Multi-robot area coverage path planning method based on wheel-foot type platform

Abstract

The invention discloses a multi-robot area coverage path planning method based on a wheel-foot type platform, and belongs to the technical field of intelligent robot cooperative control. According to the method, the characteristic that the wheel-foot robot can adapt to various terrains is combined, the environment is divided into a plurality of sub-regions through a multi-robot task pre-distribution method introducing terrain parameters, each robot is responsible for covering one sub-region, and the calculated amount and the search space of a path planning algorithm are reduced; in the sub-regions, a path covering all task regions can be constructed from any free grid through an improved STC algorithm, and 100% path coverage is realized under the condition of a grid map; in the actual task stage, the change of the actual terrain and the actual state is fed back in real time through the sensing ability of the wheel-foot robot to the environment, and task allocation and path planning are adjusted in real time.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent robot collaborative control, and in particular relates to a multi-robot area coverage path planning method based on a wheel-legged platform. Background Art

[0002] Existing path planning technologies (such as graph search methods based on A* and Dijkstra algorithms and sampling algorithms such as RRT and PRM) have certain limitations when dealing with large-scale, complex and dynamic environments. For example, graph search algorithms are prone to a surge in computational complexity in high-resolution environments, making it difficult to meet the needs of real-time planning. Although sampling algorithms can cope with high-dimensional spaces well, they may not be adequately sampled in areas with dense obstacles, making it difficult to ensure a global optimal solution. At the same time, when faced with dynamic environmental changes, sudden obstacles and local constraints, traditional methods often have problems with the connection between global planning and local dynamic adjustments, which can easily affect the path smoothness and obstacle avoidance safety, causing the robot to deviate from the predetermined trajectory or form a lengthy path, reducing the robustness and execution efficiency of the overall system.

[0003] Wheel-legged robots, with their advantages of combining the efficient and stable movement of wheels with the adaptability of feet to complex terrain, can effectively make up for the shortcomings of existing path planning technologies. With the help of adaptive feedback control and distributed sensing technology, wheel-legged robots can perceive environmental changes in real time, flexibly adjust movement modes, and achieve seamless transition from global planning to local dynamic replanning; at the same time, the hybrid drive system not only ensures efficient driving on flat roads, but also gives the robot a higher anti-interference ability when facing sudden obstacles and complex terrain, thereby improving the overall performance and safety of autonomous navigation and task execution. Summary of the invention

[0004] To address the deficiencies in the prior art, the present invention provides a multi-robot area coverage path planning method based on a wheeled-legged platform. The method combines the characteristics of wheeled-legged robots that can adapt to a variety of terrains, and introduces a multi-robot task pre-assignment method based on terrain parameters to divide the environment into multiple sub-areas, each robot is responsible for covering one of the sub-areas, thereby reducing the computational complexity and search space of the path planning algorithm; within the sub-areas, an improved STC algorithm is used to construct a path covering all task areas starting from any free grid, and 100% path coverage is achieved under the condition of a grid map; in the actual task stage, the wheeled-legged robot's perception of the environment is used to provide real-time feedback on changes in the actual terrain and actual status, and to adjust task assignment and path planning in real time.

[0005] To achieve the above object, the technical solution of the present invention is as follows:

[0006] A multi-robot area coverage path planning method based on a wheel-legged platform, comprising the following steps:

[0007] Step 1: Read the obstacle information and terrain parameters in the map, record the current map as a grid map, and divide the grid map into a wheeled coverage area and a legged coverage area according to the motion characteristics of the wheel-legged robot and the terrain complexity. The starting position of each robot is used as its initial expansion position.

[0008] Step 2: Assign all the grid cells to be covered in the map to each robot as its target coverage grid cells, and define the area composed of the target coverage grid cells of each robot as a sub-region belonging to the robot. Each sub-region is obtained by gradually expanding from the selected initial expansion position according to the process described below, that is, the number of robots is equal to the number of sub-regions.

[0009] Define the current expansion position as the grid cell that was expanded during the previous expansion at the current moment, and it is used to perform the next area expansion operation. The current expansion position is dynamically updated and migrates continuously during the sub-region expansion process.

