An AGV path planning method and system in a cold storage warehouse for agricultural products

By using an improved OMA algorithm and artificial potential field method, combined with depth point cloud and 2D laser scanning data, the problems of local minima and target unreachability in AGV path planning were solved, enabling AGVs to efficiently avoid obstacles and reach targets in unknown environments.

CN117740019BActive Publication Date: 2024-11-22ANHUI FULANDE REFRIGERATION EQUIP CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202311818748.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-12-26
Publication Date
2024-11-22
Estimated Expiration
2043-12-26

AI Technical Summary

Technical Problem

In complex environments, AGV path planning algorithms are prone to getting stuck in local minima and the problem of unreachable targets. Existing algorithms are difficult to effectively avoid obstacles in unknown environments.

Method used

An improved OMA algorithm is used for global path planning, combined with an improved artificial potential field method for local path planning. A raster map is constructed using depth point cloud and 2D laser scanning data. A target distance function and a virtual endpoint mechanism are introduced to solve the problems of local minima and target unreachability.

Benefits of technology

It improves the AGV's path planning adaptability and obstacle avoidance capabilities in unknown environments, ensuring that the vehicle can successfully reach the target point and avoid local minima and unreachable target states.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117740019B_ABST
    Figure CN117740019B_ABST
Patent Text Reader

Abstract

The present invention discloses an AGV path planning method and system in a cold storage warehouse for agricultural products. The method is that the AGV vehicle is equipped with a depth camera and a laser radar to collect warehouse map environment data, and a two-dimensional fence map is constructed by using a Gmapping algorithm. The warehouse AGV uses an improved OMA algorithm to complete global path planning. The improved OMA algorithm includes using an improved cyclic chaotic mapping initialization to replace its original random initialization, and a firefly perturbation strategy to update and iterate the optimal solution of the algorithm; introducing a virtual end point strategy and a target distance function, and using an improved artificial potential field method to complete AGV local path planning, wherein the target distance function can effectively avoid the situation where the AGV car cannot reach the target point, and the virtual end point method can help individuals quickly escape from the local extreme value and complete the obstacle avoidance function. The present invention accurately and quickly optimizes the path planning problem of agricultural products in cold storage warehouses, reduces the scheduling time of agricultural products, and reduces manual participation, thereby ensuring the safety of staff.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of AGV, and in particular relates to an AGV path planning and system in a cold storage warehouse for agricultural products. Background Art

[0002] In the fields of manufacturing and warehousing logistics, automated guided vehicles (AGVs) play a key role as unmanned intelligent handling equipment, greatly improving the automation level and production efficiency of factories. Although AGV path planning is relatively mature in static environments with known maps, designing effective path planning and obstacle avoidance algorithms in complex environments, especially when environmental information and obstacle information are unknown, is still a key challenge in the AGV field.

[0003] In terms of path planning, many scholars have proposed a variety of algorithms, but among so many algorithms, almost no algorithm can avoid the problem of falling into local minima. Among them, the artificial potential field method is the representative. The artificial potential field method path planning is a virtual force method proposed by Khatib; its basic idea is to design the movement of the AGV in the surrounding environment into a movement in an abstract artificial gravitational field. The target point generates "gravity" on the mobile AGV, and the obstacle generates "repulsion" on the mobile AGV. Finally, the movement of the mobile AGV is controlled by the resultant force. The path planned by the potential field method is generally smooth and safe, but the artificial potential field method has some shortcomings. For example, in the process of path planning, the AGV can easily fall into a local minimum, that is, it is balanced by the gravitational force of the target point and the repulsive force of the obstacle, and cannot continue to move to the target point. In addition, the artificial potential field method also faces the problem that the AGV target is unreachable. When an obstacle appears near the target, the AGV cannot reach the target point smoothly due to the large repulsive force.

[0004] In the field of AGV, improving algorithms to overcome common local minima and other problems in path planning has become a technical challenge that researchers urgently need to solve. Therefore, designing AGV path planning and obstacle avoidance algorithms to improve the system's adaptability in unknown environments is a key research task in the current field. Summary of the invention

[0005] Purpose of the invention: The present invention provides an AGV path planning method and system in a cold storage warehouse for agricultural products. The improved OMA algorithm is used for global planning of the AGV, and then the improved artificial potential field method is used for local path planning of the AGV to realize the obstacle avoidance function of the AGV.

[0006] Technical solution: The invention discloses an AGV path planning method in a cold storage warehouse for agricultural products, comprising the following steps:

[0007] (1) Construct a 2D grid map based on 3D point cloud data and 2D laser scanning data;

[0008] (2) The Gmapping algorithm is used to optimize the two-dimensional grid map to obtain the environmental map of the agricultural product cold storage warehouse;

[0009] (3) using the improved OMA algorithm to perform global path planning for the AGV vehicle based on the environment map obtained in step (2); the improved OMA algorithm includes using an improved cyclic chaotic map initialization to replace its original random initialization, and a firefly perturbation strategy to iteratively optimize the optimal solution of the algorithm;

[0010] (4) The AGV moves forward along the optimal path obtained in step (3), and uses the laser radar on board to detect whether there are obstacles in the driving path. If there are obstacles, it jumps to step (5); otherwise, it jumps to step (6);

[0011] (5) The artificial potential field method is used to complete the local path planning of the AGV to realize the basic obstacle avoidance function; when the AGV is close to the target point and there are obstacles around the target point, the target is unreachable, and the target distance function is introduced; when the AGV falls into the local optimal solution state, a new virtual end point is created near the car, and an additional attraction is applied to the AGV to attract it to the position of the virtual end point;

[0012] (6) The AGV successfully delivers the goods to the designated destination and displays them in real time through the data platform.

