Intelligent control method and system for visual storage location state recognition and task allocation

Abstract

The application discloses an intelligent control method and system for visual storage location state recognition and task allocation, belongs to the technical field of intelligent warehousing and automation control, and aims to solve the problems of low task allocation reliability and poor efficiency caused by the split of visual recognition and task allocation. The method comprises the following steps: acquiring a storage location state and a recognition confidence; generating a carrying task according to the recognition confidence; acquiring congestion heat of a target area; and comprehensively recognizing the confidence and the congestion heat to allocate a robot for the task. The system comprises a visual perception module, a cooperative control engine and a robot scheduling interface. Through cooperative control, the application can effectively avoid the wrong task allocation caused by inaccurate recognition, intelligently avoid congestion, and improve the reliability of task allocation and the system operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehousing and automated control technology, specifically to an intelligent control method and system for visual warehouse location status recognition and task allocation. Background Technology

[0002] In modern intelligent warehousing and automated production lines, mobile robots, such as automated guided vehicles (AGVs), are typically used to perform material handling tasks. Simultaneously, vision systems deployed at the work site are used to monitor the real-time usage status of storage locations, such as determining whether a location is vacant or occupied. In existing technologies, the vision recognition system responsible for status perception and the robot management system responsible for task scheduling are often two independently operating systems, lacking deep data interaction and collaborative control mechanisms.

[0003] On the one hand, while some existing vision systems can include a confidence score in their recognition results to characterize the reliability of the result, downstream robot scheduling systems often do not fully utilize this information. Scheduling systems frequently judge the validity of a recognition result based solely on a fixed threshold, ignoring the uncertainty inherent in the confidence score itself. This leads to a situation where, due to factors such as changes in lighting, partial occlusion of goods, or poor camera angles, the vision system outputs a recognition result with low confidence but is still considered valid. The scheduling system may then generate and assign handling tasks based on this unreliable data. In such cases, the robot may be incorrectly assigned to a storage location where its state has not actually changed, resulting in wasted robot resources—the so-called "empty run" phenomenon—and thus reducing the robustness and efficiency of the entire automation system.

[0004] On the other hand, while some existing robot scheduling systems consider traffic congestion to optimize routes when allocating tasks, this consideration is usually based on static map path planning or historical statistical data. This fails to reflect the dynamic "congestion intensity," determined in real-time and accurately by the robot density and the number of concurrent tasks within the work area. Therefore, during peak periods with surging order volumes, scheduling strategies cannot adaptively adjust to dynamically changing congestion conditions. This can easily lead to robots concentrating in already congested areas, exacerbating local congestion, increasing robot waiting times, and ultimately impacting the overall throughput of the warehousing system. Furthermore, the disconnect between the identification and scheduling systems, and the lack of a mechanism to correlate identification quality with allocation performance, makes problem diagnosis difficult, lacks direct data support for strategy optimization, and hinders the formation of a closed-loop continuous improvement system. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent control method and system for visual warehouse status recognition and task allocation, so as to solve the technical problems in the prior art where the visual recognition system and the robot scheduling system are isolated from each other and lack coordination, resulting in low reliability of task allocation and poor operating efficiency during peak periods.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] An intelligent control method for visual storage location status recognition and task allocation includes:

[0008] Obtain the storage location status of the target storage location and the identification confidence level corresponding to the storage location status;

[0009] Based on the identification confidence level, a transport task is generated;

[0010] Obtain the congestion heat of the area where the target storage location is located;

[0011] By combining the identification confidence level and the congestion heat level, a robot is assigned to the handling task to determine which robot will perform the handling task.

[0012] As a further aspect of the present invention, the method further includes:

[0013] When a preset recalculation trigger condition is detected, the task allocation recalculation is initiated for the already assigned transport task.

[0014] As a further aspect of the present invention: the preset recalculation triggering condition includes the congestion heat exceeding a preset threshold.

[0015] As a further aspect of the present invention: the step of generating a transport task based on the identification confidence level includes:

[0016] When the recognition confidence level is not lower than the first preset threshold, a standard handling task is generated;

[0017] When the identification confidence level is lower than the first preset threshold but higher than the second preset threshold, a transfer task to be reviewed is generated.

[0018] When the identification confidence level is not higher than the second preset threshold, no transport task is generated;

[0019] Wherein, the first preset threshold is greater than the second preset threshold.

[0020] As a further aspect of the present invention, the method further includes:

[0021] For the transport task to be reviewed, an inspection and confirmation task is automatically generated and assigned to the inspection robot for secondary confirmation.

[0022] As a further aspect of the present invention: the congestion heat is calculated based on the real-time robot density and task concurrency within the area where the target storage location is located.

