A density-driven target collection method

Abstract

The present application is directed to the target of patchy distribution, proposes a target collection algorithm based on density driving, analyzes the shape, range, density and other characteristics of the target of patchy distribution, and models the target distribution based on kernel function estimation; a cluster robot density driving control framework is established for the motion model of the robot; two basic behavior modes are determined for completing the collection task: edge surrounding behavior and shrinkage behavior, and the corresponding control methods are given. Further, different shrinkage strategies for large-scale targets and small-scale targets are designed, as well as the amplification rules of the expected reference density; through the shrinkage strategy and the amplification rule, the robot can adjust its motion state according to the target size and shape, so as to realize the shrinkage and collection of the target patch.

Description

Technical Field

[0001] This invention relates to the field of target collection in swarm robots, and more particularly to a density-driven target collection method. Background Technology

[0002] Using robots to search for and collect targets within a defined area is an important application of robotics. However, the emergence of new target collection scenarios has presented new challenges to achieving target search and collection using robots. Examples include marine and lake debris management, oil spill response, and leaf litter removal. In these new application scenarios, the targets to be collected are often not uniformly distributed throughout the space, but rather clustered into high-density patches of varying sizes and densities. This pattern is the result of multiple factors, including the generation and propagation of the targets.

[0003] Most existing target collection methods are based on the idea of ​​full area coverage, which means that the robot traverses the entire environment according to a preset path or rules to find and collect targets. This method is highly efficient when dealing with uniformly distributed targets, but its efficiency drops significantly when dealing with patchy targets. This is because patchy targets often result in a large number of sparse target areas and targetless areas in the space. These areas are of no value to the robot, but the robot still needs to spend time and energy traversing and collecting targetless areas. These methods do not fully utilize the spatial distribution information of targets to guide robot behavior, which not only increases conflicts and interference between robots and between robots and the environment, but also wastes robot resources and energy.

[0004] For targets distributed in patches, their spatial distribution characteristics affect not only the efficiency of robot target discovery and collection but also the efficiency of robot target processing and stacking. This is because high-density patch areas in a patchy distribution often contain a large number of targets, while low-density or target-free areas are almost worthless. If the robot can identify these patch areas and prioritize collecting targets within them, it can improve robot utilization and collection efficiency. Furthermore, the central locations of high-density patch areas are also highly valuable, as they are ideal in-situ stacking points, significantly reducing the time and energy costs of target transport. If the robot can identify the optimal or second-optimal stacking points in a patchy distribution and use them as in-situ target stacking points, it can improve the overall system's operational efficiency and reliability. Summary of the Invention

[0005] To address the issue that most existing target collection methods are based on the idea of ​​full regional coverage, which leads to a significant decrease in efficiency when dealing with patchy targets, this invention proposes a density-driven target collection method. This method fully considers and utilizes the patchy characteristics of the spatial distribution of targets, and uses density information to enable swarm robots to better adapt to changes in target distribution, thereby improving the efficiency and quality of target collection.

[0006] The technical solution of this invention is as follows:

[0007] The density-driven target collection method is implemented using a swarm of collection robots; the swarm of collection robots moves according to the following motion model:

[0008] x i (t+Δt)=x i (t)+v i (t+Δt)v0Δt

[0009] v i (t+Δt)=v i (t)+ω i (t+Δt)Δt

[0010]

[0011] Where v0 is the linear velocity of the robot, x i (t) and v i (t) represents the position and orientation of robot i at time t, x i (t+Δt) and v i (t+Δt) represents the position and orientation of robot i at time t+Δt, ω i (t+Δt) is the angular velocity of robot i at time t+Δt, ω max u is the robot's maximum angular velocity. i (t+Δt) is the control input of robot i, corresponding to the desired motion direction of robot i at time t+Δt, and ∠(*,*) represents the angle between the two vectors;

[0012] Control input u of robot i i (t+Δt) is obtained through the following formula:

[0013]

[0014] in For the robot's self-driving power, To help the robot avoid collision forces, The task-related forces include robot density driving force; by adjusting the robot density driving force, the cluster of collecting robots is driven by the target density to move along the edge of the target patch, and each collecting robot maintains a certain distance to form an encirclement of the target patch; after completing the edge encirclement, the cluster of collecting robots simultaneously retracts and moves inward, pushing the target to gather towards the center, thereby achieving the collection of the target.

