AGV picking path planning method and system

Abstract

The present invention relates to the field of route planning technology, and provides an AGV picking path planning method and system, which abandons the existing AGV working mode of performing sorting tasks for a single type of goods, and instead sorts multiple types of goods in an order in sequence, thereby improving the sorting efficiency of goods and shortening the travel distance of the AGV. Moreover, the parking position of the AGV after completing an order is not a parking area, but the first picking position in the previous order, which can improve the working efficiency of the AGV and save the time cost and power consumption caused by returning to the parking area. In addition, a TSP model is established in combination with the path length, path angle, and number of path inflection points of each alternative path, and the global planning path in each alternative path is determined by solving the TSP model with a genetic algorithm. Compared with the existing A* algorithm, its search efficiency is greatly improved, which can improve the path planning efficiency and improve the sorting efficiency of goods.

Description

Technical Field

[0001] The present invention relates to the technical field of route planning, and in particular to an AGV picking path planning method and system. Background Art

[0002] As one of the main equipment of the intelligent handling system in smart warehousing, Automated Guided Vehicle (AGV) plays an important role in warehouse sorting activities. The layout of existing smart warehouses is generally divided into cargo storage area, sorting area, control center, parking area and driving area according to their uses, such as Figure 1 shown. Figure 1 The blank areas where the middle dotted lines are located are all driving areas.

[0003] The working mode of AGV is usually to complete the picking of a single type of goods, starting from the parking area to the goods storage area to pick the goods, and then sending the goods to the picking table in the sorting area, and then sending the shelf back to the storage location in the goods storage area, and then the AGV returns to the parking area. In this working mode, the A* algorithm is usually used for AGV path planning.

[0004] In the AGV working mode adopted in the prior art, the AGV can only complete one type of cargo sorting in one round trip, and the sorting efficiency is not high. Moreover, due to the low search efficiency of the A* algorithm, the path planning efficiency is low, which further affects the cargo sorting efficiency. Summary of the invention

[0005] The present invention provides an AGV picking path planning method and system to solve the defects in the prior art.

[0006] The present invention provides an AGV picking path planning method, comprising:

[0007] Obtaining a current order and determining a target AGV corresponding to the current order, wherein the current order includes multiple goods;

[0008] Each alternative path is constructed with the parking position of the target AGV as the starting node, the storage positions of various goods as the intermediate nodes, and the target position of the current order as the end node; the parking position includes the first picking position in the previous order of the current order;

[0009] Based on the path length, path turning angle and the number of path inflection points of each candidate path, a TSP model is established, and a genetic algorithm is used to solve the TSP model to determine the global planning path among the candidate paths.

[0010] According to an AGV picking path planning method provided by the present invention, the genetic algorithm is used to solve the TSP model to determine the global planning path among the candidate paths, specifically including:

[0011] Taking each candidate path as an individual, taking the intermediate node of each candidate path as a gene, and determining the individual fitness based on the path length, path turning angle and the number of path inflection points of each candidate path;

[0012] Two Flink clusters are used to initialize the first and second populations respectively, and individuals of the first population are iteratively updated to obtain the third population, and individuals of the second population are iteratively updated to obtain the fourth population;

[0013] A first individual corresponding to the best individual fitness is determined from the third population and the fourth population, and the first individual is determined as the global planning path.

[0014] According to an AGV picking path planning method provided by the present invention, two Flink clusters are used to initialize a first population and a second population respectively, specifically including:

[0015] The first Flink cluster is used to initialize the first candidate population, and the second Flink cluster is used to initialize the second candidate population.

[0016] Determining the first population based on the individual fitness variance of the first candidate population;

[0017] The second population is determined based on the average similarity between each individual of the second candidate population and the second individual corresponding to the best individual fitness.

[0018] According to an AGV picking path planning method provided by the present invention, the iterative updating of the individuals of the first population to obtain the third population specifically includes:

[0019] Using the master node of the first Flink cluster, the first population is divided into multiple first sub-populations, and using the task nodes of the first Flink cluster to perform the first type of gene operation on each first sub-population in a distributed and parallel manner to obtain an updated first sub-population;

[0020] Using the master node, merging all updated first sub-populations to obtain a first new population;

[0021] The first Flink cluster is used to iteratively divide, perform first-type gene operations, and merge the first new population to obtain the third population.

[0022] According to an AGV picking path planning method provided by the present invention, the distributed parallel use of task nodes in the first Flink cluster to perform the first type of gene operation on each first sub-population specifically includes:

[0023] For any task node in the first Flink cluster, using the any task node, based on the individual fitness of any individual in the first subpopulation, determining an activation function corresponding to the any individual;

[0024] Determine a target operation probability of the first type of genetic operation corresponding to any individual based on the activation function of any individual and the operation probability range of the first type of genetic operation;

[0025] Based on the target operation probability, the first type of genetic operation is performed on any one of the individuals.

[0026] According to an AGV picking path planning method provided by the present invention, the TSP model is solved by using a genetic algorithm to determine a global planning path among the candidate paths, and then the method further includes:

[0027] Based on the dynamic window algorithm, local path planning is performed on the global planned path to obtain the local planned path of the target AGV.

[0028] According to an AGV picking path planning method provided by the present invention, the dynamic window evaluation function of the dynamic window algorithm is determined based on the evaluation function of the current simulated speed information of the target AGV, the closest distance between the simulated trajectory of the target AGV and the static obstacles around the target AGV, and the azimuth deviation between the end direction of the simulated trajectory and the target point on the global planning path, and the target point is the intersection of a circle with the current position of the target AGV as the center and the target distance as the radius and the global planning path.

[0029] According to an AGV picking path planning method provided by the present invention, the local path planning is performed on the global planning path based on the dynamic window algorithm to obtain the local planning path of the target AGV, specifically including:

[0030] During the local path planning process, if there are multiple first-class AGVs within the first preset range of the current position of the target AGV, the multiple first-class AGVs are controlled to stop running, and the global planned path of each first-class AGV is re-planned locally in descending order of the serial numbers of the multiple first-class AGVs, so that each first-class AGV can leave the first preset range.

[0031] According to an AGV picking path planning method provided by the present invention, the local path planning is performed on the global planning path based on the dynamic window algorithm to obtain the local planning path of the target AGV, specifically including:

[0032] During the local path planning process, if there are multiple second-category AGVs within a second preset range of the current position of the target AGV, the multiple second-category AGVs are controlled to decelerate at a first acceleration;

[0033] If there are multiple third-category AGVs within the third preset range of the current position of the target AGV, control the multiple third-category AGVs to decelerate at the second acceleration;

[0034] Among them, the distance between any point in the second preset range and the target AGV is greater than the distance between any point in the first preset range and the target AGV, the distance between any point in the third preset range and the target AGV is greater than the distance between any point in the second preset range and the target AGV, and the first acceleration is greater than the second acceleration.