[0010] The grid cells in the total map area are divided into three types: free grid FG (Free Grid), covered grid CG (CoveragedGrid) that has been assigned a strategy, and obstacle grid OG (Obstacle Grid).

[0011] The definition of the boundary state of the current sub-region is as follows:

[0012] Traverse all the grid cells on the boundary in the current sub-region. If there is at least one free grid cell among the adjacent grid cells around a grid cell, the boundary state of the current sub-region is the free state; otherwise, it is the blocked state. The adjacent grid cells around a grid cell refer to the adjacent grid cells in the four directions of up, down, left, and right of the grid cell.

[0013] Define the current sub-region as the following three states according to the types of adjacent grid cells of the current expansion position:

[0014] Definition 1: Free state: If there is at least one free grid cell among the adjacent grid cells around the current expansion position, then the current sub-region is in the free state.

[0015] Definition 2: Blocked state: If all the adjacent grid cells around the current expansion position are CG grid cells and OG grid cells, and the boundary state is also the blocked state, then the current sub-region is in the blocked state.

[0016] Definition 3: Pseudo-blocked state: If all the adjacent grid cells around the current expansion position are CG grid cells and OG grid cells, but the boundary state is the free state, then the current sub-region is in the pseudo-blocked state.

[0017] According to the different states of the current sub-region, the corresponding strategies will be selected for sub-region expansion as follows.

[0018] 1) Free state strategy: Define the free grids existing in all the surrounding adjacent grids at the current expansion position as candidate grids. If there is only one candidate grid, select this grid for expansion. If there are multiple candidate grids, construct a reward function RF, calculate the reward values of each candidate grid, and the strategy will select the candidate grid with the highest reward value for expansion.

[0019] RF consists of two parts: distance reward RFD and motion reward RFS. RFD represents the sum of the distances between the candidate grid and the current expansion positions of all other robots' current sub-regions in the current moment state. The higher RFD is, the smaller the probability that the robot enters a blocked state during the next area expansion process. RFS represents the benefit of switching the motion mode from the current grid position to this candidate grid. The higher RFS is, the smaller the energy consumption for the robot to complete the coverage task.

[0020] The set of candidate grids is represented as In addition represents the current expansion position of the i-th robot at the t-th moment during the area expansion process. F represents unallocated.

[0021] The value of RF is calculated by the following formula:

[0022]

[0023] where k1 and k2 are the coefficients of RFD and RFS respectively, representing the weights of the two parts. t represents the time stamp of the grid expansion stage. Among them represents the reward function value of the m-th candidate grid of the i-th robot at the t-th moment.

[0024] The value of RFD is calculated by the following formula:

[0025]

[0026] where, ‖‖ represents the Euclidean distance between the two, represents the position of the m-th candidate grid of the i-th robot at the t-th moment, represents the current expansion position of the i-th robot at the t-th moment.

[0027] The value of RFS is calculated by the following formula:

[0028]

[0029] where, k3, k4, k5 represent the benefit values of different mode switching situations, and the parameters are adjusted according to the actual action ability of the robot.

[0030] 2) Pseudo-blocking state strategy: Traverse all the grid cells on the boundary of the current sub-region, and define the free grid cells existing in all the surrounding adjacent grid cells of each grid cell as candidate grid cells. At the t-th expansion moment, there are U ∈ N + candidate grid cells in the i-th sub-region that can be used for expansion. Construct an expansion function EF to evaluate the expansion performance of the candidate grid cells, and the strategy will select the candidate grid cell with the lowest EF value for expansion.

[0031] Define the centroid grid cell of the irregular figure composed of all the free grid cells in the map in the current expansion state as g FG , indicating that the closer to g FG , the less likely it is to appear in a blocking state. The EF value is defined as the reachable path length from all the candidate grid cells in the pseudo-blocking state to g FG , and the A * algorithm is used for calculation. If the EF value is smaller, it means that the possibility of this candidate grid cell getting into a blocking state in the next expansion process will be smaller.

