Farm full coverage path planning method, cleaning navigation control system and medium

Abstract

The present invention discloses a method for full coverage path planning of a farm, a cleaning navigation control system and a medium, the method comprising: generating a grid map of a known chicken house environment through an environmental map learning mode; discretizing the grid map of the working space of a cleaning robot to obtain a neuron cell set of equal size; performing point-to-point path planning in a known grid map using an improved bio-inspired neural network algorithm; performing full coverage path planning using a fusion algorithm of a priority heuristic algorithm and an improved bio-inspired neural network point-to-point path planning algorithm; if an obstacle not marked in a known grid map appears during operation, setting a detour fuzzy logic controller to detour and avoid unknown obstacles, and updating the environmental grid map online. The present invention can include full coverage of the working area in an intelligent decision-making manner, and has low time and energy consumption costs, and can adapt to the complex working environment of the farm, while controlling the price cost of the robot.

Description

Technical Field

[0001] The present invention relates to a method for planning a full-coverage path for a breeding farm, a cleaning navigation control system and medium navigation, and belongs to the technical field of automation of livestock and poultry breeding equipment. Background Art

[0002] China is a large agricultural country, and breeding is vital to China's agriculture and economic development. Large-scale breeding methods are less able to cope with disease risks and environmental sanitation risks, so farms have high requirements for environmental cleanliness. Small mobile cleaning robots are currently a common cleaning solution used in farms. Due to their small size, cleaning robots can enter blind spots for cleaning, greatly reducing the investment in manpower and material resources in farms. However, current cleaning robots have the following main defects:

[0003] 1. Low-cost cleaning robots only have simple sensors and cannot handle complex calculations, so they can only achieve random corner or fixed route planning. The efficiency of random corner route planning is very low. Before the cleaning area is fully covered, a large number of repeated coverage areas will appear, which requires a lot of time and energy to achieve full coverage; fixed routes mainly have reciprocating and spiral planning, but both require the working area to be barrier-free and cannot adapt to situations such as farms with many obstacles and complex environments.

[0004] 2. Cleaning robots equipped with computing units and lidars can realize intelligent path planning and achieve full coverage cleaning at a lower time and energy cost. However, their disadvantage is that they must rely on high-precision lidars for comprehensive modeling of the environment and robot positioning, which is relatively expensive. Moreover, in the actual environment of farms, due to temporary stacking and other situations, it is impossible to guarantee that the cleaning robot can be modeled once and used multiple times. Comprehensive modeling needs to be performed again each time, and the complexity of comprehensive environmental modeling increases with the increase in the number of obstacles. Summary of the invention

[0005] In view of this, the present invention provides a farm cleaning navigation control system, a full coverage path planning method and a storage medium, which can use a robot to make intelligent decisions to fully cover the work area, with low time and energy costs, and can adapt to the complex working environment of the farm, while controlling the price cost of the robot.

[0006] The first object of the present invention is to provide a method for planning a path for full coverage of a farm.

[0007] The second object of the present invention is to provide a farm cleaning navigation control system.

[0008] A third object of the present invention is to provide a computer-readable storage medium.

[0009] The first object of the present invention can be achieved by adopting the following technical solutions:

[0010] A method for planning a full-coverage path for a farm, the method comprising:

[0011] Generate a grid map of the known chicken house environment through an environmental map learning model;

[0012] Discretize the grid map of the cleaning robot's workspace to obtain neuronal cell assemblies of equal size;

[0013] Use an improved bio-inspired neural network algorithm to perform point-to-point path planning in a known grid map;

[0014] Full coverage path planning is performed using a fusion algorithm of a priority heuristic algorithm and an improved bio-inspired neural network point-to-point path planning algorithm;

[0015] If an obstacle that is not marked in the known grid map appears during the operation, a detour fuzzy logic controller is set to detour the unknown obstacle and update the environmental grid map online.

[0016] Furthermore, a grid map of a known chicken house environment is generated through an environmental map learning model. The specific process includes:

[0017] Set the cleaning robot to enter the environment map learning mode, plan the traversal path, and traverse the environment;

[0018] During the traversal process, multiple ultrasonic ranging sensors arranged around the cleaning robot sense obstacles within a 360° range;

[0019] According to the real-time position coordinates of the cleaning robot, the feasible area is divided, and the grid size is set to one vehicle body length. That is, every time the robot travels a distance of one vehicle body length, the corresponding grid position is set as the feasible area in the environmental grid map;

[0020] When there is an obstacle on the traversal path and it is within the envelope of the cleaning robot, the robot will enter the obstacle surround process, otherwise it will not be surrounded;

[0021] During the obstacle circling process, the ultrasonic ranging sensor is used to obtain the boundary information of the obstacle, and the size of the obstacle occupied by the grid map and the position in the grid map are calculated in combination with the real-time position coordinates of the robot, and the obstacle is set as an infeasible area in the grid map;

[0022] After the traversal is completed, a two-dimensional grid map of the known chicken house environment is generated.

[0023] Furthermore, the grid map of the cleaning robot's workspace is discretized to obtain neuron cell assemblies of equal size. The specific process includes:

[0024] According to the grid map of the known farm environment, each discrete grid in the grid map is set to represent a neuron cell, and the distance between adjacent neurons is the distance that the cleaning robot travels in a unit time. , as follows:

[0025] l=V a ·S t Where l is the distance between two adjacent neurons; V a S is the operating speed of the cleaning robot; t For a unit of time.

[0026] If the neuron is occupied, it is marked as an obstacle area, otherwise it is marked as a feasible area;

[0027] Set the parameters of the update equation that determines the activity of each neuron cell. The activity update equation is expressed as follows:

[0028]

[0029] In the formula, x i is the activity of the ith neuron; the non-negative constant coefficients A, B, and D represent the decay rate, upper limit, and lower limit of neuron activity, respectively; k is the number of neural connections of the ith neuron in the neighborhood; w ij is the connection weight from the ith neuron to the jth neuron; i ] + and [I i ] - Represents excitatory input from the outside and inhibitory input from obstacles; [x] + is a linear threshold function, defined as [x] + = max{x,0} and [x] - =max{-x,0};

[0030] Set the target point as a strong external excitation input, and the obstacles and covered areas as strong external inhibition input, as follows:

[0031]

[0032] Wherein, E is a large positive number and satisfies E>10B.

