Automatic control system and method of storage robot

Abstract

The invention discloses an automatic control system and method for a storage robot, and the method comprises the steps: carrying out the one-hot coding of a path length through employing a data analysis and analysis algorithm based on deep learning, and carrying out the hole convolution coding of a space distribution matrix of the storage robot, and intelligently predicting the time for the first storage robot to move to the shelf according to the cross-modal space-time interaction representation between the path length low-dimensional embedding feature and the storage robot space distribution topological feature. Therefore, by comprehensively considering the position data and the path length of the storage robot and capturing the complex relation between the position data and the path length, the actual situation is comprehensively reflected, and the accuracy of time prediction is improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more specifically, to an automated control system and method for a warehousing robot. Background Art

[0002] In the context of the booming modern warehousing and logistics industry, the explosive growth of e-commerce business and consumers' stringent requirements for delivery timeliness have prompted warehouse management to move rapidly towards intelligence and efficiency. As the core equipment for improving warehouse operation efficiency, the accuracy and efficiency of warehouse robot scheduling and control are crucial.

[0003] Patent CN111232590A proposes an automated control method and device for a warehouse robot. After receiving the transport instructions from the logistics system, the system determines the shelf location of the goods to be transported and plans all paths from the shipping port to the target shelf. Then, the positions of idle and working robots on each path are analyzed, and the time when the idle robot arrives at the shelf is predicted using a pre-trained time prediction model. Finally, by comparing the predicted time, the idle robot that arrives the fastest is selected to perform the handling task, thereby optimizing the efficiency of warehouse operations.

[0004] The prediction of the arrival time at the shelf in this patent is achieved through a pre-trained time prediction model (convolutional neural network). However, the receptive field of traditional CNN is local, and the understanding of long-distance dependencies is limited, especially in complex and changing warehousing environments. This limitation may affect the accuracy of predictions. In addition, the method in the patent focuses on using the historical location data of the warehouse robot as the main input feature, but ignores the importance of other key factors such as path length, and fails to effectively integrate the interaction between multimodal features such as path length and robot spatial distribution, which may cause the model to be unable to fully utilize all available information for more accurate time predictions.

[0005] Therefore, an optimized automation control solution for warehouse robots is desired. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an automated control system and method for a warehouse robot, which performs one-hot encoding on the path length by using data analysis and analysis algorithms based on deep learning, and performs void convolution encoding on the spatial distribution matrix of the warehouse robot, so as to intelligently predict the time for the first warehouse robot to move to the shelf based on the cross-modal spatiotemporal interaction representation between the low-dimensional embedding features of the path length and the spatial distribution topological features of the warehouse robot. In this way, by comprehensively considering the location data and path length of the warehouse robot and capturing the complex relationship between the two, the actual situation is fully reflected and the accuracy of time prediction is improved.

[0007] According to one aspect of the present application, an automated control method for a warehouse robot is provided, which includes: receiving a transport instruction for goods to be transported sent by an upstream logistics system; locating the shelf where the goods to be transported are located from a plurality of shelves in a warehouse based on the transport instruction; calculating all paths from the outlet of the goods to be transported in the warehouse to the shelf; identifying the warehouse robot that is closest to the shelf and idle on each path; determining the position of the warehouse robot running between the warehouse robot that is closest to the shelf and idle on each path and the shelf; predicting the time it takes for the warehouse robot that is closest to the shelf and idle on each path to move to the shelf; and selecting the idle warehouse robot with the shortest predicted time to perform the transportation task of the goods to be transported.

[0008] According to another aspect of the present application, an automated control system for a warehouse robot is provided, which includes: a transport instruction receiving module for receiving transport instructions for goods to be transported sent by an upstream logistics system; a goods to be transported positioning module for locating the shelf where the goods to be transported are located from multiple shelves in a warehouse based on the transport instructions; a path calculation module for calculating all paths from the outlet of the goods to be transported in the warehouse to the shelf; a robot identification module for identifying the warehouse robot that is closest to the shelf and idle on each path; a robot position determination module for determining the position of the warehouse robot running between the warehouse robot that is closest to the shelf and idle on each path and the shelf; a robot movement time prediction module for predicting the time it takes for the warehouse robot that is closest to the shelf and idle on each path to move to the shelf; and a robot selection module for selecting an idle warehouse robot with the shortest predicted time to perform the transportation task of the goods to be transported.

[0009] Compared with the prior art, the present application provides an automated control system and method for a storage robot, which uses deep learning-based data analysis and analysis algorithms to perform unique hot encoding on the path length, and performs dilated convolution encoding on the storage robot spatial distribution matrix, thereby intelligently predicting the time for the first storage robot to move to the shelf based on the cross-modal spatiotemporal interaction representation between the low-dimensional embedding features of the path length and the topological features of the storage robot spatial distribution. In this way, by comprehensively considering the location data and path length of the storage robot and capturing the complex relationship between the two, the actual situation is fully reflected, and the accuracy of time prediction is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 This is a flow chart of an automated control method for a warehouse robot according to an embodiment of the present application.

