A multi-robot search method based on a packet-based collaborative decision mechanism

Abstract

The application discloses a kind of multi-robot search methods based on grouping cooperative decision mechanism, including defining robot local communication set;The search environment of robot is constructed using grid map, the task search area of robot is divided into L1*L2 Same area grid, the grid map after construction is fused with simplified biological inspired neural network, and environment information representation model is established;The environment information detected by robot in the same local communication set is fused;The robot in local communication set is divided into multiple search groups;The iteration decision sequence and decision process of the robot in each search group are determined, until the robot in group decides to complete;All robots move to next step and update environment information until the entire task area is searched out.Solve the problem that multi-robot is difficult to search task area efficiently due to limited communication in unknown environment.

Description

Technical Field

[0001] This invention belongs to the field of robot area search technology, and in particular relates to a multi-robot search method based on a group collaborative decision-making mechanism. Background Technology

[0002] With the development of robotics technology, autonomous mobile robots can now assist humans in completing specific tasks, such as battlefield reconnaissance, space exploration, disaster search and rescue, and ground cleanup. These tasks can be summarized as full-area search problems in unknown environments. An unknown environment refers to a situation where the location and distribution of targets and obstacles within the robot's search area are unknown, but the robot's search boundaries are known. This requires the robot to continuously search unexplored areas while avoiding redundant exploration. The constraints faced by autonomous robots in unknown environments are as follows: 1) The detection range of the robot's sensors is limited relative to the size of the task area; 2) The autonomous robot has no prior environmental information before the search task begins, including the distribution of targets and obstacles within the task area; 3) The autonomous robot must avoid obstacles in real time within the unknown environment.

[0003] Compared to a single autonomous robot, multi-robot systems offer advantages such as flexibility, robustness, and parallelism, improving the efficiency of area search tasks. Therefore, this paper considers using multiple autonomous robots to perform area search tasks in unknown environments. This task requires multiple robots to acquire environmental information through sensors and collaboratively complete the area search in the unknown environment with minimal cost. Furthermore, considering that communication conditions between robots may be less than ideal in harsh environments, such as disaster relief sites, multi-robot systems need to meet two additional constraints when performing area search tasks: 1) avoiding collisions between robots during the search process; 2) limited communication range between robots, allowing only local communication. Current multi-robot area search methods mainly focus on area search tasks under ideal communication conditions. For multi-robot systems, how to efficiently perform area search tasks under limited communication conditions (where robots cannot perform global communication) remains a research challenge. To address the problem of reduced area search efficiency in current multi-robot area search technology research due to the inability of multiple robots to perform global communication under limited communication conditions, a multi-robot search method based on a grouped collaborative decision-making mechanism is designed and invented. Summary of the Invention

[0004] To address the above technical problems, this invention provides a multi-robot search method based on a grouped collaborative decision-making mechanism.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A multi-robot search method based on a group collaborative decision-making mechanism includes the following steps:

[0007] S100: Treat each robot as a communication node and define the robot's local communication set;

[0008] S200: The robot's search environment is constructed using a grid map. The robot's task search area is divided into L1*L2 grids of equal area. The constructed grid map is then fused with a simplified biologically inspired neural network (SBNN) to establish an environmental information representation model.

[0009] S300: Fusion of environmental information detected by the robot within the same local communication set;

[0010] S400: Divides robots within a local communication set into multiple search groups;

[0011] S500: Determine the iterative decision-making sequence and process of the robots within each search team until the robots within the team have completed their decisions;

[0012] S600: All robots move to the next step and update environmental information until the entire task area has been searched.

