Rapid sorting and conveying method and device for logistics

Abstract

The invention provides a logistics rapid sorting and conveying method and system. The method comprises the steps that a time sequence image of goods is collected to obtain an identification result; based on the identification result, the order information and the warehouse topology data, using a multi-target path optimization model to obtain optimal path scheduling information; on the basis of the optimal path scheduling information, a sorting arm is adjusted in real time to achieve grabbing and putting operation; on the basis of the optimal path scheduling information, the grabbing position of the sorting arm and the throwing speed, the speed and direction of a conveying belt corresponding to the goods are synchronously updated at the same time, so that timely fixed-point sorting and conveying are achieved; and monitoring and calculating an error rate in the sorting process, and if the error rate exceeds a preset interval, triggering an adaptive learning module to optimize the multi-target path optimization model. According to the method, the overall response speed and processing precision of a sorting system are remarkably improved, the continuous iteration capability is achieved, sorting deviation can be actively recognized, a path planning strategy can be optimized, and therefore the manual intervention requirement is reduced, and the intelligent level and long-term stability of warehousing operation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the fields of intelligent logistics and machine learning, and in particular, to a logistics rapid sorting and conveying method and system. BACKGROUND

[0002] In the field of modern warehouse management and logistics, the logistics rapid sorting and conveying system has become one of the key technologies to improve operational efficiency. The existing technology mainly relies on automated equipment to handle the sorting and conveying process of goods. For example, the traditional sorting system usually uses a visual sensor based on two-dimensional image recognition to collect goods information, including goods type and basic location marker. Then, through a pre-set rule engine or a simple path planning algorithm, a conveying path is generated. The accuracy of this method is limited and only suitable for low-speed standardized goods processing scenarios.

[0003] Therefore, the logistics rapid sorting and conveying has become a problem to be solved.

[0004] Patent CN105032783A discloses an intelligent e-commerce warehouse goods sorting device and method, which includes: 1) shelf positioning, arranging the shelf position according to the warehouse site and inputting the shelf position information into the server for storage; 2) goods numbering, numbering the goods on the shelf, each kind of goods corresponding to a unique number, and inputting the number and the shelf position information into the server for storage; 3) inputting task instructions, inputting task instructions into the server through the input module to specify the goods to be sorted, and the server assigns the task to an idle AGV after receiving the goods information to be sorted; if there is no idle AGV, the task is queued in order; 4) goods sorting, the AGV starts working after receiving the task from the server, reaches the corresponding shelf, and takes down the corresponding goods according to the number; 5) goods conveying, the AVG that has taken the goods conveys the goods to the goods output module, puts down the goods, and the AGV returns to the idle state, and the idle AGV repeats the processes of steps 3, 4 and 5; 6) goods output, the goods output module receives the goods output by the AGV and compares the goods information with the task command of the server, and the goods with no error are pasted with a logistics single, and then enter the distribution process. This method only provides a hardware distribution method and does not involve path planning.

[0005] Based on this, the present application provides a logistics rapid sorting and conveying method and system to improve the existing technology. SUMMARY

[0006] The purpose of the present application is to provide a logistics rapid sorting and conveying method and system, which significantly improves the overall response speed and processing accuracy of the sorting system, and has a continuous iteration capability, can actively identify sorting deviation and optimize path planning strategy, thereby reducing the need for manual intervention, improving the intelligent level and long-term stability of warehouse operation.

[0007] The objective of this application is achieved through the following technical solution: In a first aspect, this application provides a method for rapid sorting and conveying of goods in logistics, which performs real-time sorting and conveying of goods in a warehouse based on multi-dimensional data, the method comprising: The time-series images of the goods are acquired to obtain recognition results, which include the goods type, initial storage location, destination storage location, and a three-dimensional point cloud model. The three-dimensional point cloud model is used to indicate the shape and size of the goods. Based on the identification results, order information, and warehouse topology data, a multi-objective path optimization model is used to obtain the optimal path scheduling information. Based on optimal path scheduling information, the sorting arm is adjusted in real time to achieve grabbing and delivery operations; Based on the optimal path scheduling information, the picking position and delivery speed of the sorting arm, the speed and direction of the corresponding conveyor belt of the goods are updated simultaneously to achieve timely fixed-point sorting and transportation. Monitor and calculate the error rate of the sorting process. If it exceeds the preset range, trigger the adaptive learning module to optimize the multi-objective path optimization model.

[0008] The beneficial effects of this technical solution are as follows: This method introduces a time-series image acquisition mechanism to acquire real-time information on cargo type, initial storage location, destination storage location, and a 3D point cloud model. This information comprehensively captures the shape and size characteristics of the cargo, avoiding blind spots in single-dimensional recognition and ensuring that sorting decisions are more aligned with actual logistics dynamics. This multi-dimensional recognition not only reduces human intervention but also adapts to the diversity of goods and the complex changes in the warehouse environment. Furthermore, a multi-objective path optimization model driven by recognition results, order information, and warehouse topology data is used to achieve intelligent path scheduling. This model comprehensively considers various constraints, such as congestion levels and resource allocation, to generate optimal path scheduling information, thereby minimizing transportation delays and resource waste. In the sorting arm adjustment stage, grasping and placement operations are executed in real-time based on scheduling information, ensuring precise positioning and efficient movement of the robotic arm and avoiding mechanical collisions or placement deviations common in traditional methods. Simultaneously, by synchronously updating the speed and direction of the conveyor belt, the method achieves seamless connection of goods in the fixed-point sorting process, resulting in smoother cargo flow and a significant increase in overall throughput. In addition, the integration of error rate monitoring and adaptive learning modules constitutes a closed-loop optimization system. The system continuously tracks abnormal indicators during the sorting process. Once a deviation exceeds a controllable range, a learning mechanism is triggered to iteratively adjust the path model. This adaptive capability not only reduces long-term operational risks but also refines the algorithm based on historical data, improving the system's robustness and scalability. Overall, this method significantly improves operational reliability and economy in highly dynamic warehouse environments, reduces reliance on labor, and drives the transformation of intelligent logistics towards automation.

[0009] In some optional implementations, the acquisition of time-series images of goods to obtain recognition results includes the type of goods, the initial acquisition of time-series images of goods to obtain recognition results, the recognition results including the type of goods, the initial storage location, the destination storage location, and a three-dimensional point cloud model, the three-dimensional point cloud model being used to indicate the shape and size of the goods; Based on the identification results, order information, and warehouse topology data, a multi-objective path optimization model is used to obtain the optimal path scheduling information. Based on optimal path scheduling information, the sorting arm is adjusted in real time to achieve grabbing and delivery operations; Based on the optimal path scheduling information, the picking position and delivery speed of the sorting arm, the speed and direction of the corresponding conveyor belt of the goods are updated simultaneously to achieve timely fixed-point sorting and transportation. Monitor and calculate the error rate of the sorting process. If it exceeds a preset range, trigger the adaptive learning module to optimize the multi-objective path optimization model, including storage location labels, destination storage location labels, and 3D point cloud models, including: Based on RFID signals and at least one frame of time-series image, obtain the cargo type, initial storage location, and target storage location of the goods; Based on subsequent temporal images, the structure-of-motion reconstruction algorithm is used to incrementally generate and optimize the 3D point cloud model of the cargo; When calculating the confidence level of the 3D point cloud model and determining whether the confidence level reaches a preset threshold, if it does, the recognition result is output.