[0023] As a further aspect of the present invention: the step of assigning a robot to the transport task further includes:

[0024] A time-segmented weight template is adopted, and the weight of congestion heat in task allocation decision is automatically increased during preset peak periods.

[0025] As a further aspect of the present invention, it also includes:

[0026] A visual operations and maintenance platform provides a linked dashboard that displays identification quality indicators related to the identification confidence level and allocation performance indicators related to task allocation in a linked manner. The dashboard also includes A / B testing steps, which include:

[0027] In response to the strategy verification command, the first allocation strategy and the second allocation strategy are used to simulate the allocation of the handling task, and the performance indicators corresponding to the first allocation strategy and the second allocation strategy in the simulated allocation are compared.

[0028] As a further aspect of the present invention, it also includes:

[0029] If, during the execution of a task, the robot is found to have increased congestion on its planned path, the robot will be dynamically replanned without revoking the task.

[0030] This invention also provides an intelligent control system for visual warehouse location status recognition and task allocation, comprising:

[0031] The visual perception module is used to acquire the storage location status of the target storage location and the recognition confidence level corresponding to the storage location status;

[0032] The collaborative control engine is used to generate handling tasks based on the identification confidence level, record the number of concurrent tasks, calculate the congestion heat of the area where the target storage location is located based on the real-time robot density obtained from the robot scheduling interface and the recorded number of concurrent tasks, and allocate a robot to the handling task by combining the identification confidence level and the congestion heat, so as to generate an allocation instruction including the identifier of the allocated robot.

[0033] The robot scheduling interface is used to provide the real-time robot density to the collaborative control engine, receive the allocation instruction, and communicate with the corresponding robot according to the identifier of the robot allocated in the allocation instruction.

[0034] Compared with the prior art, the present invention has at least the following beneficial effects:

[0035] (1) By using the confidence level of visual recognition as a key factor in task generation and assignment decisions, this application can effectively filter out invalid tasks caused by low confidence and uncertain recognition results, avoid the robot "running empty" or "misassigned", and significantly enhance the robustness and task execution success rate of the entire automation system.

[0036] (2) This application introduces a "congestion heat" index calculated based on real-time robot density and task concurrency, and combines it with a dynamic recalculation mechanism to enable the system to intelligently perceive and proactively avoid local congestion in the work area, thereby achieving adaptive adjustment of the system load. During peak work periods, tasks can be more reasonably guided to low-load areas or robots, effectively alleviating congestion bottlenecks and improving the overall work efficiency and operational stability of the system.

[0037] (3) This application provides clear data insights for system operation and maintenance and strategy optimization by linking "identification quality" and "allocation performance" at the data level. Operation and maintenance personnel can intuitively discover how identification problems affect allocation efficiency and make targeted adjustments accordingly. At the same time, it supports data-driven continuous optimization through A / B strategy verification and other methods, forming an effective closed-loop improvement and enhancing the intelligence level and maintainability of the system. Attached Figure Description

[0038] The invention will now be further described with reference to the accompanying drawings.

[0039] Figure 1 A schematic diagram of the structure of an intelligent control system for visual warehouse location status recognition and task allocation provided in an embodiment of this application;

[0040] Figure 2 A flowchart illustrating an intelligent control method for visual warehouse location status recognition and task allocation provided in an embodiment of this application;

[0041] Figure 3 This is a signaling interaction timing diagram of a typical task allocation process including dynamic recalculation in an embodiment of this application;

[0042] Figure 4 This is a schematic diagram of the collaborative control dashboard interface of the visualization operation and maintenance platform in this embodiment of the application.

[0043] In the diagram: 10. Visual perception module; 20. Collaborative control engine; 21. Task generation and credibility filter; 22. Multi-factor intelligent allocator; 23. Allocation recalculation and dynamic correction module; 30. Robot scheduling interface; 40. Visualized operation and maintenance platform; 50. Robot execution system. Detailed Implementation

[0044] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of an intelligent control system for visual warehouse location status recognition and task allocation provided in this embodiment of the invention. The intelligent control system for visual warehouse location status recognition and task allocation provided in this application aims to solve the problem of independent visual perception and robot scheduling, lacking coordinated linkage in the prior art. As an optional implementation, this intelligent control system can be deployed in intelligent warehousing, automated production lines, or any scenario requiring material handling via robots.

[0046] like Figure 1 As shown, the system includes: a visual perception module 10, a collaborative control engine 20, a robot scheduling interface 30, and an optional visual operation and maintenance platform 40. It should be noted that these modules work collaboratively and interact with the underlying robot execution system 50.