[0012] Figure 2 This is a data flow diagram of the automated control method for a warehouse robot according to an embodiment of the present application.

[0013] Figure 3 This is a flowchart of sub-step S6 of the automated control method for a warehouse robot according to an embodiment of the present application.

[0014] Figure 4 It is a block diagram of the automated control system of the storage robot according to an embodiment of the present application. DETAILED DESCRIPTION

[0015] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0016] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0017] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0018] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0019] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0020] In the technical solution of the present application, an automated control method for a warehousing robot is proposed. Figure 1 This is a flow chart of an automated control method for a warehouse robot according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the automatic control method of the storage robot according to the embodiment of the present application. Figure 1 and Figure 2 As shown, the automatic control method of the warehouse robot according to the embodiment of the present application includes the steps of: S1, receiving the transportation instruction of the goods to be transported sent by the upstream logistics system; S2, locating the shelf where the goods to be transported are located from multiple shelves in the warehouse based on the transportation instruction; S3, calculating all paths from the outlet of the goods to be transported in the warehouse to the shelf; S4, identifying the warehouse robot that is closest to the shelf and idle on each path; S5, determining the position of the warehouse robot running between the warehouse robot that is closest to the shelf and idle on each path and the shelf; S6, predicting the time for the warehouse robot that is closest to the shelf and idle on each path to move to the shelf; S7, selecting the idle warehouse robot with the shortest predicted time to perform the transportation task of the goods to be transported.

[0021] In particular, the S1 receives the transportation instructions of the goods to be transported sent by the upstream logistics system. It should be understood that the operation of the logistics system usually involves collaboration between multiple links. When an order is created or needs to be processed, the central control host responsible for managing the internal logistics of the warehouse needs to receive instructions from the upstream logistics system (such as sales platform, inventory management system, etc.) to start the transportation process. In this way, it can ensure that the operation inside the warehouse is synchronized with the external supply chain demand, so that the warehouse robot can respond to new task requests in a timely manner, improving the response speed and service level of the overall supply chain.

[0022] In one example, this process first relies on the operation of the logistics system. In an e-commerce or retail environment, once a customer places an order to purchase a product, the order information is recorded in the logistics system. This system may be a complex network service architecture that can handle and track thousands of orders and decide which products should be shipped from which warehouse based on the location, quantity, and other factors of the products. When it is determined that a product needs to be shipped from a specific warehouse, the logistics system will generate a transportation instruction. This instruction contains all the necessary information about the goods to be shipped, such as the product ID, the target shelf location, the priority, and any other data that affects the handling process.

[0023] In particular, in S2, based on the transport instruction, the shelf where the goods to be transported are located is located from multiple shelves in the warehouse. In order to accurately find and take out the correct goods, the specific location of the goods must be determined first. This step is the basis of the entire transportation process. Only by correctly identifying the location of the goods can the subsequent transportation route be effectively planned.

[0024] In one example, when a shipping instruction is received and parsed, the system first queries the internal database based on the product information contained in the instruction (such as product number, barcode or RFID tag, etc.) to determine the specific storage location of these products in the warehouse. Modern warehouse management systems usually maintain a detailed inventory record, which includes not only the basic information of each product, but also the exact location where each product is stored - that is, the specific shelf number and coordinates. Once the approximate range of the target product is determined, the next step is to accurately locate the specific shelf where it is located. This step may involve the application of multiple technologies and equipment. For example, if the warehouse is equipped with automated cargo management hardware, such as RFID readers, laser scanners or other forms of sensor networks, the system can obtain the location information of the goods in real time through these devices. For a storage robot equipped with a navigation system, it can move autonomously to the estimated shelf area after receiving the task, and use its own sensing device (such as camera, infrared sensor, etc.) for the final confirmation. In this process, the system will also take into account the current dynamic situation of the warehouse. For example, if other robots are performing tasks in the area, or if there are recent changes in the shelf layout due to the entry and exit of goods, the system needs to update the relevant data in a timely manner and adjust the search strategy. In addition, to avoid interference or collision, the system will also evaluate the traffic flow in each area and choose the most appropriate time to dispatch the robot to the designated shelf.

[0025] In particular, S3 calculates all paths from the delivery port of the goods to be transported in the warehouse to the shelf. It should be understood that the warehouse layout is complex, and there may be multiple different paths connecting the delivery port and the target shelf. Calculating all possible paths helps to select the optimal path, avoid traffic jams or unnecessary detours, and thus save time. In this way, it can provide a variety of options for the next selection, which is conducive to selecting the most suitable path, thereby improving the handling efficiency.