[0010] The beneficial effects of this technical solution are as follows: By combining RFID signals with temporal image processing and incrementally constructing a 3D point cloud model using a motion recovery structure algorithm, the robustness and model accuracy of the cargo identification process are further improved. On the one hand, the multi-source fusion of RFID and visual data enhances the reliability of cargo type and storage location tag identification, reducing misjudgments caused by occlusion and changes in lighting. On the other hand, by incrementally generating and optimizing the 3D point cloud from temporal images, the geometric features of the cargo can be gradually improved during its movement, avoiding the limitations of incomplete information in a single frame image. The introduction of a confidence assessment mechanism ensures that only highly reliable identification results are output, providing accurate input for subsequent path planning, thereby reducing the risk of sorting failures caused by identification errors and improving the system's adaptability in dynamic environments.

[0011] In some optional implementations, the multi-objective path optimization model includes a preprocessing layer, a multi-objective encoding layer, a GNN convolutional layer, and a decoding layer: The process of obtaining optimal path scheduling information based on the identification results and order information using a multi-objective path optimization model includes: A preprocessing layer is used to perform heterogeneous processing on the recognition results, order information, and sample and topology data to obtain a unified embedding vector. A multi-objective coding layer is used to fuse multi-objective constraints on a unified embedding vector to obtain multi-objective feature information; Input multi-objective feature information into the GNN convolutional layer to calculate candidate paths and their corresponding evaluation metrics; For each candidate path and its corresponding evaluation metric, the weights of the candidate paths are dynamically adjusted using the decoding layer to obtain the optimal path scheduling information.

[0012] The beneficial effects of this technical solution are as follows: The hierarchical architecture of the multi-objective path optimization model, including a preprocessing layer, a multi-objective encoding layer, a GNN convolutional layer, and a decoding layer, constructs an efficient path scheduling framework, solving the computational bottleneck of traditional optimization algorithms under complex warehouse topologies. The preprocessing layer performs heterogeneous processing on the recognition results, order information, and warehouse topology data to generate unified embedding vectors. This standardized fusion eliminates data format differences, ensures seamless integration of multi-source information, and avoids decision-making biases caused by information silos. This layer's design improves the model's input compatibility, adapting to the diverse needs of warehouses of different sizes. The multi-objective encoding layer further integrates constraints and extracts multi-objective feature information, supporting comprehensive evaluation of path planning. Subsequently, the GNN convolutional layer utilizes the representational capabilities of graph neural networks to calculate candidate paths and their evaluation metrics based on the warehouse topology. This graph structure modeling accurately captures dependencies between nodes, such as path intersections or resource competition, generating more realistic alternatives. The decoding layer dynamically adjusts path weights to select the optimal scheduling information, ensuring a balance between efficiency and cost. The end-to-end optimization of this architecture reduces iterative computation time and achieves real-time response. In practical applications, this model significantly improves the accuracy and adaptability of route planning: comprehensive evaluation of candidate routes avoids local optima traps, and the quantification of evaluation metrics guides resource allocation, reducing the incidence of transportation conflicts. Simultaneously, the convolutional mechanism of GNNs enhances the model's sensitivity to dynamic changes, such as sudden order adjustments, enabling rapid route replanning and maintaining logistics continuity. Compared to traditional heuristic methods, the hierarchical design of this model improves scalability, supporting large-scale warehouse expansion without sacrificing performance. Furthermore, it promotes energy conservation by reducing unnecessary movement through optimal routes.

[0013] In some optional implementations, the multi-objective coding layer is used to fuse multi-objective constraints on the unified embedding vector to obtain multi-objective feature information. The multi-objective coding layer incorporates a Pareto attention mechanism, including: Receive the unified embedding vector, initialize the objective function corresponding to the unified embedding vector, the objective function is used to indicate path length, transportation efficiency, transportation cost and load balancing; A Pareto front set is generated using a nondominated sorting algorithm. The Pareto front set is used to identify nondominated solutions of the objective function. Based on the Pareto front set, the attention score matrix is ​​dynamically adjusted to generate the multi-target feature information.

[0014] The beneficial effects of this technical solution are as follows: Introducing a Pareto attention mechanism into the multi-objective encoding layer further optimizes the trade-off decision-making ability and solution efficiency in multi-objective path planning. By initializing the multi-objective function and applying a non-dominated sorting algorithm, the system can quickly identify the Pareto front solution set, effectively balancing conflicting objectives such as path length, transportation cost, efficiency, and load balancing. Dynamically adjusting the attention weights based on the Pareto front allows the model to focus on key constraints, avoiding getting trapped in local optima. This mechanism not only improves the overall performance of path planning but also enhances the system's search capability in high-dimensional objective spaces, ensuring that the generated scheduling scheme achieves synergistic optimization across multiple indicators, adapting to changing order structures and warehouse operation needs.

[0015] In some optional implementations, the step of adjusting the sorting arm in real time based on optimal path scheduling information to achieve grabbing and delivery operations includes: Based on the optimal path scheduling information, extract the target grab coordinates, start time, target delivery coordinates, arrival time, and transportation priority corresponding to the current sorting task; Based on the target gripping coordinates and the current position of the sorting arm, calculate and update the movement trajectory of the end effector of the sorting arm to the optimal gripping position; Based on start time, arrival time, and transportation priority, the required end linear speed of the sorting arm during the delivery phase is dynamically calculated to generate delivery speed control instructions. The movement trajectory and the delivery speed control command are executed synchronously to drive the sorting arm to complete the grabbing and delivery operations in real time.

[0016] The beneficial effects of this technical solution are as follows: The sorting arm adjusts in real time, extracting key parameters based on optimal path scheduling information, such as target grasping coordinates, start time, delivery coordinates, arrival time, and transportation priority. This achieves precise synchronization of the robotic arm operation, overcoming common problems of coordinate deviation and timing misalignment in traditional sorting. This parameter extraction ensures fine-grained task decomposition, supporting a smooth transition of the arm-end actuator's path planning from the current position to the optimal grasping point. Obstacle avoidance and speed curve optimization are considered when calculating the movement trajectory, avoiding mechanical stress caused by sudden stops or overshoots. During the delivery phase, the end-effector linear velocity is dynamically calculated based on time windows and priorities to generate delivery speed control commands. This timing coordination mechanism ensures seamless connection between grasping and delivery, stabilizing the cargo's trajectory in the air and reducing the risk of landing deviation or collisions. Synchronous execution of movement trajectory and speed commands further enhances the reliability of real-time drive, improves the smoothness of arm movements, and adapts to high-frequency sorting requirements. The benefits of this process are extensive, manifesting in operational efficiency and safety: precise trajectory calculation shortens the arm's movement cycle, priority-guided optimization of multi-task parallelism reduces waiting time, and dynamic speed control adapts to cargo diversity, such as the gentle placement of fragile items, significantly reducing damage rates. Simultaneously, in dynamic warehouses, this method supports the coordination of the arm and conveyor belt, with real-time adjustments reducing system bottlenecks and improving overall throughput. Compared to manual or semi-automatic sorting, this adjustment mechanism reduces human error, enhances repeatability, and reduces energy consumption through energy-efficient trajectory optimization paths.

