A multi-robot cooperative regional protection method for dynamic environment

Abstract

The present application belongs to the field of dynamic control of multi-robot system, and specifically relates to a multi-robot cooperative regional protection method for dynamic environment. The method deploys a multi-robot formation in a task region to be protected, and through cooperation of a remote control terminal, a search unmanned aerial vehicle, a blocking unmanned aerial vehicle, a ground search robot and a ground blocking robot, adopts an air-ground cooperative mode to perform real-time protection on the whole region. The formation adopts a 'lattice' mode, so that the formation is easier to adjust, restore and migrate as a whole, and the robustness of the formation is enhanced; the detection of an external target adopts a man-machine interaction and air-ground cooperative mode, so that the target detection accuracy is improved; the present application proposes a formation self-repairing control algorithm based on gradient, when a blocking robot goes to a target point to block the target, the remaining robots in the region perform formation self-repairing, fill the vacancy and maintain the formation shape, so that real-time protection on the task region is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of dynamic control of multi-robot systems, and particularly relates to a multi-robot cooperative regional protection method for dynamic environment. BACKGROUND

[0002] With the rapid development of artificial intelligence, multi-robot systems have a stronger fault tolerance, better adaptability and high efficiency in the use of resources, and are more popular in the research of high-risk areas such as complex scenes, unknown maps and bad weather. However, there are few studies on the application of multi-robot cooperative formation to regional protection, and some research schemes have problems such as low efficiency, inability to restore formation, inability to migrate as a whole, low practicality and execution efficiency, and are difficult to meet the requirements of complex environment on robots and real-time protection of the region for a long time. SUMMARY

[0003] The application aims to provide a multi-robot dynamic control system for regional protection with high robustness, high efficiency and high accuracy to overcome the defects of the above multi-robot system.

[0004] The technical scheme adopted by the application to achieve the above-mentioned purpose is:

[0005] A multi-robot cooperative regional protection method for dynamic environment, comprising the following steps:

[0006] 1) According to the number of robots and the area of the region, a multi-robot protection formation is constructed in the task area;

[0007] 2) The search robots in the multi-robot protection formation monitor the task area and send the monitoring results to the remote control terminal;

[0008] 3) The remote control terminal assigns tasks to the blocking robots in the multi-robot protection formation according to the monitoring results;

[0009] 4) The blocking robots block the target according to the task instructions;

[0010] 5) The remaining robots in the multi-robot protection formation perform formation self-recovery to fill the vacancy of the blocking robots.

[0011] The multi-robot protection formation is constructed based on the Boid cluster model and consists of blocking drones, search drones, ground blocking robots and ground search robots, wherein the interval and the marching speed between adjacent robots remain the same state, the ground search robots are deployed on the four boundaries of the task area, the search drones are deployed in the center of the task area, the blocking drones are deployed inside the task area, the ground blocking robots are deployed in the entire task area in a set formation, and the requirement that each column in the formation in the task area has a ground blocking robot is met.

[0012] The step 2) is specifically:

[0013] When the foreign target enters the task area, the ground search robot will trigger the early warning state, and the ground search robot wakes up the search unmanned aerial vehicle in the area, and the search unmanned aerial vehicle identifies and monitors the type and position coordinate information of the foreign target after receiving the early warning information sent by the ground search robot, and wakes up other idle robots and sends the type and position coordinate information of the foreign target and the position information of the robots in the task area to the remote control terminal.

[0014] The step 3) includes the following steps:

[0015] 3.1) When the foreign target is about to enter the task area, each ground blocking robot broadcasts its global trajectory length list to the search unmanned aerial vehicle;

[0016] 3.2) The search unmanned aerial vehicle receives and records all the global trajectory lengths published by the ground blocking robots, and forwards them to the remote control terminal;

[0017] 3.3) The remote control terminal determines the number and ID of the ground blocking robots to be dispatched according to the global optimal condition, and constructs a cost matrix according to the broadcasted trajectory lengths;

[0018] 3.4) The cost matrix is expanded, and based on the expanded cost matrix, the Hungarian algorithm is used to assign targets to the robots.