[0047] The visual perception module 10 is used to acquire the storage location status of the target storage location and the corresponding recognition confidence level. Specifically, the visual perception module 10 may include one or more industrial cameras, depth cameras, or other types of visual sensors deployed at locations such as warehouse shelves or production lines. These visual sensors continuously or periodically acquire image or video stream data of the storage location area. The visual perception module 10 integrates image processing and analysis algorithms, such as deep learning-based object detection or image classification models. After receiving the raw image data, the visual perception module 10 preprocesses the data, such as denoising, enhancement, and correction, and then analyzes it through the algorithm model to determine the current status of the target storage location, such as "idle," "occupied," "abnormal," or "unknown." The visual perception module 10 not only outputs the status result but also calculates a quantified recognition confidence level based on various factors such as the raw score output by the algorithm model, image clarity, lighting conditions, and the presence of occlusion. This recognition confidence level is a floating-point number between 0 and 1 or a percentage between 0% and 100%, used to characterize the system's assessment of the credibility of the recognition result. Accordingly, the visual perception module 10 transmits the data stream containing the storage location identifier, the identified storage location status, and the corresponding identification confidence level to the collaborative control engine 20 in real time or near real time.

[0048] The collaborative control engine 20, as the core decision-making unit of the entire intelligent control system, is the execution entity that implements the technical solution of this application. As an middleware layer, it bridges the upstream sensing system and the downstream execution system. The collaborative control engine 20 receives data from the visual perception module 10 and is responsible for a series of complex decision-making logics, including task generation, allocation, monitoring, and dynamic adjustment. In this embodiment, the collaborative control engine 20 may consist of multiple functional sub-modules, including but not limited to a task generation and credibility filter 21, a multi-factor intelligent allocator 22, and an allocation recalculation and dynamic correction module 23.

[0049] The task generation and confidence filter 21 is responsible for generating handling tasks based on the recognition confidence level. When a storage location status change event is received from the visual perception module 10, the task generation and confidence filter 21 does not immediately generate a task, but instead uses the recognition confidence level accompanying the event to make a judgment. For example, only when the recognition confidence level reaches a certain high threshold will a standard handling task that can be immediately assigned be generated; when the recognition confidence level is in the medium range, a lower priority task that needs further confirmation may be generated; and when the recognition confidence level is extremely low, the event may be ignored directly or the storage location status may be marked as unknown without generating any automatic handling task.

[0050] The multi-factor intelligent allocator 22, as the core decision-making unit for task allocation, comprehensively considers multiple decision factors when there are unassigned handling tasks in the system. It calculates an allocation score for each available robot for that task and ultimately selects the robot with the optimal score. Decision factors include at least the recognition confidence level provided by the visual perception module 10, and a congestion heat level characterizing the busyness of the working environment. Furthermore, decision factors may also include the robot's own state (such as battery level, whether it is carrying cargo, and its current location), task priority, the expected path cost for the robot to execute the task, and the matching degree between the robot's capabilities and task requirements. Through a configurable weighted model, the multi-factor intelligent allocator 22 can achieve highly flexible and scenario-adaptive intelligent allocation.

[0051] The allocation recalculation and dynamic correction module 23 provides the system with the ability to cope with dynamic changes and abnormal situations. The module continuously monitors the system's operating status. When it detects preset recalculation trigger conditions, such as a sharp increase in congestion heat in the target area exceeding a threshold, an anomaly reported by the robot while performing a task, or an allocated task not starting execution for an extended period, the module 23 initiates the task allocation recalculation mechanism. Specifically, the recalculation process may include reclaiming allocated but not yet executed tasks from the robot and handing them over to the multi-factor intelligent allocator 22 for reallocation based on the latest system state. In some more refined control strategies, the allocation recalculation and dynamic correction module 23 can even dynamically replan the paths of robots performing tasks to avoid sudden congestion.

[0052] The robot scheduling interface 30 is a standardized communication interface between the cooperative control engine 20 and the robot execution system 50. This interface 30 is responsible for parsing the allocation instructions (including task content, target robot identifier, path suggestions, etc.) generated by the cooperative control engine 20 and translating them into a format that the robot execution system 50 can understand before issuing them. Upon receiving the instructions, the robot execution system 50 (e.g., a fleet management system for automated guided vehicles or autonomous mobile robots) schedules specific robots to perform the transport tasks. Simultaneously, the robot scheduling interface 30 is also responsible for collecting feedback information from the robot execution system 50, such as the robot's real-time location, battery level, and task execution status (start, completion, exception), and providing this information to the cooperative control engine 20 as the data basis for its decision-making and monitoring. It is understood that the real-time robot location information reported by the robot execution system 50 is a key input for the cooperative control engine 20 to calculate the robot density within the area, thereby calculating congestion intensity.

