A warehouse human-robot collaboration safety management method and system

Abstract

The application belongs to the technical field of warehouse management, and discloses a warehouse in-person machine cooperation safety management method and system, the method comprises the following steps: identifying personnel characteristics and mobile device characteristics according to environmental perception data, and determining a target combination according to the spatial correlation and motion correlation between the personnel characteristics and the mobile device characteristics; acquiring the real-time position of the target combination in the working environment map, and determining the functional area where the target combination is currently located according to the real-time position; acquiring the motion state parameters of the target combination, and determining the working intention of the target combination according to the functional area where the target combination is currently located and the motion state parameters; determining the corresponding avoidance area according to the working intention, and planning the avoidance path of the autonomous mobile robot based on the avoidance area; thereby effectively avoiding the collision risk and optimizing the avoidance behavior, and improving the safety management efficiency and accuracy of the in-warehouse person-machine cooperation.

Description

Technical Field

[0001] This application relates to the field of warehouse management technology, and more specifically, to a method and system for human-machine collaborative safety management in warehouses. Background Technology

[0002] In modern warehousing environments, automated equipment, especially autonomous mobile robots, is widely deployed to improve operational efficiency and ensure personnel safety. These robots work alongside human operators to complete tasks such as material handling and sorting. This human-robot collaboration model typically relies on a well-designed management approach, including clearly defined work areas, pre-programmed robot movement paths, and contingency planning for unexpected situations. However, in actual operation, the emergence of non-standard scenarios gradually exposes the limitations of existing management methods.

[0003] In some large automated warehouses, to improve order processing flexibility and overall throughput, warehouse management has introduced a new operating model. They've eliminated some fixed workstations, replacing them with mobile picking carts equipped with workbenches and storage space for pickers. Pickers can move more freely within the warehouse, choosing the most suitable area based on order waves, or processing multiple small batches of orders simultaneously. This change directly rendered the existing scheduling system inadequate. The robot's target point is no longer a fixed coordinate, but rather the real-time moving picking cart. The scheduling system must now dynamically plan and adjust the robot's path based on the cart's real-time location. Simultaneously, pickers are no longer confined to fixed work areas; moving picking carts through aisles has become commonplace. These two factors combined have made warehouse traffic far more complex and unpredictable than before. The frequency of encounters between robots and moving workers in main aisles has increased significantly. The original "stop and wait" safety strategy has begun to reveal serious efficiency problems. If a worker crosses the passageway, the robot queue following behind may come to a standstill, causing localized traffic congestion and significantly reducing the overall efficiency of warehouse operations.

[0004] To address the efficiency bottleneck caused by frequent obstacle avoidance, the technical team upgraded the robot's control logic. The new logic moves beyond simple stopping and waiting; it introduces a motion trend judgment mechanism. When the robot's sensors detect a worker on the path ahead, it quickly estimates the worker's likely future location based on their speed and direction. Based on this estimation, the robot attempts to calculate a safe local avoidance path in real time without stopping. This improvement has proven effective in most cases, reducing unnecessary pauses and making the robot's operation smoother.

[0005] However, after operating for a period of time, a new and more subtle problem emerged. This motion trend judgment mechanism performed well when dealing with "solo walking workers," but frequently failed when faced with "workers pushing picking carts." The system's perception and judgment model seemed only able to identify the worker as an independent moving target, failing to understand that he and the picking cart he was pushing were a unified whole. This led to two dangerous and inefficient situations. First, the system accurately judged the worker's movement trajectory and planned a tight detour, but it ignored the picking cart behind or to the side of the worker. As a result, although the robot successfully avoided the person, it scraped or even collided with the picking cart, posing a serious safety hazard. Second, when the person and cart overlapped significantly in the sensor's viewpoint, the system identified this combination as a large, irregularly shaped "obstacle" with an unpredictable movement trajectory. For safety reasons, the system plans a very wide avoidance route, with a detour distance that even exceeds the time cost of waiting in place. This makes the intelligent avoidance function, designed to improve efficiency, actually reduce efficiency. The root of the problem is that the system lacks the ability to recognize the "human-tool" combination and cannot accurately obtain the overall outline, size, and special movement limitations of this combination. For example, a person pushing a cart cannot turn or move sideways as quickly as when walking alone.

[0006] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0007] The purpose of this application is to provide a method and system for safety management of human-machine collaboration in warehouses, which can improve the efficiency and accuracy of safety management of human-machine collaboration in warehouses.

[0008] Firstly, this application provides a method for human-robot collaborative safety management in a warehouse, used to control an autonomous mobile robot. The steps of this method include:

[0009] A1. Based on environmental perception data, identify personnel characteristics and mobile device characteristics, and determine the target assembly based on the spatial and motion correlation between the personnel characteristics and the mobile device characteristics;

[0010] A2. Obtain the real-time location of the target assembly in the work environment map, and determine the functional area where the target assembly is currently located based on the real-time location;

[0011] A3. Obtain the motion state parameters of the target assembly, and determine the operational intent of the target assembly based on the current functional area of ​​the target assembly and the motion state parameters;

[0012] A4. Determine the corresponding avoidance area based on the stated task intention, and plan the avoidance path for the autonomous mobile robot based on the avoidance area.

[0013] Secondly, this application provides a warehouse human-robot collaborative safety management system for controlling autonomous mobile robots. The system includes:

[0014] The assembly recognition module is used to identify personnel features and mobile device features based on environmental perception data, and to determine the target assembly based on the spatial and motion correlation between the personnel features and the mobile device features.

[0015] The positioning module is used to obtain the real-time position of the target assembly in the work environment map, and determine the functional area where the target assembly is currently located based on the real-time position;

[0016] The task intent recognition module is used to acquire the motion state parameters of the target assembly, and determine the task intent of the target assembly based on the current functional area of ​​the target assembly and the motion state parameters.

[0017] The obstacle avoidance module is used to determine the corresponding obstacle avoidance area according to the operation intention, and to plan the obstacle avoidance path of the autonomous mobile robot based on the obstacle avoidance area.