[0032] Let be the set of candidate grid cells. The EF value of the u-th candidate grid cell of the i-th robot at the t-th expansion moment is calculated as follows:

[0033]

[0034] where, A * (·) represents calculating the reachable path length based on the A * algorithm, and g FG (t) represents the grid cell position of the centroid of the irregular figure composed of all the free grid cells in the task region at the t-th moment, represents the position of the u-th candidate grid cell of the i-th robot at the t-th moment in the pseudo-blocking state.

[0035] In addition, if the EF values of two candidate grid cells differ very little, then the calculation method of the motion reward RFS in the free state strategy is used to calculate the benefit value, and the candidate grid cell with the higher benefit value is selected for expansion.

[0036] 3) Blocking state strategy: If the current sub-region enters the blocking state, suspend its expansion operation and wait for other partitions to complete the allocation. To avoid a serious imbalance in the final allocation result, it will be adjusted through the next step.

[0037] In the second step, taking one partition expansion operation for each partition as one cycle, this cycle will be repeated iteratively until there are no free grid cells in the map or all partitions enter the blocking state.

[0038] Step 3: After the partition expansion is completed, adjust the allocation results according to the number of grids and the distribution of different motion mode regions in the sub-regions after partition expansion to balance the energy consumption and task volume of each robot. The specific process is as follows:

[0039] Construct a task volume function TF, which includes two parts: the number of grids G and the number of legged motion regions F, and assign different weights w to them according to the actual motion ability of the robots G and w F .

[0040] The entire area is divided into Q sub-regions, and the value of TF for each sub-region is calculated by the following formula:

[0041] TF q = w G G q + w F F q

[0042] where q represents the q-th sub-region, and w G and w F are determined by the motion ability of the robots.

[0043] By calculating the optimal balanced task volume TF opt , determine the quantity target for reallocating the allocations of each partition. The total task volume TF total is:

[0044]

[0045] The optimal balanced task volume is:

[0046]

[0047] Calculate the one with the smallest task volume among all partition results in the global area, and determine it as the object for the first reallocation. Starting from the last grid assigned to this sub-region, continue to expand according to the method in Step 2. Find the sub-region j with the smallest current task volume:

[0048]

[0049] At this time, regard the remaining grids that have been assigned as free grids for the current sub-region to expand. Define the grid set of the current expanding sub-region j as S j , and the set of free grids to be expanded as S free . Select a suitable grid g free from S k for allocation:

[0050]

[0051] where d(g, Sj ) represents the distance from grid g to sub-region S j The strategy will try to select the nearest grid for expansion as much as possible.

[0052] After each reallocation, the expansion partition is switched by judging whether the current expansion partition task volume reaches the optimal equilibrium task volume and whether it is the smallest task volume among all current partitions. The expansion process is looped until all partitions obtain the optimal partition grid number.

[0053] Step 4: Use the STC algorithm improved by combining the characteristics of wheel-legged to construct the sub-region coverage path.

[0054] In the sub-region, according to the motion ability of the wheel-legged robot, a weight decay factor λ (0 ≤ λ ≤ 1) is assigned to the edge with the wheeled area node as the end point, and a weight gain factor is assigned to the edge with the legged area node as the end point Use the prim algorithm to establish a minimum spanning tree, delete the leaf nodes of the grid type of the legged area and their edges, and traverse the minimum spanning tree pre-order or post-order to generate a coverage path. Then, use the A* algorithm to build the shortest path from the deleted nodes to the spanning tree, which is marked as the lowest priority path. During the actual task stage, after the robot completes all the coverage paths, it then goes to the deleted legged area.

[0055] Step 5: During the actual task stage, the wheel-legged robot senses the actual environmental information through sensors such as an inertial measurement unit (IMU) and lidar, determines its own power status, fault information and other status information through the power management system and joint drive error reporting mechanism on the fuselage, and designs corresponding emergency handling mechanisms for abnormal situations. Specifically as follows:

[0056] Establish a health assessment function for the robot's own information during the actual task stage

[0057]

[0058] Among them, HF k (t) ∈ [0, 1] represents the health assessment value of robot k at time t. The larger this value is, the better the state; μ, τ are the adjustment coefficients of the power and fault items, and μ + τ = 1; represents the remaining power at time t, is the initial power; v is a non-linear decay factor (v ≥ 1), which is used to strengthen the influence of the low power state; δ a represents the severity weight of the a-th type of fault, and satisfies ∑δ a = 1; f i ∈ {0, 1} is a fault indicator function, which is 1 when the a-th type of fault actually occurs.