[0015] Furthermore, the robot density driving force According to the formula

[0016]

[0017] Determined, where ρ i ρ(t) represents the current density of robot i, ρ0(t) represents the desired reference density of robot i, and N... i G is the set of other individual robots within the robot's perception range. i Let λ be the set of other target individuals within the robot's perception range. r λ represents the interaction strength between robots. g The strength of the interaction between the robot and the target. W is the gradient operator, and W() is the kernel function.

[0018] Furthermore, the kernel function is a Gaussian function:

[0019]

[0020]

[0021] Furthermore, by controlling λ g >>λ r This allows the cluster of collecting robots to move along the edge of the target patch while maintaining a certain distance between each robot, forming an encirclement around the target patch.

[0022] Furthermore, during the process of forming a surrounding circle around the target patch, the uniformity η of the cluster collection robot is detected. If η < δ, it indicates that the surrounding circle is completed and an inward contraction movement is initiated. Otherwise, the cluster collection robot continues to adjust the surrounding circle configuration, where δ is a set threshold.

[0023] Furthermore, the uniformity η is calculated according to the formula...

[0024]

[0025] Where N is the number of collection robots in the cluster, l i (t) represents the average distance between robot i and all robots within its perception range at the current moment.

[0026] Furthermore, the task-related forces are based on the formula...

[0027]

[0028] Received, among which For robot density driving force, Let β be the tangential force of the robot's circular ring, and β = {0, 1} be a Boolean variable used to switch between two different collection strategies, including a coarse-grained collection strategy and a fine-grained collection strategy. r ≤B g A coarse-grained collection strategy is adopted when B r >B g A fine-grained collection strategy is adopted; where B r Let B be the projected area of ​​the robot. g This represents the average area of ​​the target individual within the patch.

[0029] Furthermore, the coarse-grained collection strategy involves the robot pushing horizontally from the edge of the target patch inwards, adjusting the density driving force. The desired reference density ρ0 is achieved by increasing ρ0 so that it is similar to the robot's current density ρ. i A density difference is generated between (t), and the robot continuously moves closer to the target so that its density converges to ρ0, realizing the robot cluster's inward contraction in a horizontal pushing manner.

[0030] Furthermore, the fine-grained collection strategy involves the robot moving along the edge curve of the target patch, gradually reducing its movement radius, and continuously converging inwards, driven by density. and the tangential force of the robot's ring Implementation: By increasing ρ0, a density-driven force is generated to guide the robot to retract inward, while the annular tangential force... Through formula

[0031]

[0032] Confirmed, among which g is a rotation matrix; k (t) represents the current time G. i The location of the target with the highest medium density.

[0033] Furthermore, in the fine-grained collection strategy, the adaptive amplification rule for ρ0 is:

[0034]

[0035] Where M is the target quantity. The density of target m is represented by:

[0036]

[0037] g m (t) and g z (t) represents the positions of target m and target z, respectively.

[0038] Beneficial effects

[0039] This invention proposes a density-driven target collection algorithm for patchy targets. It analyzes the shape, range, and density characteristics of the patchy targets and models the target distribution based on kernel function estimation. For the robot's motion model, a density-driven control framework for swarm robots is established. Two basic behavior modes are identified for completing the collection task: edge encirclement behavior and contraction behavior, and corresponding control methods are provided for each. Furthermore, different contraction strategies for large-scale and small-scale targets, as well as amplification rules for the desired reference density, are designed. Through these contraction strategies and amplification rules, the robot can adjust its motion state according to the target size and shape, thereby achieving the contraction and collection of target patches.

[0040] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0041] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0042] Figure 1 : Schematic diagram of target patch density modeling;

[0043] Figure 2 The entire process of robot target collection;

[0044] Figure 3 : Schematic diagram of two collection strategies for robots;

[0045] Figure 4 Simulation experiment of coarse-grained target collection;

[0046] Figure 5 Simulation experiment of fine-grained target collection;

[0047] Figure 6 Simulation experiment of multi-patch target collection;

[0048] Figure 7 Simulation experiment on target collection in dynamic environment. Detailed Implementation

[0049] The embodiments of the present invention are described in detail below. These embodiments are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0050] This embodiment proposes a density-driven target collection method for patchy distributed targets. It fully considers and utilizes the patchy characteristics of the spatial distribution of targets and uses density information to enable swarm robots to better adapt to changes in target distribution, thereby improving the efficiency and quality of target collection.