[0018] Beneficial effects: The human-machine collaboration safety management method and system provided in this application effectively avoids collision risks and optimizes avoidance behavior by identifying the combination of personnel and mobile devices and planning avoidance paths based on their location, area and intention, thereby improving the efficiency and accuracy of human-machine collaboration safety management in warehouses. Attached Figure Description

[0019] Figure 1 A flowchart of a human-machine collaboration safety management method for warehouses provided in this application.

[0020] Figure 2 This is a schematic diagram of a human-machine collaborative safety management system for warehouses provided in this application.

[0021] Labeling Explanation: 1. Assembly Recognition Module; 2. Positioning Module; 3. Operation Intent Recognition Module; 4. Avoidance Module. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0023] 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 application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0024] Please refer to Figure 1 This application discloses a warehouse human-robot collaborative safety management method in some embodiments, used to control an autonomous mobile robot. The method includes the following steps:

[0025] A1. Based on environmental perception data, identify personnel characteristics and mobile device characteristics, and determine the target assembly based on the spatial and motion relationships between personnel characteristics and mobile device characteristics;

[0026] A2. Obtain the real-time location of the target assembly on the work environment map, and determine the functional area where the target assembly is currently located based on the real-time location;

[0027] A3. Obtain the motion state parameters of the target assembly, and determine the operational intent of the target assembly based on its current functional area and motion state parameters;

[0028] A4. Determine the corresponding avoidance area based on the task intent, and plan the avoidance path for the autonomous mobile robot based on the avoidance area.

[0029] A work environment map is a digital representation of a warehouse environment that is pre-built or updated in real time. It includes the warehouse layout, aisles, functional area divisions (such as picking areas, main aisle areas, etc.), obstacle information, etc.

[0030] This application proposes a human-robot collaborative safety management method for warehouses, used to control autonomous mobile robots. This method aims to address the problem that autonomous mobile robots cannot accurately identify and avoid combinations of people and mobile equipment in warehouse environments.

[0031] This method first identifies personnel and mobile device characteristics based on environmental perception data, and then determines the target assembly based on the spatial and motion relationships between these characteristics. In practical applications, autonomous mobile robots are typically equipped with multiple sensors, such as LiDAR and cameras. LiDAR provides precise point cloud data for identifying the 3D contours and positions of objects; cameras provide rich image information for identifying the visual features of objects. For example, when an autonomous mobile robot detects both a person and a picking cart simultaneously in a warehouse aisle using its sensors, it immediately initiates a dedicated identification program. The robot's perception system continuously scans the environment ahead. When multiple closely connected clusters of objects appear in the LiDAR point cloud data, and the visual bounding boxes of both the person and the picking cart are detected simultaneously in the camera image, the robot's internal computing unit calculates the Euclidean distance between the center point of the person's bounding box and the center point of the picking cart's bounding box. If this distance is less than a preset threshold, and the difference in the movement direction and speed of the person and the picking cart is less than a certain threshold in several consecutive frames, the system determines that they constitute a "person-cart" assembly (i.e., the target assembly). In this way, personnel and the mobile devices they operate (such as picking carts) can be identified as a whole, avoiding misjudgments or safety hazards caused by separating people from equipment.

[0032] Next, the real-time location of the target assembly on the operational environment map is obtained, and its current functional area is determined based on this location. Once the target assembly is identified, the autonomous mobile robot needs to know its specific location in the warehouse and its environmental context. For example, the robot uses its built-in localization and mapping module to map the real-time location of the "human-vehicle" assembly onto a pre-loaded warehouse area map. This map contains functional division information for different areas, such as which are "picking areas" and which are "main passageways." By comparing the assembly's location with the area boundaries on the map, the robot can accurately determine the specific functional area where the assembly is currently located. For example, if the center point coordinates of the assembly fall within the geometric area marked as "picking area" on the map, its area attribute is marked as "picking area." This helps to more accurately infer the target assembly's behavioral intentions based on the operational characteristics of different areas.

[0033] Subsequently, the motion state parameters of the target assembly are acquired, and the operational intent of the target assembly is determined based on its current functional area and motion state parameters. Simply knowing the location and area of ​​the target assembly is insufficient; it is also necessary to understand its possible behavioral intentions. For example, once the robot identifies the "human-vehicle" assembly and determines its location, it further analyzes the assembly's behavioral patterns to infer its short-term operational intent. The robot's computing unit continuously tracks the motion state of the "human-vehicle" assembly, including its instantaneous velocity, acceleration, and rate of change of direction. If the "human-vehicle" assembly is currently identified near the "picking area," and its moving speed remains below a certain threshold, or its dwell time at a certain location exceeds a preset threshold, the robot will determine its short-term operational intent as "picking." In this case, the robot will assume that the picker's next action might be irregular movement within a small range, turning in place to adjust the picking angle, or continuing to remain in place to complete the picking operation. If the human-vehicle combination is currently identified as being within the main passageway, and its movement speed is relatively stable with acceleration and rate of change of direction both below preset thresholds, the robot's intent inference module will determine its short-term operational intent as "passage." This means the picker intends to continue along the passageway, and its trajectory will be relatively smooth and predictable. By combining functional area and motion state parameters, the operational intent of the target combination can be inferred more accurately, thus providing a basis for subsequent avoidance strategies.

