Group robot coverage path planning method based on grouping and merging mechanism in local communication environment

Abstract

The invention provides a group robot coverage path planning method based on a grouping and merging mechanism in a local communication environment, which comprises the following steps: dividing a map to be covered into a plurality of sub-regions, grouping robots, and allocating a specific sub-region to each group for coverage so as to realize preliminary task allocation. And the robots in the same group keep local continuous communication and share information, and jointly cover the distributed sub-regions. According to the method, the limitation of a local communication range on intra-group cooperation is considered, and the size of the group is increased along with the enlargement of the communication range. In addition, a merging mechanism is also designed to realize inter-group cooperation, and the covered group goes to other uncovered sub-regions and is cooperatively covered with other combinations. According to the group robot coverage path planning method based on the grouping and merging mechanism, the cooperation efficiency of group robots can be improved in an unknown environment with only local communication, and a coverage task can be efficiently completed.

Description

Technical Field

[0001] The present invention relates to the field of collaborative regional coverage by swarm robots, and in particular to a swarm robot coverage path planning method based on a grouping and merging mechanism in a local communication environment. Background Art

[0002] Mobile robots are widely used in military and civilian fields to perform many repetitive or dangerous tasks. Coverage path planning is a fundamental problem for mobile robots, which aims to cover the target area except obstacles and minimize the repetition rate of the robot's path. Coverage path planning has been widely studied and applied in many applications such as automatic cleaning, post-disaster rescue, target search, and environmental monitoring.

[0003] The coverage path planning problem is related to the coverage traveling salesman problem, which is a variant of the traveling salesman problem and is proven to be an NP-hard problem. Existing coverage path planning methods can be divided into two types: offline and online, depending on whether the global environment is known. The former assumes that the global environment is known and plans a path for the robot before performing the coverage task. The latter can be applied when the global environment is unknown or changing, which updates the environmental information based on sensors and plans the path in real time. In most practical tasks, the environment is unknown or partially known. Therefore, online coverage path planning is more practical.

[0004] At present, most online coverage path planning methods for swarm robots assume reliable global communication to achieve information transmission and task allocation between robots. However, in many environments, global communication cannot be guaranteed and robots only have local communication range. When the current online coverage path planning algorithm is applied in an environment with only local communication, the coordination efficiency between robots is low or even impossible, and its coverage efficiency is greatly reduced.

[0005] Existing multi-robot collaboration methods under local communication can be divided into three categories. The first method divides the map into sub-areas and assigns the sub-areas to robots for coverage, such as BPCPP. However, due to the uneven distribution of obstacles, the coverage completion time of each sub-area is different, and the robots that have completed the coverage cannot collaborate with other robots. The second method requires robots to meet regularly to share information and assign tasks, such as AMCPP, but the meeting process consumes additional energy and time, and the time and place of each meeting need to be determined in advance. The third method maintains local communication between all robots, such as DNCPP, but this greatly restricts the movement of robots and causes a large amount of repeated coverage.

[0006] In order to complete the coverage task efficiently and quickly in an unknown environment with only local communication, there is an urgent need for a coverage path planning method that can improve the collaborative efficiency between robots under local communication. Summary of the invention

[0007] In view of the problem that the existing swarm robot online coverage path planning method has low efficiency in collaborative coverage under local communication, the present invention proposes a swarm robot coverage path planning method based on grouping and merging mechanism under local communication environment. This method divides the map to be covered into multiple sub-areas and groups the robots. Each group is assigned a specific sub-area to cover, so as to achieve preliminary task allocation. The robots in the same group maintain local continuous communication to share information and jointly cover their assigned sub-areas. In addition, a merging mechanism is designed to achieve inter-group collaboration. The group that has completed coverage goes to other sub-areas that have not been completed and combines with other groups to cover collaboratively. Accordingly, the swarm robot can greatly improve the collaborative efficiency and efficiently complete the coverage task under only local communication.