[0013] Preferably, S100 specifically comprises:

[0014] Assume robot R m The location coordinates in the raster map are (x m ,y m Robot R h The location coordinates in the raster map are (x h y h If the Euclidean distance d between two robots is less than the maximum communication radius L of the robots... c Then these two robots belong to the same communication set P. z :

[0015] P z ={{R m R h}:d(R m R h )≤L c m, h = 1, 2, ..., N r ,m≠h}

[0016] Where z = 1, 2, ..., n1, n1 represents the number of robot local communication sets, and N r For the total number of robots,

[0017] Preferably, in S200, a grid map is used to construct the robot's search environment, dividing the robot's task search area into L1*L2 grids of equal area, including:

[0018] The state of a grid includes the grid containing the target, the grid containing obstacles, the grid without obstacles, and the grid that has not been searched. The state representation of the constructed grid map is shown below:

[0019]

[0020] Where S{G(h,v)} represents the state of grid G(h,v), G(h,v) represents the grid at position (h,v), h∈[1,L1], v∈[1,L2], UA represents the unsearched grid, OA represents the grid containing the obstacle, CA represents the grid without obstacles, and TA represents the grid containing the target.

[0021] Preferably, in step S200, the constructed grid map is fused with a simplified biologically inspired neural network (SBNN) to establish an environmental information representation model, including:

[0022] Each grid corresponds to a neuron activity value x. i The SBNN model is shown below:

[0023]

[0024] The transfer function g(x) is set as follows:

[0025]

[0026] Where x i (t+1) represents the activity value of the i-th neuron at time (t+1), x j (t) represents the activity value of the j-th neuron at time t, [x j (t)] + =max[x j [t), 0], this formula indicates that only positive neurons can influence other neurons and have a global propagation effect, while negative neurons cannot propagate outwards and have a local effect, I i Let M represent the external input signal of the SBNN neuron, M represent the connection coefficient between the i-th neuron and its connected neurons, and W represent the external input signal of the SBNN neuron. ij The connection weights between neurons i and j are defined as follows:

[0027]

[0028] Where |e i -e j | represents a vector e in the state space iand e j The Euclidean distance between them, where α and r are both positive constants;

[0029] The method of combining the constructed raster map with SBNN is shown below:

[0030]

[0031] Where E >> 1, and the value range is [3, 10], it ensures that the target grid is at its peak in the neuron activity scene. Once the target is discovered, the robot will move towards the target first, while the activity value of the obstacle is at its lowest point. S{G(h, v)} represents the state of grid G(h, v).

[0032] Preferably, S300 includes:

[0033] Within the same local communication set, the neuron activity matrices of different robots are fused using a weighted average method:

[0034]

[0035] in This represents the communication set P of the robot. Z Inside, the matrix of neuron activity values ​​stored in the g-th robot, This represents the neuronal activity value of the grid G(h,v) after information fusion. It is a set PZ 内 The number of robot members, W g This represents the corresponding weighting coefficient; where W g The representation is as follows:

[0036]

[0037] in This represents the number of times the robot explores the grid G(h, v).

[0038] Preferably, S400 includes:

[0039] If two robots R m R h If both robots meet the following conditions: they belong to the same communication set and their decision spaces overlap, then these two robots will be assigned to the same search group A. i i = 1, 2, ..., n3, where n3 is the number of search groups:

[0040] R m R h ∈A i i = 1, 2, ..., n3

[0041]

[0042] Where A i This represents the i-th robot search group, n3 represents the total number of robot search groups, L represents the number of steps predicted for robot motion, and D... R This represents the farthest distance each robot can move in each step.

[0043] Preferably, determining the iterative decision-making order of the robots within each search group in S500 includes:

[0044] After the robots are grouped, iterative collaborative decision-making is carried out within each search group, and the order of robot iterative decisions is determined in a self-organizing manner:

[0045]

[0046] in This indicates that within the robot search team, its decision-making order is ranked first. This indicates the total number of robots within the robot search team.

[0047] The method for determining the search group A is as follows: Calculate the search group A i center of mass coordinate Find the distance from the center of mass The most recent robot, namely the robot that makes the first decision. The second robot to make motion decisions The method of determination is distance The latest robot, the third robot to make motion decisions Distance The most recent robot, and so on, determines the decision-making order of all robots within the same search group:

[0048]

[0049]

[0050] in This refers to the robot search team A. i The location information of the g-th robot. Represents robots With the center of mass Euclidean distance.