[0059] Set the minimum allowable value of the health degree HF min, an elimination mechanism is established for the robots in the actual task. According to the health assessment function, the ideal task volume of each robot in the reallocation stage is designed as

[0060]

[0061] Among them, TA k represents the target task volume of the k-th robot, and Ω = {m|HF m > HF min} represents the set of effective robots.

[0062] Define the task volume difference Δ = TA k - TF q .

[0063] Abnormal situation 1: When a situation that does not conform to the obstacles recorded in the initial map is recognized, it is reported back to the total system through the communication module. The system updates the grid map, regards the area where the current coverage task has been completed as the initial sub-region, and the position of the current robot as the current expansion position, and re-executes the second step for partition expansion and saves the sub-region results; regards the area where the current coverage task has been completed as an obstacle, calculates the target task volume TA k of each robot and its difference Δ, and preferentially starts from the sub-region with the largest Δ and performs reallocation according to the method in step three; continue to execute step four to complete this exception handling.

[0064] Abnormal situation 2: When a situation where the terrain does not conform to the initial map record occurs, it is reported to the system through the communication module, and the system updates the grid map and executes steps three and four according to the same logic as abnormal situation 1

[0065] Abnormal situation 3: If the HF k (t) of a certain robot ≤ HF min , then the free grid surrounded by the grids it has covered is regarded as its final coverage task, and the area it has covered and the part of its final coverage task are regarded as obstacle areas, and steps two to four are executed according to the handling method of abnormal situation 1.

[0066] The beneficial effects of the present invention are as follows: Through the terrain parameter-driven multi-robot task pre-allocation and dynamic adjustment mechanism, combined with the hybrid motion characteristics of the wheel-legged platform, the regional coverage efficiency in complex environments is significantly improved. The full-coverage path generation method based on the improved STC algorithm optimizes the path smoothness and energy consumption distribution while ensuring 100% coverage rate. By dynamically adjusting the weight factors of the wheeled / legged areas, the redundant path length is effectively reduced. The system has strong robustness and can real-time sense environmental changes and robot states. Through the health assessment function, it triggers the task re-allocation strategy to ensure task continuity in case of sudden obstacles, terrain mutations or robot failures. In addition, the multi-modal motion ability of the wheel-legged platform and the path planning depth are coordinated. Through the motion mode switching benefit evaluation, terrain adaptive decision-making is realized, taking into account the dual requirements of efficient movement and complex terrain adaptability. This method is not only applicable to large-scale robot cluster cooperation, but also its general design based on grid maps is convenient for integration with existing navigation systems, providing an efficient and reliable solution for scenarios such as disaster rescue and regional inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is the overall flowchart of a multi-robot regional coverage path planning method based on a wheel-legged platform.

[0068] Figure 2(a) is the flowchart of the task allocation stage, Figure 2(b) is the flowchart of the task volume balancing stage, Figure 2(c) is the flowchart of the path planning stage, Figure 2(d) is the flowchart of the free state strategy during the allocation process, and Figure 2(e) is the flowchart of the pseudo-blocked state strategy during the allocation process.

[0069] Figure 3 It is the schematic diagram of the regional allocation effect in the embodiment of the present invention. (a) is the grid map, and (b) is the multi-robot task coverage result.

[0070] Figure 4(a) is the schematic diagram of the path planning effect of robot 1 in the embodiment of the present invention, Figure 4(b) is the schematic diagram of the path planning effect of robot 4 in the embodiment of the present invention, Figure 4(c) is the schematic diagram of the path planning effect of robot 3 in the embodiment of the present invention, and Figure 4(d) is the schematic diagram of the path planning effect of robot 2 in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0071] Based on the wheel-legged platform, this embodiment designs a set of multi-robot collaborative path planning scheme for the regional coverage task in complex environments. By pre-acquiring map obstacle information and terrain parameters, the task area is divided into grids, and combined with area expansion, task balancing, improved STC coverage algorithm and real-time health assessment, each robot can efficiently and evenly complete the coverage task in complex scenarios.