[0033] Furthermore, the improved bio-inspired neural network algorithm is used to perform point-to-point path planning in a known grid map. The specific process includes:

[0034] Set the neuron at the initial point where the robot starts searching for the path as the starting neuron, and set the global target point so that the activity of the global target point is at the peak of the state graph;

[0035] The next position of the cleaning robot is selected based on the activity of the neuron and the previous position of the cleaning robot; as follows:

[0036]

[0037] In the formula, p j is the activity value of the jth neuron in the neighborhood, η is a constant, It is a monotonically increasing function of the difference between the current position of the robot and the next selected position. Δθ∈[0,π] represents the angle between the current moving direction and the next moving direction. The deflection angle of the next selected position is obtained through geometric calculation.

[0038] Furthermore, the calculated deflection angle and the physical parameters of the mobile robot are converted into the linear velocity and angular velocity of the mobile robot, and then converted into the rotation speed of the motor. By sending corresponding control instructions to the motor driver, the cleaning robot is successfully moved from one position to the next position, and the original position is set as the covered area;

[0039] Using the activity update equation, the activity values ​​of all connected neurons in the neighborhood of the current neuron position are calculated, and the activity of the neurons corresponding to the obstacle position is always kept at the bottom;

[0040] According to the activity update equation, the global target point information and the surrounding neuron activity information are transmitted to the current new position, and the activity values ​​of all adjacent neurons in the neighborhood of the current position are continued to be calculated. The next moving position is selected according to the above process until the set global target point position is reached.

[0041] Furthermore, the improved bio-inspired neural network algorithm simplifies the bio-inspired neural network model applied to the case where neurons cover the entire working environment. The specific process includes:

[0042] The original bio-inspired neural network model of the complete working environment is reconstructed into a small dynamic bio-inspired neural network model, and the cleaning machine is set in the center of the dynamic bio-inspired neural network model;

[0043] The distance from each neuron in the small dynamic bio-inspired neural network model to the neuron at the current position of the cleaning robot is less than the maximum measurement radius of the sensor, as follows:

[0044] 0≤D(q i ,q c )≤R

[0045] Where R is the maximum sensing distance of the multi-channel ultrasonic ranging sensor installed on the cleaning robot; D(·) is the function for calculating the Euclidean distance between the positions of two neurons in a two-dimensional plane, as follows:

[0046]

[0047] In the formula, (x i ,y i ) and (x j ,y j ) are the coordinates of the i-th and j-th positions;

[0048] During the movement of the cleaning robot, the size of the dynamic bio-inspired neural network model remains fixed, and only a small portion of the neurons in the dynamic bio-inspired neural network model relative to the global bio-inspired neural network model need to update their neuronal activity each time;

[0049] In the dynamic bio-inspired neural network model, it is necessary to set a virtual target point to gradually approach the global target point. The process includes:

[0050] Select a virtual target point, and select a neuron that can be reached by the cleaning robot from the boundary neurons in the dynamic bio-inspired neural network model. The neuron is closest to the neuron where the global target point is located, and this neuron is selected as the virtual target point.

[0051] The virtual target point is used to replace the global target point, and the most active neighboring neuron is selected as the next position to move according to the activity update equation;

[0052] After reaching the set virtual target, a new virtual target point is set again according to the above selection process until the global target point is reached.

[0053] Furthermore, a fusion algorithm of the priority heuristic algorithm and the improved bio-inspired neural network point-to-point path planning algorithm is used to perform full coverage path planning. The specific process includes:

[0054] When the robot is working normally, it uses a priority heuristic algorithm for path planning, including:

[0055] By prioritizing neurons in the neighborhood, the priorities of neighborhood neurons are divided into two grid maps corresponding to different types;

[0056] The first priority order is: east, north, west, south, southeast, northeast, northwest, southwest;

[0057] The second priority order is: southwest, northwest, northeast, southeast, south, west, north, east;

[0058] The third priority order is: southeast, northeast, northwest, southwest, north, west, south, east;

[0059] When the robot is stuck in a deadlock, an improved bio-inspired neural network point-to-point path planning algorithm is used for path planning, including:

[0060] When the cleaning robot detects neighboring neurons in order of priority, if the positions of the eight surrounding neurons are all in unreachable areas, the robot is judged to be in a "deadlock" state;

[0061] The cleaning robot scans the working area to see if there are any untraversed points, calculates the cost function to reach each untraversed point, and selects the path point with the smallest cost function output as the next traversal point of the robot. The cost function is as follows:

[0062] F i =min((εL ij +ηD ij ))

[0063] In the formula, i represents the untraversed point in the area, L and D are the indicators for evaluating the quality of the path, which are the path length and the total turning angle; ε and η are the weight coefficients of the two indicators, and j is the position point where the cleaning robot is in a "deadlock" state;

[0064] If there are no untraversed points in the scanning working area, the full coverage of all feasible areas of the cleaning robot is completed.

[0065] Furthermore, if an obstacle not marked in the known grid map appears during the operation, a detour fuzzy logic controller is set to detour the unknown obstacle to update the environmental grid map online, specifically including:

[0066] According to the ultrasonic distance sensor installed on the vehicle body in the 360° direction, the cleaning robot establishes a perception window for obstacles;

[0067] When an obstacle appears in the perception window and is determined to be an unknown obstacle, the online update process of the environment grid map is triggered, which includes:

[0068] Determine the key unknown obstacle, i.e. the detour subject, based on the shortest distance and direction of the surrounding obstacles relative to the cleaning robot;

[0069] Design fuzzy rules based on unknown key obstacles;

[0070] Convert expert experience into a rule base and build a rule base through If-then statements;

[0071] The fuzzy output from the inference engine is mapped into a clear control signal through the defuzzifier;

[0072] Determine whether a circle is completed according to the coordinate information of the cleaning robot. If so, exit the online environment update process and mark the size of the grid occupied by the unknown obstacle and its position in the map in the known environment grid map;

[0073] Among them, fuzzy rules are designed based on unknown key obstacles, including:

[0074] Set the fuzzy logic input variables to two, which are the minimum value d of the return distance of the multi-channel ultrasonic distance sensor obs and the corresponding change value Δd obs , the output variable is the speed v of the left and right wheels l and v r ;

[0075] The input variables use trapezoidal membership function and triangular membership function respectively, where the distance input is defined by the trapezoidal membership function and sets the first fuzzy subset, and the distance change input is defined by the triangular membership function and sets the second fuzzy subset;

[0076] The output variable adopts the triangular membership function and sets the third and fourth fuzzy subsets.

