serial number of a robot in the robot group be i, n represents the number of robots in the robot group, i = 1, 2, 3...n; at each time point t, the position vector corresponding to a robot i is x i (t), the direction vector is v i (t), the speed maintained is s i In each time step, the robot calculates the expected direction d according to the positions of the neighboring robots in the three sensing areas, namely the attraction area Az, the repulsion area Rz and the direction area Oz. i (t); Define a group interaction network to describe the topological structure and dynamic interaction rules of information transmission between individuals in a robot group. The social connection matrix is a mathematical abstract representation of the group interaction network, which quantifies the connection strength or interaction weight between individual robots in matrix form. And predefine the connection strength between robots, S ij (t) is the group interaction network, represented by the social connection matrix between individuals, S ij The value of (t) represents the relative influence of robot j on robot i. If robot i can be influenced by robot j, then S ij (t)≠0, otherwise S ij (t) = 0; in different robot group motion stages, the structure of the group interaction network is fixed or set to change over time; Among them, for any robot i, j, S ij When all (t) are greater than 0, the robot group is in the search phase, and the group interaction network is a fully connected network; when S ij When a part of (t) is greater than 0, the robot group is in the tracking or roundup stage, and the group interaction network is a partially connected network; In the attraction zone Az, the individual adjusts its own direction to align with the direction of the adjacent robot to prevent the robot group from dispersing; in the repulsion zone Rz, the individual actively moves away from the adjacent robot to avoid collision; in the direction zone Oz, the individual adjusts its own direction to align with the direction of the robot group; At different stages TP, if a robot j appears in the exclusion zone Rz of robot i, for each time step Δt, robot i calculates the expected direction d according to the relationship between it and the adjacent robots in the three perception areas. i (t+Δt|TP); The calculation method of the expected direction is to introduce the group interaction network s based on the couzin aggregation motion model ij (t) and noise ε i (t) is used to reflect the changes in the connection characteristics between groups at different movement stages and the influence of noise during the movement process. Therefore, the expected direction of robot i in the next step is defined as formula (2): Among them, ε i (t) is the random noise of robot i, ε i (t)~N(0,σ 2 ).

6. The method according to claim 5, wherein In the search phase, the distributed control strategy is used to calculate the expected direction when searching for the target, and the direction of the next moment is adjusted according to the calculated expected direction. In the search phase, the regional attraction intensity α1 of the attraction zone Az and the movement direction of the direction zone Oz will affect the expected direction of the robot i. At this time, the expected direction of the next step is defined as formula (3): Where α1 is the regional attraction strength of the attraction zone Az in the search phase, α1∈(0,1).

7. The method according to claim 6, wherein In the tracking phase, the robot group is organized in the form of a structural leader, and the leader determines the movement direction of the robot group and defines k i is the level of individuals in the robot group, where k i =1 represents the leader, k i =2 represents a follower. Based on whether the leader is fixed during the tracking phase and whether it changes dynamically over time, the leader's fixedness is divided into the following three cases: (1) In the tracking phase, the leader is fixed, then k i fixed; (2) In this tracking phase, the leader is randomly fixed, then k i Changes over time. At time point t, a robot is randomly selected from the group and defined as the leader; (3) In this tracking phase, the leader changes along the time and space dimensions, and the fixation mode is different at different times; In the tracking phase, the regional attraction intensity α2 of the attraction zone Az and the movement direction of the direction zone Oz will affect the expected direction of robot i, and enhance the group aggregation in the roundup phase, that is, α2>α1. At this time, the expected direction of the leader robot i in the next step is defined as formula (4): Then normalize to: In a robot group, except for the leader, the expected direction of other robots must not only consider the situation of adjacent robots within the sensing range, but also the movement direction of the leader. In this case, the preference intensity for the leader's movement direction is set to W. The larger the value of W, the greater the influence of the leader's movement direction on the next movement direction of robot i. At this time, the expected direction of follower i in the next step is defined as formula (6): Where W is the follower's preference for the leader's movement direction, W>1, v k＝1 (t) represents the velocity of the leader robot at time t.

8. The method according to claim 7, wherein In the capture phase, the robot group is organized in an informed leadership mode and consists of a leader and followers. The group movement direction is dominated by the leader and is affected by the real-time local perception information of other robots. The leader fixation method in the capture phase is the same as that in the tracking phase. There are three leader fixation methods. In the capture phase, the group aggregation is further enhanced and the regional attraction intensity α3 of the attraction zone Az in the capture phase is increased, that is, α3>α2. At this time, the expected direction of robot i in the next step is defined as formula (7): Furthermore, formula (7) is normalized to obtain formula (8):

9. The method according to claim 8, wherein In the roundup phase, in order to simulate the strategy of informed leadership, a knowledge-based preference direction g is defined for some followers. i , the leader will receive the preferred directions of the informed followers and integrate them to assist in decision-making; since the expected direction is jointly affected by the directions of the leader and the informed followers, the expected directions of the leader and the informed followers are weighted according to the weight w, which determines the degree of participation of the informed leader information in the direction decision. The smaller w is, the higher the degree of participation. The expected direction of the leader robot i in the next step in the roundup phase is updated as formula (9): Where w is the leader's preference strength for the informed follower's direction, w∈(0,1); w=0 means the leader completely relies on the informed follower's preference direction, w=1 means the leader completely relies on its own preference direction and is not affected by the informed follower's preference direction, and w between 0 and 1 means the leader partially combines the informed follower's preference direction to make decisions; at this time, the expected direction of follower i's next step is updated as formula (10):

10. The method according to claim 9, wherein When judging the capture status, the criteria for whether the capture is successful are as follows: the encirclement closure is used as an indicator to measure the integrity of the encirclement in the robot group capture task. The higher the closure, the more complete the encirclement and the lower the possibility of the captured target escaping. Let Q be the encirclement closure, LC be the actual perimeter of the encirclement, and LC be the perimeter of the ideal completely closed encirclement. The encirclement closure Q is expressed as: Assuming that the circumference of the ideal encirclement is known, which is the circumference of a perfect circle formed around the target, and R is the radius of the encirclement, the calculation formula for LC is: LC=2πR (12) Set a threshold Q_threshold for the closed boundary of the robot group. The leader determines that when all robots form a closed boundary, that is, when the encirclement closure degree Q is greater than or equal to the defined closed boundary threshold Q_threshold, the target is considered to be completely surrounded inside and the capture is successful.

Images