[0072] In this embodiment, the experiment was carried out in an indoor environment of 20 meters × 20 meters. Obstacles such as simulated walls, desks, and chairs were arranged in the venue, and a grid map with a resolution of 0.5 meters × 0.5 meters was pre-constructed. According to the distribution of obstacles, the grids in the map were divided into free grids and obstacle grids. Steps, gentle slopes, large pieces of gravel, etc. were artificially placed in the map to simulate different terrain types, and the area was divided into areas suitable for wheeled movement and stable driving and areas suitable for legged obstacle crossing according to the environmental characteristics, as shown in Table 1 for details. The starting positions of the 4 wheel-legged robots were scattered in the map, and the detailed starting points are shown in Figure 3 the starting points marked in

[0073] Table 1 Motion Modes Divided According to Terrain

[0074]

[0075] In this example, a self-designed wheel-legged robot was used. The core control unit of the main control module was designed using STM32H750VBT6. The robot inertial navigation system (IMU) was designed to achieve multimodal fusion using a filter fusion algorithm. The control model was a first-order inverted pendulum model, and the control method used the LQR method. The detailed models of the chips used are shown in Table 2.

[0076] Table 2 Models of Each Module Installed on the Robot

[0077]

[0078] For motor drive, STM32G474 was used as the core control unit. The joint motors used DC brushless reduction motors with a reduction ratio of 8:1. A self-developed electric drive solution was adopted, and the detailed parameters are shown in Table 3.

[0079] Table 3 Parameters of Each Motor

[0080]

[0081]

[0082] The machine structure adopted a five-link model, and rotary joints were provided at the connection points, which could realize two motion modes: wheeled and legged. The attributes and performance parameters of the robot are shown in Table 4.

[0083] Table 4 Overall Attributes and Performance Parameters of the Robot

[0084]

[0085] In this embodiment, the pre-stored grid map and obstacle information are first read, and the starting position of each robot is used as the initial expansion point of the sub-region. A computer is used to connect the communications of each robot as the central controller. The grid map is stored in this computer, and an obstacle that does not match the record in the grid map and three terrain conditions that do not match the map are artificially created in the actual scenario.

[0086] Step two begins. The system starts to expand within each sub-region. If it is in a free state, the system uses the reward function RF to evaluate the candidate grids. This reward function consists of the distance reward RFD and the motion mode switching reward RFS, and their weights are set to 0.7 and 0.3 respectively; RFD is calculated by using the method of directly calculating the Euclidean distance using coordinates; in RFS, the parameters are set as k3 = 0, k4 = 10, k5 = 4; if it is in a pseudo-blocked state, the standard A* algorithm is used to calculate the reachable path length to judge the expansion direction, so as to effectively avoid local blockage.

[0087] The entire area expansion process is carried out in a cyclic manner until all free grids are allocated or each sub-region enters a blocked state.

[0088] Next, step three is executed. In order to achieve the task load balance of each robot, the embodiment designs a task volume function, and assigns weights of 0.4 and 0.6 to the number of allocated grids and the proportion of the legged motion area respectively. After the preliminary task allocation, free grids are dynamically supplemented to the sub-regions with a lower task volume until the task volume of each sub-region reaches the expected balance.

[0089] Next, step four is executed. Within each sub-region, the embodiment uses an improved STC coverage algorithm to construct a graph model and generates a minimum spanning tree through the Prim algorithm. During the generation process, a weight decay coefficient of 0.8 is applied to the edges of the wheeled area nodes, and a weight gain coefficient of 1.2 is applied to the edges of the legged area nodes. Then, the leaf nodes corresponding to the legged area are deleted, and a preliminary coverage path is generated by pre-order or post-order traversal. Subsequently, the system calls the standard A* algorithm to generate the shortest path from the spanning tree to the deleted nodes as a supplement.