[0013] Furthermore, the implementation process of step (1) is as follows:

[0014] The AGV is started in a cold storage warehouse for agricultural products. The AGV is equipped with a depth camera and a lidar to collect map information and build a navigation map through the SLAM algorithm. The DC motor drives the AGV to move the position, fuse the 3D data of the depth point cloud with the 2D laser scanning data, and use the fused data to build a 2D grid map.

[0015] Furthermore, the Gmapping algorithm in step (2) is optimized based on the RBPF_SLAM algorithm; RBPF_SLAM decomposes the SLAM problem into two problems: positioning and mapping. The vehicle path is estimated first, and then the environment map is estimated based on the vehicle path; the RBPF_SLAM mapping process is divided into sampling, importance weight calculation, repositioning, and updating of landmarks; the lidar data and odometer data are input into slam_gmapping, and then openslam_gmapping is called to observe whether the data and the existing map are matched successfully. If successful, sampling is performed from the improved suggestion distribution and the importance weight is calculated. Otherwise, sampling is performed from the motion model and the importance weight is calculated; then the example validity N is calculated eff , judge N eff Is it less than the particle validity threshold? If so, resample it; otherwise, update the map. Finally, the agricultural product cold storage warehouse environment map is output through the slam_gmapping module.

[0016] Furthermore, the specific steps of improving the optical microscope algorithm described in step (3) are:

[0017] (31) Set the objective function of the optical microscope algorithm to the shortest time path and initialize the relevant parameters: population size, dimension size, upper and lower bounds of the path search space;

[0018] (32) The improved cyclic chaotic map initialization method is used to replace the random initialization method in the original OMA algorithm. The formula is as follows:

[0019]

[0020] Among them, x n represents the nth chaotic sequence number, and mod represents the remainder function

[0021] (33) Calculate the fitness value of the individual according to the objective function, and record the best path solution bestSolution corresponding to the optimal fitness value;

[0022] (34) The firefly perturbation strategy is used to iteratively optimize the path update, reduce the probability of the group falling into the local optimal value, and improve the ability to search the global value:

[0023] For a D-dimensional search space, let the total number of fireflies in the population, that is, the total number of path search ranges, be NP, then the position of the i-th firefly is represented by X i =(x i1 ,x i2 ,…,x iD ); the relative fluorescence brightness of fireflies is expressed as:

[0024]

[0025] Where I0 is the objective function value. When the I0 value is larger, the brightness of the firefly is greater and the fitness value is better. γ is the light absorption coefficient, which changes with the distance and the propagation medium. ij is the distance between each firefly, is the Euclidean distance between individuals i and j, and is determined by the following formula:

[0026]

[0027] The attractiveness of fireflies is expressed as:

[0028]

[0029] Where β0 represents the attraction of the individual light source of the firefly; the update iterative formula of the firefly disturbance position is as follows:

[0030]

[0031] The second term is attraction, and They represent the current position of the path respectively; the third item is randomization, α is the step size influencing factor, ε is a Gaussian random number, and t is the current number of iterations; for the brightest firefly x b , and its disturbance position is:

[0032]

[0033] (35) Determine whether the maximum number of iterations has been reached. If not, the OMA algorithm enters (33); otherwise, terminate the operation and output the final result.

[0034] Furthermore, the implementation process of step (4) is as follows:

[0035] The AGV drives along the best path found by the improved OMA algorithm in step 3 in the cold storage warehouse for agricultural products. It scans nearby obstacles with the laser radar it carries to determine whether there are any steps that require obstacle avoidance. If so, it proceeds to step (5); otherwise, it continues to drive.

[0036] Furthermore, the process of using the artificial potential field method to complete the local path planning of the AGV car and realize the basic obstacle avoidance function in step (5) is as follows:

[0037] In the potential field, the AGV is subject to the gravitational force of the target point; it changes as the distance between the AGV and the target point changes. When the AGV is closer to the target point, the smaller the gravitational force it receives, and when the AGV is farther from the target point, the greater the gravitational force it receives. The gravitational potential field function expression of the artificial potential field method is:

[0038]

[0039] Among them, k attr is the gain coefficient of the gravitational field, which is used to adjust the gravitational force received by the AGV; X r X is the current position coordinate of the AGV; g is the position coordinate of the target point; (X r -X g ) 2 is the square of the Euclidean distance from the current point to the target point;

[0040] The gravitational force received by the AGV is:

[0041] F attr (X r )=grad(U attr (X r ))=-k attr |X r -X g | (12)

[0042] Among them, grad(U attr (X r )) represents the gravitational field U attr In X r The gradient of the position, i.e. the gravitational field U attr Medium X r The direction in which the potential energy changes the most at the position; |X r -X g | is the distance between the AGV car point and the target point;

[0043] From formula (7), we know that the gravitational force on the AGV is linearly related to the distance to the target point. The closer the AGV is to the target point, the smaller the gravitational force.

[0044] The repulsive field function of the AGV car is:

[0045]

[0046] Where: k rep is the coefficient of the repulsive potential field, ρ(X,X0) is the Euclidean distance from the AGV to the obstacle, and the negative gradient of the repulsive potential field function is the repulsive function, expressed as:

[0047]

[0048] Among them, grad(U rep (X r )) represents the repulsive field U rep At position X r The gradient of the repulsive field U repMedium X r The direction where the potential energy changes the most at the position; ρ0 is the factor affecting the range of the obstacle repulsion, which is related to the maximum speed and maximum acceleration of the AGV car, and the relationship is as follows:

[0049] v max 2 =2a max (ρ min +s) (15)

[0050] Among them, v max is the maximum speed of the AGV; a max is the maximum acceleration of the AGV; ρ min is the lowest influencing factor of the obstacle repulsion range; s is the safe distance between the AGV and the obstacle; therefore, when ρ0ρ0 is in the following conditions, the safety of the AGV can be guaranteed:

[0051]

[0052] When the distance between the AGV and the obstacle is greater than ρ0, the AGV is not affected by the repulsive force of the obstacle; when the distance between the AGV and the obstacle is less than ρ0, the repulsive force of the obstacle on the AGV increases as the distance between the two decreases; the AGV is affected by both the gravitational field and the repulsive field in its motion trajectory; in general path planning, there is a target point and one or more obstacles, so the combined potential field of the AGV is:

[0053] U all (X r )=U attr (X r )+∑U rep (X r ) (17)

[0054] Then the resultant force on the AGV in the resultant potential field is:

[0055] F all (X r )=F attr (X r )+∑F rep (X r ) (18)

[0056] The final direction of movement of the AGV is determined by the resultant force and repulsive force it receives. As the position of the AGV changes, the magnitude and direction of the resultant force it receives also continue to change.