[0077] The second object of the present invention can be achieved by adopting the following technical solutions:

[0078] A farm cleaning navigation control system comprises a robot, the robot comprises a vehicle body, a cleaning operation mechanism and a driving system, at least two driving motors are installed inside the vehicle body, the driving motors are connected to the driving wheels of the vehicle body, and are used to drive the vehicle body to move; the cleaning operation mechanism is installed on the vehicle body; the driving system is used to control the operation of the vehicle body, and comprises a bottom control unit and an upper control unit, the bottom control unit is respectively connected to the vehicle body and the cleaning operation mechanism;

[0079] The bottom control unit includes a sensor data reading module, a bottom control unit and an upper control unit communication module, a driving control module and a cleaning control module:

[0080] The sensor data reading module includes an incremental encoder, a posture sensor and an ultrasonic sensor. The incremental encoder is installed on the output shaft of the driving motor and is used to measure the travel distance of the driving wheel of the vehicle body; the posture sensor is used to measure the heading angle of the vehicle body during movement; and the ultrasonic sensor is used to detect obstacle information around the vehicle body.

[0081] The communication module between the bottom control unit and the upper control unit is used to realize data communication between the bottom control unit and the upper control unit;

[0082] The bottom control unit determines the odometer data of the robot according to the signal detected by the incremental encoder, determines the robot's own posture according to the signal of the posture sensor, determines the obstacle distribution in the two-dimensional plane around the robot according to the signal of the ultrasonic sensor, and feeds the data back to the upper control unit;

[0083] The upper control unit is used to execute the above-mentioned full coverage path planning method for the farm, and send operation instructions for controlling the robot to perform full coverage path planning to the lower control unit.

[0084] Furthermore, the bottom control unit also includes an OLED screen display module and a power detection and alarm module:

[0085] The OLED screen display module is used to provide users with a visual interface that can grasp the real-time operating status of the vehicle body;

[0086] The power detection and alarm module is used to detect the remaining power of the vehicle body in real time and alarm when the lithium battery voltage of the vehicle body is lower than the set low voltage threshold;

[0087] Each module of the bottom control unit adopts the framework of the open source FreeRTOS embedded real-time multi-tasking operating system to perform priority scheduling between different modules.

[0088] Furthermore, the upper control unit fuses the data of the attitude sensor and the data of the incremental encoder through Kalman filtering to determine the position of the vehicle body.

[0089] The third object of the present invention can be achieved by adopting the following technical solutions:

[0090] A computer-readable storage medium stores a program, and when the program is executed by a processor, the above-mentioned farm full coverage path planning method is implemented.

[0091] The present invention has the following beneficial effects compared with the prior art:

[0092] 1. The farm full-coverage path planning method and cleaning navigation control system provided by the present invention use a Kalman filter method to obtain the precise position of the robot through a combination of an incremental encoder, a posture sensor, and an ultrasonic sensor, thereby controlling the component cost of the full-coverage cleaning robot while ensuring high-precision positioning and efficient execution of cleaning tasks.

[0093] 2. The farm full coverage path planning method and cleaning navigation control system provided by the present invention obtains the precise position of the robot through the sensor, performs full coverage path planning for the working area, and calculates and controls the motor speed based on the path planning results to achieve high-precision motion control. While completing the full coverage cleaning task, it can also achieve lower time and energy consumption costs.

[0094] 3. The farm full coverage path planning method and cleaning navigation control system provided by the present invention, facing the characteristics of the working environment position and obstacle changes in the farm cleaning task, uses sensor information to learn the environment map, and preliminarily constructs a grid map of the working environment based on the obstacle boundary information. For unknown obstacles encountered during the work process, the fuzzy logic algorithm is used to obtain the boundary of the obstacle, and the environment map is updated online. The environment is continuously learned during the operation, and the coverage path is optimized, which further improves the cleaning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0096] Figure 1 It is a flow chart of the method for full coverage path planning of a farm in Example 1 of the present invention.

[0097] Figure 2 It is a two-dimensional graph of a certain neuron and its neighboring neurons in Example 1 of the present invention.

[0098] Figure 3 It is a schematic diagram of the rasterization of the farm environment map and the D-BINN model according to Example 1 of the present invention.

[0099] Figure 4 It is a schematic diagram of the approximation of the global target point by the designed virtual target point of Example 1 of the present invention.

[0100] Figure 5 It is a schematic diagram of the full coverage process of the farm environment by combining the priority heuristic algorithm and the BINN method in Example 1 of the present invention.

[0101] Figure 6 It is a schematic diagram of online environment update for bypassing unknown obstacles according to Example 1 of the present invention.

[0102] Figure 7 It is a structural block diagram of the farm cleaning navigation control system of embodiment 2 of the present invention. DETAILED DESCRIPTION

[0103] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0104] Embodiment 1:

[0105] like Figures 1 to 6 As shown, this embodiment provides a method for full coverage path planning of a farm, which is applied to a chicken house. The method includes the following steps:

[0106] S100, generating a grid map of a known chicken house environment through an environmental map learning mode, the specific process includes:

[0107] S110, when the robot is in a new farm environment, it is necessary to rasterize the new environment and set the robot to enter an environment map learning mode;

[0108] S120, planning a traversal path as a bow-shaped path, controlling the robot to traverse the entire map through a remote controller, and during the traversal process, multiple ultrasonic ranging sensors arranged around the cleaning robot sense obstacles within a 360° range;

[0109] S130, dividing the feasible area according to the real-time position coordinates of the cleaning robot, and setting the grid size to one vehicle body length, that is, whenever the robot travels a distance of one vehicle body length, the corresponding grid position is set as the feasible area in the environmental grid map;

[0110] S140, when there is an obstacle on the traversal path and it is within the envelope of the cleaning robot, the obstacle surrounding process is started, otherwise it is not surrounded;

[0111] S150, during the obstacle circling process, the boundary information of the obstacle is obtained by using the ultrasonic ranging sensor, and the size of the obstacle is determined in combination with the real-time position coordinates of the robot, so as to obtain the size of the obstacle occupied by the grid map and the position in the grid map, and set it as an infeasible area in the grid map;

[0112] S160: After the traversal is completed, a two-dimensional grid map of the known chicken house environment is generated.