[0019] The cost matrix is specifically: the ID of the ground blocking robot is the row index of the cost matrix, the ID of the foreign target is the column index of the cost matrix, and the distance between the ground blocking robot and the foreign target is the content of the matrix.

[0020] The expansion of the cost matrix is specifically:

[0021] If a foreign target can be followed by at most N max ground blocking robots, when the number N of ground blocking robots performing tasks is greater than the number M of foreign targets, the cost matrix is expanded, and row i is the trajectory length vector of the ground blocking robot r i to all foreign targets, taking the minimum value n in N / M up rounding and N max -1, and expanding the cost matrix to n rows.

[0022] According to the distance between the foreign target and the task area and the moving speed of the foreign target, three ways of deploying ground blocking robots, blocking unmanned aerial vehicles and ground-air combination are selected to block the foreign target:

[0023] When any one of the moving speed of the foreign target or the distance between the foreign target and the task area exceeds a predetermined threshold, the blocking unmanned aerial vehicle is preferentially deployed;

[0024] When both the moving speed of the foreign target and the distance between the foreign target and the task area are within the threshold, the ground blocking robot is preferentially selected.

[0025] If the number of foreign targets exceeds the threshold, a ground-space combined deployment mode is selected to block the targets.

[0026] The step 5) comprises the following steps:

[0027] 5.1) The number of robot neighbors is selected as the degree of the robot, the ground search robot with the smallest degree in the multi-robot protection formation is selected as the gradient source robot, the gradient value of the source robot is initialized to 0, and the gradient value is diffused to the neighbor robots, the gradient of the robot directly connected with the source robot is 1, then the gradient of the robot is 2, and the gradient of the robot is sequentially increased until each robot in the formation obtains the gradient value of itself and the neighbor robots;

[0028] 5.2) When a robot in the formation performs a blocking task or fails, causing a vacancy in the formation, the neighbor with the smallest gradient among the neighbors of the vacancy robot is selected as the repair robot, which becomes the first repair robot to fill the vacancy, and if there are multiple robots with the smallest gradient, one of them is randomly selected as the repair robot according to the negotiation of the neighbor robots;

[0029] 5.3) The repair robot specifies the neighbor robot with the smallest gradient among the neighbor robots as the next repair robot to fill the vacancy left by the previous repair robot, and the iteration is performed until no repair robot can be selected, and a new formation is constructed;

[0030] 5.4) The search unmanned aerial vehicle monitors whether the repaired formation satisfies the lateral protection distance and maintains the original distance, and only shortens the longitudinal distance of the task area, if yes, step 5.5) is performed, and if not, steps 5.1) to 5.3) are repeatedly performed;

[0031] 5.5) The communication connection between the neighbors is re-established, and the formation repair is completed.

[0032] The present application has the following advantages and benefits:

[0033] 1. The present application integrates formation construction, target detection, target blocking and formation recovery, and can dynamically protect the task area in real time.

[0034] 2. The present application uses a “lattice” formation construction form, which can protect the task area in all directions, and can quickly recover the formation, improving the overall efficiency and accuracy of the system.

[0035] 3. The application adopts air-ground cooperation and man-machine cooperation to dynamically control the system, improves the robustness of the multi-robot system, and can perform tasks in a complex environment. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 Flow chart of the method;

[0037] Figure 2 Formation construction schematic diagram of the method;

[0038] Figure 3 Foreign target detection flow chart;

[0039] Figure 4 Foreign target distribution flow chart;

[0040] Figure 5 Expansion matrix diagram;

[0041] Figure 6 Robot formation recovery schematic diagram. DETAILED DESCRIPTION

[0042] The application will be further described in detail below in combination with the drawings and examples.

[0043] As shown in Figure 1 , the system constructs a multi-robot protection formation in the task area according to the number of robots and the area. When a foreign target enters the monitoring area, the ground search robot distributed on the boundary of the task area warns the foreign target and wakes up the search unmanned aerial vehicle. After receiving the warning from the ground search robot, the search unmanned aerial vehicle identifies the foreign target, wakes up all the robots in the area, and sends corresponding information to the remote control terminal and gets its reply. As the area command robot, it controls all the robots in the area, predicts the trajectory of the foreign target according to the speed and position of the foreign target, and distributes tasks to the blocking robots. When the blocking robots receive the task instructions from the search unmanned aerial vehicle, they go to the target point to block the target. At the same time, the remaining robots in the area perform formation self-recovery, fill the vacancy, maintain the formation shape, and protect the task area.