[0008] The present invention is achieved through the following technical solutions:

[0009] A group robot coverage path planning method based on grouping and merging mechanism in a local communication environment, the steps are as follows:

[0010] Establish a coverage environment, map side length L; initialize algorithm parameters: number of robots n, communication range r, robot detection range V, starting position o s .

[0011] Step 1: First, the robots are adaptively grouped according to the size of the communication range to maximize the collaboration capability. The size of the group m is calculated according to the following formula:

[0012]

[0013] where n is the number of robots, r is the radius of the robot's communication range, L is the side length of the map, and β is a constant set to 1.35. m∈[1,n], and increases with r.

[0014] Number of groups The number of robots in group i, n i The calculation is as follows:

[0015]

[0016] Step 2: Partition the map and assign a partition to each group. The map is represented as a set of target points O = {o j :j=1,2,…,z}, z is the total number of target points. The robots initially gather at the starting point o s , each target point is assigned a quadrant angle θ j , is the target point and o s The angle between the vector and the reference axis, where the reference axis is o sis the positive horizontal axis of the starting point. First, sort the target points in ascending order according to the quadrant angle to obtain a set O ′ Then, in order, ′ The target points in are assigned to each group. The set of target points assigned to the i-th group is P i , the number of target points contained in it is |P i |, respectively, represent the following:

[0017]

[0018] in Represents the total number of target points assigned to group 1 to group i.

[0019] Step 3: After grouping and division, each group of robots moves to their corresponding sub-areas for distributed collaborative coverage. During this process, the robots in the group maintain continuous local communication and share the following information: target point set O, uncovered target point set Cover target point set and the set of target points occupied by obstacles To prevent disconnection from the group, robot i selects its next destination point o i The following constraints apply when moving: i The distance to any other robot is no greater than the communication range r to maintain communication with the group:

[0020] ||o i -r j ||≤r,

[0021] where r j represents the position of robot j, G i represents the set of robots in the group to which robot i belongs.

[0022] Step 4: After each group has covered its sub-area, it returns to the starting point o s , view o s If there is no base station, a robot is left as the base station. The base station records the set of sub-areas T that have not been covered, which is initialized to the set of all sub-areas: T = {P1, P2, ..., P N}.

[0023] Step 5: The group communicates with the base station and deletes the sub-area covered by the group from the set of uncompleted covered sub-areas T. It is determined whether all sub-areas are covered (if T is an empty set, all sub-areas are covered). If so, the coverage task is completed. If not, a sub-area is selected from T and search is started.

[0024] Step 6: The search team uses the herd coverage mode to cover the distance from the starting point from near to far, and uses the following formula as the total reward function for its robot to select the next target point:

[0025] R(o j )=R H (o j )+ω D (R D (o j ))+ω G (R G (o j ))+ω S (R s (o j ))+ω B (R B (o j ))

[0026] Where R H Attracts rewards for the herd, R D For dynamic predator avoidance reward, R G is the group attraction reward, R S is the path smoothness reward, R B is the boundary reward, ω D ,ω G ,ω S ,ω B is the weighting factor.

[0027] Specifically, the reward function calculation method is as follows:

[0028] 1) Herd attraction reward

[0029]

[0030] Among them, H(o j ) is o j To the virtual herdΦ i The distance and are the candidate points of robot i to Φ i The herd attraction reward makes the robot cover the distance from the starting point from near to far.

[0031] 2) Dynamic predator avoidance reward

[0032]

[0033] in, represents the dynamic predator k of robot i, k∈{1,2,…,n}\i, Yes j arrive The distance and They are Neighbors to The maximum and minimum distances.

[0034] 3) Group attraction reward

[0035]

[0036] in, represents robot k in the same group as robot i, k∈{1,2,…,n}\i, Yes j arrive The distance and They are Neighbors to The maximum and minimum distances.

