Intelligent warehousing and sorting system and method based on AI

Abstract

The invention provides an AI-based intelligent storage sorting system and method, the system comprises a scheduling device, a plurality of robots, a plurality of intelligent storage devices, a workstation and an AI visual identification device, the scheduling device comprehensively analyzes order data, storage data of the intelligent storage devices and robot state data, and plans an optimal carrying path and a target execution robot; according to the planned path, the target robot accurately carries the target intelligent storage device with the structure adaptive to the narrow roadway environment; the AI visual identification device realizes multi-angle and anti-illumination interference identification on the article at the workstation, and outputs article attribute information; the work station grabbing device completes accurate grabbing in combination with attribute information, and the work station grabbing device and the order container are placed in a matched mode, and meanwhile the dispatching device synchronously achieves equipment state monitoring, fault early warning and data collection and analysis. The storage sorting efficiency and accuracy are improved, the error rate in the sorting process is reduced, and the adaptability of the system to complex environments and different articles is enhanced.

Description

Technical Field

[0001] This invention relates to the field of warehousing and logistics technology, and in particular to an AI-based intelligent warehousing and picking system and method. Background Technology

[0002] In the field of warehousing and logistics, especially in the pharmaceutical warehousing scenario, picking operations, as the core link connecting storage and order fulfillment, have extremely high requirements for efficiency, accuracy and compliance.

[0003] Currently, pharmacies and other pharmaceutical warehousing environments generally face multiple technological challenges. Traditional manual picking relies excessively on the memory and experience of operators, resulting in low overall operational efficiency. Furthermore, it is highly prone to mispicking or missing items when handling medicines with similar specifications and names. For example, in peak-hour scenarios with multiple orders processed concurrently, manual operation struggles to balance speed and accuracy, leading to order delays and decreased customer satisfaction. While some automated picking systems incorporate robotic equipment, they lack a unified intelligent scheduling mechanism. When multiple robots collaborate, path conflicts and resource contention frequently occur. Simultaneously, the system cannot dynamically adjust the task sequence based on order priority; for instance, emergency medication orders may not receive timely responses, significantly reducing system adaptability. In addition, the item identification process in existing picking systems primarily relies on single barcode or RFID scanning technology. When medicine packaging is damaged, obstructed, or placed in complex lighting or at multiple angles, the identification success rate drops significantly, causing subsequent mis-grabbing and operational interruptions. The lack of interoperability between warehousing equipment is also a prominent issue. Robots, shelves, and workstations lack real-time status monitoring and fault warning capabilities. Picking data is not effectively collected and analyzed, making process optimization difficult and unable to adapt to the special needs of pharmacy storage in narrow aisles and management of multiple types of medicines. For example, equipment movement is restricted in dense shelving environments, or there is a lack of targeted handling strategies for fragile medicines.

[0004] Therefore, how to improve the efficiency and accuracy of warehouse picking, reduce the error rate in the picking process, and enhance the system's adaptability to complex environments and different items are technical problems that urgently need to be solved by those skilled in the art. Summary of the Invention

[0005] This invention provides an AI-based intelligent warehouse picking system and method to improve warehouse picking efficiency and accuracy, reduce the error rate during the picking process, and enhance the system's adaptability to complex environments and different items.

[0006] On one hand, the present invention provides an AI-based intelligent warehouse picking system, which includes: The system includes a scheduling device, multiple robots, multiple intelligent warehousing devices, an AI visual recognition device, and workstations. The scheduling device is used to analyze order data and storage data of each smart warehousing device to determine the target smart warehousing device where the target item is located; based on the location data of the target smart warehousing device, the status data of the workstation and the status data of each robot, it determines the target robot and the handling path to perform the task and sends picking instructions to the target robot; The target robot is used to respond to the picking instruction and move the target intelligent warehousing device to the workstation according to the transport path; The AI ​​visual recognition device, located at the workstation, is used to perform target detection on items within the target intelligent warehousing device and identify the target items and their attribute information. The grasping device in the workstation is used to grasp the target item according to the attribute information of the target item and place the target item into the order container corresponding to the target item.

[0007] On the other hand, the present invention also provides an AI-based intelligent warehouse picking method, which is applied to the AI-based intelligent warehouse picking system described in any of the above claims, the method comprising: The scheduling device analyzes order data and storage data of each smart warehousing device to determine the target smart warehousing device where the target item is located; based on the location data of the target smart warehousing device, the status data of the workstation, and the status data of each robot, the target robot to perform the task and the handling path are determined, and picking instructions are sent to the target robot; The target robot responds to the picking instruction and moves the target intelligent warehousing device to the workstation according to the transport path; The AI ​​visual recognition device is used to detect the items in the target intelligent warehousing device and identify the target items and their attribute information. The grabbing device in the workstation grabs the target item according to its attribute information and places it into the order container corresponding to the target item.

[0008] The AI-based intelligent warehouse picking system and method provided by this invention comprehensively analyzes order data, intelligent warehouse device storage data, and robot status data through a scheduling device to plan the optimal handling path and target execution robot. The target robot accurately handles the target intelligent warehouse device according to the planned path, using a structure adapted to narrow aisle environments. An AI vision recognition device at the workstation achieves multi-angle, light-resistant recognition of items and outputs item attribute information. The workstation gripping device combines the attribute information to accurately grasp and match the items with the order container. Simultaneously, the scheduling device monitors equipment status, provides fault warnings, and collects and analyzes data, thereby improving warehouse picking efficiency and accuracy, reducing the error rate during the picking process, and enhancing the system's adaptability to complex environments and different items. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the structure of the AI-based intelligent warehouse picking system provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the AI-based intelligent warehouse picking method provided in this embodiment of the invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0012] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0013] Figure 1 This is a schematic diagram of the structure of the AI-based intelligent warehouse picking system provided in an embodiment of the present invention.