[0053] The Visualized Operations and Maintenance Platform 40 is a human-computer interaction interface for operations and maintenance managers. It acquires and integrates various data from the Collaborative Control Engine 20, displaying the overall operational status of the system in intuitive formats such as charts, reports, and heatmaps. For example, the Visualized Operations and Maintenance Platform 40 can provide a linked dashboard that correlates identification quality indicators related to identification confidence and allocation performance indicators related to task assignment, helping operations and maintenance personnel quickly locate the root cause of problems. Furthermore, the Visualized Operations and Maintenance Platform 40 provides policy configuration functions, allowing administrators to adjust the confidence threshold for task generation, modify weights in the multi-factor allocation model, set the calculation formula for congestion heat, and recalculate the threshold. Through this platform, A / B policy verification can also be performed, continuously optimizing the system's operational strategies in a data-driven manner.

[0054] Please see Figure 2This invention also provides an intelligent control method for visual warehouse location status recognition and task allocation. The method may include the following steps:

[0055] Step S101: Obtain the storage location status and recognition confidence score. In this step, the system captures images of the target storage location using visual sensors deployed on-site, and interprets the images using a built-in analysis model to determine the current status of the storage location (e.g., idle or occupied), and provides a quantified recognition confidence score for the recognition result. This step can be performed by the visual perception module 10.

[0056] Step S102: Determine Confidence Level and Generate Task. After a state change event occurs in step S101, the task generation and confidence filter 21 in the collaborative control engine 20 determines the recognition confidence level according to preset rules. For example, the recognition confidence level is compared with one or more thresholds to decide whether to generate a high-priority standard handling task, a low-priority task to be reviewed, or temporarily not generate any task due to low confidence. It is understandable that this step is a key step in incorporating perceived uncertainty into decision-making.

[0057] Step S103: Obtain regional congestion heat. For a task to be assigned, the cooperative control engine 20 needs to assess the current busy level of the region where its target storage location is located. Congestion heat is a dynamically calculated indicator, which may be based on the number of robots currently present in the region (i.e., robot density) and the number of tasks queuing to enter or pass through the region (i.e., task concurrency). This raw data can be obtained in real time from the robot execution system 50 through the robot scheduling interface 30.

[0058] Step S104: Calculate the allocation score based on multiple factors. For a task to be assigned and a set of candidate available robots, the multi-factor intelligent allocator 22 calculates a comprehensive score for each robot. This calculation is a weighted summation process, and its input factors include at least the identification confidence obtained in step S101 (e.g., the higher the confidence, the higher the initial allocation priority or score of the task) and the congestion heat obtained in step S103 (e.g., the higher the congestion heat of the path that the robot needs to take to reach the target area, the lower its allocation score).

[0059] Step S105: Select the optimal robot and assign the task. Based on the calculation results of step S104, the system typically selects the robot with the highest assignment score as the optimal choice. The cooperative control engine 20 then generates an assignment instruction containing task details and the target robot identifier, and sends it to the robot execution system 50 through the robot scheduling interface 30, thereby determining the specific robot to perform the handling task.

[0060] Step S106: Monitor recalculation trigger conditions. After task allocation, in order to cope with dynamic changes, the allocation recalculation and dynamic correction module 23 will continuously monitor a series of preset conditions, such as whether the congestion heat of the target area exceeds the safety threshold, or whether the allocated robot fails to start executing the task within the specified time.

[0061] Step S107: Trigger recalculation of allocation. If the triggering condition is met, the process proceeds to this step. At this time, the system will take corrective measures, such as reclaiming the previously allocated task and changing its status back to "pending allocation". Subsequently, the process jumps back to step S104, and a new round of allocation score calculation and decision is made for the task based on the latest system status (including updated congestion heat, robot availability, etc.). This closed loop consisting of steps S104, S105, S106, S107, and S104 enables this method to dynamically adapt to environmental changes and achieve intelligent correction and optimization. If the recalculation condition is not triggered and the task is completed normally, the process for that task ends.

[0062] Example 1

[0063] To better understand the technical solution of this application, this embodiment will be combined with Figure 1 , Figure 2 and Figure 3 This paper details a complete application scenario of an intelligent control method for visual warehouse location status recognition and task allocation. The scenario is set in the intelligent warehousing center of a large e-commerce company, where a robotic execution system 50 composed of automated guided vehicles is deployed.

[0064] Scenario: The time is 10:00 AM, the peak period for order processing. Warehouse B, being a storage area for popular goods, experiences frequent inbound and outbound tasks and high traffic volume. The system administrator has pre-configured a peak-hour strategy template through the visual operations platform 40. This strategy template automatically increases the weight of "congestion heat" in task allocation decisions. Simultaneously, the thresholds for task generation and credibility filters 21 are set as follows: the first preset threshold (high threshold) is 95%, and the second preset threshold (low threshold) is 60%.