[0035] The present invention also provides an AGV picking path planning system, comprising:

[0036] An acquisition module, used to acquire a current order and determine a target AGV corresponding to the current order, wherein the current order includes multiple goods;

[0037] An alternative path construction module is used to construct various alternative paths with the parking position of the target AGV as the starting node, the storage positions of various goods as the intermediate nodes, and the target position of the current order as the end node; the parking position includes the first picking position in the previous order of the current order;

[0038] The global path planning module is used to establish a TSP model based on the path length, path turning angle and number of path inflection points of each candidate path, and adopt a genetic algorithm to solve the TSP model to determine the global planning path among the candidate paths.

[0039] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an AGV picking path planning method as described in any one of the above is implemented.

[0040] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the AGV picking path planning method as described in any one of the above is implemented.

[0041] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-described AGV picking path planning methods.

[0042] The AGV picking path planning method and system provided by the present invention abandons the existing AGV working mode of performing sorting tasks for a single type of goods, and instead sorts multiple types of goods in an order in sequence, thereby improving the sorting efficiency of goods and shortening the travel distance of the AGV. Moreover, the parking position of the AGV after completing an order is not the parking area, but the first picking position in the previous order, which can improve the working efficiency of the AGV and save the time cost and power consumption caused by returning to the parking area. In addition, a TSP model is established in combination with the path length, path turning angle and number of path inflection points of each alternative path, and the global planning path in each alternative path is determined by solving the TSP model with a genetic algorithm. Compared with the existing A* algorithm, its search efficiency is greatly improved, which can improve the path planning efficiency and further improve the sorting efficiency of goods. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings in the following description without creative work.

[0044] Figure 1 It is a schematic diagram of the layout of a smart warehouse in the prior art;

[0045] Figure 2 It is a flow chart of the AGV picking path planning method provided by the present invention;

[0046] Figure 3 It is a schematic diagram of the reverse order of picking order and shelf return in the AGV picking path planning method provided by the present invention;

[0047] Figure 4 It is a flow chart of using two Flink clusters to execute a genetic algorithm to solve a global planning path in the AGV picking path planning method provided by the present invention;

[0048] Figure 5 It is a schematic diagram of a process in which the first Flink cluster iteratively updates the individuals of the first population to obtain the third population in the AGV picking path planning method provided by the present invention;

[0049] Figure 6 It is a schematic diagram of a curve form when the activation function used in the AGV picking path planning method provided by the present invention is a Tanh function;

[0050] Figure 7 It is a schematic diagram of the relationship between the global planning path and the target point in the AGV picking path planning method provided by the present invention;

[0051] Figure 8 It is a schematic diagram of the step-by-step emergency control strategy in the AGV picking path planning method provided by the present invention;

[0052] Fig. 9 It is a structural schematic diagram of the AGV picking path planning device provided by the present invention;

[0053] Fig.10 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Due to the AGV working mode adopted in the prior art, the AGV can only complete one type of cargo sorting in one round trip, and the sorting efficiency is not high. Moreover, the AGV returns to the parking area after completing the sorting task, and the consumption of time and electricity on this journey is unnecessary. In addition, due to the low search efficiency of the A* algorithm, the path planning efficiency is low, which further affects the cargo sorting efficiency. For this reason, an AGV picking path planning method is provided in an embodiment of the present invention.

[0056] Figure 2 FIG. 1 is a flow chart of an AGV picking path planning method provided in an embodiment of the present invention, such as Figure 2 As shown, the method includes:

[0057] S1, obtaining a current order and determining a target AGV corresponding to the current order, wherein the current order includes multiple goods;

[0058] S2, constructing various alternative paths with the parking position of the target AGV as the starting node, the storage positions of various goods as the intermediate nodes, and the target position of the current order as the end node; the parking position includes the first picking position in the previous order of the current order;

[0059] S3, based on the path length, path turning angle and number of path inflection points of each candidate path, a TSP model is established, and a genetic algorithm is used to solve the TSP model to determine a global planning path among the candidate paths.

[0060] Specifically, the AGV picking path planning method provided in the embodiment of the present invention is executed by an AGV picking path planning device, which can be configured in a control center, which can be a server, and which can be a local server, such as a computer, etc., which is not specifically limited in the embodiment of the present invention.

[0061] First, execute step S1 to obtain the current order and determine a target AGV corresponding to the current order. The current order contains multiple goods, and the multiple goods can be stored in different storage locations in the goods storage area. The target AGV refers to an AGV that needs to perform the sorting task of the goods in the current order. In other words, the working mode of the target AGV is to perform sorting tasks according to orders, that is, to complete one order first, and then execute the next order. The current order can be the first order that needs to be executed after the target AGV is turned on, or it can be the nth (n≥2) order that needs to be executed after the target AGV is turned on, and there is no specific limitation here.

[0062] Then execute step S2, construct each alternative path with the parking position of the target AGV as the starting node, the storage position of various goods as the intermediate node, and the target position of the current order as the end node. The parking position of the target AGV, that is, the departure position before the target AGV starts to pick the goods in the current order, may include the first picking position in the previous order of the current order. The previous order refers to the most recent order that the target AGV has completed before executing the sorting task of the current order, and the first picking position in the previous order refers to the storage position of the goods that the target AGV first picked when executing the previous order.

[0063] It can be seen that the working mode of the target AGV is to complete the picking of multiple goods in the order in the goods storage area in succession when executing the sorting task of each order and transport them to the target location of the order, that is, the picking platform in the sorting area. After that, the shelves of goods need to be sent back to the storage location of the goods in the goods storage area in the reverse order of the picking order. After that, the target AGV stops at the first picking location, that is, the target AGV directly stops at the storage location of the last shelf after returning to the last shelf, instead of returning directly to the parking area. This utilizes the principle of access time locality. Because the goods have just been picked, it is very likely to be picked again in the next order. The target AGV does not need to drive from the parking area to the shelf when it stays there.

[0064] It is understandable that each cargo corresponds to a shelf, which is used to carry the cargo and ensure smooth shipment of the cargo. In the embodiment of the present invention, the size of the cargo and the size of the shelf are not considered.

[0065] Therefore, the storage location of the goods is consistent with the storage location of the shelf for the goods. The reverse order of the picking sequence refers to the reverse order of the picking sequence, e.g. Figure 3 As shown in the figure, the current order contains n kinds of goods, and the picking order is the first kind of goods T1, the second kind of goods T2, ..., the nth kind of goods T n , then the reverse order is the nth type of goods T n ,…,the second type of goods T2, the first type of goods T1.