[0014] like Figure 1As shown, the AI-based intelligent warehouse picking system provided in this embodiment of the invention may include a scheduling device 11, multiple robots 12, multiple intelligent warehouse devices 13, a workstation 15, and an AI visual recognition device 14. The scheduling device 11 is used to analyze order data and the storage data of each intelligent warehouse device 13 to determine the target intelligent warehouse device where the target item is located. Based on the location data of the target intelligent warehouse device, the status data of the workstation 15, and the status data of each robot 12, the scheduling device determines the target robot and the handling path for performing the task and sends picking instructions to the target robot. The target robot is used to respond to the picking instructions and move the target intelligent warehouse device to the workstation 15 according to the handling path. The AI ​​visual recognition device 14, located at the workstation 15, is used to perform target detection on the items in the target intelligent warehouse device and identify the target items and their attribute information. The gripping device in the workstation 15 is used to grip the target items according to their attribute information and place them into the order container corresponding to the target items.

[0015] In practical applications, the scheduling device 11 can be understood as an integrated control unit whose main function is to achieve task allocation and path planning through data analysis. For example, the scheduling device 11 can sort orders by priority using preset rules or algorithm models and generate an optimal task allocation scheme by combining real-time collected equipment status data. Furthermore, the task allocation mechanism of the scheduling device 11 can be implemented in various ways, such as task queue management based on time windows and dynamic adjustment strategies based on load balancing, mainly to achieve efficient resource utilization and orderly task execution.

[0016] A target robot can be understood as an automated device with autonomous navigation capabilities, whose core function is to complete material handling tasks according to a predetermined path. Specifically, the target robot's path execution process can be achieved in various ways, such as using magnetic strip navigation, laser navigation, or visual navigation technology, to adapt to the needs of different warehousing environments. Furthermore, the target robot's task response mechanism can also be implemented in various ways, such as receiving instructions and feeding back status information based on wireless communication protocols, or automatically executing the next operation based on a locally stored task list.

[0017] The AI ​​visual recognition device 14 can be understood as an image processing device based on deep learning algorithms. Its main function is to achieve accurate recognition by extracting the visual features of target objects. For example, the AI ​​visual recognition device 14 can analyze the shape, color, texture, and other features of objects through a convolutional neural network (CNN) model, thereby distinguishing similar objects. Furthermore, the recognition process of the AI ​​visual recognition device 14 can be optimized in various ways, such as using multispectral imaging technology to improve recognition capabilities under complex lighting conditions, or using multi-angle shooting and image stitching technology to enhance recognition robustness.

[0018] A gripping device can be understood as a mechanical operating unit with adaptive capabilities. Its core function is to perform precise gripping based on the physical characteristics of the target object. For example, a gripping device can adapt to the characteristics of different objects by adjusting the gripping force, gripping position, and movement trajectory. Furthermore, the operating parameters of the gripping device can be determined in various ways, such as a gripping force mapping table based on the weight classification of the object, or by adjusting the gripping strategy through real-time sensor feedback.

[0019] Specifically, the scheduling device 11 analyzes order data and the storage data of each intelligent storage device 13 to determine the target intelligent storage device where the target item is located. This process effectively avoids the problem of mispicking or omission caused by human memory, and is especially suitable for complex situations in the pharmaceutical industry where drugs have similar specifications and names. Furthermore, based on the location data of the target intelligent storage device, the status data of the workstation 15, and the status data of each robot 12, the scheduling device 11 comprehensively considers the physical location of the equipment, the load status of the workstation 15, and the availability of the robot 12 to dynamically determine the target robot and the handling path for the task, thereby avoiding path conflicts and optimizing resource allocation. This solves the problems of congestion and task delay caused by the lack of unified scheduling when multiple robots are working together.

[0020] The target robot responds to the picking instructions sent by the scheduling device 11 and moves the target intelligent warehousing device to workstation 15 according to the planned handling path. This feature reduces the number of manual interventions and significantly improves the ability to process multiple orders in parallel during peak hours. The AI ​​visual recognition device 14 is located at workstation 15 and performs target detection on the items in the target intelligent warehousing device, identifying the target items and their attribute information. This process uses AI algorithms to replace traditional barcode or RFID identification, which can adapt to complex scenarios such as packaging damage, changes in lighting, and multi-angle placement, greatly improving the robustness of recognition and solving the problem of recognition failure caused by environmental interference.

[0021] Specifically, the gripping device in workstation 15 grips the target item based on its attribute information and places it into the corresponding order container. This feature, by binding the physical attributes of the item with the gripping action parameters, dynamically adapts to the item's fragility, weight, and other characteristics, avoiding damage or misalignment caused by fixed gripping methods, and ensuring the safe and accurate sorting of pharmaceutical supplies in narrow aisle storage environments. Thus, the entire system achieves seamless inter-device linkage and continuous process optimization, effectively addressing core challenges in pharmaceutical warehousing scenarios such as high reliance on manual labor, poor equipment coordination, and low recognition reliability. The AI-based intelligent warehouse picking system of the present invention uses a scheduling device 11 to comprehensively analyze order data, data stored in the intelligent warehouse device 13, and status data of the robot 12 to plan the optimal handling path and target execution robot 12. The target robot accurately handles the target intelligent warehouse device according to the planned path, with a structure adapted to narrow aisle environments. The AI ​​vision recognition device at workstation 15 realizes multi-angle, light-resistant recognition of items and outputs item attribute information. The grasping device at workstation 15 combines the attribute information to complete accurate grasping and matching with the order container for placement. At the same time, the scheduling device 11 simultaneously realizes equipment status monitoring, fault warning, and data collection and analysis, which improves the efficiency and accuracy of warehouse picking, reduces the error rate in the picking process, and enhances the system's adaptability to complex environments and different items.

[0022] In some embodiments, the present invention further proposes a scheduling device 11, including a robot 12 screening module and a path planning module. The robot 12 screening module is used to select robots 12 whose working state is idle and whose battery level is higher than a preset threshold as candidate robots 12. It performs a weighted summation of the distance factor, battery level factor, work efficiency factor, and workstation 15 matching factor for each candidate robot 12 to obtain a first comprehensive score for each robot 12, and selects the candidate robot 12 with the highest first comprehensive score as the target robot. The path planning module is used to perform path planning based on a warehouse map, generating multiple candidate paths from the current location of the target robot to the target intelligent warehousing device and then to the workstation 15; it evaluates the comprehensive cost of each candidate path, selects the candidate path with the lowest comprehensive cost as the transport path, and performs smoothing processing on the transport path.