[0017] In some optional implementations, the monitoring and calculation of the error rate in the sorting process, if exceeding a preset range, triggers an adaptive learning module to optimize the multi-objective path optimization model, including: Real-time collection of sorting execution data, including package misdelivery events, sorting arm timeout events, and path conflict events; Calculate the overall error rate based on sorting execution data within a preset time period; Determine whether the overall error rate is within the preset range; If the overall error rate exceeds the preset range, an optimization trigger signal is generated. In response to the optimization trigger signal, the adaptive learning module is invoked to update the parameters of the multi-objective path optimization model.

[0018] The beneficial effects of this technical solution are as follows: By designing an adaptive learning module based on error rate monitoring, the system is endowed with continuous self-diagnosis and optimization capabilities. Through real-time collection of multiple sorting anomalies (such as misdelivery, timeouts, and conflicts), the system can comprehensively assess its operational status; the calculation of the comprehensive error rate index and threshold judgment enables proactive perception of performance degradation; once the error rate exceeds the limit, the adaptive learning module triggers model parameter updates, allowing the multi-objective path optimization model to make targeted adjustments based on historical error patterns. This mechanism effectively avoids the performance degradation of traditional systems caused by environmental changes or equipment drift, improves the long-term reliability and fault tolerance of the sorting system, and reduces manual intervention and maintenance costs.

[0019] Secondly, this application provides a logistics rapid sorting and conveying device, the device including a processor configured to perform the following steps: The data acquisition module acquires time-series images of the goods to obtain recognition results. The recognition results include the type of goods, the initial storage location, the destination storage location, and a three-dimensional point cloud model. The three-dimensional point cloud model is used to indicate the shape and size of the goods. The path planning module is used to obtain the optimal path scheduling information based on the identification results, order information and warehouse topology data using a multi-objective path optimization model. The sorting module is used to adjust the sorting arm in real time based on the optimal path scheduling information to achieve grabbing and delivery operations; The transportation module is used to update the speed and direction of the corresponding conveyor belt of the goods in a timely manner based on the optimal path scheduling information, the picking position of the sorting arm and the delivery speed. The monitoring and optimization module is used to monitor and calculate the error rate of the sorting process. If it exceeds the preset range, the adaptive learning module is triggered to optimize the multi-objective path optimization model.

[0020] In some optional implementations, the processor is configured to acquire time-series images of goods to obtain recognition results, the recognition results including goods type, initial acquisition of time-series images of goods to obtain recognition results, the recognition results including goods type, initial storage location, destination storage location, and a three-dimensional point cloud model, the three-dimensional point cloud model being used to indicate the shape and size of the goods; Based on the identification results, order information, and warehouse topology data, a multi-objective path optimization model is used to obtain the optimal path scheduling information. Based on optimal path scheduling information, the sorting arm is adjusted in real time to achieve grabbing and delivery operations; Based on the optimal path scheduling information, the picking position and delivery speed of the sorting arm, the speed and direction of the corresponding conveyor belt of the goods are updated simultaneously to achieve timely fixed-point sorting and transportation. Monitor and calculate the error rate of the sorting process. If it exceeds a preset range, trigger the adaptive learning module to optimize the multi-objective path optimization model, including storage location labels, destination storage location labels, and 3D point cloud models, including: Based on RFID signals and at least one frame of time-series image, obtain the cargo type, initial storage location, and target storage location of the goods; Based on subsequent temporal images, the structure-of-motion reconstruction algorithm is used to incrementally generate and optimize the 3D point cloud model of the cargo; When calculating the confidence level of the 3D point cloud model and determining whether the confidence level reaches a preset threshold, if it does, the recognition result is output.

[0021] In some optional implementations, the processor is configured to obtain optimal path scheduling information based on the identification results and order information using a multi-objective path optimization model, including: A preprocessing layer is used to perform heterogeneous processing on the recognition results, order information, and sample and topology data to obtain a unified embedding vector. A multi-objective coding layer is used to fuse multi-objective constraints on a unified embedding vector to obtain multi-objective feature information; Input multi-objective feature information into the GNN convolutional layer to calculate candidate paths and their corresponding evaluation metrics; For each candidate path and its corresponding evaluation metric, the weights of the candidate paths are dynamically adjusted using the decoding layer to obtain the optimal path scheduling information.

[0022] In some alternative implementations, the processor is configured to fuse multi-objective constraints onto a unified embedding vector using a multi-objective coding layer to obtain multi-objective feature information, wherein the multi-objective coding layer incorporates a Pareto attention mechanism, including: Receive the unified embedding vector, initialize the objective function corresponding to the unified embedding vector, the objective function is used to indicate path length, transportation efficiency, transportation cost and load balancing; A Pareto front set is generated using a nondominated sorting algorithm. The Pareto front set is used to identify nondominated solutions of the objective function. Based on the Pareto front set, the attention score matrix is ​​dynamically adjusted to generate the multi-target feature information.

[0023] In some alternative implementations, the processor is configured to adjust the sorting arm in real time based on optimal path scheduling information to perform grabbing and delivery operations, including: Based on the optimal path scheduling information, extract the target grab coordinates, start time, target delivery coordinates, arrival time, and transportation priority corresponding to the current sorting task; Based on the target gripping coordinates and the current position of the sorting arm, calculate and update the movement trajectory of the end effector of the sorting arm to the optimal gripping position; Based on start time, arrival time, and transportation priority, the required end linear speed of the sorting arm during the delivery phase is dynamically calculated to generate delivery speed control instructions. The movement trajectory and the delivery speed control command are executed synchronously to drive the sorting arm to complete the grabbing and delivery operations in real time.

[0024] In some optional implementations, the processor is configured to monitor and calculate the error rate of the sorting process in such a way that, if the error rate exceeds a preset range, an adaptive learning module is triggered to optimize the multi-objective path optimization model, including: Real-time collection of sorting execution data, including package misdelivery events, sorting arm timeout events, and path conflict events; Calculate the overall error rate based on sorting execution data within a preset time period; Determine whether the overall error rate is within the preset range; If the overall error rate exceeds the preset range, an optimization trigger signal is generated. In response to the optimization trigger signal, the adaptive learning module is invoked to update the parameters of the multi-objective path optimization model.

[0025] Thirdly, this application provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the above methods.

[0026] Fourthly, this application provides a rapid sorting and conveying system for logistics, the system comprising: The aforementioned electronic devices.

[0027] Fifthly, this application provides a chip that stores a computer program, which, when executed by a processor, implements the steps of any of the above methods. Attached Figure Description

[0028] The present application will be further described below with reference to the accompanying drawings and embodiments.

[0029] Figure 1 This illustration shows a schematic diagram of a rapid sorting and conveying process provided in an embodiment of this application.

[0030] Figure 2 A structural block diagram of a logistics rapid sorting and conveying device provided in an embodiment of this application is shown.

[0031] Figure 3A structural framework diagram of an electronic device provided in an embodiment of this application is shown.

[0032] Figure 4 A schematic diagram of the structure of a rapid sorting and conveying system for logistics provided in an embodiment of this application is shown.

[0033] Figure 5 A schematic diagram of the structure of a program product provided in an embodiment of this application is shown. Detailed Implementation

[0034] The embodiments of this application will be further described below with reference to the accompanying drawings and specific implementation methods. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new implementation methods.