[0037] 4) Path Smoothing Rewards

[0038] In order to encourage the robot to move in a straight line and make its moving trajectory more regular, a smoothness reward is used, which is calculated as follows:

[0039]

[0040] in, Indicates the moving angle, is the goal point previously covered by robot i. and is the maximum and minimum angle of all neighboring points. Therefore, the robot will get more rewards when moving in a straight line, reducing the energy loss and wear caused by turning.

[0041] 5) Boundary Rewards

[0042] In order to reduce the time loss caused by round-trip coverage, additional rewards should be given to the target points at the boundary of the coverage area. The calculation method of the boundary reward is as follows:

[0043]

[0044] Among them, Ξ(o j ) is o j The number of neighbors to be covered, and yes Among all neighbors, j ). Using boundary rewards can also make the coverage trajectory more regular.

[0045] The search team will take the starting point as o sThe herd will cover from near to far. If it encounters another group that is covering, it will merge with it to cover together. If it does not meet, it will cover alone. After covering its sub-area, each group will separate and return to step 4.

[0046] Furthermore, in step 3 above, a distributed intra-group collaborative coverage path planning method is used to plan the moving path for each robot in real time. The next moving target point is selected from the perspective of the robot. The specific steps are as follows:

[0047] Step 3.1: Robot i calculates which target points in its neighborhood can maintain communication with any other robot according to the mobility constraints, and the set of these movable target points is represented by Q i Indicates. Further select Q i The target points not covered in the , get the candidate point set, the candidate point set is composed of W i express.

[0048] Step 3.2: If there is no uncovered target point in the neighborhood of robot i that can maintain communication with the group, that is, W i is an empty set, it will reach a dead end. To overcome this situation, first Select the one closest to r i The target point is used as the temporary target point. Then a point-to-point planner (such as A* algorithm) is used to calculate the path to the temporary target point and the next point in the path is selected as o i .if is an empty set, there is no uncovered target point in the detected area, and in P i Middle distance i The nearest undetected target point is selected as the temporary target point. Note that if the next point of the planned path loses communication with all other robots. In this case, robot i is in Q i Select the target point closest to the temporary target point for movement.

[0049] Step 3.3: If there are uncovered target points in the neighborhood of robot i that can maintain communication with the group, it will select the target point with the highest reward function value among these points to move to. The predator coverage mode is adopted, and the coverage starts from far to near the starting point. i Each candidate target point o in j Calculate the total reward function R(o j ), and select the candidate point o with the maximum reward j As the next target point o i The total reward function consists of five parts: 1) Fixed predator avoidance reward R A , 2) Dynamic predator avoidance reward R D , 3) Path smoothness reward RS , 4) Boundary reward R B , 5) Group attraction reward R G The total reward function is calculated as follows:

[0050] R(o j )=R A (o j )+ω D (R D (o j ))+ω G (R G (o j ))+ω s (R S (o j ))+ω B (R B (o j ))

[0051] where ω D ,ω G ,ω S ,ω B is the weighting factor.

[0052] Specifically, the fixed predator avoidance reward calculation method is as follows:

[0053]

[0054] Among them, Γ is a static virtual predator, set at the starting point, K(o j ) is o j The distance to Γ, and They are The maximum and minimum distances from the neighbor point to Γ. The fixed predator avoidance reward makes the robot cover the distance from the starting point from far to near.

[0055] Dynamic predator avoidance reward R D , path smoothness reward R S , boundary reward R B , Group attraction reward R G The calculation method is the same as step 6.