[0034] Finally, the corresponding avoidance zone is determined based on the task intent, and the autonomous mobile robot's avoidance path is planned based on the avoidance zone. Based on the inference of the target assembly's task intent, the autonomous mobile robot can adopt a more intelligent and efficient avoidance strategy. For example, the robot will adjust its internal prediction model and avoidance strategy according to the intent inference result. When the intent is inferred to be "picking task", the robot will predict that the assembly may move randomly or turn in place within a small area (e.g., radius 1.5 meters) around its current position. To ensure safety, the robot will set a larger "virtual safety zone" for the assembly, for example, extending its physical outline outward by 0.8 meters. When planning the avoidance path, the robot will choose a relatively spacious detour path to ensure that its own trajectory maintains a sufficient distance from this extended safety zone (e.g., a minimum safety distance of 1.5 meters). If a safe detour path cannot be planned, the robot will enter a short "waiting mode", that is, stop and continue to observe until the assembly completes the current task and its direction of movement is determined. When the intent is inferred to be "passage" (e.g., moving in a straight line on a main path), the robot predicts that the combined vehicle will maintain its current speed and direction and extrapolate linearly, resulting in a relatively smooth trajectory. In this case, the robot sets a small "virtual safety zone" for the combined vehicle, for example, extending its physical outline outward by 0.3 meters. The robot can then plan a more compact yet still safe avoidance path, for example, by fine-tuning its own speed and direction to smoothly pass behind or to the side of the combined vehicle, ensuring that its trajectory maintains a minimum safe distance (e.g., 0.8 meters) from the predicted trajectory of the combined vehicle, thus reducing unnecessary stops and detours. Throughout the avoidance process, the robot's perception, intent inference, and trajectory prediction modules operate continuously at a preset frequency (e.g., 10 times per second). This means that the robot updates its judgment of the "human-vehicle" combination's position, speed, and intent in real time and dynamically adjusts its path based on the latest information. For example, if the robot suddenly changes its intention during the avoidance process (from "passing" to "picking"), it will immediately reassess and adjust its avoidance strategy to ensure that its avoidance actions are both safe and minimize the impact on overall logistics efficiency. This intention-based avoidance strategy enables autonomous mobile robots to collaborate more flexibly and safely with human-robot assemblies.

[0035] This application introduces the concept of a "target assembly" and combines spatial and motion relationships for identification, enabling the robot to accurately treat personnel and their operated mobile devices (such as picking carts) as a whole. This solves the problem of separating people and equipment or fuzzy identification in traditional methods, thus enabling more accurate acquisition of the overall outline, size, and motion characteristics of the assembly. Furthermore, by acquiring the real-time position of the target assembly and determining its functional area, and inferring its operational intent based on motion state parameters, this application can dynamically adjust its avoidance strategy according to different operational scenarios and behavioral patterns. For example, when identifying a "picking operation intent" in the "picking operation area," the robot will reserve a larger safety area and adopt a more conservative avoidance strategy to cope with possible irregular movements of the assembly; while when identifying a "passage intent" in the "main passage area," a more compact and efficient avoidance path can be planned. This intent-based avoidance strategy significantly improves the safety and efficiency of autonomous mobile robots and human-robot assemblies in collaboration compared to traditional simple "stop and wait" or "fixed detour" strategies. It avoids unnecessary pauses and excessive detours, enabling robots to complete tasks more smoothly and intelligently while ensuring safety, thereby improving the operational efficiency of the entire warehousing system.

[0036] In some implementations, step A1 includes:

[0037] A101. Acquire environmental perception data, and identify personnel characteristics and mobile device characteristics based on the environmental perception data to obtain location data of personnel characteristics and mobile device characteristics;

[0038] A102. Based on the location data of personnel features and mobile device features, calculate the physical distance between personnel features and mobile device features, the movement speed vector of personnel features, and the movement speed vector of mobile device features;

[0039] A103. Compare the physical distance with the preset distance threshold. If the physical distance is less than the distance threshold, confirm that the spatial association condition is met.

[0040] A104. When the directional and magnitude deviations between the movement velocity vectors of personnel features and mobile device features are both within the corresponding preset consistency range, the motion association condition is confirmed to be met.

[0041] A105. When both spatial association conditions and motion association conditions are met, personnel characteristics and mobile device characteristics are marked as a target combination belonging to the same physical whole.

[0042] In step A101, environmental perception data refers to raw or preliminarily processed information about the work environment acquired through sensor systems. This data can originate from various sensing devices. For example, it can be point cloud data generated by LiDAR scanning to obtain three-dimensional geometric information and object contours of the environment; it can also be image or video data captured by cameras to identify visual features and texture information of objects; or it can be distance information acquired by ultrasonic or infrared sensors. Its purpose is to provide basic input for subsequent identification of personnel and mobile device features. Identifying personnel and mobile device features means detecting and distinguishing personnel and mobile devices from the environmental perception data using specific algorithms and models. For personnel features, deep learning-based pedestrian detection algorithms (such as YOLO, Faster R-CNN, etc.) can be used to identify the bounding boxes and pose information of personnel from image data, or clusters conforming to human body shapes can be identified from point cloud data. For mobile device features, object recognition algorithms can be used to identify the bounding boxes or contours of specific types of mobile devices (such as picking carts, forklifts, etc.), or point cloud data can be matched using a preset geometric model. The identified features typically include information such as their position, size, and shape within the sensing system's coordinate system. Location data refers to the spatial coordinates of the identified personnel and mobile device features within a specific coordinate system. This can be two-dimensional planar coordinates (e.g., x, y) or three-dimensional spatial coordinates (e.g., x, y, z), usually represented as the coordinates of the feature's center point, the corner coordinates of the bounding box, or a set of contour points. This data forms the basis for subsequent calculations of physical distance and motion status.

[0043] In step A102, physical distance refers to the straight-line distance between the person feature and the mobile device feature in space. This is typically obtained by calculating the Euclidean distance between the center points, nearest points, or bounding boxes of the two features. The calculation of physical distance is a direct basis for judging spatial correlation, reflecting the degree of proximity between the two in space. Motion velocity vector refers to the direction and magnitude of movement of the person feature or mobile device feature per unit time. Motion velocity vector can be estimated by tracking the positional changes of the feature in consecutive time frames, for example, by calculating the ratio of the displacement of the feature's center point to the time interval in adjacent frames. The velocity vector contains directional information (such as angle) and magnitude information (such as velocity value), and is a key parameter for judging motion correlation.

[0044] In step A103, the preset distance threshold refers to a pre-defined maximum permissible distance used to determine the spatial correlation between personnel characteristics and mobile device characteristics. This threshold can be empirically set or calibrated experimentally based on the actual application scenario, device size, human-machine collaboration mode, and security requirements. For example, it can be set to 0.5 meters, 0.8 meters, or 1 meter to ensure that only physically close entities are considered potentially forming a combination. The spatial correlation condition refers to the state where personnel characteristics and mobile device characteristics are spatially close enough to potentially form a physical whole. This condition is considered met when the physical distance between the two is less than the preset distance threshold.