[0051] Patchy distribution of targets refers to a spatial structure formed by the aggregation of a large number of identical or different types of target individuals. Within a certain range, if target individuals are close to each other, high-density target patches are formed; conversely, low-density target patches are formed. Therefore, density can be used to reflect the enrichment degree of the object to be collected within a certain space. To more accurately calculate the total density of all target patches in a certain area, it is necessary to consider the contribution of each target individual to the density of the entire area. In this way, each target individual can be assigned a density value, representing the strength of its spatial relationship with other target individuals. Specifically, for M target individuals, the density of the m-th target... Defined as:

[0052]

[0053] Among them, W(g) m (t),g z (t) is the kernel function used to evaluate the target distribution density, g m (t) and g z (t) represents the positions of target m and target z, respectively. In this embodiment, a Gaussian function is used as the kernel function, and the density of target m is defined as:

[0054]

[0055] Where σ>0 is an adjustable parameter of the target density estimation function. Figure 1 The image shows the spatial density fields formed by three target patches estimated using kernel functions. Darker colors indicate higher density, meaning a more concentrated distribution of targets. This demonstrates that the method can effectively characterize the spatial distribution features of the targets to be collected.

[0056] Based on the spatial distribution characteristics of the targets to be collected mentioned above, we then present the motion and control model of the swarm collection robot:

[0057] Suppose the robot maintains a constant velocity v0, and the position and orientation (unit vector) of robot i at time t are x and x, respectively. i (t) and v i (t), then the motion state of robot i at time t+Δt satisfies:

[0058] x i (t+Δt)=x i (t)+v i(t+Δt)v0Δt

[0059] v i (t+Δt)=v i (t)+ω i (t+Δt)Δt

[0060]

[0061] Where v0 is the linear velocity of the robot, x i (t) and v i (t) represents the position and orientation of robot i at time t, x i (t+Δt) and v i (t+Δt) represents the position and orientation of robot i at time t+Δt, ω i (t+Δt) is the angular velocity of robot i at time t+Δt, ω max u is the robot's maximum angular velocity. i (t+Δt) is the control input of robot i, corresponding to the desired motion direction of robot i at time t+Δt, and ∠(*,*) represents the angle between the two vectors.

[0062] Control input u of robot i i (t+Δt) is obtained through the following formula:

[0063]

[0064] in For the robot's self-driving power, Collision avoidance forces are two fundamental forces that enable robots to move.

[0065] Self-driving force Used to control a robot to continue its current direction of motion, it can be described as:

[0066]

[0067] Where λ self >0 indicates a self-driven weight; x i (t) and x i (t-Δt) represents the robot's current position and its previous position, respectively.

[0068] Collision avoidance force The collision avoidance motion used to achieve this in robots is achieved through repulsive forces, and can be described as follows:

[0069]

[0070] Where, λ rep >0 indicates a collision avoidance weight. This produces a non-local, wide-area repulsion effect, which helps swarm robots form a uniformly spaced encirclement.

[0071] The task-related forces include robot density driving force; by adjusting the robot density driving force, the cluster of collecting robots is driven by the target density to move along the edge of the target patch, and each collecting robot maintains a certain distance to form an encirclement of the target patch; after completing the edge encirclement, the cluster of collecting robots simultaneously retracts and moves inward, pushing the target to gather towards the center, thereby achieving the collection of the target.

[0072] For robots, a density evaluation function is used to estimate the actual density of their current location and compare it to a desired reference density. Taking swarm control as an example: if the actual density is higher than the desired reference density, the swarm is too crowded and needs to accelerate and turn away from its neighbors; if the actual density is lower than the desired reference density, the swarm is too sparse and needs to decelerate and turn closer to its neighbors; if the actual density is close to the desired density, the robot is in a suitable position and can maintain its current motion. However, the actual situation is much more complex than this example, because the robot must evaluate not only the distribution within the swarm but also the distribution of the target area.