[0044] As shown in Figure 2As shown, the formation construction is developed on the basis of the Boid cluster model. In the task area, all ground robots are in the form of "quasi-lattice", that is, the distance and speed between adjacent robots remain the same. The ground search robots are distributed at the edge of the task area, which facilitates multi-directional detection and early warning of external targets. The search unmanned aerial vehicle is arranged in the central area of the region, which facilitates control of all robots. The blocking unmanned aerial vehicle is arranged inside the task area, and the ground blocking robots are distributed in the entire region in a corresponding formation, and there is no straight-through channel in the region, that is, each longitudinal column in the formation has a ground blocking robot.

[0045] The topological structure of the ground multi-robot is abstracted as an undirected acyclic graph G=(v,ε) in graph theory, where v represents the vertex set of the graph, and ε represents the edge set of the graph. The neighbor relationship between the ground robots is mapped to the adjacency matrix A=[a ij ] in graph theory. If a ij ≠0, it indicates that there is an edge between the two ground robots, and they are in a neighbor relationship. q i represents the position of the ground robot i. In the cluster formation, the positions of adjacent robots satisfy the quasi-lattice algebraic constraint: The neighbor set of robot i is defined as: And the concept of the degree of robot i is introduced: The distance between two robots is represented by the σ norm of the vector ||Z|| σ . The distance is controlled by the total potential energy between the robots and the action function. When all robots are clustered, the total potential energy (Ψ(Z), defined as a non-negative function) is infinitely close to the local minimum point of Ψ(Z), and vice versa. The derivative of the potential field function is the action function, and the value of the action function determines the force between two adjacent robots, thereby controlling the distance between the robots and completing the formation construction.

[0046] As shown in Figure 3 , the ground search robot warns of external targets and wakes up the search unmanned aerial vehicle. The search unmanned aerial vehicle identifies the target, wakes up other robots, and sends target images, position coordinates, task area robot positions, and states to the remote control terminal. After entering the working state, the search unmanned aerial vehicle serves as the command robot of the task area, receives remote control terminal instructions, and controls all robots in the task area.

[0047] As shown in Figure 4As shown, when the mission area is threatened by an external target, the search drone determines the number of blocking robots to deploy based on the external target information and assigns targets to the blocking robots according to the principles of proximity and uniform distribution. Each blocking robot broadcasts a list of its global trajectory lengths to the external target to the search drone. By receiving and recording the global trajectory lengths published by all blocking robots, the search drone autonomously determines the number and IDs of the blocking robots to deploy based on global optimal conditions, and constructs a cost matrix based on the broadcast trajectory lengths: the blocking robot's ID is the row index of the cost matrix, the external target's ID is the column index of the cost matrix, and the distance between the blocking robot and the external target is the matrix content. To compensate for the one-to-one assignment limitation of the Hungarian algorithm, this method expands the cost matrix and then uses the idea of ​​the Hungarian algorithm to assign targets to the robots. That is, if a target can be assigned to at most N targets... max If there are multiple blocking robots following, and the number of blocking robots N performing the task exceeds the number of external targets M, the cost matrix will be expanded. Let row i For robot r i The trajectory length vector to all targets, rounded up to N / M, sums to N. max The minimum value n in -1 is used to expand the cost matrix so that each row contains n rows.

[0048] like Figure 5 As shown, assuming that external targets R1, R2, and R3 are about to enter the mission area, based on the information about the external targets, the search drone will decide to use four blocking robots to block these three targets. First, based on the requirement of minimizing the movement distance and time, four blocking robots, L1, L2, L3, and L4, are selected to perform the task. Now, the task is assigned using the method mentioned above: First, N... max If we choose 3, then we calculate n to be 2. The extended cost matrix changes from the original 4×3 to 4×6. We use the Hungarian algorithm to calculate the extended matrix and finally obtain the target of L1 and L4 as R1, the target of L2 as R2, and the target of L3 as R3.