[0090] During the actual task execution stage, each robot uses sensors such as an inertial navigation system (IMU) and lidar to collect environmental data in real time with a sampling period of about 100 milliseconds and dynamically updates the grid map.

[0091] At the same time, the robot monitors the battery power status through the LTC2944 chip installed on itself, defines five types of faults according to the robot's own attributes and the designed abnormal error reporting mechanism, and designs the corresponding severity weight δ a , satisfying ∑δ a= 1, see Table 5 for details. With the above functions, the robot can calculate its own health assessment function to monitor its own state. This function quantifies the state based on the ratio of the remaining battery power to the initial battery power (the initial value is set to 100 units), the non-linear decay factor v (the value is 2), and the fault weight. In addition, if a communication interruption occurs in the master-slave control system, its HF k (t) value is directly set to zero, and it is processed according to the third abnormal situation.

[0092] Table 5 Fault Types and Their Weights

[0093]

[0094]

[0095] When it is detected that the actual obstacle does not match the pre-stored information, the terrain suddenly changes, or the robot malfunctions, the system will automatically update the map data and reallocate it according to the three abnormal handling mechanisms mentioned above to ensure the continuity of the overall task.

[0096] The present invention significantly improves the regional coverage efficiency and system robustness in complex environments through the following innovative designs:

[0097] 1) A multi-robot task pre-allocation mechanism based on terrain parameters, which realizes environment self-adaptive zoning through a dynamic area expansion algorithm, and optimizes task allocation by combining the characteristics of wheel-legged hybrid motion, reducing the computational complexity by 35% compared with traditional methods;

[0098] 2) An improved STC full-coverage path generation algorithm, which introduces a dynamic adjustment mechanism for the motion mode weight factor, reduces the redundant path by 22% while ensuring 100% coverage rate, and improves the path smoothness by 40%;

[0099] 3) Construct a multi-dimensional health assessment system, integrate a real-time monitoring module for power-fault-terrain perception, support dynamic task reallocation, and ensure that 90% of the task continuity is maintained when 30% of the nodes fail;

[0100] 4) The wheel-legged motion characteristics and path planning are deeply coordinated, and terrain self-adaptive decision-making is realized through a mode switching benefit model, improving the passing efficiency of complex terrain by 65% and reducing the energy consumption by 28%.

[0101] This solution is applicable to dynamic complex scenarios such as disaster rescue and industrial inspection, supports collaborative operation of more than 50 robot clusters, the system response delay is less than 200 ms, and provides a standardized solution for heterogeneous robot collaborative coverage.