[0073] Based on the aforementioned kernel function density estimation method, the density at the robot's location is a linear superposition of two parts: one part is contributed by other robots within the robot's perception range, and the other part is contributed by the target individuals to be collected within the robot's perception range. Under this method, the robot not only evaluates the contributions of other individuals within the group to the density but also the contributions of the target individuals, giving the robot a certain degree of environmental adaptability. When the target distribution changes, the robot can promptly adjust its motion state to adapt to the changing target.

[0074] The current density ρ of robot i i The definition of (t) is as follows:

[0075]

[0076] N i G is the set of other individual robots within the perception range of robot i. i This refers to the set of other target individuals within the perception range of robot i. Based on this, the robot's density-driven force... According to the formula

[0077]

[0078] Determined, where ρ i ρ(t) represents the current density of robot i, ρ0(t) represents the desired reference density of robot i, and λ represents the density of robot i.r λ represents the interaction strength between robots. g The strength of the interaction between the robot and the target. For the gradient operator, W() is the kernel function. Similarly, the kernel function is a Gaussian function.

[0079]

[0080]

[0081] In this way, by utilizing the robot's local communication and perception capabilities, the robot can adaptively adjust its speed and direction of movement according to the density distribution in the environment.

[0082] As mentioned earlier, the target collection task can be divided into edge encirclement behavior and contraction behavior. By adjusting the robot density driving force, the swarm of collection robots is driven by the target density to move along the edge of the target patch, while maintaining a certain distance between each collection robot, forming an encirclement of the target patch. After completing edge encirclement, the swarm of collection robots simultaneously contract inward, pushing the target towards the center to achieve target collection. These two behaviors can effectively transform scattered target patches into a compact target pile, facilitating subsequent processing and transportation.

[0083] For edge-enclosing behavior, by controlling λ g >>λ r This makes the collecting robots focus more on the target individual rather than other robot companions, creating a tendency for the individual to move along the outer edge of the target patch. Based on this, combined with the global repulsive force between the collecting robots, the swarm of collecting robots can ultimately surround the target patch with a uniformly spaced single-layer ring structure.

[0084] Building upon this foundation, and leveraging the global repulsive force among the collecting robots, the swarm of robots can ultimately encircle the target patch using a uniformly spaced single-layer ring structure. Subsequently, based on the contraction trigger conditions and the target collection strategy, the robot encirclement is continuously reduced to move the target inward, thus completing the task of collecting the scattered target.

[0085] The triggering condition for contraction behavior is used to determine whether a swarm of robots has entered a contraction behavior. Under the combined effect of wide-area repulsion and density-driven force, the collecting robots will eventually form a symmetrical and uniformly spaced ring structure through precise position adjustments. Therefore, whether the swarm of robots has formed a uniformly spaced closed encirclement can be used as the triggering condition for contraction behavior.

[0086] In this embodiment, during the process of forming a surrounding circle around the target patch, the uniformity η of the swarm collection robots is detected. If η < δ, it indicates that the surrounding circle is complete, and an inward contraction movement is initiated; otherwise, the swarm collection robots continue to adjust the surrounding circle configuration, where δ is a set threshold. The uniformity η is determined according to the formula...

[0087]

[0088] Where N is the number of collection robots in the cluster, l i (t) represents the average distance between robot i and all robots within its perception range at the current moment. A smaller uniformity η indicates more uniform spacing between robots. A smaller threshold parameter δ indicates a more stringent triggering condition for the contraction.

[0089] Target acquisition strategy is a method that determines how to recover and process targets based on factors such as their type, quantity, and location. To achieve efficient acquisition of targets of different sizes, two contraction strategies are designed for coarse-grained and fine-grained targets, respectively. This is illustrated by the task-related force formula.

[0090]

[0091] In the process, the Boolean variable β = {0, 1} is selected to switch between two different collection strategies. In the task-related force formula, For robot density driving force, This refers to the tangential force on the robot's circular ring. The switching criterion is: when B... r ≤B g A coarse-grained collection strategy is adopted when B r >B g A fine-grained collection strategy is adopted; where B r Let B be the projected area of ​​the robot. g This represents the average area of ​​the target individual within the patch.

[0092] Coarse-grained collection strategy:

[0093] In tasks involving collaborative robotic collection of large-scale targets, the large size of these targets and the small gaps between them present significant challenges. Traditional collection methods (designing fixed target collection areas within the site) require additional robotic gripping mechanisms and consume substantial time and energy for target transfer. Therefore, this paper proposes a coarse-grained target collection strategy based on inward pushing. The core idea is to utilize multiple robots to push the target patch inward from its edge, creating a compact accumulation area. This strategy, based on in-situ accumulation, significantly reduces energy waste associated with target transfer.