[0113] S200, discretizing the grid map of the cleaning robot workspace in the upper control unit to obtain neuron cell assemblies of equal size, the specific process includes:

[0114] S210, according to the grid map of the known chicken house environment, each discrete grid in the grid map is set to represent a neuron cell, and the distance between adjacent neurons is the distance traveled by the cleaning robot in a unit time, as follows:

[0115] l=V a ·S t

[0116] Where l is the distance between two adjacent neurons; V a S is the operating speed of the cleaning robot; t For a unit of time.

[0117] S220. If the neuron is occupied by the obstacle grid, it is marked as an infeasible area, otherwise it is marked as a feasible area.

[0118] S230, constructing a network structure of multiple neurons according to a biologically inspired neural network (BINN) model, and vividly representing the real-time changes of the environment through changes in neuronal activity, the specific process includes:

[0119] S231, setting the parameters of the update equation that determines the activity of each neuron cell, the activity update equation is expressed as follows:

[0120]

[0121] In the formula, x i is the activity of the ith neuron; the non-negative constant coefficients A, B, and D represent the decay rate, upper limit, and lower limit of neuron activity, respectively; k is the number of neural connections of the ith neuron in the neighborhood; w ij is the connection weight from the ith neuron to the jth neuron; i ] + and [I i ] - Represents excitatory input from the outside and inhibitory input from obstacles; [x] + is a linear threshold function, defined as [x] + = max{x,0} and [x] - =max{-x,0}.

[0122] S232, set the target point as a strong external excitation input, and the obstacles and covered areas as strong external inhibition input, as follows:

[0123]

[0124] Wherein, E is a large positive number and satisfies E>10B.

[0125] In this embodiment, the following parameters are set: A=30, B=D=1, E=60, η=0.7, r0=2.

[0126] S300, using the improved BINN algorithm to perform point-to-point path planning in a known grid map, the specific process includes:

[0127] S310, improving the BINN algorithm, specifically including:

[0128] S311. Reconstruct the BINN network model of the original complete working environment into a small dynamic BINN model, and set the cleaning machine at the center of the dynamic bio-inspired neural network model.

[0129] Among them, the distance from each neuron in the small dynamic BINN model to the neuron where the cleaning robot is currently located is less than the maximum measurement radius of the sensor, as follows:

[0130] 0≤D(q i ,q c )≤R

[0131] Where R is the maximum sensing distance of the multi-channel ultrasonic ranging sensor installed on the cleaning robot; D(·) is the function for calculating the Euclidean distance between the positions of two neurons in a two-dimensional plane, as follows:

[0132]

[0133] In the formula, (x i ,y i ) and (x j ,y j ) are the coordinates of the i-th and j-th positions.

[0134] During the movement of the cleaning robot, the size of the small dynamic BINN model remains fixed, and each time only a small portion of the neurons in the small dynamic BINN model relative to the global BINN model need to update their neuronal activity;

[0135] S320, in the small dynamic BINN model, setting a virtual target point, the process includes:

[0136] S321, selecting a neuron that can be reached by the cleaning robot and is closest to the global target point neuron in the boundary neuron cells of the small dynamic BINN model as a virtual target point;

[0137] S322, using the virtual target point to replace the global target point, and selecting the most active neighboring neuron as the next position to move according to the activity update equation, so as to gradually move toward the virtual target point;

[0138] S323: After reaching the set virtual target point, a new virtual target point is set again according to the above selection process.

[0139] S330, using an improved bio-inspired neural network algorithm to perform point-to-point path planning in a known grid map, the specific process comprising:

[0140] S331, setting the neuron at the initial point where the robot starts searching for the path as the starting neuron, and setting the global target point so that the activity of the global target point is at the peak of the state graph;

[0141] S332, selecting a next position of the cleaning robot according to the activity of the neuron and the previous position of the cleaning robot;

[0142] In this embodiment, the current position of the robot in the two-dimensional Cartesian workspace is q c , its position coordinates are (x c ,y c ), the position coordinate is the precise real-time position obtained by fusing the attitude sensor and incremental encoder odometer information through Kalman filtering.

[0143] The choice of the next position of the cleaning robot is determined by the activity of the neuron at that point and the robot's previous position, as follows:

[0144]

[0145] In the formula, p j is the activity value of the jth neuron in the neighborhood, η is a constant, is a monotonically increasing function of the difference between the current position of the robot and the next selected position, Δθ∈[0,π], which represents the angle between the current moving direction and the next moving direction. Through geometric operations, we can get Δθ j as follows:

[0146] Δθ j =|θ j -θ c |=|atan2(y j -y c , x j -x c )-atan2(y c -y p , x c -x p )|

[0147] In the formula, q c is the current position of the cleaning robot, and its position coordinates are (x c ,y c ), the activity value of the neuron corresponding to this point is The robot position coordinates at the previous moment are (x p ,y p ), the position selected at the next moment in the two-dimensional Cartesian workspace is q n , whose coordinate position is (x n ,y n ), the activity value of the neuron corresponding to this point is

[0148] In this embodiment, Δθ j The driving control module of the bottom control unit will calculate the value of the Δθ between the current position and the next position. j , to control the robot to move to the next selected position.

[0149] In this embodiment, by Δθ j The value is converted into the speed command of the left and right drive motors in the bottom control unit, and the speed command is sent to the motor controllers of the left and right drive motors through the CAN bus.

[0150] S333. Every time the cleaning robot successfully moves from one position to the next, the original position is set as the covered area.

[0151] S334, using the activity update equation, calculating the activity values ​​of each connected neuron in the neighborhood of the current neuron position, and making the neuron activity at the position corresponding to the obstacle always remain at the bottom;

[0152] S335. According to the activity update equation, the global target point information and the surrounding neuron activity information are transferred to the current new position, and the activity values ​​of all adjacent neurons in the neighborhood of the current position are continued to be calculated. The next moving position is selected according to the above process until the set global target point position is reached.