[0056] Beneficial Effects

[0057] In the method proposed in the present invention, robots are first adaptively grouped according to the size of the communication range to maximize the collaborative ability. Then, the sub-areas are evenly divided for each group through the angle division method. Each group of robots moves to the corresponding sub-area for collaborative coverage, and the attraction between robots is enhanced by introducing group attraction rewards to prevent disconnection, thereby improving the coverage efficiency of robots within the group. A group merging method is proposed, which combines the herd coverage mechanism of the group that has completed coverage and the predator coverage mechanism of the group that has not completed coverage to achieve efficient merging between groups, so that the group that completes coverage first can help the group that has not completed coverage to collaboratively cover its sub-area, thereby improving the collaborative efficiency of robots between groups. The method proposed in the present invention greatly improves the collaborative ability between robots in an environment with only local communication, and can achieve rapid coverage of a specified area. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flow chart of a group robot coverage path planning method based on grouping and merging mechanism in a local communication environment of the present invention;

[0059] Figure 2 This is a schematic diagram of the coverage process of 6 robots with a communication range radius of 5 in an unknown environment of 30*30 using the method of the present invention in an embodiment of the present invention;

[0060] Figure 3 The figure is a comparison diagram of the coverage paths of the method of the present invention and three comparison methods in a complex obstacle environment. DETAILED DESCRIPTION

[0061] The technical solution provided by the present application will be further described below in conjunction with specific embodiments and accompanying drawings. The advantages and features of the present application will become more apparent with the following description.

[0062] Example 1

[0063] The method of the present invention is applied to cleaning of underground parking lots with only partial communication:

[0064] In the underground parking lot cleaning problem, the robot has only a local communication range. The group robot coverage path planning method based on the grouping and merging mechanism is applied to this problem. The group robots are first grouped according to the size of the communication range, and then the underground parking lot map is divided into multiple sub-areas according to the partitioning method. Each group is assigned a sub-area. Each group first covers and cleans the sub-area assigned to it. Then, the group that completes the coverage first will go to the sub-area that has not been covered to coordinate the coverage. When all sub-areas are covered, the cleaning task is completed.

[0065] like Figure 1As shown, a group robot coverage path planning method based on grouping and merging mechanism in a local communication environment specifically includes the following steps:

[0066] Establish a coverage environment, map side length L = 40; initialize algorithm parameters: number of robots n = 6, communication range r = 10, robot detection range V = 15, starting position o s =[20,20].

[0067] Step 1: Group the robots and calculate the group size m = 2 according to the following formula:

[0068]

[0069] where n is the number of robots, r is the radius of the robot's communication range, L is the side length of the map, and β is a constant set to 1.35. m∈[1,n], and increases with r.

[0070] Number of groups The number of robots in group i, n i The calculation is as follows:

[0071]

[0072] Step 2: Partition the map and assign a partition to each group. The map is represented as a set of target points O = {o j :j=1,2,…,z}. The robots initially gather at the starting point o s = [20,20], each target point is assigned a quadrant angle θ j , which is the target point and o s The angle between the vector and the reference axis. θ is the angle between the reference axis and the positive horizontal direction. First, sort the target points in ascending order according to the quadrant angle to obtain a set O ′ Then, in order, ′ The target points in are assigned to each group. The set of target points assigned to the i-th group is P i , the number of target points contained in it is |P i |, respectively, represent the following:

[0073]

[0074] Step 3: After grouping and division, each group of robots moves to their corresponding sub-areas for distributed collaborative coverage. During this process, the robots in the group maintain continuous local communication and share the following information: target point set O, uncovered target point set Cover target point set and the set of target points occupied by obstacles To prevent disconnection from the group, robot i selects its next destination point o i The following constraints apply when moving: i The distance to any other robot is no greater than the communication range r to maintain communication with the group:

[0075] ||o i -r j ||≤r,

[0076] where r j represents the position of robot j, G i represents the set of robots in the group to which robot i belongs.

[0077] Step 4: After each group has covered its sub-area, it returns to the starting point o s , view o s If there is no base station, a robot is left as the base station. The base station records the set of sub-areas T that have not been covered, which is initialized to the set of all sub-areas: T = {P1, P2, ..., P N}.