[0023] Specifically, the robot 12 screening module is a functional unit that performs preliminary screening of robots 12 by setting conditions and determines the optimal target robot by combining a multi-dimensional evaluation mechanism. It can be implemented using a rule engine or algorithm model, such as filtering rules based on preset thresholds combined with a weighted scoring algorithm, aiming to ensure that the selected robot 12 has the basic conditions to perform the task and optimize resource allocation. The path planning module is a functional unit that generates multiple end-to-end candidate paths by analyzing the warehouse map and comprehensively evaluates and optimizes these paths. It can be implemented using graph search algorithms (such as A* algorithm and Dijkstra's algorithm) combined with multi-objective optimization methods, aiming to improve the rationality of path selection and system operating efficiency.

[0024] In detail, the robot 12 screening module first selects idle and sufficiently powered candidate robots 12 based on their working status and battery level, avoiding the risk of task interruption due to robot 12 being busy or having insufficient power. Then, a weighted summation of distance, power, work efficiency, and workstation 15 matching factors is generated to produce a first comprehensive score for each robot 12. This multi-dimensional evaluation mechanism not only considers physical distance but also integrates energy consumption levels, historical operational efficiency, and compatibility with workstation 15, thereby achieving globally optimized resource allocation. The path planning module generates a complete journey from the current location of robot 12 through the target storage device to workstation 15 based on the warehouse map, covering end-to-end path requirements. By comprehensively evaluating multiple indicators such as time, energy consumption, and congestion risk, the path with the lowest overall cost is selected as the transport path, ensuring that path selection balances efficiency and stability. Furthermore, by smoothing the transport path and optimizing sharp turns or unnecessary stops in the trajectory, the movement of robot 12 becomes smoother, reducing mechanical wear and improving system reliability and long-term operational performance.

[0025] In the above scheme, the robot selection module and the path planning module work together to form a complete scheduling mechanism. The robot selection module ensures the rationality of the target robot selection through multi-dimensional evaluation, while the path planning module improves the scientificity and feasibility of the handling path through comprehensive cost evaluation and path optimization. The combination of the two not only solves the problems of unreasonable robot selection and frequent path conflicts, but also significantly improves the overall coordination and operational efficiency of the system. In some embodiments, the present invention further proposes a specific implementation of the path planning module. The path planning module is used to calculate a first quotient of the effective travel distance of a candidate path and the baseline travel speed corresponding to the current load of the target robot; calculate a first sum of 1 and the dynamic correction coefficient; and calculate a first product of the first quotient and the first sum to obtain the estimated time cost of each candidate path; calculate a second sum of the energy consumption of the horizontal segment and the energy consumption of the slope segment; calculate a first difference between the second sum and the supplementary energy consumption; calculate a second quotient of the first difference and the transmission efficiency; and calculate a third sum of the second quotient and the start-stop loss to obtain the energy consumption cost of each candidate path; calculate a third quotient of the number of intersections of each path segment and the number of robots that the path segment can accommodate; calculate a second product of the third quotient, the intersection probability, and the avoidance redundancy coefficient to obtain the congestion risk cost of each candidate path; score the estimated time cost, energy consumption cost, and congestion risk cost; and perform a weighted summation of the respective scores to obtain the comprehensive cost.

[0026] In practical applications, the first quotient is the ratio between the effective travel distance of the candidate path and the baseline travel speed of the target robot under the current load. This can be achieved by real-time measurement of the robot's load and combining it with a preset speed-load mapping table, aiming to accurately reflect the impact of load changes on travel speed. The first sum is the sum of 1 and a dynamic correction coefficient. This can be achieved by using environmental sensors to detect ground friction or obstacle distribution, and then dynamically adjusting the correction coefficient based on historical operating data, aiming to correct the impact of environmental disturbances on time prediction. The second sum is the cumulative energy consumption of the horizontal and sloping sections of the path. This can be achieved using an energy consumption model based on terrain analysis, aiming to distinguish energy consumption differences under different terrain conditions. The third quotient is the ratio of the number of path segment intersections to the number of robots that the path segment can accommodate. This can be achieved by statistically analyzing historical path usage data and combining it with real-time monitoring information, aiming to quantify the congestion level of path bottlenecks.

[0027] Specifically, the path planning module achieves a refined assessment of overall costs through a multi-dimensional dynamic cost quantification model. First, in calculating the estimated time cost, the ratio of effective travel distance to baseline travel speed is used as a basis, combined with a dynamic correction coefficient to adjust for environmental disturbances, ensuring the accuracy of time prediction. Second, in calculating energy consumption costs, not only are the energy consumption differences between horizontal and sloping sections differentiated, but also the additional energy consumption caused by transmission efficiency losses and start-stop operations are considered, thus comprehensively covering changes in path terrain and details of mechanical operation. Third, in calculating congestion risk costs, the ratio of the number of intersections in the path segment to the number of robots that can be accommodated reflects the degree of congestion, and by combining intersection probability and avoidance redundancy coefficients, historical conflict data and safety buffers are taken into account, effectively predicting potential conflicts in multi-robot collaboration. Finally, by scoring and weighting the costs of time, energy consumption, and congestion risks, a unified benchmark for multi-objective optimization was achieved, enabling path selection to accurately adapt to real-time changes in the warehousing environment. This significantly improves the robustness of the system and order fulfillment efficiency, especially in narrow aisle environments during peak pharmaceutical warehousing periods.

[0028] Furthermore, the path planning module and the robot 12 screening module in the scheduling device 11 work together to further optimize the rationality of task allocation and path selection by combining comprehensive cost evaluation and robot 12 status screening. For example, in the scenario of prioritizing emergency medicine orders, efficient and low-conflict handling paths can be quickly generated by adjusting the time cost weight, meeting urgent needs while avoiding resource waste. This design not only solves the problem of traditional path planning relying solely on simple distance or time indicators, but also provides reliable assurance for multi-robot collaborative operations in complex warehousing scenarios.

[0029] In some embodiments, the present invention further proposes a grasping device including a grasping strategy selection module and a grasping execution module. The grasping strategy selection module is used to select the optimal grasping strategy from multiple predefined grasping strategies based on attribute information, wherein the grasping strategy includes grasping method, grasping force, grasping position, and grasping direction; the grasping execution module is used to switch execution components according to the selected grasping strategy, grasp the target object, and integrate real-time data feedback from sensors to enable the grasping strategy selection module to adjust the grasping strategy in real time.