[0153] S400, using the fusion algorithm of the priority heuristic algorithm and the improved biologically inspired neural network point-to-point path planning algorithm to perform full coverage path planning, the specific process includes:

[0154] S410, when the robot works normally, a priority heuristic algorithm is used for path planning, specifically including:

[0155] By prioritizing neurons in the neighborhood, the priorities of neighborhood neurons are divided into two grid maps corresponding to different types;

[0156] The first priority order is: east, north, west, south, southeast, northeast, northwest, southwest;

[0157] The second priority order is: southwest, northwest, northeast, southeast, south, west, north, east;

[0158] The third priority order is: southeast, northeast, northwest, southwest, north, west, south, east;

[0159] S420, when the robot is in a "deadlock", the improved BINN algorithm is used for path planning, specifically including:

[0160] S421, when the cleaning robot detects the neighboring neurons in order of installation priority, if the positions of the eight surrounding neurons are all in an unreachable area, the robot is judged to be in a "deadlock" state;

[0161] S422, the cleaning robot scans whether there are any untraversed points in the working area, calculates the cost function of reaching each untraversed point, and selects the path point with the smallest cost function output as the next traversed point of the robot. The cost function is as follows:

[0162] F i =min((εL ij +ηD ij ))

[0163] In the formula, i represents the untraversed point in the area, L and D are the indicators for evaluating the quality of the path, which are the path length and the total turning angle; ε and η are the weight coefficients of the two indicators, and j is the position point where the cleaning robot is in a "deadlock" state;

[0164] S423. If there are no untraversed points in the scanning working area, then the cleaning robot completes full coverage of all feasible areas.

[0165] S500: If an obstacle not marked in the known grid map appears during the operation, a detour fuzzy logic controller is set to detour the unknown obstacle for online updating of the environment grid map.

[0166] Specifically, when the cleaning robot is working according to the planned full coverage path, it encounters obstacles whose distribution has not been learned through environmental map learning. The fuzzy logic algorithm is used to bypass the unknown obstacles and update the environmental map.

[0167] S510, establishing a perception window of the cleaning robot for obstacles according to the ultrasonic distance sensor installed in the 360° direction of the vehicle body;

[0168] In this embodiment, the ultrasonic distance sensor installed in the 360° direction of the vehicle body is an 8-way ultrasonic sensor;

[0169] S520: When an obstacle appears in the perception window and is determined to be an unknown obstacle, the online update process of the environment grid map is triggered, which specifically includes:

[0170] S521: During driving, if an unknown obstacle is detected within the sensing range, the shortest distance among the eight data sent back by the eight ultrasonic sensors is d. obs And the obstacles in the corresponding directions are regarded as unknown obstacles, i.e., the detour subjects.

[0171] S522, designing fuzzy rules based on unknown key obstacles, and designing the input of the fuzzy logic controller to be the shortest distance d between the robot and the unknown obstacle obs And the corresponding distance change Δd obs Based on these two quantities, the output of the fuzzy controller is derived: the wheel speed v of the robot's left and right driving wheels l and v r .

[0172] In this embodiment, the domain, scale change and membership function of the input and output variables are first determined, and the distance input d obs It is divided into the domain of [0, 0.8m] and defined by five trapezoidal membership functions. The first fuzzy subset is set as: {very close, close, medium, far, very far}; the input of the distance change Δd obs The domain is |-1, 1m], which is defined by ten triangular membership functions. The second fuzzy subset is set as: {rapid approach, medium approach, slow approach, slow approach, unchanged, slow separation, slow separation, medium separation, rapid separation}; the wheel speed v of the left and right driving wheels l and v r The output variables all use triangular membership functions, and the domain of each output variable is [-100, 100 cm / s]. Fifteen triangular membership functions are defined, and the third and fourth fuzzy subsets are set in sequence, namely, left wheel speed output: {negative left 7, negative left 6, negative left 5, negative left 4, negative left 3, negative left 2, negative left 1, left, positive left 1, positive left 2, positive left 3, positive left 4, positive left 5, positive left 6, positive left 7}; right wheel speed output: {negative right 7, negative right 6, negative right 5, negative right 4, negative right 3, negative right 2, negative right 1, right, positive right 1, positive right 2, positive right 3, positive right 4, positive right 5, positive right 6, positive right 7}.

[0173] S523. Convert expert experience into a rule base and establish the rule base through If-then statements.

[0174] In this embodiment, a fuzzy distance library and a fuzzy distance variation library of the expert system are established based on fifty If-then statement rules.

[0175] S524, mapping the fuzzy output from the inference engine into a clear control signal through the defuzzifier.

[0176] In this embodiment, the “center of mass method” is used to combine the outputs represented by the triggered fuzzy rules for fuzzy clarification, thereby obtaining the corresponding appropriate control action.

[0177] In this embodiment, when the robot is very close to the obstacle or (N), if the robot has the tendency to move away from the obstacle, the If-then statement is: If d obs is close to Δd obs To separate at medium speed, the left wheel speed is positive left 3 and the right wheel speed is positive right 1 to ensure that the robot can avoid collision and move along the boundary of the obstacle as much as possible.

[0178] S525. When it is determined whether one circle is completed based on the coordinate information of the cleaning robot, the online environment update process is exited, and the size of the grid occupied by the unknown obstacle and its position in the map are marked in the known environment grid map.

[0179] It should be noted that although the above method operations are described in a particular order in the accompanying drawings, this does not require or imply that these operations must be performed in this particular order, or that all of the operations shown must be performed to achieve the desired results. On the contrary, the steps depicted may be performed in a different order. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0180] Embodiment 2:

[0181] like Figure 7 As shown, this embodiment provides a farm cleaning navigation control system, including a robot, wherein the robot includes a body, a cleaning operation mechanism and a driving system, the body is provided with a mobile chassis, driving wheels are provided on both sides of the rear of the mobile chassis, and a universal wheel is provided in the front, two driving motors are installed inside the body, the two driving motors are connected to the two driving wheels in a one-to-one correspondence, and the two driving wheels are driven by the two driving motors to drive the body to move; the cleaning operation mechanism is installed on the body; the driving system is used to control the operation of the body, including a bottom control unit 100 and an upper control unit 200, the bottom control unit 100 is respectively connected to the body and the cleaning operation mechanism, and is used to receive instructions from the upper control unit 200, and to control the movement of the body and the manure cleaning mechanism in real time to perform manure cleaning.