[0026] In one example, first, the system loads the map data of the warehouse. This map not only contains the location information of each shelf, but also includes static elements such as aisles, safety areas, and fixed obstacles. In addition, considering the dynamic changes in the warehouse, such as other working robots, temporarily set up obstacles, etc., the system will also update the dynamic elements on the map in real time. In addition, considering the motion characteristics of the warehouse robot, path planning must also take into account factors such as the robot's size, turning radius, and speed limit. For example, some narrow passages may not be suitable for large robots to pass through; and some sharp turns require a reduction in driving speed to ensure safety. In response to this situation, the system will add additional constraints during the path calculation process to ensure that the planned path is not only theoretically feasible, but can also be successfully executed by the robot in practice.

[0027] In particular, S4 identifies the idle storage robot that is closest to the shelf on each path. That is, after selecting a specific path, it is necessary to quickly find a robot that can be put into work immediately. Selecting the nearest idle robot can minimize waiting time and make full use of existing resources. In this way, the speed of task allocation can be accelerated, ensuring that the fastest available robot can be dispatched to the designated location, thereby improving the utilization rate of the robot. In this process, various positioning technologies installed in the warehouse, such as Wi-Fi-based triangulation, Bluetooth beacons, ultra-wideband (UWB) positioning systems, or the robot's own GPS module, can provide accurate indoor positioning services, so that the system can accurately know the exact location of each robot.

[0028] In particular, S5 determines the position of the storage robot that is closest to the shelf and idle on each path and the storage robot running between the shelf. It should be understood that understanding the traffic conditions on the current path is crucial to assessing the time required for transportation. If there are other working robots on the path, delays may occur; knowing this information in advance can help predict the arrival time more accurately. In this way, the accuracy of time prediction can be improved, helping to better coordinate the activities between different robots and prevent collisions or other unexpected situations.

[0029] In particular, the step S6 predicts the time for the idle storage robot closest to the shelf on each path to move to the shelf. Figure 3As shown, S6 includes: S61, setting the storage robot that is closest to the shelf and idle on each path as the first storage robot, and obtaining the path length from the first storage robot to the shelf; S62, obtaining the position data of all the running storage robots between the first storage robot and the shelf, and constructing the storage robot spatial distribution matrix based on the position data of all the running storage robots; S63, respectively encoding the path length and the storage robot spatial distribution matrix to obtain the low-dimensional embedding feature of the path length and the storage robot spatial distribution topological feature; S64, performing cross-modal spatiotemporal dynamic analysis on the low-dimensional embedding feature of the path length and the storage robot spatial distribution topological feature to obtain the target path-spatial distribution cross-modal fine-grained interaction feature; S65, based on the target path-spatial distribution cross-modal fine-grained interaction feature, obtaining the time when the first storage robot moves to the shelf.

[0030] Specifically, in S61, the storage robot that is closest to the shelf and idle on each path is set as the first storage robot, and the path length from the first storage robot to the shelf is obtained. It should be understood that in a complex storage environment, there are many storage robots. By setting the specific title of "first storage robot", it is possible to clearly focus on the idle robot on each path that is most likely and most suitable to perform the transportation task, and to analyze the target intensively. The path length is a key basic factor that affects the time required for a storage robot to move from its current position to the shelf. Obtaining the path length from the first storage robot to the shelf provides a direct and important data basis for the subsequent prediction of the time it takes to move to the shelf.

[0031] Specifically, the S62 obtains the position data of all the running warehouse robots between the first warehouse robot and the shelf, and constructs the warehouse robot space distribution matrix based on the position data of all the running warehouse robots. It should be understood that in the warehousing and logistics environment, the operation of the warehouse robot is dynamic. In addition to the target first warehouse robot, other warehouse robots running between it and the shelf will affect its moving speed and path. These running warehouse robots may cause blockages, obstructions, etc., affecting the movement efficiency of the first warehouse robot. Therefore, it is necessary to obtain their position data so that these possible interference factors can be taken into account in time prediction. Afterwards, the purpose of constructing the warehouse robot space distribution matrix is ​​to present this position information in a structured form. The matrix is ​​an effective data representation form that can intuitively show the relative position relationship of the warehouse robot in space. In this way, considering the existence of other robots, the time prediction can be closer to the actual situation.