[0035] In the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, a and b, a and c, b and c, a and b and c, where a, b, and c can be single or multiple. It is worth noting that "at least one" can also be interpreted as "one or more".

[0036] It should also be noted that, in the embodiments of this application, the words "exemplary" or "for example" are used to indicate that they are examples, illustrations, or descriptions. Any implementation or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other implementations or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0037] Method Implementation Examples See Figure 1 , Figure 1 A schematic flowchart of a rapid sorting and conveying method for logistics provided in an embodiment of this application is shown.

[0038] This application provides a rapid sorting and conveying method for logistics, which performs real-time sorting and conveying of goods in a warehouse based on multi-dimensional data. The method includes: Step S101: Acquire time-series images of the goods to obtain recognition results. The recognition results include the type of goods, initial storage location, destination storage location, and a three-dimensional point cloud model. The three-dimensional point cloud model is used to indicate the shape and size of the goods. Step S102: Based on the identification results, order information, and warehouse topology data, use a multi-objective path optimization model to obtain the optimal path scheduling information; Step S103: Based on the optimal path scheduling information, adjust the sorting arm in real time to achieve grabbing and delivery operations; Step S104: Based on the optimal path scheduling information, the picking position of the sorting arm and the delivery speed, the speed and direction of the conveyor belt corresponding to the goods are updated simultaneously to achieve timely fixed-point sorting and transportation. Step S105: Monitor and calculate the error rate of the sorting process. If it exceeds the preset range, trigger the adaptive learning module to optimize the multi-objective path optimization model.

[0039] Therefore, this method introduces a time-series image acquisition mechanism to acquire real-time information such as cargo type, initial storage location, destination storage location, and a 3D point cloud model. This information comprehensively captures the shape and size characteristics of the cargo, avoiding blind spots in single-dimensional identification and ensuring that sorting decisions are more aligned with actual logistics dynamics. This multi-dimensional identification not only reduces human intervention but also adapts to the diversity of goods and the complex changes in the warehouse environment.

[0040] Furthermore, a multi-objective path optimization model driven by identification results, order information, and warehouse topology data was used to achieve intelligent path scheduling. This model comprehensively considers various constraints, such as congestion levels and resource allocation, to generate optimal path scheduling information, thereby minimizing transportation delays and resource waste. In the sorting arm adjustment stage, grasping and delivery operations are executed in real time based on scheduling information, ensuring precise positioning and efficient movement of the robotic arm and avoiding mechanical collisions or delivery deviations common in traditional methods.

[0041] Simultaneously, by synchronously updating the conveyor belt speed and direction, the method achieves seamless connection of goods during the fixed-point sorting process, resulting in smoother goods flow and a significant increase in overall throughput. Furthermore, the integration of error rate monitoring and adaptive learning modules constitutes a closed-loop optimization system. The system continuously tracks abnormal indicators in the sorting process; once deviations exceed controllable limits, the learning mechanism is triggered to iteratively adjust the path model. This adaptive capability not only reduces long-term operational risks but also refines the algorithm based on historical data, improving the system's robustness and scalability.

[0042] Overall, this approach significantly improves operational reliability and economy in highly dynamic warehouse environments, reduces labor dependence, and promotes the transformation of smart logistics towards automation.

[0043] In some optional implementations, the acquisition of time-series images of goods to obtain recognition results includes the type of goods, the initial acquisition of time-series images of goods to obtain recognition results, the recognition results including the type of goods, the initial storage location, the destination storage location, and a three-dimensional point cloud model, the three-dimensional point cloud model being used to indicate the shape and size of the goods; Based on the identification results, order information, and warehouse topology data, a multi-objective path optimization model is used to obtain the optimal path scheduling information. Based on optimal path scheduling information, the sorting arm is adjusted in real time to achieve grabbing and delivery operations; Based on the optimal path scheduling information, the picking position and delivery speed of the sorting arm, the speed and direction of the corresponding conveyor belt of the goods are updated simultaneously to achieve timely fixed-point sorting and transportation. Monitor and calculate the error rate of the sorting process. If it exceeds a preset range, trigger the adaptive learning module to optimize the multi-objective path optimization model, including storage location labels, destination storage location labels, and 3D point cloud models, including: Based on RFID signals and at least one frame of time-series image, obtain the cargo type, initial storage location, and target storage location of the goods; Based on subsequent temporal images, the structure-of-motion reconstruction algorithm is used to incrementally generate and optimize the 3D point cloud model of the cargo; When calculating the confidence level of the 3D point cloud model and determining whether the confidence level reaches a preset threshold, if it does, the recognition result is output.

[0044] Therefore, by combining RFID signals with temporal image processing and using motion recovery structure algorithm to incrementally construct a 3D point cloud model, the robustness and model accuracy of the cargo identification process are further improved.

[0045] On the one hand, the multi-source fusion of RFID and visual data enhances the reliability of cargo type and warehouse location label identification, and reduces misjudgments caused by occlusion and changes in lighting. On the other hand, by incrementally generating and optimizing 3D point clouds from temporal images, the geometric features of the goods can be gradually improved during their movement, avoiding the limitations of incomplete information in a single frame image. The introduction of a confidence assessment mechanism ensures that only highly reliable recognition results are output, providing accurate input for subsequent path planning, thereby reducing the risk of sorting failures caused by recognition errors and improving the system's adaptability in dynamic environments.

[0046] In one alternative embodiment, firstly, the cargo type, initial storage location, and target storage location are obtained based on RFID signals and at least one frame of time-series image. The process is initiated by the RFID reader: when the cargo enters the reading range (approximately 2-5 meters), the reader captures the EPC (Electronic Product Code) tag signal, which pre-encodes cargo metadata, including type identification (e.g., "electronic product" or "clothing") and storage location information. The signal decoding steps are: (1) filtering noise, using RSSI (Received Signal Strength Indication) thresholding to filter weak signals; (2) resolving multi-tag collisions using an anti-collision algorithm (e.g., Q algorithm) to extract a unique EPC string; (3) querying the backend database (MongoDB) to map the EPC to cargo type (based on a classification tree, such as the CIF standard) and storage location data—the initial storage location is the storage shelf coordinates (e.g., "Aisle_01_Shelf_03"), and the target storage location is the sorting slot specified in the order (e.g., "Outbound_Bin_12"). At the same time, at least one time-series image (first frame or key frame) is captured for visual verification. If it matches the RFID type, a reread is triggered if it does not.

[0047] Secondly, based on subsequent temporal images, the structure of motion recovery (SfM) algorithm is used to incrementally generate and optimize the 3D point cloud model of the cargo. Subsequent image sequences (e.g., 5-20 frames, 33ms interval) are continuously acquired from the camera to capture the motion trajectory of the cargo on the conveyor belt. The SfM process is divided into incremental iterations: (1) Feature extraction and matching: Apply SIFT or ORB detectors to extract key points (about 500-1000 points / frame) for each frame image and calculate descriptors; use FLANN matcher to find corresponding points between frames, RANSAC removes outliers, and forms trajectory chains. (2) 3D point cloud generation: Based on bundle adjustment, minimize the reprojection error E = Σ ||p_i - π(K [R|t] X_i)||^2, where p_i is the 2D observation, π is the projection function, K is the intrinsic parameter, and X_i is the 3D point. Incremental construction: Starting with the first frame of sparse point cloud (initialized from the depth map), new points are added and triangulated frame by frame. The Levenberg-Marquardt optimizer iteratively solves for the global pose and point positions, and drift is controlled by loop closure detection (e.g., DBOW2 vocabulary tree matching for inter-frame similarity). Cargo constraints are incorporated into the optimization phase: the initial / target storage location obtained from RFID is used as an anchor point to fix the boundary point cloud; ICP is applied to align the sequence point cloud, and depth information is fused to generate a dense model (voxel downsampling to 0.01m resolution). The final output point cloud model represents the complete 3D shape of the cargo, such as the convex hull volume and pose of a package, for downstream collision avoidance.