[0057] Further, in step (5), when the AGV cart approaches the target point and there are obstacles around the target point, resulting in a state where the target is unreachable, a target distance function is introduced, and the magnitude of the repulsive force potential field is restricted by the distance between the AGV cart and the target point. The implementation process is as follows:

[0058] The improved repulsive force potential field function is:

[0059]

[0060] where k rep is the coefficient of the repulsive force potential field, ρ(X, X0) is the Euclidean distance from the AGV cart to the obstacle; ρ0 is the repulsive range of the obstacle, the Euclidean distance from the AGV cart to the target point, and n is an adjustable coefficient; the corresponding repulsive force function is obtained by taking the partial derivative of the improved repulsive force potential field function:

[0061]

[0062] where F rep1 (X r ) and F rep2 (X r ) are the component forces in two directions of the repulsive force that the AGV cart will receive when it is within the repulsive range of the obstacle; F rep1 (X r ) is the force in the direction of the repulsive force of the obstacle on the AGV cart, that is, the repulsive force on the AGV cart.

[0063] Further, in step (5), when the AGV cart falls into the local optimal solution state, a new virtual end point is created near the cart, and an additional attractive force is applied to the AGV cart to attract it to the position of the virtual end point. The implementation process is as follows:

[0064] First, set a determination value of the local optimal solution as K, and then record the position of the cart at each iteration. For example, the position at the nth iteration is P n , and the position of the robot at the (n + 1)th iteration is P n+1 , where n ≥ 0. Set a detection period as T, and the robot is detected once every T iterations; the detection method is that when |P n+1 - P n | ≥ K, it is determined that the cart has not fallen into the local optimal solution and continues to move forward towards the current target point; conversely, when |P n+1 - P n | < K, it is determined that the robot has fallen into the local optimal solution state; a virtual end point is set to apply an additional gravitational force to the cart to break the current force state and get rid of the local optimal solution.

[0065] The AGV path planning system in a cold storage warehouse for agricultural products described in the present invention comprises a data acquisition and preprocessing module, a data fusion module, an AGV path planning module and a data visualization module;

[0066] The data acquisition and preprocessing module collects map data in the agricultural product cold storage warehouse in real time; and constructs a navigation map through the SLAM algorithm and optimizes the map data through the Gmapping algorithm for preprocessing;

[0067] The data fusion module fuses the deep point cloud three-dimensional data with the two-dimensional laser scanning data, performs local and global data fusion on the collected map data, and uses the fused data to construct a two-dimensional grid map to improve the integrity of the map information;

[0068] The AGV path planning module uses the improved OMA algorithm to perform global path planning for the AGV vehicle; then uses the laser radar carried by the AGV to detect whether there are obstacles in the driving path; then uses the improved artificial potential field method to complete the local path planning of the AGV to achieve the obstacle avoidance function; and realizes the path planning of the AGV;

[0069] The data visualization module displays the delivery status of the goods after the AGV vehicle path planning in real time through cloud platform data visualization.

[0070] Beneficial effects: Compared with the prior art, the present invention has the following beneficial effects:

[0071] 1. The present invention proposes an improved optical microscope (OMA) algorithm, which uses an improved cyclic chaotic mapping initialization to replace its original random initialization. Compared with the random initial point initialization, the point distribution of the cyclic mapping value is more uniform, the population search is more comprehensive, and the problem of being easily trapped in the local extreme value is effectively improved; then the firefly perturbation strategy is used to update and iteratively optimize the optimal solution of the algorithm, thereby improving the global search capability of the OMA algorithm and reducing the probability of the population falling into the local optimal value;

[0072] 2. The improved artificial potential field method proposed in the present invention solves the problem of unreachable targets by introducing a target distance function. The target distance function is introduced into the repulsive potential field function of the traditional artificial potential field method, that is, the size of the repulsive potential field is limited by the distance between the car and the target point, so that the car can truly reach the destination and get rid of the state of unreachable targets. When the problem of falling into a local minimum is encountered, the present invention proposes a virtual end point method, which sets a virtual end point to eventually guide the car to escape from the local extreme point. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a flow chart of the AGV path planning method applied in the cold storage warehouse of agricultural products;

[0074] Figure 2 It is the workflow diagram of Gmapping algorithm;

[0075] Figure 3 It is the improved OMA algorithm flow chart;

[0076] Figure 4 This is the flow chart of the virtual endpoint method for dealing with local optimal problems. DETAILED DESCRIPTION

[0077] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0078] like Figure 1 As shown, the present invention proposes an AGV path planning method in a cold storage warehouse for agricultural products, which specifically includes the following steps:

[0079] Step 1: Start the AGV. The AGV equipped with a depth camera and a laser radar begins to collect map information of the agricultural product cold storage warehouse. The navigation map is constructed through the SLAM algorithm. The DC motor drives the AGV to complete the posture movement, thereby realizing the construction of a two-dimensional grid map. Considering that a single-line laser radar can only observe the point cloud information of a certain plane, this design fuses the three-dimensional data of the depth point cloud with the two-dimensional laser scanning data, and uses the fused data to construct a two-dimensional grid map to improve the integrity of the map information.