[0049] Based on the distance between the incoming target and the mission area, and the target's movement speed, the search drone can choose from three methods to block the incoming target: deploying ground blocking robots, blocking drones, or a combination of ground and air methods. First, thresholds are set for the incoming target's movement speed, distance, and number. When the speed and distance exceed these thresholds, blocking drones will be deployed first. When the speed and distance are within the thresholds, ground blocking robots will be used first. If the number of incoming targets exceeds the threshold, a combination of ground and air methods will be used to block the target.

[0050] like Figure 6As shown, when part of the robots in the formation execute tasks or there is a robot failure in the area, the position of the robots in the formation should be adjusted in time to realize self-recovery of the robot formation in the area and real-time protection of the task area. When a robot is missing in the robot formation, a gradient diffusion mechanism can be used to improve the repair efficiency of the formation. The missing robot can be marked as R f , and the degree of the missing robot can be marked as d f . To realize diffusion of the gradient, a robot that satisfies is selected by the formation members as a source robot for generating the gradient, and a set of local interaction rules is used for diffusion of the gradient. This can form a stable gradient distribution in the formation. Finally, the formation members can estimate the distance from the source robot by the size of the gradient value of the formation members themselves, and use this information as one of the bases for selection of the repair robot.

[0051] Each robot in the formation first determines whether it is a gradient source robot according to a comparison of the gradient values with other robots. If the gradient value of the robot S is very small, the robot S is the gradient source robot, the gradient value is initialized to 0, and the gradient value is sent to the neighbor robots. The other robots receive the gradient values of the neighbors, update the gradient values of the robots according to the specified rules, and send the updated gradient values to the neighbor robots, so that each robot in the formation has a gradient value relative to the source robot, the gradient diffusion is completed, and the formation vacancy can be quickly filled to complete the formation repair.

[0052] According to the recursive switching topology rule, the basic steps of the recursive self-repair algorithm of the ground robot formation are as follows:

[0053] 1) The ground robot formation autonomously selects a ground robot with a very small degree as a gradient source robot, initializes the gradient value of the source robot to 0, and diffuses the gradient value to the neighbor robots until each robot obtains the gradient values of itself and the neighbors, and a stable gradient distribution is formed in the formation;

[0054] 2) When a robot in the formation executes a blocking task or fails, causing a vacancy in the formation, the neighbor of the vacancy robot selects the neighbor with the smallest degree as the repair robot according to the interaction information, becomes the repair robot in step 1, and fills the vacancy. If there are multiple robots with the smallest degree, one of them is randomly selected as the repair robot according to the negotiation of the neighbor robots;

[0055] 3) The repair robot specifies a robot that meets the condition (i.e., the neighbor with the smallest gradient) among the neighbor robots as the repair robot in the next step, fills the vacancy left by the repair robot in the previous step, and iterates until no repair robot can be selected, and a new formation is constructed;

[0056] 4) Search the UAV monitoring repair formation to meet the requirements of "horizontal protection distance keeps the original distance, only shortens the longitudinal distance of the task area", if it meets, execute the next step, if it does not meet, repeat the first, second and third steps;

[0057] 5) Re-establish the inter-neighbor communication connection and complete the formation repair.

Claims

1. A dynamic environment-oriented multi-robot cooperative area protection method, characterized by, The method comprises the following steps: 1) constructing a multi-robot protection formation in the task area according to the number of robots and the area of the region; 2) monitoring the task area by the search robots in the multi-robot protection formation and sending the monitoring results to the remote control terminal; 3) assigning tasks to the blocking robots in the multi-robot protection formation according to the monitoring results by the remote control terminal; 4) blocking the target by the blocking robots according to the task instructions: According to the distance between the external target and the task area and the moving speed of the external target, three ways of deploying ground blocking robots, blocking unmanned aerial vehicles and ground-air combination are selected to block the external target: When any one of the moving speed of the external target or the distance between the external target and the task area exceeds a predetermined threshold, the blocking unmanned aerial vehicle is preferentially deployed; When the moving speed of the external target and the distance between the external target and the task area are both within the threshold, the ground blocking robot is preferentially selected; If the number of external targets exceeds the threshold, the ground-air combination deployment mode is selected to block the target; 5) the remaining robots in the multi-robot protection formation perform formation self-recovery to fill the vacancy of the blocking robots.