[0066] Each alternative path is a path that the target AGV may take when executing the current order. The starting node of each alternative path is the parking position of the target AGV, and the end node is the target position of the current order, that is, the picking station in the sorting area. The difference lies in the order of the intermediate nodes, that is, the order in which the target AGV arrives at the storage location of different types of goods is different.

[0067] Finally, step S3 is performed to establish a Traveling Salesman Problem (TSP) model based on the path length, path angle, and number of path inflection points of each alternative path. The path length of each alternative path is the sum of the distances between the start node, the intermediate node, and the end node connected in sequence. The path angle can be used to characterize the smoothness of the alternative path, which refers to the sum of the angles of all nodes in the alternative path. The angle of each node is the angle between the first line and the second line of the node, the first line is the line between the node and the next node, the second line is the line between the node and the previous node, and the angle of each node is within the range of 0°-90°. The path inflection point is the node whose angle is not 0.

[0068] When establishing the TSP model, the objective function can be determined based on the path length, path turning angle, and number of path inflection points of each candidate path. The objective function can be the weighted sum of the path length, path turning angle, and number of path inflection points in each candidate path, that is:

[0069] f(T1,T2……T n )=α·l(T1,T2……T n )+β·θ(T1,T2……T n )+γ·T(T1,T2……T n )

[0070] Among them, f(T1,T2...T n ) represents the objective function, T1, T2...T nRepresents the nodes on each alternative path, there are n nodes in total, l(T1, T2...T n ) represents the length of each alternative path, θ(T1,T2…T n ) represents the path turning angle of each alternative path, α, β, and γ represent weight coefficients, and are all constants.

[0071] Then the TSP model is constructed by minimizing the objective function. The TSP model can be expressed as:

[0072] min f(T1,T2……T n )

[0073] The constraints can be expressed as:

[0074]

[0075] in, Indicates that each alternative path does not pass through obstacles, and (X, Y) represents the location set of the drivable area of ​​the target AGV.

[0076] In the embodiment of the present invention, a genetic algorithm can be used to solve the TSP model, and then the global planning path in each candidate path can be determined. When the genetic algorithm is used to solve the TSP model, each candidate path can be used as an individual, the intermediate nodes of each candidate path can be used as genes, and the objective function of the TSP model can be used as the calculation function of the individual fitness, and the conventional solution process of the genetic algorithm can be used for implementation, which is not specifically limited here.

[0077] The AGV picking path planning method provided in the embodiment of the present invention first obtains the current order and determines a target AGV corresponding to the current order, and the current order contains multiple goods; then, the parking position of the target AGV is used as the starting node, the storage position of various goods is used as the intermediate node, and the target position of the current order is used as the end node to construct each alternative path; the parking position includes the first picking position in the previous order of the current order; finally, based on the path length, path turning angle and number of path turning points of each alternative path, a TSP model is established, and a genetic algorithm is used to solve the TSP model to determine the global planning path in each alternative path. This method abandons the existing AGV working mode of performing sorting tasks for a single type of goods, but sorts multiple goods in an order in sequence, which can improve the sorting efficiency of goods and shorten the driving distance of the AGV. Moreover, the parking position of the AGV after completing an order is not the parking area, but the first picking position in the previous order, which can improve the working efficiency of the AGV and save the time cost and power consumption caused by returning to the parking area. In addition, a TSP model is established by combining the path length, path turning angle and number of path inflection points of each alternative path, and the global planning path among the alternative paths is determined by solving the TSP model with a genetic algorithm. Compared with the existing A* algorithm, its search efficiency is greatly improved, which can improve the path planning efficiency and further improve the sorting efficiency of goods.

[0078] On the basis of the above-mentioned embodiment, the AGV picking path planning method provided in the embodiment of the present invention adopts a genetic algorithm to solve the TSP model and determine the global planning path among the candidate paths, specifically including:

[0079] Taking each candidate path as an individual, taking the intermediate node of each candidate path as a gene, and determining the individual fitness based on the path length, path turning angle and the number of path inflection points of each candidate path;

[0080] Two Flink clusters are used to initialize the first and second populations respectively, and individuals of the first population are iteratively updated to obtain the third population, and individuals of the second population are iteratively updated to obtain the fourth population;

[0081] A first individual corresponding to the best individual fitness is determined from the third population and the fourth population, and the first individual is determined as the global planning path.

[0082] Specifically, in an embodiment of the present invention, when a genetic algorithm is used to solve the TSP model and determine the global planning path among each alternative path, each alternative path can be regarded as an individual, the intermediate node of each alternative path can be regarded as a gene, and the individual fitness of each individual can be determined by using an objective function determined based on the path length, path turning angle and number of path inflection points of each alternative path.

[0083] Afterwards, in view of the fact that the traditional genetic algorithm is prone to fall into the local optimal solution, in the embodiment of the present invention, Figure 4 As shown, when the AGV picking path planning device receives the current order, it starts the AGV path planning task and sends the task message to the message queue. The first Flink cluster and the second Flink cluster trigger the implementation of the genetic algorithm by monitoring the message queue. The first population is initialized by the first Flink cluster, and the individuals of the first population are iteratively updated to obtain the third population. The second population is initialized by the second Flink cluster, and the individuals of the second population are iteratively updated to obtain the fourth population.

[0084] The first population can be obtained by randomly initializing the first Flink cluster, and can be iteratively updated by using the first type of genetic operation through the first Flink cluster until a preset number of iterations is reached to obtain the third population. The first type of genetic operation can include at least one of selection, crossover, and mutation, and the probability of the operation used can be given in advance or can be adaptively adjusted, which is not specifically limited here. The preset number of iterations can be set as needed, for example, it can be set to 300 times, 500 times, 1000 times, etc.

[0085] Similarly, the second population can be obtained by randomly initializing the second Flink cluster, and can be iteratively updated by the second Flink cluster using the second type of genetic operation to reach a preset number of iterations to obtain the fourth population. The second type of genetic operation can also include at least one of selection, crossover and mutation, and in addition, can also include at least one of deletion, insertion and reversal. Deletion can be achieved by a deletion operator to delete individuals that do not meet the requirements, such as individuals corresponding to alternative paths blocked by obstacles; insertion can be achieved by an insertion operator to insert new genes into the genes of an individual; reversal can be achieved by a reversal operator to reverse the genes of an individual.

[0086] After that, the first individual corresponding to the best individual fitness can be determined from the third population and the fourth population. The first individual is the best individual in the fifth population. At this time, the first individual can be directly determined as the global planning path, that is, the global path planning is realized.