[0051] Preferably, the iterative decision-making process for determining the robots within each search group in S500 includes:

[0052] In search team Ai Within the system, each robot, when conducting a search, obtains its search efficiency based on the current neuron activity matrix H and the user-designed search performance function. L-step motion control input

[0053]

[0054]

[0055] J (L) {g a (k), u a (k)} represents the robot's predicted L-step search performance function, which is used to optimize and solve for... Future L-step motion control input U a (k) represents the robot Step k control input, g a (k) represents The state at step k, the state of the first robot to make a decision. L-step control input prediction based on neuron activity matrix Acquired Then, it sends its decision information to the next robot that makes a decision.

[0056] The aforementioned multi-robot search method based on a group collaborative decision-making mechanism designs an environmental information fusion and update model. For robots within the same communication set, the activity values ​​of neurons in a simplified bio-inspired neural network (SBNN) are used to represent the real-time environmental information detected by the robots. The environmental information detected by different robots is then fused based on a weighted fusion approach. Secondly, a group collaborative decision-making mechanism is designed, dividing robots within the same communication set into several dynamic search groups. Robots within each dynamic search group perform iterative collaborative decision-making to plan the next search path in real time, thus solving the problem that multi-robot systems struggle to efficiently search the task area in unknown environments due to limited multi-robot communication. Attached Figure Description

[0057] Figure 1 This is a flowchart of a multi-robot search method based on a grouped collaborative decision-making mechanism in one embodiment of the present invention;

[0058] Figure 2 This is a diagram of a two-dimensional SBNN structure designed in one embodiment of the present invention;

[0059] Figure 3 This is an example diagram of a robot communication set designed in one embodiment of the present invention;

[0060] Figure 4 This is a grouping example diagram of the robots designed in one embodiment of the present invention;

[0061] Figure 5 This is an example diagram of a multi-robot area search process under limited communication conditions designed in one embodiment of the present invention. Detailed Implementation

[0062] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0063] In one embodiment, such as Figure 1 As shown, a multi-robot search method based on a group collaborative decision-making mechanism includes the following steps:

[0064] S100: Treat each robot as a communication node and define the robot's local communication set.

[0065] Specifically, considering the limited communication capabilities of a single robot, robots can only establish local communication with each other. A mobile ad hoc network is a multi-hop wireless communication network composed of mobile nodes, where each node acts as both an information terminal and a router.

[0066] In one embodiment, S100 specifically includes:

[0067] Assume robot R m The location coordinates in the raster map are (x m y m Robot R h The location coordinates in the raster map are (x h y h If the Euclidean distance d between two robots is less than the maximum communication radius L of the robots... c Then these two robots belong to the same communication set P. z :

[0068] P z ={{R m R h}:d(R m R h )≤L c m, h = 1, 2, ..., N r ,m≠h}

[0069] Where z = 1, 2, ..., n1, n1 represents the number of robot local communication sets, and N r For the total number of robots,

[0070] Specifically, if one robot is within L of the maximum communication radius of another robot... c Within a given communication set, the two robots can be considered to be in a local communication set, capable of communicating with each other. Robots belonging to the same communication set can freely exchange information. As the robots move, the number and members of the robot communication set change dynamically. An example diagram of a communication set is shown below. Figure 3 As shown.

[0071] S200: The robot's search environment is constructed using a grid map. The robot's task search area is divided into L1*L2 grids of equal area. The constructed grid map is then fused with a simplified biological-inspired neural network (SBNN) to establish an environmental information representation model.