2. The method according to claim 1, wherein, The multi-robot protection formation is constructed based on the Boid cluster model and comprises blocking unmanned aerial vehicles, search unmanned aerial vehicles, ground blocking robots and ground search robots, wherein the interval and the advancing speed between adjacent robots remain the same state, the ground search robots are deployed on the four perimeter boundaries of the task area, the search unmanned aerial vehicles are deployed in the center of the task area, the blocking unmanned aerial vehicles are deployed inside the task area, and the ground blocking robots are deployed in a set formation in the entire task area, and the requirement that each longitudinal formation in the formation in the task area has a ground blocking robot is met.

3. The method of claim 1, wherein, The step 2) is specifically: When the external target enters the task area, the pre-warning state of the ground search robot is triggered, the ground search robot wakes up the search unmanned aerial vehicle in the region, the search unmanned aerial vehicle identifies and monitors the type and position coordinate information of the external target after receiving the pre-warning information sent by the ground search robot, and wakes up other idle robots and sends the type and position coordinate information of the external target and the position information of the robots in the task area to the remote control terminal.

4. The method of claim 1, wherein, The step 3) comprises the following steps: 3.1) when the external target is about to enter the task area, each ground blocking robot broadcasts the global trajectory length list of the external target to the search unmanned aerial vehicle; 3.2) the search unmanned aerial vehicle receives and records the global trajectory length published by all ground blocking robots and forwards it to the remote control terminal; 3.3) the remote control terminal determines the number and ID of the ground blocking robots to be activated according to the global optimal condition and constructs a cost matrix according to the broadcasted trajectory length; 3.4) the cost matrix is expanded, and the Hungarian algorithm is used to assign targets to the robots based on the expanded cost matrix.

5. The method of claim 4, wherein, The cost matrix is specifically: the ID of the ground blocking robot is the row index of the cost matrix, the ID of the external target is the column index of the cost matrix, and the distance between the ground blocking robot and the external target is the content of the matrix.

6. The method of claim 4, wherein, The expansion of the cost matrix is specifically: If an extraneous target can be followed by at most ground blocking robots, when the number of ground blocking robots performing a task is more than the number of extraneous targets , the cost matrix is extended by letting be the trajectory length vector of the ground blocking robots to all extraneous targets, taking the ceiling of the minimum of and , and extending the cost matrix by .​ 7. The method of claim 1, wherein, The step 5) comprises the following steps; 5.1) The number of robot neighbors is selected as the degree of the robot, the ground search robot with the minimum degree in the multi-robot protective formation is selected as the gradient source robot, the gradient value of the source robot is initialized to 0, and is diffused to the neighbor robots, the gradient of the robot directly connected to the source robot is 1, then the gradient of the robot connected to the robot with the gradient of 1 is 2, and so on, until each robot in the formation obtains its own and neighbor gradient values; 5.2) When there is a robot in the formation performing a blocking task or a fault occurs, causing the formation to have a vacancy, the neighbor with the minimum gradient among the neighbors of the vacancy robot is selected as the repair robot, which becomes the repair robot in the first step, to fill the vacancy, if there are multiple robots with the minimum gradient, one of them is randomly selected as the repair robot according to the negotiation of the neighbor robots; 5.3) The repair robot specifies the neighbor robot with the minimum gradient among its neighbors as the repair robot in the next step, to fill the vacancy left by the repair robot in the previous step, and the iteration is performed until no repair robot can be selected, and a new formation is constructed; 5.4) The search unmanned aerial vehicle monitors whether the repaired formation meets the horizontal protective distance, keeps the original distance, and only shortens the longitudinal distance of the task area, if it meets, step 5.5) is performed, if it does not meet, steps 5.1) to 5.3) are repeatedly performed; 5.5) Re-establish the communication connection between the neighbors, and complete the formation repair.

Images