[0045] In step A104, directional deviation refers to the angular difference between the movement velocity vector of the person feature and the movement velocity vector of the mobile device feature. This can be obtained by calculating the angle between the two velocity vectors. Directional deviation reflects the consistency of their movement directions. Magnitude deviation refers to the difference in magnitude (i.e., velocity value) between the movement velocity vector of the person feature and the movement velocity vector of the mobile device feature. This can be obtained by calculating the absolute or relative difference in magnitude of the two velocity vectors. Magnitude deviation reflects the consistency of their movement speeds. The preset consistency range refers to the allowable range set for directional deviation and magnitude deviation respectively. For example, the preset consistency range for directional deviation can be 0 to 15 degrees, indicating that the movement directions of the two are roughly the same; the preset consistency range for magnitude deviation can be 0 to 0.2 m / s, indicating that the movement speeds of the two are not significantly different. These ranges are used to quantify the synchronicity of the movement of the person and the mobile device. The motion association condition refers to the state in which the person feature and the mobile device feature exhibit a high degree of consistency in movement direction and speed, thus potentially forming a physical whole. This condition is considered satisfied when both the directional deviation and magnitude deviation of their velocity vectors are within their respective preset consistency ranges.

[0046] In step A105, the target assembly marked as belonging to the same physical whole refers to the entity into which the system logically binds personnel characteristics and mobile device characteristics into a single entity with holistic attributes after simultaneously satisfying spatial association conditions and motion association conditions. This target assembly will be treated as an indivisible unit in subsequent path planning and avoidance strategies, and its kinematic and dynamic characteristics will comprehensively consider the attributes of personnel and mobile devices.

[0047] Through the above technical solution, this application can accurately identify combinations of personnel and mobile devices in the warehousing environment, avoiding the problem of misjudging the combination as an independent obstacle or ignoring its correlation. This enables autonomous mobile robots to more accurately perceive complex human-machine interactions in the working environment, thereby optimizing the overall characteristics of the combination in subsequent obstacle avoidance path planning, effectively avoiding safety hazards such as scratches or collisions with mobile devices. At the same time, due to the more accurate identification of combinations, the robot does not need to plan overly conservative obstacle avoidance paths, reducing unnecessary detours and stops, significantly improving the operating efficiency of autonomous mobile robots and the overall smoothness of warehousing operations.

[0048] In some implementations, step A2 includes:

[0049] A201. Obtain location data of personnel and mobile device characteristics within the target assembly;

[0050] A202. Calculate the real-time location of the target assembly in the work environment map based on the location data of personnel characteristics and mobile device characteristics in the target assembly;

[0051] A203. Determine the current functional area of ​​the target assembly based on its real-time location.

[0052] This scheme aims to ensure the accuracy of location calculation and functional area determination for the target assembly, thereby solving the problem of misjudgment of functional areas caused by inaccurate location calculation. Specifically, step A201 involves acquiring location data of personnel and mobile device characteristics within the target assembly. This step aims to obtain precise spatial location information of the two core elements constituting the target assembly—personnel characteristics and mobile device characteristics. This location data forms the basis for subsequent calculations of the overall real-time location of the target assembly. Specifically, visual sensors (such as cameras) combined with image processing algorithms can be used to identify the bounding boxes or key points of personnel and mobile devices, and their coordinates in three-dimensional space can be calculated using depth information or triangulation. Alternatively, point cloud data can be acquired through LiDAR scanning, and then point cloud clusters of personnel and mobile devices can be identified through clustering algorithms and feature matching, calculating their centroids or geometric centers as location data. UWB (Ultra-Wideband) positioning systems can also be used, with UWB tags worn on personnel and mobile devices, and their precise location coordinates calculated by receiving signals from base stations.

[0053] In step A202, the real-time position of the target assembly in the work environment map is calculated based on the location data of personnel and mobile device features within the target assembly. This step utilizes the acquired location data of personnel and mobile device features to comprehensively calculate the real-time position representing the entire target assembly. This comprehensive calculation can more accurately reflect the overall spatial occupancy and motion state of the assembly, avoiding biases caused by single features. Specifically, the location data of personnel and mobile device features can be weighted and averaged, with weights set according to their physical size, importance, or relative position within the assembly, to obtain the center point position of the assembly. Alternatively, the location data of personnel and mobile device features can be used as input, and a geometric fusion algorithm (e.g., calculating the center point of their bounding rectangles or minimum bounding boxes) can be used to determine the overall real-time position of the target assembly. Furthermore, state estimation algorithms such as Kalman filtering or extended Kalman filtering can be used, with the location data of personnel and mobile devices as observations, fusing historical data and motion models to more robustly estimate the real-time position of the target assembly.

[0054] In step A203, the functional area where the target assembly is currently located is determined based on its real-time location. This step aims to match the calculated real-time location of the target assembly with functional areas in a predefined work environment map, thereby clarifying the current work environment context of the assembly. This is crucial for subsequent work intent recognition and avoidance strategy formulation. Specifically, the work environment map can be divided into multiple functional areas with clear boundaries (e.g., picking area, main aisle area, packing area, etc.). The area to which the target assembly belongs can be determined by judging whether its real-time location falls within the geometric range of a certain functional area. Alternatively, a grid-based or raster map method can be used, where each grid cell is labeled with a specific functional area attribute. The functional area can be determined by querying the attribute of the grid cell where the real-time location of the target assembly is located. Furthermore, a machine learning classifier can be used, taking the real-time location coordinates of the target assembly and surrounding environmental features (e.g., whether there are shelves nearby, aisle width, etc.) as input, and outputting its corresponding functional area.

[0055] Through the aforementioned technical solution, this application ensures a comprehensive consideration of the overall target assembly by explicitly acquiring location data of personnel and mobile device characteristics, avoiding information loss due to reliance on a single feature. By comprehensively calculating the location data of these features, a more accurate real-time position of the target assembly can be obtained, effectively overcoming the problem of inaccurate position calculation in traditional methods. Based on this high-precision real-time position, the system can accurately determine the functional area currently occupied by the target assembly, thereby eliminating the risk of misjudgment of functional areas. This accurate position and area determination provides reliable contextual information for subsequent work intention recognition and lays a solid foundation for autonomous mobile robots to plan avoidance paths, significantly improving the efficiency and reliability of safety management in human-machine collaborative environments.