[0182] Furthermore, the drive motor is a DC motor, and the two drive motors are respectively a left drive motor and a right drive motor. The bottom control unit 100 is connected to the left drive motor and the right drive motor through a CAN bus to achieve independent control of the left drive motor and the right drive motor.

[0183] The bottom-level control unit 100 includes a driving control module 110, a cleaning control module 120, a sensor data reading module 130, an OLED screen display module 140, a power monitoring and alarm module 150, and a bottom-level control unit and upper-level control unit communication module 160. The sensor data reading module 130 includes an incremental encoder 131, a posture sensor 132, and an ultrasonic sensor 133.

[0184] In the underlying control unit 100, in order to cope with the scenario of processing multiple tasks simultaneously, the control system is written using the open source FreeRTOS embedded real-time operation multi-tasking operating system framework to perform priority scheduling between different tasks, that is, each module of the underlying control unit 100 uses the open source FreeRTOS embedded real-time operation multi-tasking operating system framework to perform priority scheduling between different modules.

[0185] In this embodiment, the bottom-level control unit and upper-level control unit communication module 160 is used to implement data communication between the bottom-level control unit 100 and the upper-level control unit 200 .

[0186] In order to cope with possible emergencies in the farm environment, the driving control module 110 and the cleaning control module 120 both include a remote control mode and an automatic operation mode.

[0187] Specifically, the remote control mode adopts the sbus protocol, and parses the remote control command received by the receiver in the form of DMA direct memory access. The remote control mode adopts a dual-channel hybrid control method to realize on-site manual control of the robot's driving, with the purpose of directly manually controlling the robot in case of an emergency; the automatic operation mode is to execute the farm full coverage path planning method of Example 1, and issue instructions through the bottom control unit and the upper control unit communication module 160 to control the robot to perform automatic operation, specifically including:

[0188] Generate a grid map of the known chicken house environment through an environmental map learning model;

[0189] Discretize the grid map of the cleaning robot's workspace to obtain neuronal cell assemblies of equal size;

[0190] Use an improved bio-inspired neural network algorithm to perform point-to-point path planning in a known grid map;

[0191] Full coverage path planning is performed using a fusion algorithm of a priority heuristic algorithm and an improved bio-inspired neural network point-to-point path planning algorithm;

[0192] If an obstacle that is not marked in the known grid map appears during the operation, a detour fuzzy logic controller is set to detour the unknown obstacle and update the environmental grid map online.

[0193] In this embodiment, the incremental encoder 131 is installed on the output shaft of the drive motor, and the signal line of the incremental encoder 131 is connected to the electric drive controller pre-installed on the drive motor, which is used to measure the travel distance of the driving wheels of the vehicle body. The pulse feedback signal of the incremental encoder 131 of the electric drive controller connected to the corresponding address on the CAN bus can be read in the bottom control unit 100, and the real-time mileage information of the vehicle body can be obtained by calculation in the program, and the real-time mileage information of the vehicle body is fed back to the upper control unit 200.

[0194] In this embodiment, the attitude sensor 132 is used to measure the turning angle of the vehicle body during movement. It adopts a nine-axis acceleration sensor, including a three-axis accelerometer, a three-axis gyroscope and a three-axis magnetometer. The accelerometer can detect the acceleration of the vehicle body and detect the movement states of the vehicle body such as tilt, impact, vibration, etc. The gyroscope can detect the attitude (roll angle, pitch angle and heading angle) and angular velocity of the vehicle body. The magnetometer can detect the yaw angle of the vehicle body and locate the direction of the vehicle body. The attitude sensor 132 transmits the real-time heading angle of the vehicle body to the underlying control unit 100 through the serial port in the form of DMA direct memory access.

[0195] In this embodiment, the ultrasonic sensor 133 is used to detect obstacle information around the vehicle body. It adopts a 360° ultrasonic matrix sensor with a total of 8 channels, which are arranged in a certain way without mutual echo interference. The 360° ultrasonic matrix sensor can obtain obstacle distance information on the two-dimensional plane around the vehicle body during operation and driving, and update the obstacle distribution of the environment in the surrounding direction in real time. The 360° ultrasonic sensor 133 transmits the real-time distance of the surrounding area to the underlying control unit 100 through the RS458 communication interface.

[0196] In this embodiment, the bottom-level control unit 100 adopts an ARM microcontroller, and the upper-level control unit 200 adopts an NVIDIA TX2 industrial computer. The bottom-level control unit 100 determines the odometer data of the robot (real-time mileage information of the vehicle body) according to the signal detected by the incremental encoder 131, determines the robot's own posture (real-time heading angle of the vehicle body) according to the signal of the posture sensor 132, and determines the obstacle distribution in the two-dimensional plane around the robot (obstacle distance information on the two-dimensional plane around the vehicle body during operation) according to the signal of the ultrasonic sensor; the bottom-level control unit 100 uploads the obtained sensor data to the upper-level control unit 200 through the communication module 160 between the bottom-level control unit and the upper-level control unit; the upper-level control unit 200 stores a grid map of a known working environment formed by edge learning, and uses the full coverage algorithm and fuzzy logic algorithm of the biologically inspired neural network to send operation instructions to the bottom-level control unit 100 to control the robot to perform full coverage path planning.

[0197] Furthermore, the upper control unit 200 fuses the data of the attitude sensor and the data of the incremental encoder through Kalman filtering to determine the precise position of the vehicle body. The specific process includes:

[0198] 1) Prediction:

[0199]

[0200]

[0201] 2) Update:

[0202] First calculate the Kalman gain K:

[0203]

[0204] Then calculate the distribution of the posterior probability:

[0205]

[0206]

[0207] The above-mentioned That is the corrected location information.

[0208] in, is the state variable of the system at time k, u k is the control variable of the system at time k; B is the system input control matrix, is the estimated state of the system at time k-1; A k and C k are the system state transfer matrix and measurement matrix respectively; R and Q k are the process noise matrix and the measurement noise matrix respectively; represents the last covariance matrix, is the current state covariance matrix; z k is the measurement vector of the sensor at time k.

[0209] In this embodiment, the OLED screen display module 140 uses I2C for communication to display the current operating mode, remaining power and other operating information, providing the user with a visual interface that can grasp the real-time operating status of the vehicle body.