[0087] In the embodiment of the present invention, two Flink clusters are used to solve the TSP model through a genetic algorithm, which solves the problem that the traditional genetic algorithm is prone to fall into a local optimal solution, and makes the obtained global planning path more accurate.

[0088] On the basis of the above-mentioned embodiment, the AGV picking path planning method provided in the embodiment of the present invention adopts two Flink clusters, and respectively initializes the first population and the second population, specifically including:

[0089] The first Flink cluster is used to initialize the first candidate population, and the second Flink cluster is used to initialize the second candidate population.

[0090] Determining the first population based on the individual fitness variance of the first candidate population;

[0091] The second population is determined based on the average similarity between each individual of the second candidate population and the second individual corresponding to the best individual fitness.

[0092] Specifically, in the embodiment of the present invention, when two Flink clusters are used to initialize the first population and the second population, the first Flink cluster can be used to initialize the first candidate population, and the second Flink cluster can be used to initialize the second candidate population. Both the first candidate population and the second candidate population can be randomly initialized.

[0093] After that, the first candidate population can be determined based on the individual fitness variance of the first candidate population. Assume that the size of the first candidate population is n, that is, there are n individuals, f i (i=1,2,…n) represents the individual fitness of each individual in the first candidate population, f avg Represents the average fitness of all individuals in the first candidate population, then the individual fitness variance of the first candidate population can be expressed as:

[0094]

[0095] Combined with the individual fitness variance, the first population can be determined. For example, if S 2 >σ1, it means that the first candidate population meets the population diversity requirement, and the first candidate population can be used as the first population; otherwise, it means that the first candidate population does not meet the population diversity requirement, so several individuals can be randomly selected from the first candidate population and their chromosomes can be regenerated until S is met. 2 >σ1, then the first candidate population is the first population. Wherein, σ1 is the first threshold constant, which can be set as needed.

[0096] For the second population, it can be determined based on the mean similarity between each individual in the second candidate population and the second individual corresponding to the best individual fitness. Assume that the size of the second candidate population is n, that is, there are n individuals, C i (i=1,2,…n) represents the i-th individual in the second candidate population, C i =(S i1 ,S i2 ,…S im ) represents the chromosome set of the i-th individual. o =(S o1 ,S o2 ,…S om ) represents the chromosome set of the second individual corresponding to the best individual fitness in the second alternative population.

[0097] Each individual C in the second alternative population i With the second individual C o The similarity between i The Euclidean distance can be used for calculation. The final similarity mean of the second candidate population can be expressed as When sim avg When σ > 2, it means that the second candidate population meets the population diversity requirement, and the second candidate population can be determined as the second population. Otherwise, it means that the second candidate population does not meet the population diversity requirement. At this time, several individuals are randomly selected from the second candidate population and their chromosomes are regenerated until sim avg >σ2, then the second candidate population is the second population. Wherein, σ2 is the second threshold constant, which can be set as needed.

[0098] In the embodiment of the present invention, a first Flink cluster and a second Flink cluster are used. In the process of initializing the first population and the second population, the individual fitness variance and the mean similarity between each individual and the second individual corresponding to the best individual fitness are used as criteria for determining the first population and the second population, respectively, to ensure that the first population and the second population meet the individual diversity requirements, further optimize the genetic algorithm, and make the results obtained by the genetic algorithm more accurate.

[0099] On the basis of the above-mentioned embodiment, the AGV picking path planning method provided in the embodiment of the present invention, wherein the iterative updating of the individuals of the first population to obtain the third population specifically includes:

[0100] Using the master node of the first Flink cluster, the first population is divided into multiple first sub-populations, and using the task nodes of the first Flink cluster to perform the first type of gene operation on each first sub-population in a distributed and parallel manner to obtain an updated first sub-population;

[0101] Using the master node, merging all updated first sub-populations to obtain a first new population;

[0102] The first Flink cluster is used to iteratively divide, perform first-type gene operations, and merge the first new population to obtain the third population.

[0103] Specifically, in an embodiment of the present invention, the first Flink cluster and the second Flink cluster may each include a master node (ResourceManager) and multiple task nodes (TaskManager). For example, the first Flink cluster and the second Flink cluster may each include 10 task nodes, which can support 100 to 200 AGVs for picking path planning of 10 types of goods per order in a short time. The specific number of AGVs supported and the types of goods are related to the size of the warehouse space and the number and computing performance of the task nodes in the first Flink cluster and the second Flink cluster.

[0104] The first Flink cluster can iteratively update the individuals of the first population to obtain the third population. In this process, Figure 5 As shown, the master node of the first Flink cluster can be used to divide the first population into k first sub-populations, and the k first sub-populations are sent to the message queue so that each task node that monitors the message queue obtains a first sub-population. Then, the task node of the first Flink cluster is used to perform the first type of gene operation on each first sub-population in a distributed and parallel manner to update the first sub-population, obtain the updated first sub-population, calculate the individual fitness of the updated first sub-population, and return the updated first sub-population and its individual fitness to the message queue. The master node merges all the updated first sub-populations in the message queue to obtain the first new population, and iterates the above process under the condition that the number of iterations is not reached. Finally, under the condition that the number of iterations is reached, the first new population obtained is the third population.

[0105] After the master node merges all the updated first sub-populations in the message queue to obtain the first new population, it can adopt one of the following two strategies a or b, or select a or b by setting the probability for the next iteration to exchange information between populations, thereby increasing the diversity of the population and enhancing the global optimization ability of the algorithm.

[0106] a. The k updated first sub-populations are randomly shuffled to obtain the first new population, and then k new first sub-populations are selected from the first new population for the next iteration.

[0107] b. Divide the k updated first subpopulations into two parts, select the best half of the individuals from one part of the updated first subpopulation, and then select the worst half of the individuals from the other part of the updated first subpopulation to form the first new population. After that, select k new first subpopulations from the first new population for the next iteration.

[0108] Similarly, the processing method of the second Flink cluster is the same as that of the first Flink cluster, which is not repeated here.

[0109] In the embodiment of the present invention, the first Flink cluster is used to divide the first population into multiple first sub-populations for parallel processing, which can speed up the genetic algorithm to solve the global planning path.

[0110] On the basis of the above embodiment, the AGV picking path planning method provided in the embodiment of the present invention, wherein the distributed parallel use of the task nodes in the first Flink cluster to perform the first type of gene operation on each first sub-population specifically includes:

[0111] For any task node in the first Flink cluster, using the any task node, based on the individual fitness of any individual in the first subpopulation, determining an activation function corresponding to the any individual;

[0112] Determine a target operation probability of the first type of genetic operation corresponding to any individual based on the activation function of any individual and the operation probability range of the first type of genetic operation;

[0113] Based on the target operation probability, the first type of genetic operation is performed on any one of the individuals.