[0065] First, let's take a task generation event based on a medium recognition confidence level as an example. A camera deployed in area B by the visual perception module 10 captures an image of storage location B-05 showing a change, seemingly indicating that a new material bin has been placed there. Because another robot passes through the aisle at that moment, causing a momentary partial occlusion and light change at the storage location, the analysis model of the visual perception module 10, after processing, outputs a recognition result: the storage location status changes from "idle" to "occupied," but its recognition confidence level is only 85%. This result is reported to the collaborative control engine 20.

[0066] Upon receiving this event, the task generation and confidence filter 21 in the collaborative control engine 20 determines that the confidence level is 85%, which is lower than the first preset threshold of 95% but higher than the second preset threshold of 60%. According to the preset strategy, this indicates an event with moderate confidence, which may be correct or incorrect. Therefore, the system does not immediately generate a standard inbound task but instead generates a special "inbound task pending review." The priority of this inbound task pending review is automatically set to low, and on the interface of the visual operation and maintenance platform 40, storage location B-05 is highlighted with a "low confidence" alarm, prompting manual operation and maintenance to perform remote image confirmation or on-site inspection. In this way, the system successfully avoids a potential robot error or wasted trip due to inaccurate identification, improving the system's reliability.

[0067] Almost simultaneously, the system received a high-priority outbound task, requiring the retrieval of a material box from storage location B-02 in area B. This was a standard handling task, and the multi-factor intelligent allocator 22 immediately began searching for the most suitable robot to perform the task.

[0068] During the allocation decision-making process, the collaborative control engine 20 first executes step S103 to obtain the congestion heat of the target area. It queries the robot scheduling interface 30 to find that there are currently 5 robots operating in area B, and another 3 tasks are queuing to enter area B. Based on the preset formula congestion heat H = α * D_robot + β * N_task (where α and β are weighting coefficients, D_robot is the robot density, and N_task is the number of concurrent tasks), the congestion heat of area B is calculated to be 0.8 (assuming the heat range is 0 to 1), which belongs to the "high" level.

[0069] Next, proceed to step S104 to calculate and assign scores to the candidate robots. At that time, there are two candidate robots, AGV-06 and AGV-08. AGV-06 has the shortest physical path to the target storage location B-02, only 30 meters away. However, its planned path must cross the busiest central aisle in Zone B. AGV-08 is slightly further from the target storage location B-02, at 50 meters, but its location allows it to bypass the edge aisles of Zone B, where the congestion heat index is only 0.3.

[0070] The formula for calculating the allocation score of the multi-factor intelligent allocator 22 can be simplified as follows:

[0071] Score =w1*P_task-w2*D_path-w3*C_path,

[0072] Where P_task is the task priority, D_path is the path distance, C_path is the total congestion cost of the path, and w1, w2, and w3 are their respective weights. Since we are currently in a peak period, the weight w3 for congestion intensity has been automatically increased.

[0073] For AGV-06: the path distance score is relatively high (due to its short distance), but the congestion cost score is extremely high, resulting in a low overall score.

[0074] For AGV-08: the path distance score is low, but the congestion cost score is very low. After comprehensive calculation, its total score is actually the highest.

[0075] Therefore, in step S105, the system ultimately selects AGV-08 to perform this outbound task and generates an allocation instruction, which is then sent to AGV-08 through the robot scheduling interface 30. This decision demonstrates the system's ability to ensure efficiency during peak periods by sensing and avoiding congestion.

[0076] During the task execution, the dynamic recalculation mechanism was triggered. On its way to B-02, AGV-08 experienced an unexpected congestion in the central alleyway of area B (for example, a robot temporarily malfunctioned), causing the overall congestion heat index of area B to surge from 0.8 to 0.95 in a short period of time, exceeding the system's set recalculation threshold of 0.9.

[0077] The allocation recalculation and dynamic correction module 23 detects this situation in step S106 and immediately triggers allocation recalculation in step S107. This process can be referenced. Figure 3 The signaling interaction timing diagram is shown. The collaborative control engine 20 detects a congestion alarm and initiates recalculation. It first sends a signal to AGV-08 (equivalent to...) through the robot scheduling interface 30. Figure 3 The robot AGV-A sends a command to cancel the task, removing the outbound task from the task queue of AGV-08 and rolling the task status back to "pending assignment".