[0056] In some implementations, the functional area includes at least a picking area and a main aisle area;

[0057] Step A3 includes:

[0058] A301. Obtain the motion state parameters of the target assembly within a preset time window; the motion state parameters include start / stop status and movement speed; the preset time window is a window of preset duration with the current moment as the end point;

[0059] A302. Based on the motion state parameters of the target assembly within a preset time window, calculate the average moving speed, number of starts and stops, and moving speed fluctuation of the target assembly within the preset time window.

[0060] A303. When the target assembly is located in the picking operation area, if the average moving speed is lower than the preset speed threshold or the number of starts and stops exceeds the preset number of starts and stops, the operation intention is determined to be a picking operation intention; otherwise, the operation intention is determined to be an uncertain intention.

[0061] A303. When the target assembly is located in the main passage area, if the fluctuation range of the moving speed is less than the preset range threshold, the operation intention is determined to be the passage intention; otherwise, the operation intention is determined to be the uncertain intention.

[0062] The functional areas mentioned in this application refer to specific physical spaces within a warehousing environment that are divided according to their primary operational type or passage purpose.

[0063] In step A301, the preset time window refers to a continuous time period with a fixed duration, ending at the current moment. It is used to collect and analyze the motion state data of the target assembly within this time period to reflect its recent behavioral patterns. The introduction of the preset time window ensures that the statistics of motion state parameters are based on recent behavior, reflecting the dynamic changes of the target assembly in a timely manner and avoiding the use of outdated data that could lead to delayed or inaccurate intent judgments. Motion state parameters are quantitative indicators describing the motion characteristics of the target assembly over a period of time. Start-stop status reflects whether the target assembly is in motion or stationary, while movement speed quantifies the speed of its motion. Start-stop status can be determined by whether the change in position within a continuous time period is less than a certain minimum threshold; for example, if the displacement is less than 0.05 meters within 0.5 seconds, it is considered stationary. Movement speed can be calculated using the time difference of continuous position data. Another implementation method is to directly acquire acceleration data through an inertial measurement unit (IMU), determining stationary when the acceleration is close to zero for an extended period, and obtaining the velocity by integrating the acceleration. These parameters are the basic data for identifying the behavioral patterns of target assemblies. They are directly related to the typical characteristics of picking and passage operations, providing raw input for subsequent statistical analysis.

[0064] In step A302, the average movement speed, number of starts and stops, and movement speed fluctuation amplitude are quantitative indicators obtained from the statistics of the original motion state parameters within a preset time window. The average movement speed reflects the overall speed of the target assembly within that time period; the number of starts and stops reflects the degree of intermittency or frequent pauses in its movement; and the movement speed fluctuation amplitude reflects the smoothness or instability of its speed changes. The average movement speed can be calculated by summing the absolute values ​​of all instantaneous speeds within the preset time window and then dividing by the number of sampling points or the time length. The number of starts and stops can be counted by detecting the number of events where the speed changes from non-zero to zero (or below a very small threshold). The movement speed fluctuation amplitude can be quantified by calculating the standard deviation, the difference between the maximum and minimum values, or the mean absolute deviation of the instantaneous speeds within the preset time window. These statistics transform the complex motion behavior patterns of the target assembly into quantifiable values, providing an objective and concise basis for intent recognition, enabling the system to distinguish different operational intentions by comparing these values ​​with preset thresholds.

[0065] In step A303, the preset speed threshold, preset frequency threshold, and preset amplitude threshold are reference values ​​used for intent judgment. They are set based on the typical operational characteristics and experience of different functional areas. The preset speed threshold distinguishes between low-speed picking and normal passage; the preset frequency threshold distinguishes between picking behavior with frequent starts and stops and passage behavior with continuous movement; and the preset amplitude threshold distinguishes between passage behavior with stable speed and picking behavior with unstable speed. These thresholds can be determined through statistical analysis and machine learning training on a large amount of actual warehouse operation data. For example, they can be set by observing the average speed and pause frequency of pickers in the picking area, and the speed stability of passage personnel in the main aisle. These thresholds are the core components of the intent recognition rules. They provide quantitative judgment criteria, enabling the system to automatically and accurately classify the operational intent of target combinations.

[0066] Through the above technical solution, this application effectively solves the problem that traditional methods in warehouse human-machine collaborative safety management suffer from inaccurate intent recognition due to the lack of specific judgment rules and quantitative indicators for different functional areas. This leads to an inability to effectively distinguish between picking and passage intentions, resulting in avoidance decision errors or low efficiency. Specifically, by dividing the functional area into a picking area and a main aisle area, and setting different intent recognition rules for different areas, this solution provides precise context for intent judgment. In the picking area, by monitoring the average moving speed and number of starts and stops of the target assembly, the low-speed and frequent pause characteristics of the picking operation can be accurately captured, thereby identifying the picking intent. In the main aisle area, by monitoring the fluctuation range of the moving speed, the stability of the passage behavior can be effectively identified, thereby determining the passage intention. This area-specific and quantitative intent recognition method significantly improves the accuracy of judging the operational intent of the target assembly. This precise intent recognition capability enables the autonomous mobile robot to dynamically adjust its avoidance strategy according to the true intent of the target assembly. For example, when a robot is identified as intending to perform a picking operation, it can adopt a more cautious and flexible avoidance strategy to prevent interfering with the picker's operation. When it is identified as intending to pass through, the robot can plan a more compact and efficient avoidance path, reducing unnecessary stopping, waiting, and detour distance. This not only effectively ensures the safety of human-robot collaboration and avoids the risk of collisions caused by misjudgment, but also significantly improves the operating efficiency of autonomous mobile robots and the overall throughput of warehouse operations, solving the efficiency bottleneck caused by frequent avoidance.

[0067] In some implementations, step A4 includes:

[0068] A401. Based on the operational intent, combined with the physical boundary information of the target assembly and the preset safety distance parameters, determine the initial avoidance zone;

[0069] A402. Obtain the current motion state information of the autonomous mobile robot; the current motion state information includes the current speed, current acceleration, and current turning radius;

[0070] A403. Based on the current motion state information of the autonomous mobile robot and the preset kinematic constraints, the initial avoidance area is adjusted to obtain the adjusted avoidance area;

[0071] A404. Based on the adjusted avoidance area, plan the avoidance path of the autonomous mobile robot so that the avoidance path meets the preset energy consumption optimization target and motion smoothness requirements.