[0114] Specifically, in the embodiment of the present invention, when the task nodes in the first Flink cluster are used to perform the first type of gene operation on each first sub-population in a distributed and parallel manner, the target operation probability used by the first type of gene operation can be adaptively determined by the following method, that is, for any task node in the first Flink cluster, the activation function corresponding to any individual in the first sub-population can be determined by using the task node based on the individual fitness of any individual in the first sub-population. The activation function can be in the form of a Sigmoid function, a Logistic function, a Hyperbolic tangent (Tanh) function, or a Rectified linear units (ReLu) activation function, etc.

[0115] Taking the activation function as Tanh function as an example, its curve form is as follows Figure 6 As shown. Then the activation function corresponding to any individual can be expressed as:

[0116]

[0117] Among them, f i represents the individual fitness of the ith individual, f iavg Represents the average fitness of the first sub-population.

[0118] Afterwards, the target operation probability of the first type of genetic operation corresponding to any individual is determined according to the activation function of any individual and the operation probability range of the first type of genetic operation. The first type of genetic operation may include selection, crossover and mutation, and each operation corresponds to an operation probability range, which can be given in advance. For example, [P s_min ,P s_max ] represents the selection probability range, P s_min , P s_max are the minimum and maximum values ​​of the selection probability, respectively. c_min ,P c_max ] represents the crossover probability range, P c_min , P c_max are the minimum and maximum values ​​of the crossover probability, respectively. m_min ,P m_max ] represents the mutation probability range, P m_min ,P m_max are the minimum and maximum values ​​of the mutation probability respectively.

[0119] The target operation probability of the first type of genetic operation corresponding to any individual can be expressed as:

[0120]

[0121] Among them, P s is the selection probability of the ith individual, P c is the crossover probability of the ith individual, P m is the mutation probability of the ith individual.

[0122] In the embodiment of the present invention, an activation function is introduced to realize adaptive determination of the target operation probability, which can enable the generation of new individuals while ensuring that population evolution will not damage individuals with high fitness, thereby retaining elite individuals, further realizing optimization of the genetic algorithm, and ensuring the accuracy of the genetic algorithm solution.

[0123] On the basis of the above embodiments, the AGV picking path planning method provided in the embodiments of the present invention adopts a genetic algorithm to solve the TSP model to determine the global planning path among the candidate paths, and then further includes:

[0124] Based on the dynamic window algorithm, local path planning is performed on the global planned path to obtain the local planned path of the target AGV.

[0125] Specifically, in the embodiment of the present invention, since the global planning path obtained by the genetic algorithm is static, when multiple AGVs are traveling, the target AGV is likely to encounter other moving AGVs, and dynamic path planning is required at this time. In the embodiment of the present invention, the dynamic window approach (DWA) is used for local path planning to obtain the local planning path of the target AGV.

[0126] DWA performs spatial sampling of the speed of the target AGV in the window area and simulates the motion trajectory of the target AGV. The motion state of the target AGV is determined by the linear velocity v of the target AGV at time t. t and angular velocity w t To feedback, (v t ,w t ) represents the trajectory of the target AGV. Since there are several groups of sampling speeds in the speed space that are feasible, the best trajectory can be selected as the local planning path among all feasible trajectories through the evaluation function. In the time interval Δt, assuming that the target AGV moves in a uniform straight line, the kinematic model is:

[0127]

[0128] Among them, (x, y) is the coordinate of the target AGV, v x v t The component in the x direction, v y v t The component in the y direction, θ t is the azimuth.

[0129] In the embodiment of the present invention, the evaluation function is used for local path planning. The target AGV should avoid obstacles as much as possible and reach the end point of the local planned path in the shortest time. The existing evaluation function can be used for implementation and is not specifically limited here.

[0130] In the embodiment of the present invention, DWA is used to perform dynamic local path planning based on the global planned path, so as to avoid collision between the target AGV and other moving AGVs.

[0131] On the basis of the above embodiments, in the AGV picking path planning method provided in the embodiments of the present invention, the dynamic window evaluation function of the dynamic window algorithm is determined based on the evaluation function of the current simulated speed information of the target AGV, the closest distance between the simulated trajectory of the target AGV and the static obstacles around the target AGV, and the azimuth deviation between the end direction of the simulated trajectory and the target point on the global planning path, and the target point is the intersection of a circle with the current position of the target AGV as the center and the target distance as the radius and the global planning path.

[0132] Specifically, in the embodiment of the present invention, the dynamic window evaluation function of the dynamic window algorithm used can be expressed as:

[0133] G(v,w)=σ(α·goal(v,w,G i )+β·obstdist(v,w,O)+γ·value(v,w))

[0134] Where obstdist(v,w,O) is the shortest distance between the simulated trajectory of the target AGV and the static obstacle O around the target AGV, value(v,w) is the evaluation function of the current simulated speed information; σ is the smoothing coefficient; α, β, and γ are weighting coefficients. goal(v,w,G i ) is the end point direction of the simulated trajectory of the target AGV and the target point G on the global planning path i The azimuth deviation between the target point G i It is a point on the global planning path that is at a target distance (R) from the current position of the target AGV in the forward direction of the global planning path. Figure 7 As shown, the global planning path is SD, and the target point is the end point of the arrow. It can be understood that the target distance is the local path planning radius, which can be adaptively adjusted according to the number of static obstacles around the target AGV. The static obstacles around the target AGV can be detected by sensors installed on the target AGV.

[0135] Here, the end point of the simulated trajectory of the target AGV can be randomly selected from the global planning path, so that the target AGV can maintain the dynamic obstacle avoidance function of the local path during driving, and ensure that the actual path of the target AGV does not deviate too far from the global planning path.

[0136] In the embodiment of the present invention, the determined evaluation function can prevent the dynamic window algorithm from falling into the local optimum, so that the local path planning follows the sequence points of the global planning path, thereby ensuring that the global planning path is optimal.

[0137] On the basis of the above-mentioned embodiment, the AGV picking path planning method provided in the embodiment of the present invention, based on the dynamic window algorithm, performs local path planning on the global planning path to obtain the local planning path of the target AGV, specifically comprising:

[0138] During the local path planning process, if there are multiple first-class AGVs within the first preset range of the current position of the target AGV, the multiple first-class AGVs are controlled to stop running, and the global planned path of each first-class AGV is re-planned locally in descending order of the serial numbers of the multiple first-class AGVs, so that each first-class AGV can leave the first preset range.