[0030] Specifically, the grasping strategy selection module is a functional unit that makes intelligent decisions based on attribute information. It can be implemented using methods such as rule matching, machine learning models, or fuzzy logic reasoning. Its purpose is to quickly select the most suitable grasping strategy by analyzing the physical characteristics of the target item. The grasping execution module is a system combining hardware and software responsible for the specific operation. It can be implemented through the dynamic switching and collaborative control of execution components such as robotic arms, grippers, and suction cups. Its purpose is to ensure that hardware resources can flexibly adapt to the needs of different grasping strategies.

[0031] In detail, this technical solution forms a closed-loop control system through the close cooperation between the gripping strategy selection module and the gripping execution module. The gripping strategy selection module first performs quantitative analysis based on the target item's attribute information (such as fragility, weight, size, and surface material), and then selects the optimal strategy from multiple predefined gripping strategies. These strategies cover multiple dimensions, including gripping method, gripping force, gripping position, and gripping direction, thus ensuring accurate adaptation to diverse items. The gripping execution module dynamically switches execution components according to the selected strategy; for example, switching to a flexible gripper when handling fragile items, or activating a suction cup configuration when handling smooth surface items. Simultaneously, this module captures dynamic changes during the gripping process by integrating sensor feedback data (such as real-time output from force and vision sensors) and transmits this data to the gripping strategy selection module for real-time adjustments. This dynamic optimization mechanism effectively addresses challenges in complex scenarios such as contaminated pharmaceutical packaging, multi-angle placement, or identification information deviations, significantly improving gripping success rate and system reliability. In addition, the design of the grasping strategy selection module and the grasping execution module fully considers the linkage requirements with other system components such as the scheduling device 11 and the AI ​​visual recognition device 14, thereby enhancing the overall adaptability and intelligence level of the system and solving problems such as low efficiency and damage to items caused by fixed or single grasping strategies.

[0032] In some embodiments, the present invention further proposes a grasping strategy selection module, including an attribute analysis unit, a candidate strategy generation unit, a strategy scoring unit, and an optimal strategy selection unit. The attribute analysis unit performs multi-dimensional quantitative analysis on attribute information to obtain quantitative analysis results; the quantitative analysis results include fragility grading, weight grading, size classification, and surface material identification results. Based on the quantitative analysis results, the candidate strategy generation unit selects preliminarily matching grasping strategies from multiple predefined grasping strategies as candidate grasping strategies. The strategy scoring unit scores each candidate strategy on multi-dimensional indicators to obtain a second comprehensive score for each candidate strategy; the multi-dimensional indicators include safety, stability, efficiency, and energy consumption. The optimal strategy selection unit selects the grasping strategy with the highest second comprehensive score as the optimal grasping strategy and outputs the specific grasping parameters.

[0033] The attribute analysis unit refers to a module that systematically processes the attribute information of a target item, extracts key characteristics, and transforms them into structured data. This can be achieved using image recognition technology combined with machine learning algorithms, such as classifying the surface material of an item using a deep learning model, or grading weight and size using a rule engine. Its purpose is to transform vague item characteristics into quantifiable data, providing an objective basis for subsequent decision-making.

[0034] The candidate strategy generation unit can be understood as a filtering mechanism. Its core function is to quickly locate a set of strategies that match the characteristics of the target item from a predefined crawling strategy library. This can be achieved using feature vector matching algorithms, such as calculating the similarity between the feature vector of the target item and the feature vectors of each crawling strategy, thereby filtering out strategies with a matching degree higher than a preset threshold as candidates. The aim is to narrow the search range and improve the efficiency and accuracy of strategy selection.

[0035] The strategy scoring unit is a multi-dimensional evaluation tool that comprehensively assesses candidate strategies by weighted and integrating indicators such as safety, stability, efficiency, and energy consumption. It can be implemented using scoring models, such as calculating a safety score using a risk assessment model, a stability score using clamping force requirements and friction coefficient estimates, an efficiency score using action time and path complexity, and an energy consumption score using gripping force and movement distance. Its purpose is to dynamically balance various indicators to ensure a balanced strategy selection.

[0036] The optimal strategy selection unit is an intelligent decision-making tool that automatically selects the optimal strategy based on scoring results and outputs specific parameters. Its implementation can be based on ranking algorithms, such as sorting all candidate strategies in descending order of their second comprehensive scores and selecting the strategy with the highest score as the final solution. Its purpose is to make intelligent decisions based on quantitative data, avoiding the uncertainty brought about by experience-based selection.

[0037] Specifically, the aforementioned modules together constitute a complete grasping strategy selection mechanism. The attribute analysis unit first transforms the characteristics of the target item into quantitative analysis results. These results provide the candidate strategy generation unit with a screening basis, enabling it to efficiently locate a preliminary matching set of strategies. Subsequently, the strategy scoring unit performs multi-dimensional evaluation of the candidate strategies, deriving a comprehensive score through weighted fusion of various indicators, ensuring a balance between stability, efficiency, and energy consumption while prioritizing safety. Finally, the optimal strategy selection unit selects the optimal strategy based on the scoring results and outputs specific parameters, allowing the grasping device to dynamically adjust its operation according to actual needs.

[0038] This mechanism is particularly suitable for handling complex items in pharmaceutical warehousing scenarios. For example, when dealing with fragile medicines, the attribute analysis unit can accurately identify their fragility and surface material characteristics, the candidate strategy generation unit quickly selects suitable grasping strategies, and the strategy scoring unit ensures the reliability of the grasping process through a high-weight safety score. Simultaneously, this mechanism can adapt to the needs of narrow aisle storage and multi-category medicine management, significantly improving the efficiency and accuracy of picking operations, thereby effectively solving the problems of medicine damage, grasping failures, or low operational efficiency caused by inaccurate strategy selection in traditional methods.

[0039] In some embodiments, the present invention further proposes a specific implementation of the candidate strategy generation unit. The candidate strategy generation unit constructs a feature vector of the target item based on the quantitative analysis results; calculates the matching degree between the feature vector of the target item and the strategy feature vectors of each predefined grasping strategy; and selects grasping strategies with a matching degree higher than a preset threshold as candidate grasping strategies.