[0072] In step A401, the step of determining the initial avoidance zone based on the operational intent, the physical boundary information of the target assembly, and preset safety distance parameters aims to provide the autonomous mobile robot with a preliminary safe space to avoid collisions with the target assembly. The operational intent can indicate the future behavior pattern of the target assembly, such as whether to perform picking operations or to pass through. The physical boundary information of the target assembly refers to its actual occupied area in space, which can be obtained through sensor data (such as LiDAR, depth camera). The preset safety distance parameters are set according to safety regulations and practical operating experience, and are used to delineate an additional buffer zone around the target assembly. The determination of the initial avoidance zone can be based on the target assembly's current position and predicted future position, combined with its physical boundary and safety distance. For example, different expansion coefficients can be selected according to the operational intent, or the maximum possible space occupied by the target assembly in the future can be predicted based on the intent, and a safety margin can be added accordingly.

[0073] Step A402, which involves acquiring the current motion state information of the autonomous mobile robot, including its current speed, current acceleration, and current turning radius, is used to monitor the robot's motion capabilities and limitations in real time. Current speed refers to the robot's current linear and angular velocity, reflecting its instantaneous speed and direction. Current acceleration refers to the robot's linear and angular acceleration, reflecting the trend of its speed change. Current turning radius refers to the radius of curvature of the robot's current trajectory, reflecting its turning ability. This information can be directly acquired through the robot's internal sensors (such as encoders and inertial measurement units, IMUs) and motion controllers, or obtained by smoothing and differentiating historical motion data. Acquiring this information is fundamental for subsequent adjustments to the avoidance area, ensuring that the planned path is actually executable by the robot.

[0074] Step A403, adjusting the initial avoidance area based on the current motion state information of the autonomous mobile robot and preset kinematic constraints to obtain the adjusted avoidance area, is one of the core steps of this solution. This aims to make the avoidance area more consistent with the actual motion capabilities of the autonomous mobile robot. The preset kinematic constraints include physical limitations such as the robot's maximum speed, maximum acceleration, maximum deceleration, and minimum turning radius. For example, if the robot's current speed is high and a longer deceleration distance is required, the boundary of the avoidance area may need to be expanded outwards to provide sufficient deceleration space; if the robot's current turning radius is large, a larger turning space is required, and the shape of the avoidance area may need to be adjusted to adapt to this limitation. The adjustment process can employ methods such as geometric expansion / contraction and shape transformation to ensure that the adjusted avoidance area can accommodate the robot's avoidance actions while satisfying the kinematic constraints.

[0075] In step A404, the step of planning an avoidance path for the autonomous mobile robot based on the adjusted avoidance area, ensuring that the avoidance path meets preset energy consumption optimization goals and motion smoothness requirements, aims to generate a safe and efficient avoidance path. After obtaining the adjusted avoidance area, a path planning algorithm (such as A* algorithm, RRT algorithm, DWA algorithm, etc.) will search for a path from the robot's current position to the target position outside this area. Energy consumption optimization goals typically refer to minimizing path length, the number of turns, or speed changes to reduce energy consumption. Motion smoothness requirements refer to small curvature changes and continuous speed changes on the path to ensure the stability and comfort of the robot's movement and avoid sudden stops and turns. By comprehensively considering these goals, a path can be generated that effectively avoids the target assembly while ensuring the robot's own operational efficiency and stability.

[0076] Through the above technical solution, this method fully considers the robot's real-time motion capabilities when determining the avoidance zone of the autonomous mobile robot. Based on the initial avoidance zone determined according to the operational intent of the target assembly, the method further acquires motion state information such as the current speed, current acceleration, and current turning radius of the autonomous mobile robot. Based on this information and preset kinematic constraints, the initial avoidance zone is dynamically adjusted, resulting in an adjusted avoidance zone that better reflects the robot's actual motion capabilities. This adjustment avoids planning avoidance paths that exceed the robot's physical limits, effectively solving the problem of avoidance paths not conforming to robot kinematic constraints in traditional methods. Furthermore, an avoidance path is planned based on the adjusted avoidance zone, satisfying preset energy consumption optimization goals and motion smoothness requirements. This enables the autonomous mobile robot to operate more efficiently and smoothly when performing avoidance actions, reducing unnecessary sudden stops and turns, thereby lowering energy consumption, improving motion smoothness and safety, and ultimately enhancing the overall efficiency of human-robot collaboration within the warehouse. For example, when the robot is traveling at high speed, the adjusted avoidance zone will allow for a longer deceleration distance to avoid sudden braking; when the robot's turning ability is limited, the adjusted zone will ensure sufficient turning space to avoid collisions. This dynamic and adaptive avoidance strategy significantly enhances the intelligent avoidance capabilities of autonomous mobile robots in complex warehouse environments.

[0077] Preferably, step A401 may include:

[0078] B1. Based on the task intent, retrieve the expected behavior pattern corresponding to the task intent from the preset intent behavior library; the expected behavior pattern includes the expected movement direction, the expected movement speed range, and the expected stopping area;

[0079] B2. Based on the expected behavior pattern and combined with the physical boundary information of the target assembly, determine the expected space occupied by the target assembly;

[0080] B3. Based on the expected space to be occupied and combined with the preset safety distance parameters, the expected space to be occupied is expanded to obtain the initial avoidance zone.