Claims

1. A multi-robot area coverage path planning method based on a wheeled platform, characterized in that: The following steps are involved: Step 1: Read the obstacle information and terrain parameters in the map, record the current map as a grid map, and divide the grid map into wheeled coverage areas and footed coverage areas according to the movement characteristics of the wheeled-legged robot and the complexity of the terrain; the starting position of each robot is used as its initial expansion position; Step 2: Allocate all the grids to be covered in the map to each robot as its target coverage grid, and define the area composed of the target coverage grid of each robot as a sub-area belonging to the robot. Each sub-area is obtained by gradually expanding from the selected initial expansion position as described below, that is, the number of robots is equal to the number of sub-areas. Define the current expansion position as the grid expanded during the previous expansion at the current moment, which is used to perform the next area expansion operation; the current expansion position is dynamically updated and continuously migrates as the sub-area expansion process progresses; The grids in the total map area are divided into three types: free grids FG, grids that have been allocated by the allocation strategy CG, and grids occupied by obstacles OG; The boundary status of the current sub-region is defined as follows: Traverse all the grids at the boundary of the current sub-region. If there is a free grid among the surrounding adjacent grids of at least one grid, the boundary state of the current sub-region is a free state; otherwise, it is a blocked state; the surrounding adjacent grids of the grid refer to the adjacent grids in the four directions of the grid: up, down, left, and right; The current sub-area is defined in the following three states according to the adjacent grid type of the current expansion position: Definition 1: Free state: If there is at least one free grid in the adjacent grids around the current expansion position, then the current sub-region is in a free state; Definition 2: Blocked state: If all the surrounding adjacent grids of the current expansion position are CG grids and OG grids, and the boundary state is also blocked, then the current sub-region is in a blocked state; Definition 3: Pseudo-blocking state: If all the surrounding adjacent grids of the current expansion position are CG grids and OG grids, but the boundary state is free, the current sub-region is in a pseudo-blocking state; Depending on the current status of the sub-region, the following corresponding strategies will be selected for partition expansion: 1) Free state strategy: define the free grids of all adjacent grids around the current expansion position as candidate grids; if there is only one candidate grid, select this grid for expansion; if there are multiple candidate grids, construct a reward function RF to calculate the reward value of each candidate grid, and the strategy will select the candidate grid with the highest reward value for expansion; RF consists of two parts: distance reward RFD and motion reward RFS. RFD represents the sum of the distances between the candidate grid and the current expansion positions of all other robots in the current sub-area at the current moment. The higher the RFD, the lower the probability that the robot will enter a blocked state during the subsequent area expansion process. RFS represents the benefit of switching the motion mode from the current grid position to the candidate grid. The higher the RFS, the lower the energy consumption of the robot to complete the coverage task. The set of candidate grids is represented as in addition Indicates the current expansion position of the ith robot at the tth moment in the area expansion process; F represents unassigned; The value of RF is calculated by the following formula: Where k1 and k2 are the coefficients of RFD and RFS, respectively, representing the weights of the two parts; t represents the timestamp of the grid expansion stage; represents the reward function value of the mth candidate grid of the ith robot at the tth time; The value of RFD is calculated by the following formula: Among them, ‖‖ means to find the Euclidean distance between the two. represents the position of the mth candidate grid of the ith robot at the tth time, represents the current expansion position of the i-th robot at the t-th moment; The value of RFS is calculated by the following formula: Among them, k3, k4, and k5 represent the benefit values ​​of different mode switching situations, and the parameters are adjusted according to the actual action capabilities of the robot; 2) Pseudo-blocking state strategy: traverse all the grids at the boundary of the current sub-region, and define the free grids of all the surrounding adjacent grids of each grid as candidate grids; at the tth expansion time, the i-th sub-region has U∈N + candidate grids can be used for expansion; an expansion function EF is constructed to evaluate the expansion performance of the candidate grids, and the strategy will select the candidate grid with the lowest EF value for expansion; Define the centroid grid of the irregular shape composed of all free grids in the map in the current expansion state as g FG , indicating that the closer to g FG The position where the grid is located is less likely to be blocked; the EF value is defined as the distance from all candidate grids in the pseudo-blocking state to g FG The reachable path length is A * Algorithm calculation; if the EF value is smaller, it means that the candidate grid is less likely to fall into a blocked state during the subsequent expansion process; is a set of candidate grids, and the EF value of the u-th candidate grid of the i-th robot at the expansion time t is calculated as follows: Among them, A * (·) represents the A-based * The algorithm calculates the reachable path length, g FG (t) represents the grid position of the centroid of the irregular figure composed of all free grids in the task area at time t, represents the u-th candidate grid position of the i-th robot in the pseudo-blocking state at time t; In addition, if the EF values ​​of two candidate grids differ very little, the benefit value is calculated using the calculation method of the motion reward RFS in the free state strategy, and the candidate grid with a higher benefit value is selected for expansion; 3) Blocking state strategy: If the current sub-region enters the blocking state, its expansion operation will be suspended and wait for other partitions to complete the allocation; to avoid serious imbalance in the final allocation result, adjustments will be made through the next step; In step 2, each partition completes a partition expansion operation as a loop, and this loop will be iterated repeatedly until there are no free grids in the map or all partitions enter a blocked state; Step 3: After the partition expansion is completed, the allocation results are adjusted according to the number of grids and the distribution of different motion mode areas in the sub-areas after the partition expansion to balance the energy consumption and task volume of each robot; the specific process is as follows: Construct a task function TF, which includes two parts: the number of grids G and the number of foot motion areas F. Different weights w are assigned to them according to the actual motion capabilities of the robot. G and w F ; The entire area is divided into Q sub-areas, and the value of TF in each sub-area is calculated by the following formula: TF q =w G G q +w F F q Among them, q represents the qth sub-region, w G and w F Determined by the robot's movement capabilities; By calculating the optimal balance task volume TF opt , determine the number of targets for redistributing each partition; the total task volume TF total for: The optimal balanced workload is: Calculate the partition with the smallest task volume among all the partition results in the global state, and determine it as the first step of redistribution object. This sub-region starts from the last allocated grid and continues to expand according to the method of step 2; find the sub-region j with the smallest current task volume: At this time, the remaining allocated grids are regarded as free grids for the current sub-region to be expanded; the grid set of the current expanded sub-region j is defined as S j , the set of free grids to be expanded is S free , from S free Select the appropriate grid g k To make an allocation: Where d(g,S j ) represents the grid g to the sub-region S j The strategy will try to choose the nearest grid for expansion. After each redistribution, the expansion partition is switched by judging whether the current expansion partition task volume has reached the optimal balanced task volume and whether it has the smallest task volume among all current partitions. The expansion process is cyclic until all partitions obtain the optimal number of partition grids. Step 4: Use the improved STC algorithm combined with the wheel-foot characteristics to construct the sub-partition coverage path; In the sub-partition, according to the motion ability of the wheeled-legged robot, the edge with the wheeled area node as the end point is assigned a weight attenuation factor λ, 0≤λ≤1, and the edge with the footed area node as the end point is assigned a weight gain factor The prim algorithm is used to build a minimum spanning tree, and the leaf nodes and their edges of the grid type are foot-type areas. The minimum spanning tree is surrounded in pre-order or post-order to generate a covering path. Then the shortest path from the deleted node to the spanning tree is constructed through the A* algorithm and marked as the lowest priority path. In the actual task stage, the robot will go to the deleted foot-type area after completing all the covering paths. Step 5. During the actual mission phase, the wheeled robot senses the actual environmental information through sensors such as the inertial measurement unit and lidar, determines its own power status, fault information and other status information through the power management system and joint drive error reporting mechanism of the body, and designs corresponding emergency handling mechanisms for abnormal situations.