[0210] In this embodiment, the power detection and alarm module 150 is used to detect the remaining power of the vehicle body in real time to monitor the voltage of the vehicle body lithium battery. By setting the voltage threshold, if it is lower than the set low voltage threshold, the low voltage signal is uploaded to the upper control unit 200, and the upper control unit 200 uploads the current position information of the vehicle body and the low power alarm information to the external terminal 300 through the wireless communication unit; at the same time, the alarm signal is controlled by the bottom control unit 100. Based on this, the remote monitoring of the vehicle body power can be realized, and the vehicle body with insufficient power can be located faster and more accurately. At the same time, the bottom control unit 100 controls the alarm to generate an alarm and drives the vehicle body to automatically return to charge according to the planned route.

[0211] Furthermore, the farm cleaning navigation control system of this embodiment also includes a power module, which supplies power to the upper control unit, the lower control unit, the posture sensor, the robot's movement control component, and the cleaning operation mechanism motion control component.

[0212] Furthermore, the farm cleaning navigation control system of this embodiment also includes an external terminal 300, which communicates with the upper control unit 200 through a wireless communication unit, and the voltage output end of the power module is connected to the bottom control unit through a voltage divider circuit. The power information is detected by the power detection and alarm module of the bottom control unit. When the lithium battery voltage of the vehicle body is lower than the set low voltage threshold, the bottom control unit 100 sends a low voltage signal and sends the remaining power to the upper control unit 200, and the upper control unit 200 uploads the robot's current position information and low power alarm information to the external terminal 300 through the wireless communication unit; at the same time, the bottom control unit 100 controls the issuance of an alarm signal, receives the next step instruction from the external terminal, and executes instructions including automatic return and immediate stop according to the remaining power.

[0213] It can be seen that the farm full coverage path planning method and cleaning navigation control system provided by embodiments 1 and 2 of the present invention use a combination of an incremental encoder, a posture sensor, and an ultrasonic sensor to obtain the precise position of the robot using a Kalman filter method, while ensuring high-precision positioning and efficient execution of cleaning tasks, and controlling the component cost of the full coverage cleaning robot. At the same time, the farm full coverage path planning method and cleaning navigation control system provided by embodiments 1 and 2 of the present invention obtain the precise position of the robot through a sensor, perform full coverage path planning on the working area, and calculate and control the motor speed based on the path planning results to achieve high-precision motion control. While completing the full coverage cleaning task, lower time and energy consumption costs can also be achieved. In addition, the farm full coverage path planning method and cleaning navigation control system provided by embodiments 1 and 2 of the present invention, facing the characteristics of the working environment position and obstacle changes in the farm cleaning task, use sensor information to learn the environment map, and preliminarily construct a grid map of the working environment based on the obstacle boundary information. For unknown obstacles encountered during the work process, fuzzy logic algorithms are used to obtain the boundaries of the obstacles, update the environmental map online, continuously learn the environment during the operation, optimize the coverage path, and further improve the cleaning efficiency.

[0214] Embodiment 3:

[0215] This embodiment provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for planning a full coverage path for a farm in the above-mentioned embodiment 1 is implemented, which is specifically as follows:

[0216] Generate a grid map of the known chicken house environment through an environmental map learning model;

[0217] Discretize the grid map of the cleaning robot's workspace to obtain neuronal cell assemblies of equal size;

[0218] Use an improved bio-inspired neural network algorithm to perform point-to-point path planning in a known grid map;

[0219] Full coverage path planning is performed using a fusion algorithm of a priority heuristic algorithm and an improved bio-inspired neural network point-to-point path planning algorithm;

[0220] If an obstacle that is not marked in the known grid map appears during the operation, a detour fuzzy logic controller is set to detour the unknown obstacle and update the environmental grid map online.

[0221] It should be noted that the computer-readable storage medium of the present embodiment may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0222] In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In this embodiment, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium may also be any computer-readable storage medium other than a computer-readable storage medium, which may send, propagate, or transmit a program used by or in combination with an instruction execution system, device or device. The computer program contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0223] The computer readable storage medium can be written in one or more programming languages ​​or a combination thereof to execute the computer program of the present embodiment, and the programming language includes an object-oriented programming language, such as Java, Python, C++, and a conventional procedural programming language, such as C or a similar programming language. The program can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect through the Internet).

[0224] The above is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solution and inventive concept of the present invention within the scope disclosed by the present invention, which shall fall within the protection scope of the present invention.

Claims

1. A method for planning a full coverage path for a farm, characterized in that: The following steps are involved: Generate a raster map of the known farm environment through the environmental map learning model; Discretize the grid map of the cleaning robot's workspace to obtain neuronal cell assemblies of equal size; Use an improved bio-inspired neural network algorithm to perform point-to-point path planning in a known grid map; Full coverage path planning is performed using a fusion algorithm of a priority heuristic algorithm and an improved bio-inspired neural network point-to-point path planning algorithm; If an obstacle that is not marked in the known grid map appears during the operation, a detour fuzzy logic controller is set to detour the unknown obstacle and avoid it, which is used to update the environmental grid map online. If an obstacle that is not marked in the known grid map appears during the operation, a detour fuzzy logic controller is set to detour the unknown obstacle to update the environmental grid map online, including: According to the ultrasonic distance sensor installed on the vehicle body in the 360° direction, the cleaning robot establishes a perception window for obstacles; When an obstacle appears in the perception window and is determined to be an unknown obstacle, the online update process of the environment grid map is triggered, which includes: Determine the key unknown obstacle, i.e. the detour subject, based on the shortest distance and direction of the surrounding obstacles relative to the cleaning robot; Design fuzzy rules based on unknown key obstacles; Convert expert experience into a rule base and build a rule base through If-then statements; The fuzzy output from the inference engine is mapped into a clear control signal through the defuzzifier; Determine whether a circle is completed according to the coordinate information of the cleaning robot. If so, exit the online environment update process and mark the size of the grid occupied by the unknown obstacle and its position in the map in the known environment grid map; Among them, fuzzy rules are designed based on unknown key obstacles, including: Set the fuzzy logic input variables to two, which are the minimum value d of the return distance of the multi-channel ultrasonic distance sensor obs and the corresponding change Δd obs , the output variable is the speed v of the left and right wheels l and v r ; The input variables use trapezoidal membership function and triangular membership function respectively, where the distance input is defined by the trapezoidal membership function and sets the first fuzzy subset, and the distance change input is defined by the triangular membership function and sets the second fuzzy subset; The output variable adopts a triangular membership function and sets the third and fourth fuzzy subsets.