[0081] Step B1 aims to predict the future behavioral trends of the target assembly by analyzing its operational intentions. Operational intentions are high-level behavioral abstractions, while expected behavioral patterns are concrete descriptions of these abstractions, including spatial and temporal behavioral characteristics. The pre-defined intention behavior library can be a predefined database or lookup table storing the mapping relationships between different operational intentions (e.g., "picking intention" and "travel intention") and their corresponding expected behavioral patterns. For example, "picking intention" could correspond to "expected movement direction is random or a small-range oscillation, expected movement speed range is 0-0.5 m / s, and expected stopping area is a circular area with a radius of 1.5 meters around the current position"; while "travel intention" could correspond to "expected movement direction is the current direction, expected movement speed range is 0.5-1.5 m / s, and expected stopping area is none or a linear area along the passageway." Alternatively, a machine learning model can be trained, taking operational intentions as input and outputting the corresponding expected behavioral patterns. Expected behavior patterns are key information for predicting the future movement and dwell time of a target assembly. Expected movement direction refers to the main direction of movement that the target assembly may take in the future, which can be a single direction vector or a range of directions. Expected movement speed range refers to the possible speed range of the target assembly in the future, reflecting its dynamic nature. Expected dwell time area refers to the spatial range in which the target assembly may dwell or frequently appear in the future, reflecting its static or quasi-static occupancy.

[0082] Step B2 combines the predicted behavioral trends with the actual physical dimensions of the target assembly to generate a more accurate representation of the space it may occupy in the future. Physical boundary information provides the static dimensions and shape of the assembly, while the expected behavioral pattern gives it dynamism. This expected occupied space can be calculated geometrically by translating and / or rotating the physical boundaries of the target assembly (e.g., a rectangle or polygon) along the expected direction of movement, and calculating the maximum space it may cover in the future based on the expected range of movement speeds and a preset time step. Simultaneously, the expected dwell area is overlaid with the physical boundaries to determine the static occupied space. Alternatively, a probabilistic raster map can be used to represent the expected occupied space. Based on the expected behavioral pattern, probabilities are assigned to the occurrence of the target assembly in different future locations and directions, and its physical boundary information is projected onto the raster map, accumulating to obtain a map representing the expected occupancy probabilities. The expected occupied space is a dynamic concept; it not only considers the current physical dimensions of the target assembly but also predicts all the spaces it may occupy in the future, including its movement trajectory and possible dwell areas.

[0083] Step B3 adds a safety margin to the predicted expected space to ensure the autonomous mobile robot has sufficient buffer space during avoidance, mitigating potential collision risks, and taking into account the robot's kinematic constraints and perception errors. This expansion can be achieved by uniformly extending the geometric boundaries of the expected space outwards; the extension distance is the preset safety distance parameter. For example, if the expected space is a polygon, all its edges are shifted outwards by a preset distance. Furthermore, the preset safety distance parameter can be dynamically adjusted based on factors such as the complexity of the operating environment, the speed of the autonomous mobile robot, and the confidence level of the target assembly's intent. For example, the safety distance can be appropriately increased in high-speed traffic areas or when the intent is uncertain. The initial avoidance zone is the direct basis for the autonomous mobile robot's path planning; it is a larger area than the expected space, including a safety buffer, ensuring the robot can safely detour.

[0084] This application's solution accurately predicts the space occupancy of a target assembly by using expected behavior patterns based on operational intent, and generates an initial avoidance zone accordingly, effectively solving the problem of inaccurate initial avoidance zones. Specifically, firstly, expected behavior patterns, including expected movement direction, expected movement speed range, and expected dwell area, are obtained from a pre-defined intention behavior library based on operational intent. This step uses operational intent as input to match corresponding behavioral features from the predefined library, ensuring that behavior prediction closely matches the actual operational scenario of the assembly, avoiding deviations caused by random or general assumptions, thus providing a reliable basis for subsequent space prediction. Next, the expected occupied space is determined based on the expected behavior patterns combined with the physical boundary information of the target assembly. By fusing dynamic behavioral trends (such as movement direction and speed range) with static physical dimensions (such as the assembly outline), the future space occupancy state of the assembly is accurately simulated, overcoming the deficiency of ignoring motion dynamics when relying solely on physical boundary information, and ensuring that the predicted space occupancy is more in line with actual behavioral constraints. Then, based on the expected space occupied and preset safety distance parameters, the initial avoidance zone is obtained by expansion. A safety buffer is added to the accurately predicted space, avoiding the problem of the zone being too large or too small due to directly expanding the physical boundary or safety distance. This ensures that the initial avoidance zone is both safe and efficient. Overall, this technical solution achieves refined generation of the initial avoidance zone by progressively associating operational intent, behavioral patterns, and physical information, providing an accurate foundation for subsequent path planning. Combined with the aforementioned technique of determining the corresponding avoidance zone based on operational intent, this solution provides autonomous mobile robots with a more accurate and dynamically adaptable avoidance zone, thereby significantly improving the robot's operational efficiency and path planning flexibility while ensuring safety.

[0085] Further, step B2 includes:

[0086] B201. Based on the expected direction of movement and the expected range of movement speed, and combined with the current position and physical boundary information of the target assembly, calculate the expected displacement vector and expected rotation angle of the target assembly within the preset time step;

[0087] B202. Based on the expected displacement vector and expected rotation angle, and combined with the physical boundary information of the target assembly, generate the expected trajectory segment of the target assembly within a preset time step, and determine the expected dynamic occupied space covered by the expected trajectory segment.

[0088] B203. Based on the geometry and spatial extent of the expected dwelling area, and combined with the physical boundary information of the target assembly, determine the expected static occupied space of the target assembly within the expected dwelling area;

[0089] B204. Merge the expected dynamic occupied space with the expected static occupied space to obtain the expected occupied space of the target combination.

[0090] In step B201, calculating the expected displacement vector and expected rotation angle aims to accurately predict the motion trend of the target assembly within a short period, providing a basis for subsequent trajectory generation. This calculation can be based on state estimation algorithms such as Kalman filtering or extended Kalman filtering, combining historical motion data and current sensing data to predict the displacement and rotation of the target assembly within a preset time step. Alternatively, a kinematic model-based prediction method can be used, deriving the expected displacement vector and expected rotation angle within a preset time step based on the target assembly's current velocity, angular velocity, and its physical boundary information (e.g., turning radius limitations when pushing the cart).