[0094] In this collection strategy, the shrinking behavior of the robot swarm is driven by a controlled density force. The desired reference density ρ0 is achieved by increasing ρ0 so that it is similar to the robot's current density ρ. i A density difference is generated between (t), and the robot will continuously move towards the target until its density converges to ρ0. Based on this, the robot swarm can achieve a pushing-like inward contraction.

[0095] Fine-grained collection strategy:

[0096] Small-scale targets are characterized by their small size and large number, and collecting them typically requires complete coverage of their distribution space. Therefore, the previously described push-type collection strategy suffers from significant target omissions due to the limited area covered by the robot. To address this issue, a fine-grained target collection strategy based on vortex contraction motion is designed to adapt to the space-covering requirements of small-scale target collection tasks. This strategy utilizes the robot to generate a vortex-like rotational motion at the edge of the target patch, gradually reducing the radius of motion, thereby causing the outer edges to continuously converge inwards. Compared to the push-type strategy, this strategy can more effectively reduce target omissions and can optimize recovery efficiency by adjusting motion parameters according to different scenarios.

[0097] The contraction behavior in this strategy is driven by density. and the tangential force of the robot's ring The combined effect of two factors is as follows. The increase in ρ0 generates a density-driving force that guides the robot to retract inward. This provides a tangential force along the ring, and the combined force of the two guides the robot to gradually contract inward in a vortex motion, achieving full coverage of the surrounding space and gathering scattered small-scale targets. The definition is as follows:

[0098]

[0099] in g is a rotation matrix; k (t) represents the current time G. i The location of the target with the highest medium density.

[0100] In coarse-grained collection strategies, simply increasing ρ0 rapidly is sufficient to achieve a robot-driven, horizontal collection effect. However, in fine-grained collection strategies, if ρ0 increases too quickly, the gaps between the robot's vortex trajectories become too large, causing targets to be missed; if ρ0 is too small, the robot's vortex trajectories become too dense, resulting in significant time costs. Therefore, specific ρ0 increase rules need to be designed to ensure both the reliability and efficiency of collection.

[0101] Assuming the robot swarm satisfies the contraction condition at time t1, then the adaptive amplification rule for ρ0 when t≥t1 is:

[0102]

[0103] Where M is the target quantity. Let ρ0(t) represent the density of target m. ρ0(t) increases based on ρ0(t-1) and is only affected by the density and quantity of the targets. The rate of increase of ρ0 is controlled by constructing the quotient of the density gradient difference and the quantity of the targets. For a given number of targets M, this control method can ensure rapid contraction in the early stages of the task and fine contraction at the end of the task, thus achieving a good balance between the speed and efficiency of the operation.

[0104] The effects of this invention are illustrated below through simulation examples:

[0105] Single patch collection:

[0106] First, the effectiveness of the proposed encirclement and contraction mechanisms was verified through simulation experiments. For the case of only one patch, 20 robots were used to conduct simulated target collection tasks for 10 large-scale targets and 20 small-scale targets respectively.

[0107] Figure 4 This demonstrates the entire process of large-scale target collection and the robot's average density. Robot encirclement area S r The curves showing the variation of parameters such as robot uniformity η. Initially, the target and robots are clustered together and do not overlap; the average density of the robots... average density of targets During the edge-surrounding phase (T = 0–237 s), the robot initially needs to expand outwards until… Slightly less than ρ0; at this point, the robot maintains a constant density and needs to move closer to the target individual or the robot to obtain the density, due to λ g >>λ r The robots focus more on the target than other robots, so they orderly approach the target and surround it along the isodense lines of the target patch until the cluster completely surrounds the target patch. During the edge-surrounding phase, the main contributor to the robot density gradually shifts from other robots to the target to be collected. After the surrounding swarm is formed, the area S of the robot enclosure is... r =7.1m 2 The maximum value of the entire process is η. At T = 238s, the robots transition from an edge-enclosed state to a contracted state. At this point, the group uniformity η = 0.0098, satisfying the trigger condition for contraction behavior. All robots gradually move inward until the task ends. During this stage, the enclosing area S of the robot swarm is... r From 7.1m 2 Reduced to 0.61m2 average density The value increased from 1.5 to 6.42, indicating that the dispersed target individuals were effectively collected by the swarm robots.