[0048] In some optional implementations, the multi-objective path optimization model includes a preprocessing layer, a multi-objective encoding layer, a GNN convolutional layer, and a decoding layer: The process of obtaining optimal path scheduling information based on the identification results and order information using a multi-objective path optimization model includes: A preprocessing layer is used to perform heterogeneous processing on the recognition results, order information, and sample and topology data to obtain a unified embedding vector. A multi-objective coding layer is used to fuse multi-objective constraints on a unified embedding vector to obtain multi-objective feature information; Input multi-objective feature information into the GNN convolutional layer to calculate candidate paths and their corresponding evaluation metrics; For each candidate path and its corresponding evaluation metric, the weights of the candidate paths are dynamically adjusted using the decoding layer to obtain the optimal path scheduling information.

[0049] Therefore, the hierarchical architecture of the multi-objective path optimization model, including a preprocessing layer, a multi-objective encoding layer, a GNN convolutional layer, and a decoding layer, constructs an efficient path scheduling framework, solving the computational bottleneck of traditional optimization algorithms under complex warehouse topologies. The preprocessing layer performs heterogeneous processing on the recognition results, order information, and warehouse topology data to generate a unified embedding vector. This standardized fusion eliminates data format differences, ensures seamless integration of multi-source information, and avoids decision-making biases caused by information silos.

[0050] This layer's design enhances the model's input compatibility, adapting to the diverse needs of warehouses of different sizes. The multi-objective encoding layer further integrates constraints, extracts multi-objective feature information, and supports comprehensive evaluation of path planning. Subsequently, the GNN convolutional layer leverages the representational capabilities of graph neural networks to calculate candidate paths and their evaluation metrics based on warehouse topology. This graph structure modeling accurately captures dependencies between nodes, such as path intersections or resource competition, generating more realistic alternatives. The decoding layer dynamically adjusts path weights, selecting the optimal scheduling information to ensure a balance between efficiency and cost.

[0051] The end-to-end optimization of this architecture reduces iterative computation time, enabling real-time response. In practical applications, this model significantly improves the accuracy and adaptability of route planning: comprehensive evaluation of candidate routes avoids local optima traps, and the quantification of evaluation metrics guides resource allocation, reducing the incidence of transportation conflicts. Simultaneously, the convolutional mechanism of GNN enhances the model's sensitivity to dynamic changes, such as sudden order adjustments, enabling rapid route replanning and maintaining logistics continuity. Compared to traditional heuristic methods, the hierarchical design of this model improves scalability, supporting large-scale warehouse expansion without sacrificing performance. Furthermore, it promotes energy conservation by reducing unnecessary movement through optimal routes.

[0052] In an optional embodiment, firstly, a preprocessing layer is used to heterogeneously process the recognition results, order information, and sample and topological data to obtain a unified embedding vector. This layer is designed as a multimodal embedding module to process heterogeneous inputs: (1) Goods recognition result embedding: The type and warehouse location are converted into sequence vectors using an encoder and projected into a 128-dimensional space. The order information is encoded and embedded into 64 dimensions; (2) Sample data processing: Historical path samples (sequence point sets) are captured by an LSTM sequence encoder to capture temporal patterns and output dynamic embeddings; (3) Topological data embedding: The warehouse graph structure is generated by random walk to capture adjacency relationships. Heterogeneous fusion adopts an attention gating mechanism: the cross-modal similarity matrix S = softmax(QK^T / sqrt(d)) is calculated, where Q / K are query / key vectors (derived from different inputs), and weighted summation is used to generate a unified embedding vector U (256 dimensions, normalized L2 norm).

[0053] A multi-objective encoding layer is used to fuse multi-objective constraints on a unified embedding vector to obtain multi-objective feature information. The core of this layer is the Pareto attention mechanism, which is expanded into a multi-head self-attention (8 heads, 32 dimensions) after input U: (1) Multi-objective constraint injection: Define constraints such as path length, time delay, and energy consumption, and map them to the query vector through a learnable projection matrix W_c; (2) Pareto front calculation: Evaluate the non-dominated solution set for each objective function and store it as a set P; (3) Dynamically adjust attention score: Based on P, modify the standard attention formula A = softmax((QK^T + αP) / sqrt(d)), where α is an adaptive weight to highlight the score of non-dominated paths. Output multi-objective feature information M (512 dimensions), and fuse constraints such as high priority orders to increase the weight of time objectives. The mechanism is trained by gradient descent to ensure Pareto balance.

[0054] Then, multi-target feature information is input into the GNN convolutional layer to calculate the candidate paths and their corresponding evaluation metrics. The convolution process consists of 3 layers of GAT: (1) Node update: for each layer h_i^{(l+1)} = σ(∑{j∈N(i)} α{ij} W h_j^{(l)}), attention coefficient α_{ij} = LeakyReLU( (W h_i || W h_j) ), which is integrated into M as the global context; (2) Path generation: K candidate paths (K=20) are decoded from the source node (initial storage location) using Beam Search, expanding the neighborhood and pruning low-scoring branches at each step; (3) Evaluation metric calculation: a multi-dimensional metric vector is calculated for each path p, such as length L(p)=∑dist(e), delay D(p)=∑ wait_time(v), energy E(p)=∑ power_cost(e), which is predicted by MLP regression (MSE loss). Output candidate set C={p1: [L1,D1,E1], ..., pK}.

[0055] Finally, for each candidate path and its corresponding evaluation metric, the weights of the candidate paths are dynamically adjusted using the decoding layer to obtain the optimal path scheduling information. This layer is a Transformer decoder structure. After inputting C: (1) Weight calculation: The metric vector is normalized using softmax w_k = exp(β * score(ind_k)) / ∑ exp(...), and the score is obtained by multi-objective weighted sum (the weights are derived from user preferences or learning, such as time-dominant 0.6); (2) Dynamic adjustment: Reinforcement feedback is incorporated, and Q-learning updates the value function V(p) = r + γ max V(p'), where r is the immediate reward (negative error rate); (3) Path selection: Weighted sampling or argmax selects the optimal p*, and the output scheduling information S={path: [seq_nodes],metrics: [opt_L, opt_D], actions: [arm_move, agv_route]}.

[0056] In some optional implementations, the multi-objective coding layer is used to fuse multi-objective constraints on the unified embedding vector to obtain multi-objective feature information. The multi-objective coding layer incorporates a Pareto attention mechanism, including: Receive the unified embedding vector, initialize the objective function corresponding to the unified embedding vector, the objective function is used to indicate path length, transportation efficiency, transportation cost and load balancing; A Pareto front set is generated using a nondominated sorting algorithm. The Pareto front set is used to identify nondominated solutions of the objective function. Based on the Pareto front set, the attention score matrix is ​​dynamically adjusted to generate the multi-target feature information.