[0080] The specific implementation process of the SLAM algorithm to construct a navigation map is as follows: First, the data of the depth camera and the two-dimensional lidar are preprocessed, and the fused data is passed to the front end of the algorithm to represent a complete data frame. At the same time, the local map is constructed through the information of continuous frames, and the local map is optimized using the nonlinear least squares method. In this step, the IMU data obtained through the Kalman filter is used as the initial value of the optimization algorithm to fine-tune the position and posture of the automatic guided vehicle (AGV) in the local map. This helps to build the optimal local map. In large-scale scenes, nonlinear least squares can be established through a closed loop, and the branch and bound method can be used to accelerate the matching efficiency between the current frame and the historical frame. Through this process, we can optimize the global map and improve the overall performance of the system.

[0081] Step 2: Use the Gmapping algorithm to optimize the two-dimensional grid map, such as Figure 2 As shown in the figure, the map data is locally and globally fused. The Gmapping algorithm is optimized based on the RBPF_SLAM algorithm; RBPF_SLAM decomposes the SLAM problem into two problems: positioning and mapping. It first estimates the vehicle path and then estimates the environment map based on the vehicle path.

[0082] The RBPF_SLAM mapping process is mainly divided into four steps: sampling, importance weight calculation, relocalization, and updating landmarks.

[0083] Sampling: Sampling is to use the particle set of the previous moment To estimate the particle set at this moment The generation process formula of the particle set at this moment is as follows:

[0084] P(x k ,m|Z 0:k ,U 0:k ,x0) (1)

[0085] Among them, x k represents the current AGV car posture state, m represents the map, Z 0:k Represents historical environmental observation information, U 0:k Represents the historical motion information of the AGV, and x0 represents the position state of the AGV at time 0.

[0086] Importance weight calculation: In the previous sampling step, the credibility of each particle is different; each estimated trajectory of the vehicle is represented by a particle, and the importance weight needs to be calculated. The importance weight is represented by the difference between the proposed distribution and the target distribution. The importance weight calculation formula is:

[0087]

[0088] in, represents the particle set from time 1 to the previous time, z 1:t Represents the historical environmental observation information from time 1 to the previous time, u 1:t-1 Indicates the historical movement information of the AGV from moment 1 to the previous moment. is the target distribution, is the recommended distribution.

[0089] Resampling: In the initial sampling stage, only the vehicle control information and the vehicle state at the previous moment are considered, and the observation information is ignored, resulting in the obtained particle set only conforming to the proposed distribution but not the target distribution; in order to improve the accuracy of the estimation, the collected particle set needs to be resampled according to the importance weight; particles with high importance weights have higher credibility, while particles with low weights have lower credibility; during the resampling process, particles with low weights are screened out, and particles with high weights are retained and copied to ensure that the total number of particles before and after is consistent; this helps the particle set better reflect the actual system state.

[0090] Update the roadmap: Each particle point is updated according to p(m|x 1:t ,z 1:t ) can draw corresponding maps, and match the maps drawn at historical moments to obtain updated maps.

[0091] Input the lidar data and odometer data into the slam_gmapping module, then call the openslam_gmapping module to observe whether the data matches the existing map successfully. If successful, sample from the improved suggestion distribution and calculate the importance weight. Otherwise, sample from the motion model and calculate the importance weight. Then calculate the example validity N eff , judge N eff Is it less than the particle validity threshold? If so, resample it. Otherwise, update the map and output the map through the slam_gmapping module.

[0092] RBPF_SLAM usage As a proposed distribution, Gmapping introduces the observation data z t and control parameter u t Improve the distribution of suggestions. The distribution of improved suggestions is shown as follows:

[0093]

[0094] In the resampling process of RBPF_SLAM, continuous resampling will lead to the dissipation of particle diversity. In the end, particles may be copied from one or several particles, and the particle diversity is greatly reduced. At the same time, due to the existence of noise, bad particles may be copied repeatedly, good particles may be deleted, and finally only bad particles are left, resulting in reduced particle effectiveness. In the future, to ensure particle diversity and effectiveness, Gmapping adopts an adaptive resampling method; the specific method is to calculate the particle effectiveness N eff , by setting a threshold to determine whether to resample, when the effectiveness is lower than a threshold N th When the validity is higher than the threshold, resampling is performed; when the validity is higher than the threshold, no resampling is performed. The particle validity calculation formula is:

[0095]

[0096] Step 3: If Figure 3 As shown in the figure, the improved optical microscope algorithm is used to optimize the path of the AGV car and search for the optimal path. The specific implementation steps are as follows:

[0097] Step 3.1: Set the objective function of the optical microscope (OMA) algorithm to the shortest time path and initialize the relevant parameters: population size, dimension size, upper and lower bounds of the path search space.

[0098] Step 3.2: Use the improved cyclic chaotic map initialization method to replace the random initialization method in the original OMA algorithm. The formula is as follows:

[0099]

[0100] Among them, x n represents the nth chaotic sequence number, and mod represents the remainder function

[0101] Step 3.3: Calculate the fitness value of the individual according to the objective function, and record the best path solution bestSolution corresponding to the optimal fitness value.

[0102] Step 3.4: Use the firefly perturbation strategy to iteratively optimize the path update, reduce the probability of the group falling into the local optimal value, and improve the ability to search the global value;

[0103] The principle of the firefly perturbation strategy is to regard each point in space as a firefly. The firefly population is randomly distributed, and the best position is found by using the characteristic that high-brightness fireflies attract low-brightness fireflies.