[0040] Specifically, feature vectors refer to the transformation of the multi-dimensional quantitative attributes of a target item into a structured mathematical representation. This can be achieved through vector space models or tensor expressions, aiming to uniformly encode item characteristics and avoid matching biases caused by attribute discretization in traditional methods. Matching degree refers to the degree of fit between the target item and a predefined grasping strategy, evaluated using mathematical algorithms. This can be achieved using cosine similarity, Euclidean distance, or other vector similarity algorithms, aiming to accurately quantify the correlation between item characteristics and strategy parameters. Preset thresholds are dynamically set lower limits for matching degree, used to filter low-matching options. These thresholds can be adjusted based on historical data statistics or experimental verification, aiming to ensure the quality and reliability of candidate strategies.

[0041] In detail, the candidate strategy generation unit constructs feature vectors for target items, transforming multi-dimensional quantitative data such as fragility classification, weight classification, size classification, and surface material identification results output by the attribute analysis unit into a unified mathematical representation, providing a foundation for subsequent matching calculations. When calculating the matching degree, mathematical methods are used to evaluate the degree of fit between the target item and the predefined grasping strategy, replacing the ambiguity of traditional empirical rule matching. This ensures that the screening process objectively reflects the intrinsic relationship between the physical properties of the drug and the strategy parameters. By setting a reasonable matching degree threshold, redundant scoring calculations caused by too many candidate strategies are avoided, and key strategies are prevented from being overlooked. This focuses on a highly reliable candidate set, significantly improving the targeting of subsequent strategy scoring and the robustness of the overall grasping operation. This vector-driven matching mechanism is particularly well-suited to the stringent requirements of pharmaceutical warehousing for safe drug grasping, effectively reducing the grasping risks caused by attribute misjudgments, while also adapting to the high-efficiency operational requirements of special scenarios such as narrow aisle storage.

[0042] Furthermore, the candidate strategy generation unit works closely with other units in the grasping strategy selection module to form a complete grasping strategy screening system. By combining the quantitative analysis results of the attribute analysis unit and the multi-dimensional index scoring of the strategy scoring unit, the entire process from initial screening to final strategy determination is optimized, significantly improving the adaptability and execution efficiency of the grasping strategy and meeting the complex and ever-changing picking needs in pharmaceutical warehousing scenarios.

[0043] In some embodiments, the present invention further proposes a strategy scoring unit, specifically used for: Safety score calculation: Based on the fragility classification of the target item, surface material identification results, and the grasping force and grasping position of the candidate grasping strategies, combined with historical grasping success rate data, the risk probability of item damage during grasping is calculated through a pre-trained risk assessment model and mapped to a safety score. The lower the risk probability, the higher the safety score. Specifically, based on the fragility classification of the target item, surface material identification results, and the grasping force and grasping position of the candidate grasping strategies, grasping records of similar items are extracted from the historical grasping database to construct a training dataset. A machine learning algorithm is used to train the risk assessment model. The risk assessment model takes the item attribute vector and the grasping strategy parameter vector as input and outputs the risk probability of item damage during grasping. The risk probability is mapped to the [0,1] interval through the sigmoid function and converted into a safety score, where the safety score = (1 - risk probability) × 100. Stability score calculation: Based on the size classification, weight classification and surface material identification results of the target item, calculate the safety factor of the theoretical clamping force and the minimum clamping force required by the candidate gripping strategy, and combine it with the estimated value of the friction coefficient of the gripping contact surface to obtain the stability score. The higher the safety factor and the friction coefficient, the higher the stability score. Efficiency score calculation: Based on the switching time of the execution components required by the candidate crawling strategy, the estimated time of the crawling action, and the estimated movement path complexity from the crawling position to the order container, the total time of a single operation of the strategy is calculated and mapped to an efficiency score. The shorter the total time, the higher the efficiency score. Calculate the energy consumption score: Based on the grasping force of the candidate grasping strategy, the movement distance and complexity of the execution component, and the weight classification of the target item, estimate the comprehensive energy consumption of the candidate strategy to perform one grasping operation, and map it to the energy consumption score. The lower the comprehensive energy consumption, the higher the energy consumption score. The four scoring items are weighted and summed to obtain the second comprehensive score for each candidate strategy, with the security score having the highest weight.

[0044] In practical applications, fragility grading refers to classifying items into different levels based on their probability of breakage under stress. This can be achieved through material strength testing or historical breakage statistics. Surface material identification results can be understood as the detection results of the physical properties of the item's surface, which can be obtained through optical or tactile sensors. Grip force and grip position are key parameters determining grip stability, and can be adjusted in real time through force and position sensors on the robotic arm's end effector. Historical grip success rate data refers to the records of past successful and failed gripping operations, aiming to provide training for risk assessment models and improve scoring accuracy. Theoretical clamping force refers to the minimum force required to grip the target item, which can be determined through mechanical simulation or experimental calibration. The estimated coefficient of friction is the quantitative result of the frictional characteristics of the gripping contact surface, which can be obtained through material comparison tables or real-time measurement. Execution component switching time refers to the switching time between different gripping tools, which can be obtained through statistical analysis of historical data from the device controller. Estimated movement path complexity refers to the difficulty of path planning from the gripping position to the order container, which can be quantified by indicators such as the number of path nodes or turns.

[0045] Specifically, the above solution effectively addresses the issues of drug damage and operational instability caused by inaccurate scoring mechanisms in pharmaceutical warehousing scenarios by constructing a multi-dimensional scoring mechanism. In safety scoring, based on the fragility classification of the target item, surface material identification results, and the gripping force and position of candidate gripping strategies, combined with historical gripping success rate data, a pre-trained risk assessment model dynamically quantifies gripping risk, ensuring the scientific rigor and adaptability of the scoring. In stability scoring, the reliability of the gripping process is ensured by combining the safety coefficient of theoretical gripping force and minimum gripping force with the estimated friction coefficient, making it particularly suitable for complex placement scenarios in narrow aisle storage environments. In efficiency scoring, the overall operational efficiency is optimized by precisely decomposing the switching time of execution components, the estimated time of gripping actions, and the complexity of movement paths, meeting the demands of parallel processing of multiple orders during peak periods. In energy consumption scoring, by comprehensively considering gripping force, movement distance and complexity, and the weight classification of the target item, a low-energy-consumption strategy is selected, extending the robot's 12-cell battery life. Finally, by weighting and summing the four scoring items of safety, stability, efficiency, and energy consumption, with safety having the highest weight, the design principle of prioritizing safety in pharmaceutical warehousing is reflected. This ensures that the integrity of medicines is prioritized when selecting strategies, avoiding the sacrifice of safety in pursuit of efficiency or energy saving, thereby comprehensively improving the reliability and compliance of the picking system.