[0057] Therefore, a Pareto attention mechanism is introduced into the multi-objective encoding layer to further optimize the trade-off decision-making ability and solution efficiency in multi-objective path planning. By initializing the multi-objective function and applying a non-dominated sorting algorithm, the system can quickly identify the Pareto front solution set and effectively balance conflicting objectives such as path length, transportation cost, efficiency, and load balancing. Based on the Pareto front, the attention weights are dynamically adjusted to focus the model on key constraints and avoid getting trapped in local optima.

[0058] This mechanism not only improves the overall performance of path planning, but also enhances the system's search capability in high-dimensional target space, ensuring that the generated scheduling scheme achieves synergistic optimization across multiple indicators and adapts to changing order structures and warehouse operation needs.

[0059] In some alternative embodiments, the system first receives a unified embedding vector from the upstream module. This vector is formed by fusing multidimensional cargo data (such as image features, RFID tags, and path coordinates) through a Transformer encoder. It is a 512-dimensional floating-point vector representing information such as the cargo's current location, speed, and type.

[0060] In the initialization phase, the multi-objective coding layer defines four objective functions to quantify path optimization constraints. First, the path length function L is calculated as the sum of Euclidean distances: L = Σ √[(x_{i+1} - x_i)^2 + (y_{i+1} - y_i)^2], where (x_i, y_i) are the coordinates of the goods on the conveyor belt. The objective is to minimize L to shorten transport time. Second, transport efficiency E is defined as the number of goods processed per unit time: E = N / T, where N is the sorting volume and T is the response time. The objective is to maximize E to improve throughput. Third, transport cost C includes energy consumption and maintenance costs: C = α * power + β * number of collision events, where α and β are weighting coefficients. The objective is to minimize C to control operating expenses. Finally, load balancing B uses a variance metric: B = Var(load distribution), where load distribution refers to the density of goods in each conveyor segment. The objective is to minimize B to avoid local congestion. These functions are pre-trained and initialized based on historical warehouse data, stored in the model parameters, and represented by PyTorch tensors.

[0061] Next, a non-dominated sorting algorithm is applied to generate the Pareto front set. This algorithm iteratively processes the candidate path solution set corresponding to the unified embedding vector. The sorting process consists of two steps: First, the dominance relationship is calculated for each solution. If solution A is not inferior to solution B on all objectives and is superior to B on at least one objective, then A dominates B. Second, based on the crowding distance sorting, non-dominated solutions with high diversity are retained first to form the Pareto front set. This set contains 20-50 Pareto optimal solutions. For example, one solution may have a short path but high cost, while another is efficient but has uneven load. The algorithm complexity is O(MN^2), where M is the number of objectives (4) and N is the size of the solution set, which is controlled within milliseconds in a single forward propagation.

[0062] Based on the Pareto front set, the attention score matrix is ​​dynamically adjusted to generate multi-objective feature information. The attention mechanism employs multi-head attention with 8 heads. The initial score matrix A = softmax(QK^T / √d_k), where Q and K are the query and key projections of the unified embedding vector, and d_k is the dimension. Pareto adjustment introduces front weights: a score w_i = 1 / (1 + rank_i) is assigned to each front solution, where rank_i is the sorting position; then, the fusion matrix A' = A ⊙ W, where W is the Pareto weight matrix, weighted through a broadcast operation. Finally, the multi-objective feature information F = A'V + front embedding, where V is the value projection, and the front embedding is the average pooled representation of the Pareto set.

[0063] In some optional implementations, the step of adjusting the sorting arm in real time based on optimal path scheduling information to achieve grabbing and delivery operations includes: Based on the optimal path scheduling information, extract the target grab coordinates, start time, target delivery coordinates, arrival time, and transportation priority corresponding to the current sorting task; Based on the target gripping coordinates and the current position of the sorting arm, calculate and update the movement trajectory of the end effector of the sorting arm to the optimal gripping position; Based on start time, arrival time, and transportation priority, the required end linear speed of the sorting arm during the delivery phase is dynamically calculated to generate delivery speed control instructions. The movement trajectory and the delivery speed control command are executed synchronously to drive the sorting arm to complete the grabbing and delivery operations in real time.

[0064] Therefore, the sorting arm adjusts in real time, extracting key parameters based on optimal path scheduling information, such as target grasping coordinates, start time, delivery coordinates, arrival time, and transportation priority. This achieves precise synchronization of the robotic arm's operation, overcoming common problems of coordinate deviation and timing misalignment in traditional sorting. This parameter extraction ensures fine-grained task decomposition, supports a smooth transition of the arm-end actuator's path planning from the current position to the optimal grasping point, and considers obstacle avoidance and speed curve optimization when calculating the movement trajectory, avoiding mechanical stress caused by sudden stops or overshoots.

[0065] During the delivery phase, the dynamic calculation of the end-effector linear velocity based on time windows and priorities generates delivery speed control commands. This timing coordination mechanism ensures seamless connection between grasping and delivery, stabilizing the cargo's trajectory in the air and reducing the risk of landing deviations or collisions. Synchronous execution of movement trajectory and speed commands further enhances the reliability of real-time drive, improves the smoothness of arm movements, and adapts to high-frequency sorting requirements.

[0066] The benefits of this process are extensive in terms of operational efficiency and safety: precise trajectory calculation shortens the arm movement cycle, priority guidance optimizes multi-task parallelism, and reduces waiting time; the dynamic nature of speed control adapts to the diversity of goods, such as the gentle delivery of fragile items, significantly reducing the damage rate.

[0067] Meanwhile, in a dynamic warehouse, this method supports the coordination of the arm and conveyor belt, and real-time adjustments reduce system bottlenecks and improve overall throughput. Compared to manual or semi-automatic sorting, this adjustment mechanism reduces human error, enhances repeatability, and reduces energy consumption through energy-saving paths optimized for trajectory optimization.

[0068] In some optional implementations, the monitoring and calculation of the error rate in the sorting process, if exceeding a preset range, triggers an adaptive learning module to optimize the multi-objective path optimization model, including: Real-time collection of sorting execution data, including package misdelivery events, sorting arm timeout events, and path conflict events; Calculate the overall error rate based on sorting execution data within a preset time period; Determine whether the overall error rate is within the preset range; If the overall error rate exceeds the preset range, an optimization trigger signal is generated. In response to the optimization trigger signal, the adaptive learning module is invoked to update the parameters of the multi-objective path optimization model.

[0069] Therefore, by designing an adaptive learning module based on error rate monitoring, the system is endowed with continuous self-diagnosis and optimization capabilities. By collecting multiple types of sorting anomalies in real time (such as misdelivery, timeout, and conflict), the system can comprehensively assess its operational status; the calculation of the comprehensive error rate index and threshold judgment realize the proactive perception of performance degradation; once the error rate exceeds the limit, the adaptive learning module triggers the model parameter update, enabling the multi-objective path optimization model to make targeted adjustments based on historical error patterns.

[0070] This mechanism effectively avoids the performance degradation of traditional systems caused by environmental changes or equipment drift, improves the long-term reliability and fault tolerance of the sorting system, and reduces manual intervention and maintenance costs.