[0139] Specifically, in the embodiment of the present invention, since the dynamic window evaluation function of DWA is composed of three parts, the weighted coefficients of α, β, and γ are difficult to determine. In some cases, the selected simulation trajectory ignores the importance of β obstacle avoidance, and the local planning path finally obtained may cause dynamic collision events. Therefore, in the embodiment of the present invention, when multiple first-class AGVs enter the first preset range of the current position of the target AGV, each first-class AGV is controlled to stop running and wait for operation. Then, the global planning path of the first-class AGV with the largest serial number in the first preset range can be re-planned locally, so that the first-class AGV with the largest serial number executes the new local planning path to pick up goods and leave the first preset range. Thereafter, the global planning path of the first-class AGV with the new largest serial number in the first preset range continues to be re-planned locally until there is no first-class AGV in the first preset range.

[0140] It is understandable that the first preset range can be understood as the collision critical area of ​​the target AGV, and its range can be set as needed, for example, it can be an area within a preset radius R1 around the current position of the target AGV. The serial numbers of the first type of AGVs can be pre-randomly sorted.

[0141] In an embodiment of the present invention, by controlling multiple first-class AGVs within a first preset range of the current position of the target AGV to stop running and leave the first preset range in order from large to small serial numbers, not only can the collision between the first-class AGV and the target AGV be avoided, but also the collision can be made impossible by destroying the deadlock.

[0142] On the basis of the above-mentioned embodiment, the AGV picking path planning method provided in the embodiment of the present invention, based on the dynamic window algorithm, performs local path planning on the global planning path to obtain the local planning path of the target AGV, specifically comprising:

[0143] During the local path planning process, if there are multiple second-category AGVs within a second preset range of the current position of the target AGV, the multiple second-category AGVs are controlled to decelerate at a first acceleration;

[0144] If there are multiple third-category AGVs within the third preset range of the current position of the target AGV, control the multiple third-category AGVs to decelerate at the second acceleration;

[0145] Among them, the distance between any point in the second preset range and the target AGV is greater than the distance between any point in the first preset range and the target AGV, the distance between any point in the third preset range and the target AGV is greater than the distance between any point in the second preset range and the target AGV, and the first acceleration is greater than the second acceleration.

[0146] Specifically, in the embodiment of the present invention, during the local path planning process, if there are multiple second-class AGVs within the second preset range of the current position of the target AGV, the multiple second-class AGVs are controlled to decelerate at the first acceleration; if there are multiple third-class AGVs within the third preset range of the current position of the target AGV, the multiple third-class AGVs are controlled to decelerate at the second acceleration. Figure 8 As shown, the second preset range can be a ring formed by a circle with R2 as the radius and a circle with R1 (R2>R1) as the radius, with the current position of the target AGV as the center, and the third preset range can be a ring formed by a circle with R3 as the radius and a circle with R2 (R3>R2) as the radius, with the current position of the target AGV as the center.

[0147] The first acceleration is greater than the second acceleration. The first acceleration can be determined by the distance between the second type AGV and the target AGV, for example, r1 2 a1. Among them, r1 is the distance between the second type AGV and the target AGV, and a1 is the initial acceleration, which is a constant value.

[0148] The second acceleration can be determined by the distance between the third type AGV and the target AGV, for example, r2a2. Among them, r2 is the distance between the second type AGV and the target AGV, and a2 is the initial acceleration, which is a constant value. a1 and a2 can be equal or unequal, which is not specifically limited here.

[0149] Figure 8In the example, if AGV1 is the target AGV, then within the first preset range of radius R1, AGV8 and AGV7 are both first-class AGVs. First, re-plan the local path of the global planning path of AGV8, and control AGV8 to leave the first preset range. Then, re-plan the local path of the global planning path of AGV7, and control AGV7 to leave the first preset range. Within the second preset range of radius R1-R2, AGV5 and AGV6 are both second-class AGVs, then AGV5 and AGV6 can be controlled to decelerate at the first acceleration at the same time. Within the third preset range of radius R2-R3, AGV2, AGV3 and AGV4 are all third-class AGVs, then AGV2, AGV3 and AGV4 can be controlled to decelerate at the second acceleration at the same time.

[0150] In the embodiment of the present invention, a step-by-step emergency control strategy is adopted to prevent the target AGV from colliding with other surrounding AGVs.

[0151] On the basis of the above-mentioned embodiments, the AGV picking path planning method provided in the embodiments of the present invention is implemented by using the grid method to establish a map model for simulation and verification, which will not be described in detail here.

[0152] In summary, a large-scale AGV picking path planning method based on Flink, DWA and genetic algorithm is provided in the embodiment of the present invention. First, a new working mode of AGV is provided, that is, multiple goods in an order are picked up by one AGV at one time, and after the AGV picking task is completed and all shelves are returned to their positions, the AGV does not return to the parking area, but stays next to the last shelf, that is, the first picking position. The advantage of this is that due to the principle of access time locality, the goods just accessed are more likely to appear in the next order. This saves the time for the AGV to return to the parking area and the time to go from the parking area to the shelf next time. Secondly, in view of the defects that the traditional genetic algorithm cannot guarantee the diversity of population individuals and is easy to fall into the local optimal solution, the embodiment of the present invention proposes to use two Flink clusters to initialize two populations in parallel in two ways, and calculate the optimal individual at the same time. The first Flink cluster uses the individual fitness variance to judge the population diversity, and adopts adaptive probability to perform the first type of gene operations such as selection, crossover and mutation. The second Flink cluster uses the average value of the similarity between the individual and the optimal individual fitness to judge the population diversity, and adopts insertion, deletion and reversal operators to solve the optimal individual. Secondly, in order to speed up the solution of the global path planning problem of the genetic algorithm, the population is set to be relatively large in the embodiment of the present invention, and then the population is divided into multiple sub-populations and distributed to the first Flink cluster and the second Flink cluster for parallel calculation respectively, and finally the population is merged for the next iteration operation. Finally, since the path obtained by the genetic algorithm is a static global planning path, when multiple AGVs are driving, it is easy to encounter other moving AGVs. At this time, the dynamic window algorithm is used for dynamic local path planning. In addition, since the dynamic window evaluation function of DWA consists of three parts, the weighted coefficients of α, β, and γ are difficult to determine. In some cases, the selected simulation trajectory ignores the importance of β obstacle avoidance, and the local planning path finally obtained may have dynamic collision events. At this time, the embodiment of the present invention provides a dynamic obstacle avoidance method for large-scale AGV systems, that is, static waiting and sequential restart strategies. At the last checkpoint, it is ensured that no collision events will occur during the large-scale AGV picking process.