[0072] In one embodiment, S200 uses a grid map to construct the robot's search environment, dividing the robot's task search area into L1*L2 grids of equal area, including:

[0073] The state of a grid includes the grid containing the target, the grid containing obstacles, the grid without obstacles, and the grid that has not been searched. The state representation of the constructed grid map is shown below:

[0074]

[0075] Where S{G(h,v)} represents the state of grid G(h,v), G(h,v) represents the grid at position (h,v), h∈[1,L1], v∈[1,L2], UA represents the unsearched grid, OA represents the grid containing the obstacle, CA represents the grid without obstacles, and TA represents the grid containing the target.

[0076] In one embodiment, S200 involves fusing the constructed grid map with a simplified biologically inspired neural network (SBNN) to establish an environmental information representation model, including:

[0077] Each grid corresponds to a neuron activity value x. i The SBNN model is shown below:

[0078]

[0079] The transfer function .9(x) is set as follows:

[0080]

[0081] Where x i (t+1) represents the activity value of the i-th neuron at time (t+1), xj (t) represents the activity value of the j-th neuron at time t, [x j (t)] + =max[x j [t), 0], this formula indicates that only positive neurons can influence other neurons and have a global propagation effect, while negative neurons cannot propagate outwards and have a local effect, I i Let M represent the external input signal of the SBNN neuron, M represent the connection coefficient between the i-th neuron and its connected neurons, and W represent the external input signal of the SBNN neuron. ij The connection weights between neurons i and j are defined as follows:

[0082]

[0083] Where |e i -e j | represents a vector e in the state space i and e j The Euclidean distance between them, where α and r are both positive constants;

[0084] The method of combining the constructed raster map with SBNN is shown below:

[0085]

[0086] Where E >> 1, and the value range is [3, 10], it ensures that the target grid is at its peak in the neuron activity scene. Once the target is discovered, the robot will move towards the target first, while the activity value of the obstacle is at its lowest point. S{G(h, v)} represents the state of grid G(h, v).

[0087] Specifically, the structural diagram of SBNN is as follows: Figure 2 As shown.

[0088] S300: Fusion of environmental information detected by the robot within the same local communication set.

[0089] Specifically, since each robot operates independently and possesses its own SBNN, and the robot's detection range is limited, it can only acquire environmental information within its detection range. Therefore, it is necessary to integrate the environmental information collected by different robots to improve search efficiency.

[0090] In one embodiment, S300 includes:

[0091] Within the same local communication set, the neuron activity matrices of different robots are fused using a weighted average method:

[0092]

[0093] in This represents the communication set P of the robot. Z Inside, the matrix of neuron activity values ​​stored in the g-th robot, This represents the neuronal activity value of the grid G(h,v) after information fusion. It is set P Z The number of robot members inside, W g This represents the corresponding weighting coefficient; where W g The representation is as follows:

[0094]

[0095] in This represents the number of times the robot explores the grid G(h, v).

[0096] S400: Divides robots within a local communication set into multiple search groups.

[0097] Specifically, after constructing the environmental information fusion and update model, the robots within the local communication set are divided into multiple search groups. An example diagram of the search groups is shown below. Figure 4 As shown, Figure 4 (a) represents the grouping of the 9 robots at a certain moment. Figure 4 (b) shows the grouping of the 9 robots in the next moment. It can be seen that the number of groups and members of the multiple robots are dynamic.

[0098] In one embodiment, S400 includes:

[0099] If two robots R m R h If both robots meet the following conditions: they belong to the same communication set and their decision spaces overlap, then these two robots will be assigned to the same search group A. i i = 1, 2, ..., n3, where n3 is the number of search groups:

[0100] R m R h ∈A i i = 1, 2, ..., n3

[0101]

[0102] Where A i This represents the i-th robot search group, n3 represents the total number of robot search groups, L represents the number of steps predicted for robot motion, and D... R This represents the farthest distance each robot can move in each step.

[0103] S500: Determine the iterative decision-making sequence and process of the robots within each search team until the robots within the team have completed their decisions.