2. The method for planning a full coverage path for a farm according to claim 1, characterized in that: The grid map of the known farm environment is generated through the environmental map learning model. The specific process includes: Set the cleaning robot to enter the environment map learning mode, plan the traversal path, and traverse the environment; During the traversal process, multiple ultrasonic ranging sensors arranged around the cleaning robot sense obstacles within a 360° range; According to the real-time position coordinates of the cleaning robot, the feasible area is divided, and the grid size is set to one vehicle length. That is, every time the robot travels a distance of one vehicle length, the corresponding grid position is set as the feasible area in the environmental grid map; When there is an obstacle on the traversal path and it is within the envelope of the cleaning robot, the robot will enter the obstacle surround process, otherwise it will not be surrounded; During the obstacle circling process, the ultrasonic distance sensor is used to obtain the boundary information of the obstacle, and the size of the obstacle occupied by the grid map and the position in the grid map are calculated in combination with the real-time position coordinates of the robot, and the infeasible area is set in the grid map; After the traversal is completed, a two-dimensional grid map of the known farm environment is generated.

3. The method for planning a full coverage path for a farm according to claim 1, characterized in that: Discretize the grid map of the cleaning robot's workspace to obtain neuron cell assemblies of equal size. The specific process includes: According to the grid map of the known farm environment, each discrete grid in the grid map is set to represent a neuron cell, and the distance between adjacent neurons is the distance traveled by the cleaning robot in a unit time; If the neuron is occupied by the obstacle grid, it is marked as an infeasible area, otherwise it is marked as a feasible area; Set the parameters of the activity update equation that determines the activity of each neuron cell. The activity update equation expression is: In the formula, A represents the decay rate of neuronal activity, B represents the upper limit of neuronal activity, D represents the lower limit of neuronal activity, and x i is the activity of the ith neuron, k is the number of neural connections of the ith neuron in the neighborhood; w ij is the connection weight from the ith neuron to the jth neuron; i ] + and [I i ] - Represents excitatory input from the outside and inhibitory input from obstacles; [x] + is a linear threshold function, defined as [x] + = max{x,0} and [x] - =max{-x,0}; The target point is set as a strong external excitation input, and the obstacles and covered areas are set as strong external inhibition input. The specific expression is: In the formula, I i is the input of the i-th neuron, E is a positive number, and satisfies E>10B.

4. The method for planning a full coverage path for a farm according to claim 3, characterized in that: The improved bio-inspired neural network algorithm is used to perform point-to-point path planning in a known grid map. The specific process includes: Set the neuron at the initial point where the robot starts searching for the path as the starting neuron, and set the global target point so that the activity of the global target point is at the peak of the state graph; The next position of the cleaning robot is selected based on the activity of the neurons and the previous position of the cleaning robot; Every time the cleaning robot successfully moves from one position to the next, the previous position is set as the covered area; Using the activity update equation, the activity values ​​of all connected neurons in the neighborhood of the current neuron position are calculated, and the activity of the neurons corresponding to the obstacle position is always kept at the bottom; According to the activity update equation, the global target point information and the surrounding neuron activity information are transmitted to the current new position, and the activity values ​​of all adjacent neurons in the neighborhood of the current position are continued to be calculated. The next moving position is selected according to the above process until the set global target point position is reached.

5. The method for planning a full coverage path for a farm according to claim 4, characterized in that: The improved bio-inspired neural network algorithm simplifies the bio-inspired neural network model applied to the case where neurons cover the entire working environment. The specific process includes: The original bio-inspired neural network model of the complete working environment is reconstructed into a small dynamic bio-inspired neural network model, and the cleaning robot is set in the center of the dynamic bio-inspired neural network model; The distance from each neuron in the small dynamic bio-inspired neural network model to the neuron where the cleaning robot is located is smaller than the maximum measurement radius of the sensor; During the movement of the cleaning robot, the size of the small dynamic bio-inspired neural network model remains fixed, and only the neurons in the small dynamic bio-inspired neural network model need to update their neuronal activity each time; In a small dynamic bio-inspired neural network model, it is necessary to set a virtual target point to gradually approach the global target point. The process includes: Select a virtual target point, and select a neuron that can be reached by the cleaning robot from the boundary neurons in the small dynamic bio-inspired neural network model. The neuron is closest to the neuron where the global target point is located, and this neuron is selected as the virtual target point. The virtual target point is used to replace the global target point, and the most active neighboring neuron is selected as the next position to move according to the aforementioned activity update equation, so as to gradually move toward the virtual target point; After arriving at the set virtual target point, a new virtual target point is reset according to the above selection process.

6. The method for planning a full coverage path for a farm according to claim 1, characterized in that: The full coverage path planning is performed using the fusion algorithm of the priority heuristic algorithm and the improved biologically inspired neural network point-to-point path planning algorithm. The specific process includes: When the robot is working normally, it uses a priority heuristic algorithm for path planning, including: By prioritizing neurons in the neighborhood, the priorities of neighborhood neurons are divided into two grid maps corresponding to different types; The first priority order is: east, north, west, south, southeast, northeast, northwest, southwest; The second priority order is: southwest, northwest, northeast, southeast, south, west, north, east; The third priority order is: southeast, northeast, northwest, southwest, north, west, south, east; When the robot is stuck in a "deadlock", an improved bio-inspired neural network point-to-point path planning algorithm is used for path planning, including: When the cleaning robot detects neighboring neurons in order of priority, if the positions of the eight surrounding neurons are all in unreachable areas, the robot is judged to be in a "deadlock" state; The cleaning robot scans the working area to see if there are any untraversed points, calculates the cost function to reach each untraversed point, and selects the path point with the minimum cost function output as the next traversal point of the robot; If there are no untraversed points in the scanning working area, the full coverage of all feasible areas of the cleaning robot is completed.

7. A computer-readable storage medium storing a program, characterized in that: When the program is executed by the processor, the farm full coverage path planning method described in any one of claims 1-6 is implemented.

Images