[0032] Specifically, the S63 performs feature encoding on the path length and the storage robot spatial distribution matrix to obtain the path length low-dimensional embedding feature and the storage robot spatial distribution topological feature. That is, in the technical solution of the present application, first, the path length is uniquely encoded to obtain the path length low-dimensional embedding coding vector as the path length low-dimensional embedding feature; considering that the path length can usually be regarded as a discrete classification feature, especially when it is divided into multiple categories or intervals, and the original form of data may not be well understood and processed by the model during model processing, therefore, in order to be able to convert it into a form that can be processed by the computer, in the technical solution of the present application, the path length is uniquely encoded to convert the original data into a vector form to obtain a path length low-dimensional embedding coding vector. In particular, this vector form can be better understood and analyzed by the model, so as to be integrated into the overall model calculation and prediction process. Further, the spatial topological feature extractor based on the void convolutional neural network model is used to convolutionally encode the storage robot spatial distribution matrix to obtain the storage robot spatial distribution topological feature map. It should be understood that although the warehouse robot spatial distribution matrix can represent the positional relationship of the robot, it is a relatively primitive data structure. Each element in the matrix only represents the position information of the robot, lacking a high-level abstract description of the spatial structure and distribution pattern. In particular, in a warehousing environment, the spatial relationship between robots (such as the degree of aggregation, the interval distance, whether a blocked channel is formed, etc.) is complex and has a topological structure. Based on this, in the technical solution of the present application, a spatial topological feature extractor based on a void convolutional neural network model is used to convolutionally encode the warehouse robot spatial distribution matrix to obtain the warehouse robot spatial distribution topological feature map, and obtain the warehouse robot spatial distribution topological feature map. It should be understood that the void convolution can expand the receptive field of the convolution kernel without adding too many parameters by introducing "holes" or skipping certain positions in the convolution kernel, thereby better capturing the long-distance correlation between the robot positions and the overall characteristics of the spatial distribution.

[0033] Specifically, the S64 performs a cross-modal spatiotemporal dynamic analysis on the low-dimensional embedding feature of the path length and the spatial distribution topological feature of the storage robot to obtain the target path-spatial distribution cross-modal fine-grained interactive feature. Considering that in the task of predicting the moving time of the storage robot, the path length and the spatial distribution of the storage robot are two key influencing factors, and they belong to different modal information. The low-dimensional embedding feature of the path length represents the distance-related information, while the spatial distribution topological feature of the storage robot reflects the spatial layout-related information. Moreover, the path length and the spatial distribution of the storage robot do not exist in isolation, but there is a complex interactive relationship. For example, if a longer path encounters an area with dense spatial distribution, it may cause a significant increase in the moving time. Therefore, in order to comprehensively and accurately predict the time when the robot moves to the shelf, the present application performs a cross-modal spatiotemporal dynamic analysis on the low-dimensional embedding feature of the path length and the spatial distribution topological feature of the storage robot to deeply explore the interactive relationship between the two modal information, thereby capturing this complex mutual influence, and obtaining the target path-spatial distribution cross-modal fine-grained interactive feature, which provides more powerful support for subsequent time prediction.

[0034] Specifically, first, the spatial distribution topological feature map of the warehouse robot is feature decoupled and feature flattened to obtain a set of local feature vectors of the spatial distribution topology of the warehouse robot. Through feature decoupling, the system can separate the different attributes of these positions, such as the relative distance between robots, orientation, and obstacles around them. The flattening process converts these multi-dimensional information into one-dimensional vectors, which can simplify subsequent calculations and ensure that all robot position information can be effectively analyzed by the system. This process is crucial to eliminating redundant information and enhancing the adaptability of the system, so that the system can focus on factors that really affect path selection. In one example, the spatial distribution topological feature map of the warehouse robot is feature decoupled and feature flattened using the following feature decoupling and feature flattening formulas to obtain a set of local feature vectors of the spatial distribution topology of the warehouse robot; wherein the feature decoupling and feature flattening formulas are: ;in, is the spatial distribution topological characteristic diagram of the storage robot, It is a feature decoupling and feature flattening operation. is the set of local feature vectors of the spatial distribution topology of the warehouse robot, and are the first, second, and third local feature vectors in the set of spatial distribution topology of the warehouse robot. , and The local feature vector of the spatial distribution topology of storage robots.