[0046] In some embodiments, the present invention further proposes a process for calculating the stability score using a strategy scoring unit, comprising: Based on the estimated center of gravity position determined by the weight classification and size classification of the target item, the minimum theoretical clamping force required is obtained by querying a predefined clamping force mapping table. Obtain the rated clamping force that the candidate grasping strategy can provide under the current execution component configuration; calculate the fourth quotient of the rated clamping force and the minimum theoretical clamping force as the safety factor; Based on the surface material identification results, consult the material and friction coefficient comparison table to obtain the estimated static friction coefficient of the gripping contact surface; The safety factor and friction coefficient estimates are weighted and summed to obtain a fourth sum. The fourth sum is then mapped to obtain a stability score.

[0047] Specifically, the estimated center of gravity position refers to the point where the center of gravity of the target item is calculated based on its weight classification and size classification. This can be achieved through geometric modeling, physical formula derivation, or historical data fitting. The purpose of introducing the estimated center of gravity position is to avoid inaccurate clamping force calculations due to center of gravity positioning deviations, thereby improving the reliability of the gripping process. The clamping force mapping table can be a lookup table built based on historical gripping data or a physical model. Its function is to provide an objective and consistent reference for clamping force requirements for different item attributes, ensuring the scientific nature of the clamping force calculation. The rated clamping force refers to the maximum clamping capacity that the candidate gripping strategy can provide under the current execution component configuration. It can be obtained through equipment specifications or real-time sensor feedback, and its purpose is to assess clamping redundancy in conjunction with the actual state of the equipment. The safety factor is the ratio of the rated clamping force to the minimum theoretical clamping force, used to quantify clamping redundancy; a higher value indicates stronger clamping safety. The estimated static friction coefficient refers to the friction characteristic parameter obtained from a lookup table based on the surface material identification results. It can be constructed through experimental measurement or literature citation, aiming to avoid the applicability defects of a general friction coefficient on different materials. Weighted summation is a comprehensive evaluation method that assigns different weights to the estimated safety and friction coefficients, highlighting the contribution of key factors and making the scoring more closely reflect actual grasping scenarios. The mapping process standardizes the weighted summation result into comparable values, facilitating integration with other scoring items for decision-making.

[0048] In detail, the above technical solution achieves accurate calculation of stability scores through the organic combination of several key steps. First, the estimated center of gravity position is determined based on the weight and size classification of the target item. This process utilizes multi-dimensional attribute data to replace empirical estimation, providing a reliable benchmark for subsequent clamping force calculations. Second, the minimum theoretical clamping force is obtained by querying a predefined clamping force mapping table, ensuring the objectivity and consistency of clamping force requirements. Next, a safety factor is calculated by combining the rated clamping force of candidate gripping strategies. This step fully considers the real-time status of the equipment, avoiding deviations between theoretical values ​​and actual capabilities. Subsequently, an estimated static friction coefficient is obtained based on the surface material identification results and weighted and summed with the safety factor. This design highlights the contribution of key mechanical factors, making the score more consistent with actual gripping needs. Finally, the weighted summation result is mapped and transformed into a standardized stability score, facilitating integration with other scoring items for decision-making. Overall, these steps constitute a systematic stability scoring framework, significantly improving the reliability of gripping strategies for fragile items in pharmaceutical warehousing. Furthermore, this solution works closely with other functional modules in the gripping device, such as the gripping strategy selection module and the gripping execution module, to further enhance the overall performance of the system and effectively solve the problems of item damage or gripping failure caused by deviations in center of gravity estimation, ambiguity in clamping force requirements, or neglect of friction characteristics.

[0049] In some embodiments, the present invention further proposes a process for calculating the stability score by a strategy scoring unit, comprising: determining the estimated center of gravity position based on the weight classification and size classification of the target item; obtaining the required minimum theoretical clamping force by querying a predefined clamping force mapping table; obtaining the rated clamping force that the candidate gripping strategy can provide under the current execution component configuration; calculating a fourth quotient of the rated clamping force and the minimum theoretical clamping force as a safety factor; obtaining an estimated static friction coefficient of the gripping contact surface by querying a material and friction coefficient comparison table based on the surface material identification result; performing a weighted summation of the safety factor and the friction coefficient estimate to obtain a fourth sum; and mapping the fourth sum to obtain the stability score. The process of calculating the efficiency score by the strategy scoring unit includes: recording the historical average time of switching between different execution components through the controller of the grasping device, establishing a component switching time library, and obtaining the execution component switching time according to the type of execution component of the selected candidate strategy; based on the size classification of the target item and the grasping position, calculating the motion angle and speed of each joint of the grasping execution module through forward and inverse kinematics, and accumulating the motion time of each joint to obtain the estimated time of the grasping action; dividing the path from the grasping position to the order container into straight segments and turning segments, where the time of the straight segment is the fifth quotient of distance and movement speed, and the time of the turning segment is the third product of turning angle and turning coefficient, and accumulating them to obtain the path time; summing the execution component switching time, the estimated time of the grasping action, and the path time to obtain the fifth sum, and mapping the fifth sum to obtain the efficiency score. The process of calculating the energy consumption score by the strategy scoring unit includes: calculating the product of the grasping force of the candidate grasping strategy, the movement distance of the execution component, and the grasping energy consumption coefficient to obtain the grasping energy consumption; calculating the product of the target item's weight class, item mass, movement energy consumption coefficient, and the movement distance from the grasping position to the order container to obtain the movement energy consumption; summing the grasping energy consumption and the movement energy consumption to obtain the sixth sum value, and mapping the sixth sum value to obtain the energy consumption score.