[0078] Subsequently, the collaborative control engine 20 immediately reassigned this high-priority outbound task. At this time, AGV-10, located on the other side of Zone B and having just completed its previous task, became available, and its path to B-02 was completely unaffected by the newly emerging congestion point. After recalculating the allocation scores for all available robots, AGV-10 (equivalent to...) Figure 3 The AGV-B robot was chosen as the new optimal option. The system then issued a new task assignment instruction to the AGV-10.

[0079] Through this series of collaborative control operations, this embodiment demonstrates how the system comprehensively utilizes identification confidence to avoid erroneous order assignment, and how it ensures the execution efficiency of high-priority tasks in complex dynamic environments by real-time perception of congestion heat, the adoption of a weighted allocation model, and a dynamic recalculation mechanism, ultimately improving the throughput and stability of the entire warehousing system.

[0080] Example 2

[0081] This embodiment is a variant of the method for handling medium-confidence events in Embodiment 1, aiming to further improve the system's automation capabilities and reduce reliance on manual intervention. This embodiment primarily extends the logic of task generation and confidence filter 21 in the collaborative control engine 20.

[0082] In this embodiment, when the system encounters the same scenario as in Embodiment 1, that is, when the visual perception module 10 identifies a change in the status of storage location B-05, but the confidence level of the identification is moderate at 85%, the task generation and confidence filter 21 no longer simply generates a low-priority task waiting for manual review.

[0083] Instead, the system will automatically perform the following series of operations:

[0084] The system will still first create an internal "task to be reviewed and entered into the database", but its status will be marked as "pending automatic confirmation" instead of "pending manual confirmation".

[0085] Next, based on the "request for verification and warehousing task", the system will automatically generate a new task type that can be executed by the robot, called "inspection and confirmation task". The target location of this task is storage location B-05, and the task content is "to perform high-precision secondary identification of the target storage location".

[0086] The "inspection and confirmation task" is submitted to the multi-factor intelligent allocator 22 for allocation. During allocation, the system strategy prioritizes specific types of robots. For example, the robot execution system 50 may include some "inspection robots" equipped with higher resolution cameras or adjustable gimbals. When calculating the allocation score, these types of robots will receive a very high capability matching score for the "inspection and confirmation task".

[0087] Suppose the system assigns this task to the inspection robot AGV-12. After receiving the task, AGV-12 will autonomously navigate to the preset "optimal observation point" in front of storage location B-05.

[0088] Upon arrival, the AGV-12 will use its built-in high-precision camera to take pictures of storage location B-05 and transmit the image data back to the visual perception module 10 or its integrated recognition unit for analysis.

[0089] The result of the secondary recognition will determine the subsequent process:

[0090] If the secondary identification result is "occupied" and the identification confidence level is higher than 95% (high threshold), then the system determines that the initial identification is basically correct. The collaborative control engine 20 will automatically update the original "pending review and warehousing task" status to "confirmed" and upgrade it to a "standard warehousing task," placing it in the normal task allocation queue to wait for the handling robot to process it.

[0091] If the secondary identification result is "idle" or any other non-"occupied" state, the system determines that the initial identification was a false alarm. The collaborative control engine 20 will cancel the original "pending review and entry task" and record an identification anomaly event in the log for subsequent analysis. At the same time, the status of storage location B-05 will remain "idle" in the system.

[0092] The solution described in this embodiment enables the system to initiate an automated closed-loop confirmation process for uncertain perceived events, transforming the original manual intervention into the autonomous behavior of the robot. This greatly accelerates the response and processing speed for uncertain events, and further improves the automation level and operational efficiency of the entire warehousing system.

[0093] Example 3

[0094] This embodiment is a deepening and variation of the congestion heat application method in Embodiment 1, demonstrating how congestion heat can be used for more refined in-transit task management, rather than just initial task allocation and recalculation. This embodiment mainly involves an advanced function of the allocation recalculation and dynamic correction module 23 in the cooperative control engine 20: dynamic path replanning.

[0095] Continuing with the scenario in Example 1: The high-priority outbound task was assigned to robot AGV-08, whose planned initial path was P1-P2-P3, in order to bypass the known congestion area at that time.

[0096] AGV-08 began traveling along path P1. However, just as it was about to complete section P1 and enter section P2, a new change occurred in the warehouse environment: another robot performing other tasks triggered an emergency stop on section P2 due to a cargo falling off, causing a sudden blockage on section P2.

[0097] The collaborative control engine 20 continuously updates the global congestion heatmap of the warehouse in real time by acquiring robot position and status information from the robot execution system 50. It immediately detected a sharp increase in robot density in the P2 section, with task transit time being extended indefinitely, causing the congestion heatmap of the P2 section to surge to an extremely high value instantly.