[0035] Then, in order to further explore the spatial patterns contained in these flattened local feature vectors, the path length-spatial distribution prior cluster center encoding vector between the low-dimensional embedding encoding vector of the path length and the set of the local feature vectors of the spatial distribution topology of the storage robot is calculated. Specifically, in an embodiment of the present application, first, the set of the local feature vectors of the spatial distribution topology of the storage robot is subjected to spatial distribution modal kernel feature extraction to obtain the spatial distribution topology feature kernel encoding vector of the storage robot; in this way, the key features of the spatial distribution of the storage robot can be captured, and the generated path length-spatial distribution prior cluster center encoding vector provides a solid foundation for subsequent steps, helping the system to better understand and predict the behavior of the robot in different situations. Then, the path length low-dimensional embedding encoding vector and the spatial distribution topology feature kernel encoding vector of the storage robot are cascaded and linearly transformed to obtain the path length-spatial distribution prior cluster center encoding vector. Through cascade processing and linear transformation, the system creates a path length-spatial distribution prior cluster center encoding vector to capture robot groups that have similarities in spatial distribution and path selection. In a specific implementation, the path length-spatial distribution prior cluster center encoding vector combines the advantages of the two modes of path length and spatial distribution. By comparing or associating the two features, the cluster center focusing on the core related information of the path length and spatial distribution can be predicted and set. Furthermore, the path length-spatial distribution prior cluster center encoding vector and the set of the warehouse robot spatial distribution topology local feature vectors are subjected to cross-modal fine-grained clustering aggregation to obtain the target path-spatial distribution cross-modal fine-grained interaction encoding vector as the target path-spatial distribution cross-modal fine-grained interaction feature. In one example, the path length-spatial distribution prior cluster center encoding vector between the path length low-dimensional embedding encoding vector and the set of the warehouse robot spatial distribution topology local feature vectors is calculated using the following kernel calculation formula; wherein the kernel calculation formula is: ;in, For Perform kernel feature extraction, represents the one-norm of a vector, The number of vectors in the set of local feature vectors representing the spatial distribution topology of the storage robot is reduced by one, yes The corresponding semantic difference value of the spatial distribution of the storage robot, Indicated by natural constant The exponential function with base , is the number of vectors in the set of local feature vectors of the spatial distribution topology of the storage robot, is the kernel encoding vector of the spatial distribution topological features of the warehouse robot, is the path length low-dimensional embedding encoding vector, For cascade operation, is the weight transformation matrix, is the bias vector, is the path length-spatial distribution prior cluster center encoding vector.

[0036] Finally, the path length-spatial distribution prior cluster center encoding vector and the set of the warehouse robot spatial distribution topology local feature vectors are subjected to cross-modal fine-grained clustering aggregation to obtain the target path-spatial distribution cross-modal fine-grained interaction encoding vector as the target path-spatial distribution cross-modal fine-grained interaction feature. Specifically, in an embodiment of the present application, first, a semantic metric based on an inverse hyperbolic cosine function is used to calculate the Poincare distances of the path length-spatial distribution prior cluster center encoding vector and each warehouse robot spatial distribution topology local feature vector in the set of the warehouse robot spatial distribution topology local feature vector to obtain a set of target path-spatial distribution semantic metric values; that is, by calculating the Poincare distances of the path length-spatial distribution prior cluster center encoding vector and each warehouse robot spatial distribution topology local feature vector in the set of the warehouse robot spatial distribution topology local feature vector, the degree of deviation of each robot from its ideal position is quantified, that is, the target path-spatial distribution semantic metric value, so as to help the system accurately identify which robots are in the correct position and which need to be adjusted. Next, a binary function is used to cluster the set of target path-spatial distribution semantic metric values ​​to obtain a set of target path-spatial distribution clustering coefficients; a high clustering coefficient means that the robots in the group work closely together and share similar behavior patterns; a low clustering coefficient may indicate that there are problems that need to be solved or room for improvement. Furthermore, based on the set of target path-spatial distribution clustering coefficients, the path length-spatial distribution prior clustering center encoding vector and the set of the warehouse robot spatial distribution topology local feature vectors are cross-domain fine-grained fused to obtain the target path-spatial distribution cross-modal fine-grained interaction encoding vector. That is, the actions of the warehouse robot are guided by fusing the path length information and the spatial distribution features. This method not only integrates information from many aspects, but also achieves a high degree of sensitivity to environmental changes, making warehouse management more intelligent and dynamic. In one example, the path length-spatial distribution prior cluster center encoding vector and the set of the spatial distribution topology local feature vector of the warehouse robot are subjected to cross-modal fine-grained clustering aggregation using the following cross-modal fine-grained clustering aggregation formula to obtain the target path-spatial distribution cross-modal fine-grained interaction encoding vector; wherein the cross-modal fine-grained clustering aggregation formula is: ;in, is the square of the Euclidean norm of the vector, is the inverse hyperbolic cosine function, for and The target path-space distribution semantic metric between yes The corresponding target path-spatial distribution clustering coefficient, is the preset threshold, is the target path-space distributed cross-modal fine-grained interaction encoding vector.