[0050] Specifically, the estimated center of gravity position refers to the point where the center of gravity is located after analyzing the weight distribution and shape characteristics of the target item. This can be achieved through 3D modeling technology or an empirical model based on historical data. The purpose is to accurately reflect the physical characteristics of the item and avoid instability caused by a shift in the center of gravity. The clamping force mapping table can be understood as a standardized data table used to store the minimum clamping force requirements corresponding to different item specifications. It can be generated through experimental testing or simulation to reduce subjective errors and provide a reliable reference for clamping force. The safety factor is the ratio between the rated clamping force and the minimum theoretical clamping force. It can be optimized by dynamically adjusting weighting factors to quantify the safety margin during the gripping process. The estimated coefficient of friction is the static friction coefficient calculated based on the surface material properties of the target item. It can be obtained through experimental measurement or by querying a material database to account for the impact of packaging materials on gripping stability.

[0051] In detail, the strategy scoring unit comprehensively evaluates the performance of candidate grasping strategies through precise calculations across multiple dimensions. In stability scoring, the estimated center of gravity position is first determined based on the target item's weight and size classification. The minimum theoretical grasping force is then obtained using a gripping force mapping table, and the safety margin of the grasp is quantified by calculating a safety factor. Simultaneously, an estimated static friction coefficient is obtained using a material-friction coefficient comparison table, comprehensively considering the influence of mechanical factors during the grasping process. In efficiency scoring, a component switching time database is established using historical data recorded by the controller, making the component switching time more realistic. Joint motion time and path consumption are calculated based on forward and inverse kinematics, ensuring more accurate time estimations for grasping actions and path planning. In energy consumption scoring, grasping energy consumption and movement energy consumption are calculated separately, and the specific impact of grasping force, movement distance, and item mass on energy consumption is quantified through multi-parameter multiplication. These calculation methods not only solve the problem of inaccurate scoring, but also significantly improve the scientific nature of picking strategy selection. Especially in the pharmaceutical warehousing scenario, they can effectively cope with special needs such as the high fragility of medicines, diverse specifications, and narrow aisle operation restrictions, thereby greatly improving the accuracy and overall efficiency of picking operations.

[0052] Furthermore, the aforementioned solution, by refining the scoring calculation process, works closely with functional units such as the picking strategy selection module and the picking execution module. For example, after the attribute analysis unit completes multi-dimensional quantitative analysis, the strategy scoring unit can quickly select the optimal picking strategy based on the quantitative results, ensuring the rationality of the picking parameters and execution efficiency. This design not only enhances the system's intelligence level but also provides a reliable guarantee for efficient picking in complex warehousing environments.

[0053] Based on the same general inventive concept, this invention also protects an AI-based intelligent warehouse picking method. The AI-based intelligent warehouse picking method provided by this invention will be described below. The AI-based intelligent warehouse picking method described below can be referred to in correspondence with the AI-based intelligent warehouse picking system described above.

[0054] Figure 2 This is a flowchart illustrating an AI-based intelligent warehouse picking method provided in an embodiment of the present invention, wherein the method is applied to the control device of an AI-based intelligent warehouse picking system. Figure 2 As shown, the AI-based intelligent warehouse picking method in this embodiment includes the following steps: 201. Analyze the order data and the storage data of each intelligent warehousing device through the scheduling device to determine the target intelligent warehousing device where the target item is located; based on the location data of the target intelligent warehousing device, the status data of the workstation and the status data of each robot, determine the target robot to perform the task and the handling path, and send picking instructions to the target robot; 202. The target robot responds to the picking instruction and moves the target intelligent warehousing device to the workstation according to the transport path; 203. Using an AI visual recognition device, target detection is performed on the items within the target intelligent warehousing device to identify the target items and their attribute information; 204. Using the grasping device in the workstation, the target item is grasped according to its attribute information, and placed into the order container corresponding to the target item.