[0098] At this point, the "Dynamic Path Replanning" submodule in the allocation recalculation and dynamic correction module 23 is activated. This submodule continuously monitors the planned paths of all robots in transit and compares the real-time congestion intensity of each segment on the path with the robot's current position.

[0099] This submodule detected the following:

[0100] The AGV-08 robot is performing a task.

[0101] The congestion level of the next segment P2 of its planned path has exceeded the preset replanning threshold.

[0102] The robot AGV-08 has not yet entered section P2.

[0103] Based on the above judgment, the system considers path intervention for AGV-08 to be effective and necessary. Unlike the "reassignment" strategy in Example 1, which reclaims the entire task and reassigns it to other robots, this example adopts a more lightweight and efficient "replanning" strategy.

[0104] The collaborative control engine 20 performs the following operations:

[0105] It retains ownership of the AGV-08 for the outbound task, meaning the task itself is not reclaimed.

[0106] It invokes the path planning algorithm, requiring a new path to be recalculated for AGV-08 from its current location to the target storage location B-02. At the same time, in the cost function of the path calculation, the passage cost of segment P2 is set to a maximum value to force the algorithm to avoid that segment.

[0107] The path planning algorithm returns a new path, such as P1-P4-P5-P3, where P4 and P5 are alternative, currently more accessible lanes.

[0108] The collaborative control engine 20 immediately sends a "path update" command to AGV-08 through the robot scheduling interface 30. This command contains new path information.

[0109] After receiving the new route, the AGV-08's onboard control system seamlessly switched its navigation target, abandoning entry into P2 and instead heading towards P4 to continue performing its mission.

[0110] Through this dynamic path replanning mechanism, the system can help robots on the way to intelligently and in real time avoid sudden congestion without interrupting the task or increasing the overhead of relocation and scheduling. This reduces the robot's ineffective waiting time and improves the driving efficiency of a single robot and the smoothness and stability of the entire system.

[0111] Example 4

[0112] This embodiment will combine Figure 4 The focus is on how to use the visual operation and maintenance platform 40 to achieve closed-loop optimization of the allocation strategy, especially through the A / B strategy verification function, to fine-tune the weight parameters of the multi-factor intelligent allocator 22 in a data-driven manner.

[0113] Scenario Background: After a period of operation, the operations administrator discovered through the collaborative control dashboard of the visual operations platform 40 that although the system had applied a congestion-based allocation strategy, the average task waiting time in area B remained excessively long during the two extreme peak periods of 10:00-11:00 AM and 3:00-4:00 PM daily. The area congestion heatmap 403 also showed that this area remained consistently in a deep red state. The administrator suspected that the weight of the congestion heat factor in the current peak period template (assuming it is currently 0.3) was still set too low, resulting in insufficient system deflection.

[0114] In order to scientifically verify this hypothesis and find a better weight value, the administrator decided to use the A / B policy verification function provided by the Visual Operation and Maintenance Platform 40.

[0115] 1. Strategy Configuration: The administrator created a one-week A / B test plan on the platform interface.

[0116] Group A (Control Group): Maintain the existing strategy. During peak hours (10:00-11:00) on Mondays, Wednesdays, and Fridays, the weight w3 of the congestion heat factor will remain at 0.3 when the multi-factor intelligent allocator 22 calculates the allocation score.

[0117] Group B (Experimental Group): Applying a new strategy. During the peak hours of Tuesday and Thursday (10:00-11:00), the weight w3 of the congestion heat factor was increased to 0.45.

[0118] In response to this strategy verification command, the system will automatically assign actual or simulated transportation tasks from area B within the specified time period using the corresponding allocation strategy.

[0119] 2. Data collection and monitoring: During the one-week experimental period, the collaborative control engine 20 and the visual operation and maintenance platform 40 automatically collected and recorded the key performance indicators of the two groups of strategies A and B during their effective periods in the background.

[0120] 3. Results Analysis and Decision-Making: After the experiment period, the administrator opens the A / B test comparison analysis report on the collaborative control dashboard of the visual operation and maintenance platform 40. The report interface is as follows: Figure 4 As shown, the core indicators under the two strategies are clearly displayed side by side.

[0121] In performance monitoring area 402, the report shows:

[0122] The average task waiting time for Group B (weight 0.45) was 2.5 minutes, a decrease of 16.7% compared to 3.0 minutes for Group A (weight 0.3).

[0123] Group B's "average idle time of robots" decreased by 10%, indicating that the robots were being utilized more efficiently.

[0124] The average task execution path length in Group B increased by 5% compared to Group A, which is in line with expectations, because increasing the congestion weight will make the robot more inclined to take a longer route to avoid congestion.

[0125] In the quality monitoring zone 401, there were no significant differences in indicators such as the "proportion of low-confidence events" between the two groups, indicating that the strategy adjustment did not affect the upstream identification process.