[0153] like Fig. 9 As shown, based on the above embodiments, an AGV picking path planning system is provided in an embodiment of the present invention, including:

[0154] An acquisition module 91 is used to acquire a current order and determine a target AGV corresponding to the current order, wherein the current order includes multiple goods;

[0155] The alternative path construction module 92 is used to construct various alternative paths with the parking position of the target AGV as the starting node, the storage positions of various goods as the intermediate nodes, and the target position of the current order as the end node; the parking position includes the first picking position in the previous order of the current order;

[0156] The global path planning module 93 is used to establish a TSP model based on the path length, path turning angle and number of path turning points of each candidate path, and use a genetic algorithm to solve the TSP model to determine the global planning path among the candidate paths.

[0157] On the basis of the above-mentioned embodiments, in the AGV picking path planning system provided in the embodiments of the present invention, the global path planning module is specifically used for:

[0158] Taking each candidate path as an individual, taking the intermediate node of each candidate path as a gene, and determining the individual fitness based on the path length, path turning angle and the number of path inflection points of each candidate path;

[0159] Two Flink clusters are used to initialize the first and second populations respectively, and individuals of the first population are iteratively updated to obtain the third population, and individuals of the second population are iteratively updated to obtain the fourth population;

[0160] A first individual corresponding to the best individual fitness is determined from the third population and the fourth population, and the first individual is determined as the global planning path.

[0161] On the basis of the above-mentioned embodiments, in the AGV picking path planning system provided in the embodiments of the present invention, the global path planning module is specifically used for:

[0162] The first Flink cluster is used to initialize the first candidate population, and the second Flink cluster is used to initialize the second candidate population.

[0163] Determining the first population based on the individual fitness variance of the first candidate population;

[0164] The second population is determined based on the average similarity between each individual of the second candidate population and the second individual corresponding to the best individual fitness.

[0165] On the basis of the above-mentioned embodiments, in the AGV picking path planning system provided in the embodiments of the present invention, the global path planning module is specifically used for:

[0166] Using the master node of the first Flink cluster, the first population is divided into multiple first sub-populations, and using the task nodes of the first Flink cluster to perform the first type of gene operation on each first sub-population in a distributed and parallel manner to obtain an updated first sub-population;

[0167] Using the master node, merging all updated first sub-populations to obtain a first new population;

[0168] The first Flink cluster is used to iteratively divide, perform first-type gene operations, and merge the first new population to obtain the third population.

[0169] On the basis of the above-mentioned embodiments, in the AGV picking path planning system provided in the embodiments of the present invention, the global path planning module is specifically used for:

[0170] For any task node in the first Flink cluster, using the any task node, based on the individual fitness of any individual in the first subpopulation, determining an activation function corresponding to the any individual;

[0171] Determine a target operation probability of the first type of genetic operation corresponding to any individual based on the activation function of any individual and the operation probability range of the first type of genetic operation;

[0172] Based on the target operation probability, the first type of genetic operation is performed on any one of the individuals.

[0173] On the basis of the above embodiment, the AGV picking path planning system provided in the embodiment of the present invention further includes a local path planning module for:

[0174] Based on the dynamic window algorithm, local path planning is performed on the global planned path to obtain the local planned path of the target AGV.

[0175] On the basis of the above embodiments, in the AGV picking path planning system provided in the embodiments of the present invention, the dynamic window evaluation function of the dynamic window algorithm is determined based on the evaluation function of the current simulated speed information of the target AGV, the closest distance between the simulated trajectory of the target AGV and the static obstacles around the target AGV, and the azimuth deviation between the end direction of the simulated trajectory and the target point on the global planning path, and the target point is the intersection of a circle with the current position of the target AGV as the center and the target distance as the radius and the global planning path.

[0176] On the basis of the above-mentioned embodiment, in the AGV picking path planning system provided in the embodiment of the present invention, the local path planning module is specifically used for:

[0177] During the local path planning process, if there are multiple first-class AGVs within the first preset range of the current position of the target AGV, the multiple first-class AGVs are controlled to stop running, and the global planned path of each first-class AGV is re-planned locally in descending order of the serial numbers of the multiple first-class AGVs, so that each first-class AGV can leave the first preset range.

[0178] On the basis of the above-mentioned embodiment, in the AGV picking path planning system provided in the embodiment of the present invention, the local path planning module is specifically used for:

[0179] During the local path planning process, if there are multiple second-category AGVs within a second preset range of the current position of the target AGV, the multiple second-category AGVs are controlled to decelerate at a first acceleration;

[0180] If there are multiple third-category AGVs within the third preset range of the current position of the target AGV, control the multiple third-category AGVs to decelerate at the second acceleration;

[0181] Among them, the distance between any point in the second preset range and the target AGV is greater than the distance between any point in the first preset range and the target AGV, the distance between any point in the third preset range and the target AGV is greater than the distance between any point in the second preset range and the target AGV, and the first acceleration is greater than the second acceleration.

[0182] Specifically, the functions of each module in the AGV picking path planning system provided in the embodiment of the present invention correspond one-to-one to the operating procedures of each step in the above-mentioned method embodiment, and the effects achieved are also consistent. Please refer to the above-mentioned embodiment for details, and no further details will be given in the embodiment of the present invention.

[0183] Fig.10 An example of a physical structure diagram of an electronic device is shown in FIG. Fig.10As shown, the electronic device may include: a processor (Processor) 110, a communication interface (Communications Interface) 120, a memory (Memory) 130 and a communication bus 140, wherein the processor 110, the communication interface 120, and the memory 130 complete mutual communication through the communication bus 140. The processor 110 can call the logic instructions in the memory 130 to execute the AGV picking path planning method provided in the above embodiments, the method comprising: obtaining a current order and determining a target AGV corresponding to the current order, wherein the current order contains a variety of goods; constructing each alternative path with the parking position of the target AGV as the starting node, the storage positions of various goods as the intermediate nodes, and the target position of the current order as the end node; the parking position includes the first picking position in the previous order of the current order; based on the path length, path turning angle and number of path inflection points of each alternative path, a TSP model is established, and a genetic algorithm is used to solve the TSP model to determine the global planning path in each alternative path.

[0184] In addition, the logic instructions in the above-mentioned memory 130 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0185] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the AGV picking path planning method provided in the above embodiments, the method including: obtaining a current order and determining a target AGV corresponding to the current order, wherein the current order contains multiple goods; constructing each alternative path with the parking position of the target AGV as the starting node, the storage positions of various goods as the intermediate nodes, and the target position of the current order as the end node; the parking position includes the first picking position in the previous order of the current order; based on the path length, path turning angle and number of path inflection points of each alternative path, a TSP model is established, and a genetic algorithm is used to solve the TSP model to determine the global planning path in each alternative path.