[0104] The relative fluorescence brightness (I) of fireflies is expressed as:

[0105]

[0106] In the formula, I0 is the objective function value. When the I0 value is larger, the brightness of the firefly is greater and the fitness value is better; γ is the light intensity absorption coefficient, which changes with the change of distance and the different propagation media; r ij is the distance between each firefly, is the Euclidean distance between individuals i and j, and is determined by the following formula:

[0107]

[0108] The attractiveness of fireflies (β) is expressed as:

[0109]

[0110] In the formula, β0 represents the attraction of individual fireflies to the light source;

[0111] The update iterative formula for the firefly disturbance position is as follows:

[0112]

[0113] The second term is attraction, and They represent the current position of the path respectively; the third item is randomization, α is the step size influencing factor, ε is a Gaussian random number, and t is the current number of iterations; for the brightest firefly x b , since there are no other brighter fireflies, to follow the second expression in the above equation, this term will be omitted, as shown in the following equation:

[0114]

[0115] Step 3.5: Determine whether the maximum number of iterations has been reached. If not, the OMA algorithm proceeds to step 3.3; otherwise, terminate the operation and output the final result.

[0116] Step 4: The AGV drives along the global optimal path found by the improved OMA algorithm in step 3, and scans nearby obstacles with the binocular camera and lidar to determine whether there are any steps that need to be avoided. If so, it goes to step 5, otherwise it continues to drive.

[0117] Specifically, the binocular camera obtains depth information through the difference between the perspectives of the two cameras. The binocular camera matches the images captured by the left and right cameras to calculate the position relationship of the object in space, and can also work normally under outdoor light exposure. Then it determines whether there is a step that requires obstacle avoidance. If so, the AGV car starts to use the improved artificial potential field method for local path planning. Otherwise, it will continue to drive along the path of step 3 until the goods are delivered to the target point.

[0118] Step 5: Use the improved artificial potential field method to complete the local path planning of the AGV to achieve the obstacle avoidance function; the artificial potential field method is to add a virtual end point method and introduce a target distance function in the algorithm, which greatly improves the performance of the AGV path finding and obstacle avoidance.

[0119] The gravitational potential field function of the artificial potential field method is as follows: in the two-dimensional motion space, the AGV is not only subject to the repulsive force given by the obstacles in the space, but also to the gravitational force from the target point that the AGV wants to reach; in the potential field, the AGV is subject to the gravitational force of the target point; as the distance between the AGV and the target point changes, the gravitational force it receives is smaller when the AGV is closer to the target point, and the gravitational force it receives is larger when the AGV is farther from the target point; the gravitational potential field function expression of the artificial potential field method is:

[0120]

[0121] Among them, k attr is the gain coefficient of the gravitational field, which is used to adjust the gravitational force received by the car; X r is the current position coordinate of the car; X g is the position coordinate of the target point; (X r -X g ) 2 It is the square of the Euclidean distance from the current point to the target point.

[0122] The gravitational force received by the AGV is:

[0123] F attr (X r )=grad(Uattr (X r ))=-k attr |X r -X g | (12)

[0124] Among them, grad(U attr (X r )) represents the gravitational field U attr In X r The gradient of the position, i.e. the gravitational field U attr Medium X r The direction in which the potential energy changes the most at the position; |X r -X g | is the distance between the AGV car point and the target point.

[0125] From formula (7), we can see that the gravitational force on the AGV is linearly related to the distance to the target point. The closer the AGV is to the target point, the smaller the gravitational force is.

[0126] The repulsive field function of the AGV car is:

[0127]

[0128] Among them, k rep is the coefficient of the repulsive potential field, and ρ(X,X0) is the Euclidean distance from the AGV to the obstacle. The negative gradient of the repulsive potential field function is the repulsive function, expressed as:

[0129]

[0130] Among them, grad(U rep (X r )) represents the repulsive field U rep At position X r The gradient of the repulsive field U rep Medium X r The direction where the potential energy changes the most at the position; ρ0 is the factor affecting the range of the obstacle repulsion, which is related to the maximum speed and maximum acceleration of the AGV car, and the relationship is as follows:

[0131] v max 2 =2a max (ρ min +s) (15)

[0132] Among them, v max is the maximum speed of the AGV; a max is the maximum acceleration of the AGV; ρ minis the lowest influencing factor of the obstacle repulsion range; s is the safe distance between the AGV and the obstacle; therefore, when ρ0 is in the following conditions, the safety of the AGV can be guaranteed:

[0133]

[0134] It can be seen from formula (10) that when the distance between the AGV and the obstacle is greater than ρ0, the AGV is not affected by the repulsive force of the obstacle; when the distance between the AGV and the obstacle is less than ρ0, the repulsive force of the obstacle on the AGV increases as the distance between the two decreases; the AGV is affected by both the gravitational field and the repulsive field in its motion trajectory; in general path planning, there is a target point and one or more obstacles, so the combined potential field on which the AGV is subject is:

[0135] U all (X r )=U attr (X r )+∑U rep (X r ) (17)

[0136] Then the resultant force on the AGV in the resultant potential field is:

[0137] F all (X r )=F attr (X r )+∑F rep (X r ) (18)

[0138] According to the principle of artificial potential field method, the final direction of movement of AGV is determined by the resultant force and repulsive force on AGV. As the position of AGV changes, the magnitude and direction of the resultant force on it also change continuously. Therefore, the algorithm has a good real-time obstacle avoidance effect, and the operation is simple and easy to implement, and it is widely used in path planning. The complexity of the environment that actually requires path planning varies. The resultant force may be zero at non-target points. If AGV is trapped in this position, it cannot reach the target point. This is caused by the principle of artificial potential field method, so this defect needs to be improved.