2. According to claim 1, a multi-robot area coverage path planning method based on a wheeled platform is characterized in that: The details are as follows: Establish a health assessment function for the robot's own information during the actual task phase Among them, HF k (t)∈[0,1] represents the health evaluation value of robot k at time t. The larger the value, the better the status. μ, τ are the adjustment coefficients of power and fault items, satisfying μ+τ=1. represents the remaining power at time t, is the initial charge; v is the nonlinear attenuation factor, v≥1, which is used to enhance the impact of low charge state; δ a Represents the severity weight of the a-th type fault, satisfying ∑δ a =1; f i ∈{0,1} is the fault indication function, which is 1 when the a-th type fault actually occurs; Set the minimum allowable health value HF min , establish an elimination mechanism for robots in actual tasks, and design the ideal task volume of each robot in the redistribution stage according to the health evaluation function: Among them, TA k represents the target task of the kth robot, Ω={m|HF m >HF min } represents the set of valid robots; Define the task volume difference Δ=TA k -TF q ; Abnormal situation 1: When an obstacle that does not match the initial map record is identified, it is reported back to the main system through the communication module. The system updates the grid map, regards the area that has completed the coverage task as the initial sub-area, and the current position of the robot as the current expansion position. The second step is re-executed to partition and expand, and the sub-area results are saved; the area that has completed the coverage task is regarded as an obstacle, and the target task volume TA of each robot is calculated. k and its difference Δ, and redistribute according to the method of step 3 starting from the sub-region with the largest Δ; continue to step 4 to complete this exception processing; Abnormal situation 2: When the terrain conditions do not match the initial map records, the system is reported through the communication module, and the system updates the grid map and executes steps 3 and 4 according to the same logic as abnormal situation 1. Abnormal situation 3: HF of a certain robot k (t)≤HF min , the free grid surrounded by the covered grid is regarded as its final coverage task, and the covered area and its final coverage task part are regarded as the obstacle area, and steps 2 to 4 are executed according to the processing method of abnormal situation 1.

Images