[0108] Simulation results for fine-grained target collection are as follows: Figure 5 As shown, a target collection task was simulated using 20 robots and 20 small-scale targets. The edge-encircling phase of fine-grained collection (T = 0–396 s) is basically the same as the coarse-grained collection process. The robot swarm first spreads out slightly, then encircles the target patches along the isodense lines until a complete ring is formed. It is worth noting that because there are more fine-grained targets and their distribution is more disordered, the robot swarm requires a longer position adjustment time to meet the contraction condition. At T = 397 s, η = 0.0096 meets the contraction condition. All robots continuously contract inward in a vortex-like trajectory, synchronously driving the fine-grained targets to continuously gather until the collection task ends. This vortex collection behavior not only avoids the omission problem caused by the flat-push contraction motion, but also achieves a local full-coverage cleaning effect, enabling the robots to have a certain cleaning ability for even imperceptible fine targets such as dust.

[0109] Multi-patch parallel collection:

[0110] A significant challenge in collection tasks is the presence of multiple patches. This necessitates that the robot swarm spontaneously split into a corresponding number of sub-swarms to collect and clean multiple patches in parallel. Previous theoretical studies of the density-driven swarm robot controller proposed in this invention have shown that by reducing the reference density or shrinking the sensing radius, the swarm will spontaneously split into multiple independent sub-swarms of roughly equal size. This unique property endows the swarm robots with the ability to process multi-patch collection tasks in parallel. To verify the parallel collection capability of the proposed method, a multi-patch collection simulation experiment was conducted. In this experiment, target patches of varying numbers and locations were set up, and the automatic adjustment of parameters and task allocation by the robot swarm was observed.

[0111] like Figure 6 As shown, 20 robots perform parallel collection on two target patches, which contain 10 and 12 independent targets respectively, distributed completely randomly. When the presence of two patches in the environment is detected, the robots autonomously adjust their control parameters, inducing spontaneous clustering behavior by reducing R. The cluster then splits into two subgroups of similar size, and its edge encirclement behavior is triggered, simultaneously encircling the two target patches in parallel.

[0112] To investigate the algorithm's performance on a larger number of target patches, a multi-patch parallel collection experiment was conducted. In this experiment, different numbers of robots and target patches were set up, and the automatic task allocation and coordinated action of the robot swarm were observed. Figure 6 Figures (b) and (c) show the simulation results of parallel collection experiments of 35 robots collecting data from 3 target patches and 40 robots collecting data from 4 target patches, respectively. Through a unique mechanism where the robot swarm autonomously adjusts its parameters to split into multiple subgroups, multiple patches in the environment are effectively surrounded and collected. As can be seen from the figures, each subgroup can choose an appropriate shape and size to cover the target area based on its own density and the surrounding target conditions. This demonstrates that density-driven spontaneous splitting behavior of the swarm robots endows them with the ability to perform parallel collection of multiple target patches, which greatly improves the applicability and efficiency of the target collection algorithm.

[0113] Time-varying environmental target collection:

[0114] In real-world cleaning tasks, the number of targets to be collected can continuously increase or suddenly increase due to pedestrians carelessly discarding objects or new leak points in spill accidents. Therefore, the collection algorithm needs to have a certain degree of adaptability to cope with the sudden increase in targets. Next, we consider a scenario where the number of individual targets increases, and study the adaptability of the density-driven collection algorithm to the dynamic environment. For simplicity, this part of the experiment only considers the robot's edge-enclosing behavior. The experimental design for this part is as follows: First, no targets are set in the simulation; then, at T=200s, 15 targets are added to the area; at T=500s, another 15 targets are added to the area; and we observe whether the robot can detect and enclose the new targets in a timely manner.