[0104] In one embodiment, determining the iterative decision-making order of the robots within each search group in S500 includes:

[0105] After the robots are grouped, iterative collaborative decision-making is carried out within each search group, and the order of robot iterative decisions is determined in a self-organizing manner:

[0106]

[0107] in This indicates that within the robot search team, its decision-making order is ranked first. This indicates the total number of robots within the robot search team.

[0108] The method for determining the search group A is as follows: Calculate the search group A i center of mass coordinate Find the distance from the center of mass The most recent robot, namely the robot that makes the first decision. The second robot to make motion decisions The method of determination is distance The latest robot, the third robot to make motion decisions Distance The most recent robot, and so on, determines the decision-making order of all robots within the same search group:

[0109]

[0110]

[0111] in This refers to the robot search team A. i The location information of the g-th robot. Represents robots With the center of mass Euclidean distance.

[0112] In one embodiment, the iterative decision-making process for determining the robots within each search group in S500 includes:

[0113] In search team A i Within the system, each robot, when conducting a search, obtains its search efficiency based on the current neuron activity matrix H and the user-designed search performance function. L-step motion control input

[0114]

[0115]

[0116] J (L) {g a (k), u a (k)} represents the robot's predicted L-step search performance function, which is used to optimize and solve for... Future L-step motion control input U a (k) represents the robot Step k control input, g a (k) represents The state at step k, the first robot to make a decision. L-step control input prediction based on neuron activity matrix Acquired Then, it sends its decision information to the next robot that makes a decision.

[0117] S600: All robots move to the next step and update environmental information until the entire task area has been searched.

[0118] Specifically, the robots in each search team are selected As its motion control input, it moves to the next step and updates environmental information until all targets in the area have been searched.

[0119] The following examples will be used to verify and analyze this, with the experimental parameters set as follows:

[0120] 1) 20*20 grid map; 2) Number of robots set to 8; 3) Number of targets set to 10; 4) Communication radius of robots is 4.

[0121] Figure 5 An example of a multi-robot area search process is shown, in which Figure 5 (a) shows the area search results of 8 robots moving 2 steps. Figure 5 (b) The area search results of 8 robots moving 5 steps. Figure 5 (c) shows the area search results of 8 robots moving 9 steps. Figure 5 (d) shows the area search results of 8 robots moving 18 steps. It can be seen that the multi-robot system can quickly complete the area search task of a specified area under local communication conditions and complete the search of 10 targets.

[0122] The aforementioned multi-robot search method based on a group collaborative decision-making mechanism designs an environmental information fusion and update model. For robots within the same communication set, the activity values ​​of neurons in a simplified bio-inspired neural network (SBNN) are used to represent the real-time environmental information detected by the robots. The environmental information detected by different robots is then fused based on a weighted fusion approach. Secondly, a group collaborative decision-making mechanism is designed, dividing robots within the same communication set into several dynamic search groups. Robots within each dynamic search group perform iterative collaborative decision-making to plan the next search path in real time, thus solving the problem that multi-robot systems struggle to efficiently search the task area in unknown environments due to limited multi-robot communication.

[0123] The above provides a detailed description of a multi-robot search method based on a grouped collaborative decision-making mechanism provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the core ideas of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this invention.

Claims

1. A multi-robot search method based on a grouped collaborative decision-making mechanism, characterized in that, The method includes the following steps: S100: Treat each robot as a communication node and define the robot's local communication set; S200: Uses a grid map to construct the robot's search environment, dividing the robot's task search area into... A grid of equal area is used to construct a grid map and fuse it with a simplified biologically inspired neural network (SBNN) to establish an environmental information representation model. S300: Fusion of environmental information detected by the robot within the same local communication set; S300 includes: Within the same local communication set, the neuron activity matrices of different robots are fused using a weighted average method: in Represents the communication set of the robot within, no. A matrix of neuron activity values ​​stored in each robot. Represents the raster after information fusion Neuron activity values, It is a set The number of robot members inside, This represents the corresponding weight coefficients; where The representation is as follows: in This refers to the robot's relationship with the grid. Number of explorations; S400: Divides robots within a local communication set into multiple search groups; S500: Determine the iterative decision-making sequence and process of the robots within each search team until the robots within the team have completed their decisions; S600: All robots move to the next step and update environmental information until the entire task area has been searched.