[0078] Step 5: The group communicates with the base station and deletes the sub-area covered by the group from the set of uncompleted covered sub-areas T. It is determined whether all sub-areas are covered (if T is an empty set, all sub-areas are covered). If so, the coverage task is completed. If not, a sub-area is selected from T and search is started.

[0079] Step 6: The search team uses the herd coverage model and adopts the following formula as the total reward function for its robot to select the next target point:

[0080] R(o j )=R H (o j )+ω D (R D (o j ))+ω G (R G (o j ))+ω S (R S (o j ))+ω B (R B (o j ))

[0081] Where R H Attracts rewards for the herd, R D For dynamic predator avoidance reward, R S is the path smoothness reward, R B is the boundary reward, R Gis the group attraction reward, ω D ,ω G ,ω S ,ω B are weighting factors, which are set to 0.5, 0.5, 0.3, and 0.1 respectively.

[0082] Let the starting point be o s The herd will cover from near to far. If it encounters another group that is covering, it will merge with it to cover together. If it does not meet, it will cover alone. After covering its sub-area, each group will separate and return to step 4.

[0083] Furthermore, in step 3 above, a distributed intra-group collaborative coverage path planning method is used to plan the moving path for each robot in real time. The next moving target point is selected from the perspective of the robot. The specific steps are as follows:

[0084] Step 3.1: Robot i calculates which target points in its neighborhood can maintain communication with any other robot according to the mobility constraints, and the set of these movable target points is represented by Q i Indicates. Further select Q i The target points not covered in the , get the candidate point set, the candidate point set is composed of W i express.

[0085] Step 3.2: If there is no uncovered target point in the neighborhood of robot i that can maintain communication with the group, that is, W i is an empty set, it will reach a dead end. To overcome this situation, first Select the one closest to r i The target point is used as the temporary target point. Then a point-to-point planner (such as A* algorithm) is used to calculate the path to the temporary target point and the next point in the path is selected as o i .if is an empty set, there is no uncovered target point in the detected area, and in P i Middle distance i The nearest undetected target point is selected as the temporary target point. Note that if the next point of the planned path loses communication with all other robots. In this case, robot i is in Q i Select the target point closest to the temporary target point for movement.

[0086] Step 3.3: If there are uncovered target points in the neighborhood of robot i that can maintain communication with the group, it will choose the target point with the highest reward function value among these points to move to. i Each candidate target point o in j Calculate the total reward function R(o j ), and select the candidate point o with the maximum rewardj As the next target point o i The total reward function consists of five parts: 1) Fixed predator avoidance reward R A , 2) Dynamic predator avoidance reward R D , 3) Smoothness reward R S , 4) Boundary reward R B , 5) Group attraction reward R G The total reward function is calculated as follows:

[0087] R(o j )=R A (o j )+ω D (R D (o j ))+ω G (R G (o j ))+ω S (R S (o j ))+ω B (R B (o j ))

[0088] where ω D ,ω G ,ω S ,ω B are weighting factors, which are set to 0.5, 0.5, 0.3, and 0.1 respectively.

[0089] Example 2

[0090] Figure 2 The following is a coverage process of 6 robots with a communication range radius of 5 in a 30*30 unknown environment using the method proposed in the present invention. There are three groups, group g1 with robots 1 and 2, group g2 with robots 3 and 4, and group g3 with robots 5 and 6. Figure (a) shows that g3 has completed the coverage of its sub-area and leaves robot 5 at the starting point as a base station. Robot 6 goes to the sub-area where group g1 is located. Figure (b) shows that robot 6 and robot 1 establish communication, and robot 6 merges with g1. Figure (c) shows that g2 has completed the coverage of its sub-area and communicated with the base station, and learned that the sub-area where group g1 is located has not been covered, and moved to the sub-area where group g1 is located. Communication is established between robot 3 and robot 1. Robots 3 and 4 merge with g1 for joint coverage. Figure (d) shows that all target points have been covered and the coverage task is completed.