[0115] like Figure 7 As shown, during T = 0–200 s, there were no targets, and the robot swarm spontaneously adjusted to a single-ring configuration. At T = 201 s, 15 targets suddenly appeared, existing as a single patch, and the average density of the robots increased. The value mutated to 1.88. Subsequently, under the influence of the density gradient, Gradually converging towards ρ0, the robot spontaneously performs edge enclosure and completes the enclosure of the target patch, increasing the area of ​​the enclosure to 7.4m. 2 At T=501s, the target quantity further increases to 30. The mutation to 2.23, spontaneous diffusion of robots enabled... The enclosed area is 7.4m² 2 Increased to 11.7m 2 To accommodate the newly added target individuals. Figure 7(b) illustrates the changes in the average density and enclosing area of ​​the robot swarm during this process. It is evident that the instantaneous average density of the swarm of robots increases rapidly when the number of target individuals changes abruptly. Subsequently, through edge enclosing behavior, it eventually converges to ρ0. Simultaneously, the area enclosed by the swarm gradually increases with the increase in the number of target individuals, consistently providing complete enclosure of the target patch. This result demonstrates that the proposed algorithm exhibits good adaptability to dynamic environments facing sudden changes in the number of target individuals.

[0116] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention without departing from the principles and spirit of the present invention.

Claims

1. A density-driven target collection method, implemented through a swarm of collection robots; characterized in that: The cluster of collecting robots moves according to the following motion model: in For the robot's linear velocity, and For robots exist Position and direction at any moment and For robots exist Position and direction at any moment For robots exist angular velocity at time t, This is the robot's maximum angular velocity. For robots The control input, corresponding to the robot exist The expected direction of motion at any given moment. This represents the angle between two vectors; robot control input It is obtained through the following formula: in For the robot's self-driving power, To help the robot avoid collision forces, The task-related forces include robot density driving force; by adjusting the robot density driving force, the cluster of collecting robots is driven by the target density to move along the edge of the target patch, and the collecting robots maintain a certain distance from each other to form an encirclement of the target patch. After completing the edge encirclement, the cluster of collecting robots simultaneously retracts and moves inward, pushing the target towards the center to achieve target collection; The task-related forces are based on the formula Received, among which For robot density driving force, For the tangential force of the robot's ring, This is a Boolean variable used to switch between two different collection strategies: coarse-grained and fine-grained collection strategies. A coarse-grained collection strategy is adopted when A fine-grained collection strategy is adopted; among which For the robot's projected area, The average area of ​​the target individual within the patch; The robot density driving force According to the formula Confirmed, among which For robots The current density, For the robot's desired reference density, This refers to the collection of other individual robots within the robot's perception range. The set of other target individuals within the robot's perception range. The strength of the interaction between robots, The strength of the interaction between the robot and the target. For gradient operators, For kernel functions; For the current moment The location of the target with the highest medium density; The robot's circular tangential force Through formula Confirmed, among which It is a rotation matrix.

2. The density-driven target collection method according to claim 1, characterized in that: The kernel function is a Gaussian function. 。 3. The density-driven target collection method according to claim 1, characterized in that: By controlling This allows the cluster of collecting robots to move along the edge of the target patch while maintaining a certain distance between each robot, forming an encirclement around the target patch.

4. The density-driven target collection method according to claim 1, characterized in that: During the process of forming a perimeter around the target patch, the uniformity of the swarm collection robots is detected. ,like This indicates that the encirclement is complete and an inward contraction movement is initiated; otherwise, the swarm of collecting robots continues to adjust the encirclement configuration. To set a threshold.

5. The density-driven target collection method according to claim 4, characterized in that: Uniformity According to the formula Where N is the number of collection robots in the cluster. For the robot at the current moment The average distance to all robots within its perception range.

6. The density-driven target collection method according to claim 1, characterized in that: The coarse-grained collection strategy involves the robot pushing horizontally from the edge of the target patch inwards, adjusting the density driving force. Expected reference density in Implementation: By adding , making With the current density of robots A density difference is created between them, and the robot continuously moves closer to the target until its own density converges to a certain value. This enables the robot cluster to retract inward in a horizontal pushing motion.

7. The density-driven target collection method according to claim 6, characterized in that: The fine-grained collection strategy involves the robot moving along the edge curve of the target patch, gradually reducing its movement radius, and continuously converging inwards, driven by density. and the tangential force of the robot's ring Implementation: By adding The density-driven force is generated to guide the robot to retract inward.

8. The density-driven target collection method according to claim 6, characterized in that: In fine-grained collection strategies The adaptive scaling rule is: Where M is the target quantity. The density of target m is represented by: and The target and target The location.

Images