[0055] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI-based intelligent warehouse picking system, characterized in that, The scheduling device, a plurality of robots, a plurality of intelligent storage devices, an AI visual recognition device, and a workstation are included. The scheduling device is configured to analyze order data and storage data of each intelligent storage device, and determine a target intelligent storage device where a target item is located. According to the position data of the target intelligent storage device, the state data of the workstation, and the state data of each robot, a target robot for executing a task and a carrying path are determined, and a picking instruction is sent to the target robot. The target robot is configured to respond to the picking instruction and carry the target intelligent storage device to the workstation according to the carrying path. The AI visual recognition device is located at the workstation and is configured to perform target detection on items in the target intelligent storage device, and identify the target item and attribute information of the target item. The grabbing device in the workstation is configured to grab the target item according to the attribute information of the target item and place the target item into an order container corresponding to the target item. 2.The AI-based intelligent warehouse picking system according to claim 1, characterized in that, The scheduling device comprises: A robot screening module is configured to screen robots in an idle state and with a battery level higher than a preset threshold as candidate robots, weight and sum distance factors, power factors, work efficiency factors, and workstation matching factors of each candidate robot to obtain a first comprehensive score of each robot, and screen a candidate robot with the highest first comprehensive score as the target robot. A path planning module is configured to plan a path based on a warehouse map, generate a plurality of candidate paths from a current position of the target robot to the target intelligent storage device and then to the workstation, evaluate a comprehensive cost of each candidate path, select a candidate path with the lowest comprehensive cost as a carrying path, and perform smoothing processing on the carrying path. 3.The AI-based intelligent warehouse picking system according to claim 2, characterized in that, The path planning module is specifically configured to: Calculate a first quotient value of an effective moving distance of a candidate path and a reference moving speed corresponding to a current load of the target robot, a first sum value of 1 and a dynamic correction coefficient, and a first product value of the first quotient value and the first sum value to obtain an estimated time cost of each candidate path; Calculate a second sum value of a path horizontal section energy consumption and a path slope section energy consumption, a first difference value of the second sum value and a supplementary energy consumption, a second quotient value of the first difference value and a transmission efficiency, and a third sum value of the second quotient value and a start-stop loss to obtain an energy consumption cost of each candidate path Calculate a third quotient value of a number of intersections of each path segment of a candidate path and a number of robots that can be accommodated by the path segment, a second product value of the third quotient value, an intersection probability, and an avoidance redundancy coefficient to obtain a congestion risk cost of each candidate path; Score the estimated time cost, the energy consumption cost, and the congestion risk cost, and weight and sum the respective scores obtained to obtain a comprehensive cost. 4.The AI-based intelligent warehouse picking system according to claim 1, wherein, The grabbing device comprises: A grabbing strategy selection module is configured to select an optimal grabbing strategy from a plurality of predefined grabbing strategies according to the attribute information, wherein the grabbing strategy includes a grabbing method, a grabbing force, a grabbing position, and a grabbing direction. The grabbing execution module is configured to switch the execution component according to the selected grabbing strategy, perform grabbing on the target object, and integrate sensor feedback real-time data, so that the grabbing strategy selection module adjusts the grabbing strategy in real time. 5.The AI-based intelligent warehouse picking system according to claim 4, characterized in that, The grabbing strategy selection module comprises: The attribute analysis unit is configured to perform multi-dimensional quantitative analysis on the attribute information to obtain a quantitative analysis result, wherein the quantitative analysis result comprises a fragility classification, a weight classification, a size classification, and a surface material identification result. The candidate strategy generation unit is configured to select a preliminarily matched grabbing strategy from a plurality of predefined grabbing strategies as a candidate grabbing strategy based on the quantitative analysis result. The strategy scoring unit is configured to score a plurality of multi-dimensional indexes of each candidate strategy to obtain a second comprehensive score of each candidate strategy, wherein the plurality of multi-dimensional indexes comprise safety, stability, efficiency, and energy consumption. The optimal strategy selection unit is configured to select a grabbing strategy with the highest second comprehensive score as an optimal grabbing strategy and output specific grabbing parameters. 6.The AI-based intelligent warehouse picking system according to claim 5, characterized in that, The candidate strategy generation unit is specifically configured to: construct a feature vector of the target object based on the quantitative analysis result; calculate a matching degree between the feature vector of the target object and a strategy feature vector of each predefined grabbing strategy; and select a grabbing strategy with a matching degree higher than a preset threshold as a candidate grabbing strategy. 7.The AI-based intelligent warehouse picking system according to claim 5, characterized in that, The strategy scoring unit is specifically configured to: calculate a safety score based on the fragility classification, the surface material identification result of the target object, and the grabbing force and grabbing position of the candidate grabbing strategy, and combine historical grabbing success rate data to calculate a risk probability of damage to the object in the grabbing process by using a pre-trained risk assessment model, and map the risk probability to the safety score, wherein the lower the risk probability, the higher the safety score; calculate a stability score based on the size classification, the weight classification, and the surface material identification result of the target object, calculate a safety factor of a theoretical clamping force provided by the candidate grabbing strategy and a minimum required clamping force, and combine an estimated value of a friction coefficient of a grabbing contact surface to comprehensively obtain the stability score, wherein the higher the safety factor and the friction coefficient, the higher the stability score; calculate an efficiency score based on execution component switching time, grabbing action estimated time consumption, and estimated moving path complexity from the grabbing position to an order container of the candidate grabbing strategy, calculate total time consumption of a single operation of the strategy, and map the total time consumption to the efficiency score, wherein the shorter the total time consumption, the higher the efficiency score; calculate an energy consumption score based on the grabbing force, the motion distance and complexity of the execution component of the candidate grabbing strategy, and the weight classification of the target object, estimate comprehensive energy consumption of the candidate strategy for performing a grabbing operation once, and map the comprehensive energy consumption to the energy consumption score, wherein the lower the comprehensive energy consumption, the higher the energy consumption score; perform weighted summation on the four score items to obtain the second comprehensive score of each candidate strategy, wherein the safety score has the highest weight. 8.The AI-based intelligent warehouse picking system according to claim 7, characterized in that, The process of calculating the stability score by the strategy scoring unit comprises: obtaining the minimum required theoretical clamping force by querying a predefined clamping force mapping table according to the estimated center of gravity position determined based on the weight classification and the size classification of the target object; and Obtain the rated clamping force provided by the candidate grabbing strategy under the current execution component configuration; calculate the fourth quotient value of the rated clamping force and the minimum theoretical clamping force as a safety factor; According to the surface material identification result, query the material and friction coefficient table to obtain the static friction coefficient estimate value of the grabbing contact surface; The fourth sum value is obtained by weighting and summing the safety factor and the friction coefficient estimate value, and the stability score is obtained by mapping the fourth sum value. 9.The AI-based intelligent warehouse picking system according to claim 7, wherein, The process of calculating the efficiency score by the strategy scoring unit includes: Record the historical average time of switching different execution components through the controller of the grabbing device, establish a component switching time library, and obtain the execution component switching time according to the execution component type matching of the selected candidate strategy; Based on the size classification and grabbing position of the target object, calculate the joint motion angle and speed of the grabbing execution module through kinematics forward and inverse solution, and accumulate the joint motion time to obtain the estimated time consumption of grabbing action; Divide the path from the grabbing position to the order container into straight line segments and turning segments, wherein the time consumption of the straight line segment is the fifth quotient value of the distance and the moving speed, and the time consumption of the turning segment is the third product value of the turning angle and the turning coefficient, and the path time consumption is accumulated; Sum the execution component switching time, the estimated time consumption of grabbing action, and the path time consumption to obtain the fifth sum value, and map the fifth sum value to obtain the stability score; The process of calculating the energy consumption score by the strategy scoring unit includes: Calculate the product of the grabbing force of the candidate grabbing strategy, the motion distance of the execution component, and the grabbing energy consumption coefficient to obtain the grabbing energy consumption; Calculate the product of the weight classification of the target object, the object mass, the motion energy consumption coefficient, and the moving distance from the grabbing position to the order container to obtain the moving energy consumption; Sum the grabbing energy consumption and the moving energy consumption to obtain the sixth sum value, and map the sixth sum value to obtain the energy consumption score.

10. An AI-based intelligent warehouse picking method, characterized in that, The method is applied to the AI-based intelligent warehouse picking system in any one of claims 1-9, and the method comprises: Analyze the order data and the storage data of each intelligent warehouse device through the scheduling device to determine the target intelligent warehouse device where the target object is located; determine the target robot performing the task and the carrying path according to the position data of the target intelligent warehouse device, the state data of the workstation, and the state data of each robot, and send a picking instruction to the target robot; Carry the target intelligent warehouse device to the workstation according to the carrying path through the target robot in response to the picking instruction; Detect the target object and the attribute information of the target object through AI visual recognition device to detect the target object in the target intelligent warehouse device; Grab the target object according to the attribute information of the target object through the grabbing device in the workstation, and place the target object into the order container corresponding to the target object.

Images