[0091] Example 3

[0092] In order to more intuitively verify the performance of the present invention in the coverage path planning problem, we use a map with complex obstacles to simulate the environment to be covered, where the robot has only local communication in the environment and needs to cover all the target points in the map that are not occupied by obstacles. Assume that the robot needs one time step to move to another target point in its neighborhood. t represents the total time steps for the swarm robots to cover all the target points in the map that are not occupied by obstacles. The performance of the algorithm is expressed in N t To evaluate, N t The smaller it is, the better the performance of the algorithm.

[0093] Three comparison methods are set up, which are the most outstanding robot collaboration methods under current local communication. The first is a dynamic network method, called DNCPP, in which all robots form a group and maintain local continuous connections between robots to obtain global information. The second is an adaptive meeting method, which is named AMCPP. Once the robots meet, they share their exploration and coverage target points, assign the next coverage task, and determine the location and time of the next meeting. The third is a balanced partitioning method, named BPCPP, which uses a partitioning method to divide the map into multiple sub-areas and assign them to each robot, and each robot covers its own sub-area.

[0094] Figure 3 The coverage paths of each algorithm are shown. t They are 228, 269, 286 and 260 respectively. Compared with DNCPP, AMCPP and BPCPP, the performance of the proposed method is improved by 15.2%, 20.3% and 12.3% respectively.

[0095] The above content is a further detailed description of the present invention. It cannot be determined that the specific implementation methods of the present invention are limited to these. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as belonging to the scope of protection of the invention determined by the claims of the present invention.

Claims

1. A group robot coverage path planning method based on grouping and merging mechanism in a local communication environment, characterized in that: The following steps are involved: Establish a coverage environment, map side length L; initialize algorithm parameters: number of robots n, communication range r, robot detection range V, starting position o s ; Step 1: First, the robots are adaptively grouped according to the size of the communication range to maximize the collaboration capability. The size of the group m is calculated according to the following formula: Where n is the number of robots, r is the radius of the robot's communication range, L is the side length of the map, β is a constant, m∈[1,n], and increases with the increase of r; Number of groups The number of robots in group i, n i The calculation is as follows: Step 2: Partition the map and assign a partition to each group; the map is represented as a set of target points O = {o j :j=1,2,…,z}, z is the total number of target points; the robots initially gather at the starting point o s , each target point is assigned a quadrant angle θ j , is the target point and o s The angle between the vector and the reference axis, where the reference axis is o s is the positive horizontal axis of the starting point; first, sort the target points in ascending order according to the quadrant angle to obtain a set O′; then, assign the target points in O′ to each group in order; the set of target points assigned to the i-th group is P i , the number of target points contained in it is |P i |, respectively, represent the following: in represents the total number of target points assigned to group 1 to group i; Step 3: After grouping and division, each group of robots moves to their corresponding sub-areas for distributed collaborative coverage; during this process, the robots in the group maintain continuous local communication and share the following information: target point set O, uncovered target point set Cover target point set and the set of target points occupied by obstacles To prevent disconnection from the group, robot i selects its next destination point o i The following constraints apply when moving: i The distance to any other robot is no greater than the communication range r to maintain communication with the group: where r j represents the position of robot j, G i represents the set of robots in the group to which robot i belongs; Step 4: After each group has covered its sub-area, it returns to the starting point o s , view o s If there is no base station, a robot is left as the base station. The base station records the set of sub-areas T that have not been covered, which is initialized to the set of all sub-areas: T = {P1, P2, ..., P N }; Step 5: The group communicates with the base station and deletes the sub-area covered by the group from the set of uncompleted covered sub-areas T; determines whether all sub-areas are covered. If T is an empty set, then all sub-areas are covered. If so, then the coverage task is completed. If not, then select a sub-area from T and proceed to search; Step 6: The search team uses the herd coverage mode to cover the distance from the starting point from near to far, and uses the following formula as the total reward function for its robot to select the next target point: R(o j )=R H (the j )+ω D (R D (the j ))+ω G (R G (the j ))+ω S (R S (the j ))+ω B (R B (the j )) Where R H Attracts rewards for the herd, R D For dynamic predator avoidance reward, R G is the group attraction reward, R S is the path smoothness reward, R B is the boundary reward, ω D ,ω G ,ω S ,ω B is the weighting factor; The search team will take the starting point as o s The herd covers from near to far. If it encounters other groups that are covering during the process, it will merge with them to cover together. If it does not meet them, it will cover alone. After covering its sub-area, each group will separate and return to step 4.