[0071] Device Examples See Figure 2 , Figure 2 A structural block diagram of a logistics rapid sorting and conveying device provided in an embodiment of this application is shown.

[0072] Secondly, this application provides a logistics rapid sorting and conveying device, the device including a processor configured to perform the following steps: The data acquisition module 101 acquires time-series images of goods to obtain recognition results. The recognition results include goods type, initial storage location, destination storage location, and a three-dimensional point cloud model. The three-dimensional point cloud model is used to indicate the shape and size of the goods. The path planning module 102 is used to obtain the optimal path scheduling information based on the identification results, order information and warehouse topology data using a multi-objective path optimization model. The sorting module 103 is used to adjust the sorting arm in real time based on the optimal path scheduling information to achieve grabbing and delivery operations; The transportation module 104 is used to update the speed and direction of the conveyor belt corresponding to the goods simultaneously based on the optimal path scheduling information, the picking position of the sorting arm and the delivery speed, so as to achieve timely fixed-point sorting and transportation. The monitoring and optimization module 105 is used to monitor and calculate the error rate of the sorting process. If it exceeds the preset range, the adaptive learning module is triggered to optimize the multi-objective path optimization model.

[0073] In some optional implementations, the processor is configured to acquire time-series images of goods to obtain recognition results, the recognition results including goods type, initial acquisition of time-series images of goods to obtain recognition results, the recognition results including goods type, initial storage location, destination storage location, and a three-dimensional point cloud model, the three-dimensional point cloud model being used to indicate the shape and size of the goods; Based on the identification results, order information, and warehouse topology data, a multi-objective path optimization model is used to obtain the optimal path scheduling information. Based on optimal path scheduling information, the sorting arm is adjusted in real time to achieve grabbing and delivery operations; Based on the optimal path scheduling information, the picking position and delivery speed of the sorting arm, the speed and direction of the corresponding conveyor belt of the goods are updated simultaneously to achieve timely fixed-point sorting and transportation. Monitor and calculate the error rate of the sorting process. If it exceeds a preset range, trigger the adaptive learning module to optimize the multi-objective path optimization model, including storage location labels, destination storage location labels, and 3D point cloud models, including: Based on RFID signals and at least one frame of time-series image, obtain the cargo type, initial storage location, and target storage location of the goods; Based on subsequent temporal images, the structure-of-motion reconstruction algorithm is used to incrementally generate and optimize the 3D point cloud model of the cargo; When calculating the confidence level of the 3D point cloud model and determining whether the confidence level reaches a preset threshold, if it does, the recognition result is output.

[0074] In some optional implementations, the processor is configured to obtain optimal path scheduling information based on the identification results and order information using a multi-objective path optimization model, including: A preprocessing layer is used to perform heterogeneous processing on the recognition results, order information, and sample and topology data to obtain a unified embedding vector. A multi-objective coding layer is used to fuse multi-objective constraints on a unified embedding vector to obtain multi-objective feature information; Input multi-objective feature information into the GNN convolutional layer to calculate candidate paths and their corresponding evaluation metrics; For each candidate path and its corresponding evaluation metric, the weights of the candidate paths are dynamically adjusted using the decoding layer to obtain the optimal path scheduling information.

[0075] In some alternative implementations, the processor is configured to fuse multi-objective constraints onto a unified embedding vector using a multi-objective coding layer to obtain multi-objective feature information, wherein the multi-objective coding layer incorporates a Pareto attention mechanism, including: Receive the unified embedding vector, initialize the objective function corresponding to the unified embedding vector, the objective function is used to indicate path length, transportation efficiency, transportation cost and load balancing; A Pareto front set is generated using a nondominated sorting algorithm. The Pareto front set is used to identify nondominated solutions of the objective function. Based on the Pareto front set, the attention score matrix is ​​dynamically adjusted to generate the multi-target feature information.

[0076] In some alternative implementations, the processor is configured to adjust the sorting arm in real time based on optimal path scheduling information to perform grabbing and delivery operations, including: Based on the optimal path scheduling information, extract the target grab coordinates, start time, target delivery coordinates, arrival time, and transportation priority corresponding to the current sorting task; Based on the target gripping coordinates and the current position of the sorting arm, calculate and update the movement trajectory of the end effector of the sorting arm to the optimal gripping position; Based on start time, arrival time, and transportation priority, the required end linear speed of the sorting arm during the delivery phase is dynamically calculated to generate delivery speed control instructions. The movement trajectory and the delivery speed control command are executed synchronously to drive the sorting arm to complete the grabbing and delivery operations in real time.

[0077] In some optional implementations, the processor is configured to monitor and calculate the error rate of the sorting process in such a way that, if the error rate exceeds a preset range, an adaptive learning module is triggered to optimize the multi-objective path optimization model, including: Real-time collection of sorting execution data, including package misdelivery events, sorting arm timeout events, and path conflict events; Calculate the overall error rate based on sorting execution data within a preset time period; Determine whether the overall error rate is within the preset range; If the overall error rate exceeds the preset range, an optimization trigger signal is generated. In response to the optimization trigger signal, the adaptive learning module is invoked to update the parameters of the multi-objective path optimization model.

[0078] Equipment Implementation Examples This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of any of the above methods. The specific implementation method and the technical effects achieved are the same as those described in the above method embodiments, and some contents will not be repeated.

[0079] See Figure 3 , Figure 3 A structural framework diagram of an electronic device provided in an embodiment of this application is shown.

[0080] The electronic device includes at least one memory 210, at least one processor 220, and a bus 230 connecting different platform systems.

[0081] The memory 210 may include a readable medium in the form of volatile memory, such as random access memory (RAM) 211 and / or cache memory 212, and may further include read-only memory (ROM) 213.

[0082] The memory 210 also stores a computer program, which can be executed by the processor 220 to enable the processor 220 to implement the steps of any of the above methods.

[0083] The memory 210 may also include a utility 214 having at least one program module 215, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0084] Accordingly, processor 220 can execute the aforementioned computer program, and can also execute utility 214.

[0085] The processor 220 may employ one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0086] Bus 230 can be one or more of several types of bus structures, including a memory bus or memory controller, peripheral bus, graphics acceleration port, processor, or a local bus using any bus structure with multiple bus structures.

[0087] The electronic device can also communicate with one or more external devices 240, such as a keyboard, pointing device, Bluetooth device, etc., and with one or more devices capable of interacting with the electronic device, and / or with any device that enables the electronic device to communicate with one or more other computing devices (e.g., a router, modem, etc.). This communication can be performed through input / output interface 250. Furthermore, the electronic device can communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 260. Network adapter 260 can communicate with other modules of the electronic device via bus 230. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.

[0088] System Implementation Examples See Figure 4 , Figure 4 A schematic diagram of the structure of a rapid sorting and conveying system for logistics provided in an embodiment of this application is shown.

[0089] This application embodiment also provides a rapid logistics sorting and conveying system, the system comprising: The aforementioned electronic devices.

[0090] Medium Examples This application also provides a chip that stores a computer program. When the computer program is executed by a processor, it implements the steps of any of the above methods. The specific implementation method and the technical effects achieved are the same as those described in the above method embodiments, and some details will not be repeated.

[0091] See Figure 5 , Figure 5 A schematic diagram of the structure of a program product provided in an embodiment of this application is shown.