[0139] The situation where the combined force of gravity and repulsion is zero can be divided into two types: unreachable target and local minimum. The following will improve the algorithm for these two problems respectively;

[0140] Improvement method for the target unreachable problem: When the AGV car approaches the target point, if there are obstacles around the target point, the target may be unreachable. To solve this problem, this paper introduces the target distance function. The target distance function is introduced into the repulsive potential field function of the traditional artificial potential field method, that is, the size of the repulsive potential field is limited by the distance between the AGV car and the target point. The closer the AGV car is to the target point, the smaller the target distance function is, and the value decreases exponentially, so the restriction on the repulsive potential field is also greater. When the AGV car approaches the target point, the repulsive force it is subjected to will infinitely approach 0, so that the influence of the repulsive force can be ignored at the most critical point when the AGV car is about to reach the target point, so that the AGV car can truly reach the destination and get rid of the target unreachable state.

[0141] The improved repulsive potential field function is:

[0142]

[0143] Among them, k rep is the coefficient of the repulsive potential field, ρ(X,X0) is the Euclidean distance from the AGV to the obstacle, ρ0 is the repulsive range of the obstacle, the Euclidean distance from the AGV to the target point, and n is an adjustable coefficient. By taking the partial derivative of the improved repulsive potential field function, the corresponding repulsive function can be obtained:

[0144]

[0145] Among them, F rep1 (X r ) and F rep2 (X r ) When the AGV is in the repulsive force range of the obstacle, the repulsive force it receives will be decomposed into two directions of force.

[0146] F rep1 (X r ) is the force in the direction of the obstacle's repulsive force on the AGV, that is, the repulsive force on the AGV. From the formula, it can be seen that the distance from the AGV to the target point exists in the form of an exponent in the formula, which means that the distance from the AGV to the target point can affect the entire F rep1 (X r ) component has a huge impact.

[0147]

[0148] Among them, F rep2 (X r) is the gravitational force in the direction of the gravitational force generated by the target point pair on the AGV vehicle, which generates a gravitational force on the AGV vehicle. Since the distance from the AGV vehicle to the target point also exists in the form of an exponent in this formula, it can also have a great impact on the entire F rep2 (X r ) component force.

[0149]

[0150] In order to enable the target distance function to achieve the best effect, the value of the adjustable parameter n becomes particularly important because different values of the parameter n will result in different effects of the entire target distance function on the repulsive force function. When 0 < n < 1, if the distance between the AGV vehicle and the target point is greater than 1, then as the distance between the AGV vehicle and the target point decreases, the repulsive force function will also become smaller; and if the distance between the AGV vehicle and the target point is less than 1, then as the distance between the AGV vehicle and the target point shrinks, the repulsive force function will become larger. When n > 1, if the distance between the AGV vehicle and the target point is greater than 1, then as the distance between the AGV vehicle and the target point decreases, the repulsive force function will also become smaller; if the distance between the AGV vehicle and the target point is less than 1, then as the distance between the AGV vehicle and the target point shrinks, the repulsive force function will become smaller. It can be seen from formulas (16) and (17) that when the AGV vehicle is close to the target point, the repulsive force it receives will be closer to 0, which limits the repulsive effect of obstacles on the AGV vehicle when the AGV vehicle approaches the target point, enabling the AGV vehicle to enter the target location smoothly without being interfered by the repulsive force, thereby solving the problem of target inaccessibility.

[0151] Improvement method for local minimum: During the operation of the AGV vehicle, it will first be detected whether the AGV vehicle falls into the local optimal solution trap. According to the concept of the local optimal solution, when the gravitational force and the repulsive force received by the AGV vehicle are in opposite directions and equal in magnitude, the AGV vehicle will be in a stationary state, and at this time it will be regarded as falling into the local optimal solution. However, in actual operation, it rarely occurs that the gravitational force and the repulsive force received by the AGV vehicle are exactly in opposite directions and equal in magnitude. Therefore, in more cases, after the robot falls into the local optimal state, it will oscillate repeatedly in a region and cannot escape from this region. When the AGV vehicle falls into the local optimal solution state, a new virtual end point is created near the AGV vehicle in this paper. This virtual end point will additionally apply an attractive force to the robot and attract it towards the position of the virtual end point. This suddenly appearing force will break the existing force balance of the robot and enable the robot to get rid of the local optimal solution state and continue to move towards the real end point.

[0152] Therefore, the following detections are made on whether the AGV vehicle falls into the local optimal state, and the flow chart is as Figure 4Shown as follows: First, set a determination value of the local optimal solution as K, and then record the position of the AGV vehicle at each iteration. For example, the position at the nth iteration is P n , and the position of the robot at the (n + 1)th iteration is P n+1 , where n ≥ 0. Set a detection period as T, and perform a detection on the robot every T iterations. The detection method is that when |P n+1 - P n | ≥ K, it is determined that the AGV vehicle has not fallen into the local optimal solution and can continue to move forward towards the current target point. On the contrary, when |P n+1 - P n | < K, it is determined that the robot has fallen into the local optimal solution state. Because at this time it is oscillating repeatedly within a small range, the path distance between its each iteration will be extremely short. Based on the above determination method of the local optimal solution, a detection will be performed every T time during the operation of the AGV vehicle. If the detection is passed, it proves that it has not fallen into the local optimal solution and can continue to move forward towards the target point. If the detection is not passed, it proves that it has fallen into the local optimal solution. At this time, a virtual end point needs to be set to apply additional gravitational force to the AGV vehicle to break the current force state and get rid of the local optimal solution until it successfully reaches the target point position.

[0153] Based on the same inventive concept, the present invention also proposes an AGV path planning in an agricultural product cold storage warehouse, including a data acquisition and preprocessing module, a data fusion module, an AGV path planning module, and a data visualization module; the data acquisition and preprocessing module performs real-time acquisition of the map data in the agricultural product cold storage warehouse; and preprocesses the map data by constructing a navigation map through the SLAM algorithm and optimizing the map data through the Gmapping algorithm; the data fusion module fuses the depth point cloud three-dimensional data with the two-dimensional laser scanning data, performs local and global data fusion on the collected map data, and constructs a two-dimensional grid map using the fused data to improve the integrity of the map information; the AGV path planning module uses the improved OMA algorithm to perform global path planning for the AGV vehicle; then detects whether there are obstacles in the driving path through the lidar carried by the AGV vehicle; and then adopts the improved artificial potential field method to complete the local path planning of the AGV vehicle to achieve the obstacle avoidance function; realizing the path planning of the AGV vehicle; the data visualization module visually displays the delivery situation of the goods after the AGV vehicle path planning in real time through the cloud platform data.