[0126] By replaying the regional congestion heatmap 403 during peak hours using the timeline selector 404, it can be clearly seen that during the period when the B group strategy is in effect, the dark red area in area B is smaller and lasts for a shorter period of time.

[0127] 4. Closed-Loop Optimization: Based on the aforementioned clear data comparison, the operations administrator concluded that increasing the congestion heat weight from 0.3 to 0.45, although slightly increasing the robot's travel distance, significantly improved the system's overall task processing efficiency and throughput by more effectively managing traffic and reducing congestion waiting times. Therefore, the administrator officially updated the peak-hour weight template to 0.45 through the platform, completing one iteration of strategy optimization.

[0128] This embodiment demonstrates how the system, by providing observable data dashboards and scientific experimental tools, correlates "identification quality" with "allocation performance" and empowers operations and maintenance personnel to make data-driven decisions, thereby forming a continuous improvement closed loop of "monitoring-analysis-verification-optimization". This is a concrete manifestation of the beneficial effect of this application, which is "enhancing the interpretability and self-optimization capability of the system".

[0129] The preferred embodiments of the present invention have been described in detail above and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An intelligent control method for visual warehouse location status recognition and task allocation, characterized in that, include: Obtain the storage location status of the target storage location and the identification confidence level corresponding to the storage location status; Based on the identification confidence level, a transport task is generated; Obtain the congestion heat of the area where the target storage location is located; By combining the identification confidence level and the congestion heat level, a robot is assigned to the handling task to determine which robot will perform the handling task.

2. The intelligent control method for visual warehouse location status recognition and task allocation according to claim 1, characterized in that, The method further includes: When a preset recalculation trigger condition is detected, the task allocation recalculation is initiated for the already assigned transport task.

3. The intelligent control method for visual warehouse location status recognition and task allocation according to claim 2, characterized in that, The preset recalculation trigger condition includes the congestion heat exceeding a preset threshold.

4. The intelligent control method for visual warehouse location status recognition and task allocation according to claim 1, characterized in that, The step of generating a transport task based on the recognition confidence level includes: When the recognition confidence level is not lower than the first preset threshold, a standard handling task is generated; When the identification confidence level is lower than the first preset threshold but higher than the second preset threshold, a transfer task to be reviewed is generated. When the identification confidence level is not higher than the second preset threshold, no transport task is generated; Wherein, the first preset threshold is greater than the second preset threshold.

5. The intelligent control method for visual warehouse location status recognition and task allocation according to claim 4, characterized in that, The method further includes: For the transport task to be reviewed, an inspection and confirmation task is automatically generated and assigned to the inspection robot for secondary confirmation.

6. The intelligent control method for visual warehouse location status recognition and task allocation according to claim 1, characterized in that, The congestion heat is calculated based on the real-time robot density and task concurrency within the area where the target storage location is located.

7. The intelligent control method for visual warehouse location status recognition and task allocation according to claim 1, characterized in that, The step of assigning a robot to the transport task further includes: A time-segmented weight template is adopted, and the weight of congestion heat in task allocation decision is automatically increased during preset peak periods.

8. The intelligent control method for visual warehouse location status recognition and task allocation according to claim 1, characterized in that, Also includes: A visual operations and maintenance platform provides a linked dashboard that displays identification quality indicators related to the identification confidence level and allocation performance indicators related to task allocation in a linked manner. The dashboard also includes A / B testing steps, which include: In response to the strategy verification command, the first allocation strategy and the second allocation strategy are used to simulate the allocation of the handling task, and the performance indicators corresponding to the first allocation strategy and the second allocation strategy in the simulated allocation are compared.

9. The intelligent control method for visual warehouse location status recognition and task allocation according to claim 1, characterized in that, Also includes: If, during the execution of a task, the robot is found to have increased congestion on its planned path, the robot will be dynamically replanned without revoking the task.

10. An intelligent control system for visual warehouse location status recognition and task allocation, characterized in that, include: The visual perception module is used to acquire the storage location status of the target storage location and the recognition confidence level corresponding to the storage location status; The collaborative control engine is used to generate handling tasks based on the identification confidence level, record the number of concurrent tasks, calculate the congestion heat of the area where the target storage location is located based on the real-time robot density obtained from the robot scheduling interface and the recorded number of concurrent tasks, and allocate a robot to the handling task by combining the identification confidence level and the congestion heat, so as to generate an allocation instruction including the identifier of the allocated robot. The robot scheduling interface is used to receive the allocation instruction from the collaborative control engine regarding the real-time robot density, and to communicate with the corresponding robot based on the identifier of the robot to be allocated in the allocation instruction.

Images