[0092] The program product is used to implement any of the methods described above. The program product may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto. In embodiments of this application, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device. The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0093] The chip may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium capable of sending, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, or any suitable combination thereof. The program code for performing the operations of this invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C, Python, or similar programming languages. The program code may be executed entirely on a user computing device, partially on a user device, as a standalone software package, partially on a user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing devices can be connected to user computing devices via any type of network, including local area networks (LANs) or wide area networks (WANs), or they can be connected to external computing devices (e.g., via the Internet using an Internet service provider).

[0094] This application describes the invention from the perspectives of purpose, performance, progress, and novelty, and it meets the functional enhancement and use requirements emphasized by the Patent Law. The above description and drawings are merely preferred embodiments of this application and are not intended to limit this application. Therefore, all structures, devices, features, etc., that are similar to or identical to those of this application, i.e., all equivalent substitutions or modifications made in accordance with the scope of this patent application, shall fall within the scope of protection of this patent application.

Claims

1. A method for rapid sorting and conveying of logistics, characterized in that, The method for real-time sorting and transporting of goods in a warehouse based on multidimensional data includes: The time-series images of the goods are acquired to obtain recognition results, which include the goods type, initial storage location, destination storage location, and a three-dimensional point cloud model. The three-dimensional point cloud model is used to indicate the shape and size of the goods. Based on the identification results, order information, and warehouse topology data, a multi-objective path optimization model is used to obtain the optimal path scheduling information. Based on optimal path scheduling information, the sorting arm is adjusted in real time to achieve grabbing and delivery operations; Based on the optimal path scheduling information, the picking position and delivery speed of the sorting arm, the speed and direction of the corresponding conveyor belt of the goods are updated simultaneously to achieve timely fixed-point sorting and transportation. Monitor and calculate the error rate of the sorting process. If it exceeds the preset range, trigger the adaptive learning module to optimize the multi-objective path optimization model.

2. The rapid sorting and conveying method for logistics according to claim 1, characterized in that, The process involves collecting time-series images of the goods to obtain recognition results. These results include the goods type, the initial storage location, the destination storage location, and a 3D point cloud model. The 3D point cloud model is used to indicate the shape and size of the goods. Based on the identification results, order information, and warehouse topology data, a multi-objective path optimization model is used to obtain the optimal path scheduling information. Based on optimal path scheduling information, the sorting arm is adjusted in real time to achieve grabbing and delivery operations; Based on the optimal path scheduling information, the picking position and delivery speed of the sorting arm, the speed and direction of the corresponding conveyor belt of the goods are updated simultaneously to achieve timely fixed-point sorting and transportation. Monitor and calculate the error rate of the sorting process. If it exceeds a preset range, trigger the adaptive learning module to optimize the multi-objective path optimization model, including storage location labels, destination storage location labels, and 3D point cloud models, including: Based on RFID signals and at least one frame of time-series image, obtain the cargo type, initial storage location, and target storage location of the goods; Based on subsequent temporal images, the structure-of-motion reconstruction algorithm is used to incrementally generate and optimize the 3D point cloud model of the cargo; When calculating the confidence level of the 3D point cloud model and determining whether the confidence level reaches a preset threshold, if it does, the recognition result is output.

3. The rapid sorting and conveying method for logistics according to claim 1, characterized in that, The multi-objective path optimization model includes a preprocessing layer, a multi-objective encoding layer, a GNN convolutional layer, and a decoding layer. The process of obtaining optimal path scheduling information based on the identification results and order information using a multi-objective path optimization model includes: A preprocessing layer is used to perform heterogeneous processing on the recognition results, order information, and sample and topology data to obtain a unified embedding vector. A multi-objective coding layer is used to fuse multi-objective constraints on a unified embedding vector to obtain multi-objective feature information; Input multi-objective feature information into the GNN convolutional layer to calculate candidate paths and their corresponding evaluation metrics; For each candidate path and its corresponding evaluation metric, the weights of the candidate paths are dynamically adjusted using the decoding layer to obtain the optimal path scheduling information.

4. The rapid sorting and conveying method for logistics according to claim 3, characterized in that, The method utilizes a multi-objective coding layer to fuse multi-objective constraints on a unified embedding vector to obtain multi-objective feature information. This multi-objective coding layer incorporates a Pareto attention mechanism, including: Receive the unified embedding vector, initialize the objective function corresponding to the unified embedding vector, the objective function is used to indicate path length, transportation efficiency, transportation cost and load balancing; A Pareto front set is generated using a non-dominated sorting algorithm. The Pareto front set is used to identify non-dominated solutions of the objective function. Based on the Pareto front set, the attention score matrix is ​​dynamically adjusted to generate the multi-target feature information.

5. The rapid sorting and conveying method for logistics according to claim 1, characterized in that, The method of adjusting the sorting arm in real time based on optimal path scheduling information to achieve grabbing and delivery operations includes: Based on the optimal path scheduling information, extract the target grab coordinates, start time, target delivery coordinates, arrival time, and transportation priority corresponding to the current sorting task; Based on the target gripping coordinates and the current position of the sorting arm, calculate and update the movement trajectory of the end effector of the sorting arm to the optimal gripping position; Based on start time, arrival time, and transportation priority, the required end linear speed of the sorting arm during the delivery phase is dynamically calculated to generate delivery speed control instructions. The movement trajectory and the delivery speed control command are executed synchronously to drive the sorting arm to complete the grabbing and delivery operations in real time.

6. The rapid sorting and conveying method for logistics according to claim 1, characterized in that, The monitoring and calculation of the error rate in the sorting process, if exceeding a preset range, triggers an adaptive learning module to optimize the multi-objective path optimization model, including: Real-time collection of sorting execution data, including package misdelivery events, sorting arm timeout events, and path conflict events; Calculate the overall error rate based on sorting execution data within a preset time period; Determine whether the overall error rate is within the preset range; If the overall error rate exceeds the preset range, an optimization trigger signal is generated. In response to the optimization trigger signal, the adaptive learning module is invoked to update the parameters of the multi-objective path optimization model.

7. A rapid sorting and conveying device for logistics, characterized in that, The rapid sorting and conveying device for logistics includes: The data acquisition module acquires time-series images of the goods to obtain recognition results. The recognition results include the type of goods, the initial storage location, the destination storage location, and a three-dimensional point cloud model. The three-dimensional point cloud model is used to indicate the shape and size of the goods. The path planning module is used to obtain the optimal path scheduling information based on the identification results, order information and warehouse topology data using a multi-objective path optimization model. The sorting module is used to adjust the sorting arm in real time based on the optimal path scheduling information to achieve grabbing and delivery operations; The transportation module is used to update the speed and direction of the corresponding conveyor belt of the goods in a timely manner based on the optimal path scheduling information, the picking position of the sorting arm and the delivery speed. The monitoring and optimization module is used to monitor and calculate the error rate of the sorting process. If it exceeds the preset range, the adaptive learning module is triggered to optimize the multi-objective path optimization model.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program and the processor being configured to perform the steps of the method according to any one of claims 1-6.

9. A rapid sorting and conveying system for logistics, characterized in that, The rapid sorting and conveying system for logistics includes: The electronic device according to claim 8.

10. A chip, characterized in that, The chip stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1-6.

Images