2. The method according to claim 1, characterized in that, S100 specifically refers to: Assuming a robot The location coordinates in the raster map are ,robot The location coordinates in the raster map are If the Euclidean distance between the two robots Less than the robot's maximum communication radius If the two robots belong to the same communication set, then they are in the same communication set. : in, , This indicates the number of local communication sets of the robot. For the total number of robots, .

3. The method according to claim 2, characterized in that, The S200 uses a grid map to construct the robot's search environment, dividing the robot's task search area into... A grid of equal area, including: The state of a grid includes the grid containing the target, the grid containing obstacles, the grid without obstacles, and the grid that has not been searched. The state representation of the constructed grid map is shown below: in Represents grid state, Indicates the position as The grid, , , Indicates no raster was searched. Indicates the grid cell where the obstacle is located. Indicates an obstacle-free grid. Indicates the grid cell where the target is located.

4. The method according to claim 3, characterized in that, In S200, the constructed grid map is fused with a simplified biologically inspired neural network (SBNN) to establish an environmental information representation model, including: Each grid corresponds to a neuron activity value. The SBNN model is shown below: Where the transfer function Set to: in It indicates the first One neuron in Activity value at time, It indicates the first One neuron in Activity value at time, This formula indicates that only positive neurons can influence other neurons and have a global propagation effect, while negative neurons cannot propagate outwards and have a local effect. This represents the external input signal of an SBNN neuron. Indicates the first The connection coefficients between a neuron and its connected neurons. Represents neurons With neurons The connection weights are defined as follows: in Represents a vector in the state space and The Euclidean distance between them and All are positive constants; The method of combining the constructed raster map with SBNN is shown below: in 1. The value range is [3, 10]. This ensures that the target grid is at its peak activity level in the neuron activity scene. Once the target is detected, the robot prioritizes moving towards the target, while the activity value of obstacles is at its lowest point. S{ } represents a grid The state.

5. The method according to claim 4, characterized in that, The S400 includes: If two robots If both robots meet the following conditions—that they belong to the same communication set and their decision spaces overlap—then they will be assigned to the same search team. , Number of search teams: in It indicates the first A robot search team, This indicates the total number of robot search teams. This indicates the number of steps predicted for the robot's motion. This represents the farthest distance each robot can move in each step.

6. The method according to claim 5, characterized in that, The S500 determines the iterative decision-making order of the robots within each search team, including: After the robots are grouped, iterative collaborative decision-making is carried out within each search group, and the order of robot iterative decisions is determined in a self-organizing manner: in This indicates that within the robot search team, its decision-making order is ranked first. This indicates the total number of robots within the robot search team. The method for determining the search group is as follows: Calculate the search group center of mass coordinate( , Find the distance from the center of mass. The most recent robot, namely the robot that makes the first decision. The second robot to make motion decisions The method of determination is distance The latest robot, the third robot to make motion decisions Distance The most recent robot, and so on, determines the decision-making order of all robots within the same search group: in( , () indicates the robot search team The Middle Location information of each robot Represents robots With the center of mass Euclidean distance.

7. The method according to claim 6, characterized in that, The S500 defines the iterative decision-making process for the robots within each search team, including: In the search team Within the system, each robot performs a search based on its current neuron activity matrix. And user-designed search performance functions to obtain of L Step motion control input : in This refers to robot predictions. L Step-by-step search for the performance function, and then optimize the solution by searching for the performance function. The Future Step motion control input ; Represents robots No. Step control input, express No. k The first robot to make a decision in the current state. Predictions based on neuron activity matrix Step control input , obtained Then, it sends its decision information to the next robot that makes a decision. .

Images