[0154] The above embodiments are only used to illustrate the technical concept and features of the present invention, and the purpose is to enable those who are familiar with this technology to understand the content of the present invention and implement it accordingly, and cannot be used to limit the protection scope of the present invention. All equivalent transformations or modifications made according to the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. An AGV path planning method in a cold storage warehouse for agricultural products, characterized in that: The following steps are involved: (1) Construct a 2D grid map based on 3D point cloud data and 2D laser scanning data; (2) The Gmapping algorithm is used to optimize the two-dimensional grid map to obtain the environmental map of the agricultural product cold storage warehouse; (3) using the improved OMA algorithm to perform global path planning for the AGV vehicle based on the environment map obtained in step (2); the improved OMA algorithm includes using an improved cyclic chaotic map initialization to replace its original random initialization, and a firefly perturbation strategy to iteratively optimize the optimal solution of the algorithm; (4) The AGV moves forward along the optimal path obtained in step (3), and uses the laser radar on board to detect whether there are obstacles in the driving path. If there are obstacles, it jumps to step (5); otherwise, it jumps to step (6); (5) The artificial potential field method is used to complete the local path planning of the AGV to realize the basic obstacle avoidance function; when the AGV is close to the target point and there are obstacles around the target point, the target is unreachable, and the target distance function is introduced; when the AGV falls into the local optimal solution state, a new virtual end point is created near the car, and an additional attraction is applied to the AGV to attract it to the position of the virtual end point; (6) The AGV successfully delivers the goods to the designated destination and displays them in real time through the data platform; The specific steps of improving the optical microscope algorithm described in step (3) are: (31) Set the objective function of the optical microscope algorithm to the shortest time path and initialize the relevant parameters: population size, dimension size, upper and lower bounds of the path search space; (32) The improved cyclic chaotic map initialization method is used to replace the random initialization method in the original OMA algorithm. The formula is as follows: Among them, x n represents the nth chaotic sequence number, and mod represents the remainder function (33) Calculate the fitness value of the individual according to the objective function, and record the best path solution bestSolution corresponding to the optimal fitness value; (34) The firefly perturbation strategy is used to iteratively optimize the path update, reduce the probability of the group falling into the local optimal value, and improve the ability to search the global value: For a D-dimensional search space, let the total number of fireflies in the population, that is, the total number of path search ranges, be NP, then the position of the i-th firefly is represented by X i =(x i1 ,x i2 ,…,x iD ); the relative fluorescence brightness of fireflies is expressed as: Where I0 is the objective function value. When the I0 value is larger, the brightness of the firefly is greater and the fitness value is better. γ is the light absorption coefficient, which changes with the distance and the propagation medium. ij is the distance between each firefly, is the Euclidean distance between individuals i and j, and is determined by the following formula: The attractiveness of fireflies is expressed as: Where β0 represents the attraction of the individual light source of the firefly; the update iterative formula of the firefly disturbance position is as follows: The second term is attraction, and They represent the current position of the path respectively; the third item is randomization, α is the step size influencing factor, ε is a Gaussian random number, and t is the current number of iterations; for the brightest firefly x b , and its disturbance position is: (35) Determine whether the maximum number of iterations has been reached. If not, the OMA algorithm enters (33); otherwise, terminate the operation and output the final result; The implementation process of step (4) is as follows: The AGV drives along the best path found by the improved OMA algorithm in step 3 in the cold storage warehouse of agricultural products. It scans nearby obstacles with the laser radar and determines whether there are any obstacles to avoid. If so, it goes to step (5). Otherwise, it continues to drive. The process of using the artificial potential field method to complete the local path planning of the AGV car and realize the basic obstacle avoidance function in step (5) is as follows: In the potential field, the AGV is subject to the gravitational force of the target point; it changes as the distance between the AGV and the target point changes. When the AGV is closer to the target point, the smaller the gravitational force it receives, and when the AGV is farther from the target point, the greater the gravitational force it receives. The gravitational potential field function expression of the artificial potential field method is: Among them, k attr is the gain coefficient of the gravitational field, which is used to adjust the gravitational force received by the AGV; X r X is the current position coordinate of the AGV; g is the position coordinate of the target point; (X r -X g ) 2 is the square of the Euclidean distance from the current point to the target point; The gravitational force on the AGV is: F attr (X r )=grad(U attr (X r ))=-k attr |X r -X g | (12) Among them, grad(U attr (X r )) represents the gravitational field U attr In X r The gradient of the position, i.e. the gravitational field U attr Medium X r The direction in which the potential energy changes the most at the position; |X r -X g | is the distance between the AGV car point and the target point; From formula (12), we know that the gravitational force on the AGV is linearly related to the distance to the target point. The closer the AGV is to the target point, the smaller the gravitational force. The repulsive field function of the AGV car is: Among them, k rep is the coefficient of the repulsive potential field, ρ0 is the obstacle repulsive range influence factor, ρ(X,X0) is the Euclidean distance from the AGV to the obstacle, and the negative gradient of the repulsive potential field function is the repulsive function, expressed as: Among them, grad(U rep (X r )) represents the repulsive field U rep At position X r The gradient of the repulsive field U rep Medium X r The direction where the potential energy changes the most at the position; ρ0 is the factor affecting the range of the obstacle repulsion, which is related to the maximum speed and maximum acceleration of the AGV car, and the relationship is as follows: v max 2 =2a max (ρ min +s)(15) Among them, v max is the maximum speed of the AGV; a max is the maximum acceleration of the AGV; ρ min is the lowest influencing factor of the obstacle repulsion range; s is the safe distance between the AGV and the obstacle; therefore, when ρ0 is in the following conditions, the safety of the AGV can be guaranteed: When the distance between the AGV and the obstacle is greater than ρ0, the AGV is not affected by the repulsive force of the obstacle; when the distance between the AGV and the obstacle is less than ρ0, the repulsive force of the obstacle on the AGV increases as the distance between the two decreases; the AGV is affected by both the gravitational field and the repulsive field in its motion trajectory; in general path planning, there is a target point and one or more obstacles, so the combined potential field of the AGV is: U all (X r )=U attr (X r )+∑U rep (X r ) (17) Then the resultant force on the AGV in the resultant potential field is: F all (X r )=F attr (X r )+ΣF rep (X r ) (18) The final movement direction of the AGV is determined by the resultant force and repulsive force on the AGV. As the position of the AGV changes, the magnitude and direction of the resultant force on it also change continuously. In step (5), when the AGV car approaches the target point and there are obstacles around the target point, the target is unreachable. The target distance function is introduced to limit the size of the repulsive potential field by the distance between the AGV car and the target point. The implementation process is as follows: The improved repulsive potential field function is: Among them, k rep is the coefficient of the repulsive potential field, ρ(X,X0) is the Euclidean distance from the AGV to the obstacle; ρ0 is the influence factor of the obstacle repulsive force range, is the Euclidean distance from the AGV to the target point, and n is an adjustable coefficient. The corresponding repulsive force function is obtained by taking the partial derivative of the improved repulsive force potential field function: Among them, F rep1 (X r ) and F rep2 (X r ) is the force that the AGV receives when it is within the repulsive force range of the obstacle, and it will decompose the repulsive force into two directions; F rep1 (X r ) is the force in the direction of the obstacle's repulsive force on the AGV, that is, the repulsive force on the AGV; F rep2 (X r ) is the force of the target point on the AGV in the direction of gravity, that is, the gravity on the AGV; In step (5), when the AGV car falls into the local optimal solution state, a new virtual end point is created near the car, and an additional attraction is applied to the AGV car to attract it to the position of the virtual end point. The implementation process is as follows: First, set a decision value for the local optimal solution as K, and then record the position of the trolley at each iteration. For example, the position at the nth iteration is P n , and the position of the robot at the (n + 1)th iteration is P n+1 , where n ≥ 0. Set a detection period as T, and perform a detection on the robot every T iterations; the detection method is when |P n+1 - P n | ≥ K, it is determined that the trolley has not fallen into the local optimal solution and continues to move forward towards the current target point; otherwise, when |P n+1 - P n | < K, it is determined that the robot has fallen into the local optimal solution state; set a virtual end point to apply additional gravitational force to the trolley to break the current force state and get rid of the local optimal solution.