[0037] Specifically, the S65 obtains the time when the first warehouse robot moves to the shelf based on the target path-spatial distribution cross-modal fine-grained interaction features. That is, in the technical solution of the present application, the target path-spatial distribution cross-modal fine-grained interaction coding vector is input into the arrival time estimator to obtain the time when the first warehouse robot moves to the shelf. That is, the target path-spatial distribution cross-modal fine-grained interaction features obtained by cross-modal spatiotemporal dynamic analysis using the path length low-dimensional embedding coding vector and the warehouse robot spatial distribution topological feature map are decoded and processed, so that through the operation of the decoder, time-related information can be extracted from the coding vector and converted into an understandable time measurement. In particular, this predicted time is crucial for the subsequent selection of the optimal warehouse robot to perform the transportation task. Only accurate time prediction can ensure the selection of the robot that can reach the shelf the fastest, thereby improving the transportation efficiency of warehousing logistics.

[0038] In a preferred example, the target path-spatial distribution cross-modal fine-grained interaction encoding vector is input into a decoder-based arrival time estimator to obtain the time for the first storage robot to reach the shelf, including: determining the mean eigenvalue corresponding to the target path-spatial distribution cross-modal fine-grained interaction encoding vector and the standard deviation of the eigenvalues .

[0039] The target path-spatial distribution cross-modal fine-grained interaction encoding vector is multiplied by the point-subtraction vector of the eigenvalue mean and the eigenvalue standard deviation to obtain a first target path-spatial distribution cross-modal fine-grained interaction encoding fair target vector: ;in, represents the standard deviation of the eigenvalue, represents the target path-space distribution cross-modal fine-grained interaction encoding vector, represents the mean of the eigenvalues, Indicates point reduction, represents dot product, Representing the first target path-space distribution encodes a fair target vector for cross-modal fine-grained interaction.

[0040] The target path-spatial distribution cross-modal fine-grained interaction encoding vector and the point-wise subtraction vector of the eigenvalue standard deviation are multiplied by the eigenvalue mean to obtain a second target path-spatial distribution cross-modal fine-grained interaction encoding fair target vector: ;in, The second target path-space distribution encodes a fair target vector with fine-grained interactions across modalities.

[0041] After multiplying the bit-by-bit inverse of the second target path-spatial distribution cross-modal fine-grained interactive coding fair target vector by the first target path-spatial distribution cross-modal fine-grained interactive coding fair target vector, take the bit-by-bit logarithm with base 2 to obtain the target path-spatial distribution cross-modal fine-grained interactive coding information correction vector: ;in, represents the bit-wise inverse of the second target path-space distribution cross-modal fine-grained interaction encoding fair target vector, represents the logarithm operation with base 2, Represents the target path-space distribution cross-modal fine-grained interaction encoding information correction vector.

[0042] The eigenvalue mean Divide by the standard deviation of the eigenvalue The square root of the quotient is multiplied by the weight hyperparameter, and then the optimized target path-spatial distribution cross-modal fine-grained interaction encoding vector is obtained by adding the target path-spatial distribution cross-modal fine-grained interaction encoding information correction vector point to obtain the optimized target path-spatial distribution cross-modal fine-grained interaction encoding vector: ;in, represents the weight hyperparameter, Indicates point addition, A cross-modal fine-grained interaction encoding vector representing the target path-space distribution of the optimization.

[0043] The optimized target path-spatial distribution cross-modal fine-grained interaction encoding vector is input into a decoder-based arrival time estimator to obtain the time it takes for the first warehouse robot to arrive at the shelf.

[0044] Here, since the path length low-dimensional embedded coding vector and the warehouse robot spatial distribution topological feature map respectively represent the embedded coding features of the path length from the first warehouse robot to the shelf and the local spatial correlation topological features of the spatial distribution matrix of all warehouse robots, when performing cross-domain fine-grained aggregation analysis based on modal prior, the feature population attributes corresponding to different modal data will have differences in cross-domain fine-grained aggregation fairness based on the modal prior level, thereby affecting the interactive inclusiveness of the aggregated feature distribution of the target path-spatial distribution cross-modal fine-grained interaction coding vector, and reducing the accuracy of the decoding result obtained by inputting the decoder-based arrival time estimator.

[0045] Therefore, considering the attribute level fairness differences of the data population corresponding to the sequence fusion features of the target path-spatial distribution cross-modal fine-grained interaction coding vector, in order to improve the aggregated interaction inclusiveness under the feature distribution diversity of the target path-spatial distribution cross-modal fine-grained interaction coding vector, the group feature information interaction propagation of the target path-spatial distribution cross-modal fine-grained interaction coding vector is corrected by taking the cross probability value constraint based on the target path-spatial distribution cross-modal fine-grained interaction coding vector as the interactive fairness target representation, and taking the unified statistical feature response interaction based on the target path-spatial distribution cross-modal fine-grained interaction coding vector as the feature distribution multi-level fairness target bias to achieve the robust distribution fairness unified representation of the target path-spatial distribution cross-modal fine-grained interaction coding vector, forming a fair collaboration paradigm under the feature distribution framework of the target path-spatial distribution cross-modal fine-grained interaction coding vector, and improving the accuracy of the time from the first storage robot to the shelf obtained by the arrival time estimator based on the decoder.