[0186] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the AGV picking path planning method provided in the above-mentioned embodiments, the method comprising: obtaining a current order and determining a target AGV corresponding to the current order, wherein the current order contains multiple goods; constructing each alternative path with the parking position of the target AGV as the starting node, the storage positions of various goods as the intermediate nodes, and the target position of the current order as the end node; the parking position includes the first picking position in the previous order of the current order; establishing a TSP model based on the path length, path turning angle and number of path inflection points of each alternative path, and using a genetic algorithm to solve the TSP model to determine the global planning path in each alternative path.

[0187] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0188] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0189] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AGV picking path planning method, characterized in that: include: Obtaining a current order and determining a target AGV corresponding to the current order, wherein the current order includes multiple goods; Each alternative path is constructed with the parking position of the target AGV as the starting node, the storage positions of various goods as the intermediate nodes, and the target position of the current order as the ending node; the parking position includes the first picking position in the previous order of the current order; the first picking position refers to the storage position of the goods first picked by the target AGV when executing the previous order; Based on the path length, path turning angle and number of path inflection points of each candidate path, a TSP model is established, and a genetic algorithm is used to solve the TSP model to determine a global planning path among the candidate paths; The adopting of a genetic algorithm to solve the TSP model and determine a global planning path among the candidate paths specifically includes: Taking each candidate path as an individual, taking the intermediate node of each candidate path as a gene, and determining the individual fitness based on the path length, path turning angle and the number of path inflection points of each candidate path; Two Flink clusters are used to initialize the first and second populations respectively, and individuals of the first population are iteratively updated to obtain the third population, and individuals of the second population are iteratively updated to obtain the fourth population; A first individual corresponding to the best individual fitness is determined from the third population and the fourth population, and the first individual is determined as the global planning path.

2. The AGV picking path planning method according to claim 1, characterized in that: Two Flink clusters are used to initialize the first population and the second population respectively, specifically including: The first Flink cluster is used to initialize the first candidate population, and the second Flink cluster is used to initialize the second candidate population. Determining the first population based on the individual fitness variance of the first candidate population; The second population is determined based on the mean similarity between each individual of the second candidate population and a second individual corresponding to the best individual fitness of the second candidate population.

3. The AGV picking path planning method according to claim 1, characterized in that: The iterative updating of the individuals of the first population to obtain the third population specifically includes: Using the master node of the first Flink cluster, the first population is divided into multiple first sub-populations, and using the task nodes of the first Flink cluster to perform the first type of gene operation on each first sub-population in a distributed and parallel manner to obtain an updated first sub-population; Using the master node, merging all updated first sub-populations to obtain a first new population; The first Flink cluster is used to iteratively perform the above operation on the first new population to obtain the third population.

4. The AGV picking path planning method according to claim 3, characterized in that: The step of using the task nodes in the first Flink cluster to perform the first type of gene operation on each first sub-population in a distributed and parallel manner specifically includes: For any task node in the first Flink cluster, using the any task node, based on the individual fitness of any individual in the first subpopulation, determining an activation function corresponding to the any individual; Determine a target operation probability of the first type of genetic operation corresponding to any individual based on the activation function of any individual and the operation probability range of the first type of genetic operation; Based on the target operation probability, the first type of genetic operation is performed on any one of the individuals.

5. The AGV picking path planning method according to any one of claims 1 to 4, characterized in that: The genetic algorithm is used to solve the TSP model to determine the global planning path among the candidate paths, and then the following steps are further included: Based on the dynamic window algorithm, local path planning is performed on the global planned path to obtain the local planned path of the target AGV.

6. The AGV picking path planning method according to claim 5, characterized in that: The dynamic window evaluation function of the dynamic window algorithm is determined based on the evaluation function of the current simulated speed information of the target AGV, the closest distance between the simulated trajectory of the target AGV and the static obstacles around the target AGV, and the azimuth deviation between the end direction of the simulated trajectory and the target point on the global planning path. The target point is the intersection of a circle with the current position of the target AGV as the center and the target distance as the radius and the global planning path.

7. The AGV picking path planning method according to claim 5, characterized in that: The method of performing local path planning on the global planned path based on the dynamic window algorithm to obtain the local planned path of the target AGV specifically includes: During the local path planning process, if there are multiple first-class AGVs within the first preset range of the current position of the target AGV, the multiple first-class AGVs are controlled to stop running, and the global planned path of each first-class AGV is re-planned locally in descending order of the serial numbers of the multiple first-class AGVs, so that each first-class AGV can leave the first preset range.

8. The AGV picking path planning method according to claim 7, characterized in that: The method of performing local path planning on the global planned path based on the dynamic window algorithm to obtain the local planned path of the target AGV specifically includes: During the local path planning process, if there are multiple second-category AGVs within a second preset range of the current position of the target AGV, the multiple second-category AGVs are controlled to decelerate at a first acceleration; If there are multiple third-category AGVs within the third preset range of the current position of the target AGV, control the multiple third-category AGVs to decelerate at the second acceleration; Among them, the distance between any point in the second preset range and the target AGV is greater than the distance between any point in the first preset range and the target AGV, the distance between any point in the third preset range and the target AGV is greater than the distance between any point in the second preset range and the target AGV, and the first acceleration is greater than the second acceleration.

9. An AGV picking path planning system, characterized in that: include: An acquisition module, used to acquire a current order and determine a target AGV corresponding to the current order, wherein the current order includes multiple goods; An alternative path construction module is used to construct various alternative paths with the parking position of the target AGV as the starting node, the storage positions of various goods as the intermediate nodes, and the target position of the current order as the end node; the parking position includes the first picking position in the previous order of the current order; the first picking position refers to the storage position of the goods first picked by the target AGV when executing the previous order; A global path planning module, for establishing a TSP model based on the path length, path turning angle and number of path inflection points of each candidate path, and solving the TSP model using a genetic algorithm to determine a global planning path among the candidate paths; The adopting of a genetic algorithm to solve the TSP model and determine a global planning path among the candidate paths specifically includes: Taking each candidate path as an individual, taking the intermediate node of each candidate path as a gene, and determining the individual fitness based on the path length, path turning angle and the number of path inflection points of each candidate path; Two Flink clusters are used to initialize the first and second populations respectively, and individuals of the first population are iteratively updated to obtain the third population, and individuals of the second population are iteratively updated to obtain the fourth population; A first individual corresponding to the best individual fitness is determined from the third population and the fourth population, and the first individual is determined as the global planning path.

Images