2. The AGV path planning method in a cold storage warehouse for agricultural products according to claim 1 is characterized in that: The implementation process of step (1) is as follows: The AGV is started in a cold storage warehouse for agricultural products. The AGV is equipped with a depth camera and a lidar to collect map information and build a navigation map through the SLAM algorithm. The DC motor drives the AGV to move the position, fuse the 3D data of the depth point cloud with the 2D laser scanning data, and use the fused data to build a 2D grid map.

3. The AGV path planning method in a cold storage warehouse for agricultural products according to claim 1 is characterized in that: Step (2) The Gmapping algorithm is optimized based on the RBPF_SLAM algorithm; RBPF_SLAM decomposes the SLAM problem into two problems: positioning and mapping. The vehicle path is estimated first, and then the environment map is estimated based on the vehicle path; the RBPF_SLAM mapping process is divided into sampling, importance weight calculation, repositioning, and updating of landmarks; the lidar data and odometer data are input into slam_gmapping, and then openslam_gmapping is called to observe whether the data and the existing map are matched successfully. If successful, sampling is performed from the improved suggestion distribution and the importance weight is calculated. Otherwise, sampling is performed from the motion model and the importance weight is calculated; Then calculate the example validity N eff , judge N eff Is it less than the particle validity threshold? If so, resample it; otherwise, update the map. Finally, the agricultural product cold storage warehouse environment map is output through the slam_gmapping module.

4. An AGV path planning system in a cold storage warehouse for agricultural products using the method as claimed in any one of claims 1 to 3, characterized in that: It includes data acquisition and preprocessing module, data fusion module, AGV path planning module and data visualization module; The data acquisition and preprocessing module collects map data in the agricultural product cold storage warehouse in real time; and constructs a navigation map through the SLAM algorithm and optimizes the map data through the Gmapping algorithm for preprocessing; The data fusion module fuses the deep point cloud three-dimensional data with the two-dimensional laser scanning data, performs local and global data fusion on the collected map data, and uses the fused data to construct a two-dimensional grid map to improve the integrity of the map information; The AGV path planning module uses the improved OMA algorithm to perform global path planning for the AGV vehicle; then uses the laser radar carried by the AGV to detect whether there are obstacles in the driving path; then uses the improved artificial potential field method to complete the local path planning of the AGV to achieve the obstacle avoidance function; and realizes the path planning of the AGV; The data visualization module displays the delivery status of the goods after the AGV vehicle path planning in real time through cloud platform data visualization.

Citation Information

Patent Citations

  • Method for scheduling UAVs based on chaotic adaptive firefly algorithm

    AU2020101065A4

  • Path planning method based on improved A* algorithm

    CN113359768A

  • Robot path planning method in complex environment

    CN113467472A

  • Path planning method based on fusion of improved firefly algorithm and dynamic window method

    CN115129064A

  • Near-earth distribution unmanned aerial vehicle path planning method based on BOA-TSAR algorithm

    CN115562347A