[0046] In particular, in S7, an idle storage robot with the shortest predicted time is selected to perform the transportation task of the goods to be transported. By selecting an idle storage robot with the shortest predicted time to perform the transportation task of the goods to be transported, not only can the requirements of immediacy be met, but also the most efficient cargo handling can be achieved, ensuring the rapid response capability in the logistics process, while also reducing operating costs and improving customer experience.

[0047] In summary, the automated control method of the storage robot according to the embodiment of the present application is explained, which uses data analysis and analysis algorithms based on deep learning to perform one-hot encoding on the path length, and performs void convolution encoding on the storage robot spatial distribution matrix, so as to intelligently predict the time for the first storage robot to move to the shelf based on the cross-modal spatiotemporal interaction representation between the low-dimensional embedding features of the path length and the topological features of the spatial distribution of the storage robot. In this way, by comprehensively considering the location data and path length of the storage robot and capturing the complex relationship between the two, the actual situation is fully reflected and the accuracy of time prediction is improved.

[0048] Furthermore, an automated control system for a storage robot is also provided.

[0049] Figure 4 FIG. 1 is a block diagram of an automated control system for a storage robot according to an embodiment of the present application. Figure 4As shown, the automated control system 300 of the warehouse robot according to the embodiment of the present application includes: a transport instruction receiving module 310, which is used to receive the transport instruction of the goods to be transported sent by the upstream logistics system; a goods to be transported positioning module 320, which is used to locate the shelf where the goods to be transported are located from multiple shelves in the warehouse based on the transport instruction; a path calculation module 330, which is used to calculate all paths from the outlet of the goods to be transported in the warehouse to the shelf; a robot identification module 340, which is used to identify the warehouse robot that is closest to the shelf and idle on each path; a robot position determination module 350, which is used to determine the position of the warehouse robot running between the warehouse robot that is closest to the shelf and idle on each path and the shelf; a robot movement time prediction module 360, which is used to predict the time for the warehouse robot that is closest to the shelf and idle on each path to move to the shelf; a robot selection module 370, which is used to select the idle warehouse robot with the shortest predicted time to perform the transportation task of the goods to be transported.

[0050] As described above, the automated control system 300 of the storage robot according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with an automated control algorithm for the storage robot. In a possible implementation, the automated control system 300 of the storage robot according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the automated control system 300 of the storage robot can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the automated control system 300 of the storage robot can also be one of the many hardware modules of the wireless terminal.

[0051] Alternatively, in another example, the warehouse robot's automation control system 300 and the wireless terminal may also be separate devices, and the warehouse robot's automation control system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0052] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. An automatic control method for a storage robot, comprising: Receive transportation instructions for goods to be transported from the upstream logistics system; Based on the transport instruction, locate the shelf where the goods to be transported are located from multiple shelves in the warehouse; Calculate all paths from the shipping port of the goods to be transported in the warehouse to the shelf; identify the storage robot that is closest to the shelf and is idle on each path; Determine the position of the storage robot that is closest to the shelf and is idle on each path and the storage robot running between the shelf; predict the time for the storage robot that is closest to the shelf and is idle on each path to move to the shelf; select the idle storage robot with the shortest predicted time to perform the transportation task of the goods to be transported; it is characterized in that predicting the time for the storage robot that is closest to the shelf and is idle on each path to move to the shelf includes: setting the storage robot that is closest to the shelf and is idle on each path as the first storage robot, and obtaining the path length from the first storage robot to the shelf; obtaining the first storage robot The position data of all the running warehouse robots between the robot and the shelf are obtained, and the warehouse robot spatial distribution matrix is ​​constructed based on the position data of all the running warehouse robots; the path length and the warehouse robot spatial distribution matrix are feature encoded respectively to obtain the path length low-dimensional embedding feature and the warehouse robot spatial distribution topological feature; the path length low-dimensional embedding feature and the warehouse robot spatial distribution topological feature are cross-modal spatiotemporal dynamic analysis is performed on the path length low-dimensional embedding feature and the warehouse robot spatial distribution topological feature to obtain the target path-spatial distribution cross-modal fine-grained interaction feature; based on the target path-spatial distribution cross-modal fine-grained interaction feature, the time when the first warehouse robot moves to the shelf is obtained.