2. The method according to claim 1, characterized in that: In step 3, a distributed intra-group collaborative coverage path planning method is used to plan the movement path for each robot in real time; the next moving target point is selected from the robot's perspective. The specific steps are as follows: Step 3.1: Robot i calculates which target points in its neighborhood can maintain communication with any other robot according to the mobility constraints, and the set of these movable target points is represented by Q i Indicates; further select Q i The target points not covered in the , get the candidate point set, the candidate point set is composed of W i express; Step 3.2: First, Select the one closest to r i The target point is used as a temporary target point; Then use the point-to-point planner to calculate the path to the temporary goal point and choose the next point on the path as o i ;if is an empty set, there is no uncovered target point in the detected area, and in P i Middle distance i The nearest undetected target point is selected as the temporary target point; If the next point of the planned path loses communication with all other robots, robot i is in Q i Select the target point closest to the temporary target point to move; Step 3.3: If there are uncovered target points in the neighborhood of robot i that can maintain communication with the group, it will select the target point with the highest reward function value among these points to move to; adopt the predator coverage mode, starting from the farthest to the closest distance from the starting point; W i Each candidate target point o in j Calculate the total reward function R(o j ), and select the candidate point o with the maximum reward j As the next target point o i ; The total reward function consists of five parts: 1) Fixed predator avoidance reward R A , 2) Dynamic predator avoidance reward R D , 3) Path smoothness reward R S , 4) Boundary reward R B , 5) Group attraction reward R G ; The total reward function is calculated as follows: R(o j )=R A (the j )+ω D (R D (the j ))+ω G (R G (the j ))+ω S (R S (the j ))+ω B (R B (the j )) where ω D ,ω G ,ω S ,ω B is the weighting factor.

3. The method according to claim 2, characterized in that: Fixed predator avoidance reward calculation as follows: Among them, Γ is a static virtual predator, set at the starting point, K(o j ) is o j The distance to Γ, and They are The maximum and minimum distances from the neighbor point to Γ; the fixed predator avoidance reward makes the robot cover the distance from the starting point from far to near.

4. The method according to claim 1, characterized in that: In step 6, the reward function is calculated as follows: 1) Herd attraction reward Among them, H(o j ) is o j To the virtual herdΦ i The distance and are the candidate points of robot i to Φ i The maximum and minimum distances of the robot; the herd attraction reward makes the robot cover the distance from the starting point from near to far; 2) Dynamic predator avoidance reward in, represents the dynamic predator k of robot i, k∈{1,2,...,n}\i, Yes j arrive The distance and They are Neighbors to The maximum and minimum distances; 3) Group attraction reward in, represents the robot k in the same group as robot i, k∈{1,2,...,n}\i, Yes j arrive The distance and They are Neighbors to The maximum and minimum distances; 4) Path Smoothing Rewards Smoothness bonus, calculated as follows: in, Indicates the moving angle, is the target point previously covered by robot i; and is the maximum and minimum angle of all neighboring points; therefore, the robot will get more rewards when moving in a straight line, reducing the energy loss and wear caused by turning; 5) Boundary Rewards The calculation method of the boundary reward is as follows: Among them, Ξ(o j ) is o j The number of neighbors to be covered, and yes Among all neighbors, j )’s maximum and minimum values; using boundary rewards can also make the coverage trajectory more regular.

Images