2. The automatic control method of the storage robot according to claim 1, characterized in that: The path length and the warehouse robot spatial distribution matrix are feature encoded respectively to obtain a low-dimensional embedding feature of the path length and a warehouse robot spatial distribution topological feature, including: performing one-hot encoding on the path length to obtain a low-dimensional embedding coding vector of the path length as the low-dimensional embedding feature of the path length; performing convolution encoding on the warehouse robot spatial distribution matrix to obtain a warehouse robot spatial distribution topological feature map as the warehouse robot spatial distribution topological feature.

3. The automatic control method of the storage robot according to claim 2, characterized in that: The spatial distribution matrix of the warehouse robot is convolutionally encoded to obtain a spatial distribution topological feature map of the warehouse robot, including: using a spatial topological feature extractor based on a void convolutional neural network model to convolutionally encode the spatial distribution matrix of the warehouse robot to obtain the spatial distribution topological feature map of the warehouse robot.

4. The automatic control method of the storage robot according to claim 3, characterized in that: The path length low-dimensional embedding feature and the warehouse robot spatial distribution topology feature are subjected to cross-modal spatiotemporal dynamic analysis to obtain target path-spatial distribution cross-modal fine-grained interaction features, including: feature decoupling and feature flattening of the warehouse robot spatial distribution topology feature graph to obtain a set of warehouse robot spatial distribution topology local feature vectors; calculating the path length-spatial distribution prior clustering center encoding vector between the path length low-dimensional embedding coding vector and the set of the warehouse robot spatial distribution topology local feature vectors; cross-modal fine-grained clustering aggregation is performed on the path length-spatial distribution prior clustering center encoding vector and the set of the warehouse robot spatial distribution topology local feature vectors to obtain the target path-spatial distribution cross-modal fine-grained interaction coding vector as the target path-spatial distribution cross-modal fine-grained interaction feature.

5. The automatic control method of the storage robot according to claim 4, characterized in that: Calculate the path length-spatial distribution prior clustering center encoding vector between the path length low-dimensional embedding encoding vector and the set of the spatial distribution topology local feature vectors of the warehouse robot, including: performing spatial distribution modal kernel feature extraction on the set of the spatial distribution topology local feature vectors of the warehouse robot to obtain the spatial distribution topology feature kernel encoding vector of the warehouse robot; perform cascade processing and linear transformation on the path length low-dimensional embedding encoding vector and the spatial distribution topology feature kernel encoding vector of the warehouse robot to obtain the path length-spatial distribution prior clustering center encoding vector.

6. The automatic control method of the storage robot according to claim 5, characterized in that: The path length-spatial distribution prior cluster center encoding vector and the set of the warehouse robot spatial distribution topology local feature vectors are subjected to cross-modal fine-grained clustering aggregation to obtain a target path-spatial distribution cross-modal fine-grained interaction encoding vector, including: using a semantic metric based on an inverse hyperbolic cosine function to respectively calculate the Poincare distance of the path length-spatial distribution prior cluster center encoding vector and each storage robot spatial distribution topology local feature vector in the set to obtain a set of target path-spatial distribution semantic measurement values; using a binary function to perform clustering judgment on the set of target path-spatial distribution semantic measurement values ​​to obtain a set of target path-spatial distribution clustering coefficients; based on the set of target path-spatial distribution clustering coefficients, cross-domain fine-grained fusion is performed on the path length-spatial distribution prior cluster center encoding vector and the set of the warehouse robot spatial distribution topology local feature vector to obtain the target path-spatial distribution cross-modal fine-grained interaction encoding vector.

7. The automatic control method of a storage robot according to claim 6, characterized in that: Based on the target path-spatial distribution cross-modal fine-grained interaction features, the time for the first warehouse robot to move to the shelf is obtained, including: inputting the target path-spatial distribution cross-modal fine-grained interaction encoding vector into an arrival duration estimator to obtain the time for the first warehouse robot to move to the shelf.

8. The automatic control method of the storage robot according to claim 7, characterized in that: The arrival time estimator is a decoder-based arrival time estimator.

9. An automatic control system for a storage robot, characterized in that: include: A transport instruction receiving module is used to receive transport instructions for goods to be transported sent by an upstream logistics system; A module for locating goods to be transported, used to locate the shelf where the goods to be transported are located from a plurality of shelves in the warehouse based on the transport instruction; A path calculation module, used to calculate all paths from the delivery port of the goods to be transported in the warehouse to the shelf; a robot identification module, used to identify the storage robot that is closest to the shelf and is idle on each path; A robot position determination module, used to determine the position of the storage robot running between the storage robot closest to the shelf and idle on each path and the storage robot running between the shelf; a robot movement time prediction module, used to predict the time for the storage robot closest to the shelf and idle on each path to move to the shelf; The robot selection module is used to select an idle storage robot with the shortest predicted time to perform the transportation task of the goods to be transported.

Images