[0091] In step B202, the expected trajectory segment is generated, and the expected dynamic occupied space covered by the expected trajectory segment is determined. Its purpose is to transform the predicted motion into actual spatial occupation, ensuring that the avoidance planning considers the complete outline of the composite object during its movement. This can be achieved using geometric modeling methods, translating and rotating the physical boundary of the target composite object (e.g., a composite shape composed of multiple rectangles or circles) along the calculated expected displacement vector and expected rotation angle. Within a preset time step, the occupied space of the composite object in each small time interval is unioned to generate the dynamic occupied space covered by the expected trajectory segment.

[0092] In step B203, the expected static occupancy space within the expected dwell area is determined. This aims to accurately identify the fixed space occupied by the target assembly when it may dwell, avoiding collisions caused by ignoring static occupancy. This process can be based on the geometry (e.g., rectangle, circle, or polygon) and spatial extent of the expected dwell area defined in a pre-defined intention behavior library, combined with the physical boundary information of the target assembly. The target assembly is placed within this area, and its occupancy space in the most unfavorable or typical dwell position is calculated as the expected static occupancy space. Alternatively, historical data can be analyzed to statistically analyze the typical dwelling postures and positional distribution of the target assembly within a specific expected dwell area. Then, based on these statistical data and the physical boundaries of the target assembly, a probabilistic or deterministic static occupancy space model can be generated.

[0093] In step B204, the expected dynamic occupation space and the expected static occupation space are spatially merged. This integrates all possible occupancy of the target assembly in both moving and stationary states, forming a comprehensive expected occupation space. This can be achieved by using a geometric union operation to merge the expected dynamic and static occupation spaces in two-dimensional or three-dimensional space, resulting in a final occupied area containing all points from both. Alternatively, raster map or point cloud fusion techniques can be used to map the dynamic and static occupation information to a unified spatial representation, and then perform a logical OR operation on this information to generate the merged expected occupation space.

[0094] This application's solution refines the calculation process for expected space occupation, enabling autonomous mobile robots to more accurately predict the future space occupation of a target assembly (e.g., a person pushing a picking cart) in a warehouse environment. Specifically, after retrieving the corresponding expected behavior pattern from a pre-set intention behavior library based on the operational intent, this solution first calculates the expected displacement vector and expected rotation angle of the target assembly within a pre-set time step, based on the expected movement direction and expected movement speed range, combined with the target assembly's current position and physical boundary information. These parameters accurately describe the dynamic movement trend of the assembly in a short time, taking into account its physical constraints, such as the turning radius constraint when pushing the cart. Next, using these calculated displacement vectors and rotation angles, the expected trajectory segment of the target assembly is generated within the pre-set time step, and the expected dynamic space occupation covered by this trajectory segment is determined. This ensures that the complete physical contour of the assembly is taken into account during its movement, avoiding the safety hazards caused by predicting only the center point while ignoring its dimensions. Simultaneously, this solution also determines the expected static space occupation within the expected dwell area based on the geometry and spatial range of the expected dwell area, combined with the target assembly's physical boundary information. This addresses the issue of insufficient consideration being given to the fixed space occupied by the combined robot when it may stop in a specific area. Ultimately, the expected dynamic and static occupancy spaces are spatially merged to obtain a comprehensive and accurate expected occupancy space. This accurate expected occupancy space is then used to expand the expected occupancy space in conjunction with preset safety distance parameters, thereby obtaining the initial avoidance zone. In this way, this solution provides more reliable and comprehensive spatial information for planning avoidance paths for autonomous mobile robots, making the determination of the avoidance zone more accurate, thus effectively avoiding collisions and optimizing the efficiency of the avoidance path.

[0095] Preferably, step A403 may include:

[0096] Based on the current speed, current acceleration, and preset acceleration / deceleration capability constraints, calculate the minimum distance required for the autonomous mobile robot to decelerate or accelerate at the initial avoidance zone boundary.

[0097] Based on the current turning radius and the preset minimum turning radius constraint, calculate the minimum turning space required for the autonomous mobile robot to perform a turning operation at the initial avoidance zone boundary;

[0098] Based on the minimum distance and minimum turning space, the boundary of the initial avoidance zone is adjusted by shrinking or expanding it to obtain the adjusted avoidance zone.

[0099] Specifically, when calculating the minimum distance required for an autonomous mobile robot to decelerate or accelerate at the initial boundary of the avoidance zone, it can be based on classical kinematic formulas. For example, when the robot needs to decelerate to a stop, the minimum distance d can be determined by the formula d=v^2 / (2a_max), where v is the current velocity and a_max is the maximum deceleration. Alternatively, this minimum distance can also be obtained by consulting a pre-established performance curve or lookup table generated based on actual robot test data. This lookup table records the minimum distance required for the robot to complete deceleration or acceleration at different speeds.

[0100] When calculating the minimum turning space required for an autonomous mobile robot to turn at the initial avoidance zone boundary, it can be determined geometrically based on the robot's geometric model and its preset minimum turning radius constraints. For example, for a differential drive robot, its minimum turning radius is usually determined by the wheelbase and maximum steering angle, from which the minimum lateral and longitudinal space required to complete a turn at a specific angle can be calculated. Alternatively, this minimum turning space can also be determined by pre-calibrating the actual spatial envelope occupied by the robot's body at different speeds and turning radii, thus obtaining a dynamic minimum turning area that considers the robot's size and motion characteristics.

[0101] When adjusting the boundary of the initial avoidance zone by shrinking or expanding it, the geometric boundary of the initial avoidance zone can be modified based on the calculated minimum distance and minimum turning space. For example, if the calculated minimum distance required for deceleration is greater than the reserved space in the robot's forward direction of the initial avoidance zone, the zone is expanded outward in the forward direction to ensure the robot has sufficient deceleration buffer. Conversely, if the reserved space is too large, it can be appropriately shrunk inward to improve efficiency. Similarly, for turning operations, if the calculated minimum turning space requires the robot to occupy a larger lateral area, the lateral boundary of the initial avoidance zone is expanded outward; if the robot can complete the operation with a smaller turning radius, the lateral boundary can be appropriately shrunk inward. This adjustment can be based on vector translation, scaling, or more complex geometric deformations to ensure that the adjusted avoidance zone accurately encompasses all the space required for the robot to complete the corresponding operation.

[0102] This solution effectively solves the problem of inaccurate adjustment by precisely quantifying the key kinematic parameters of an autonomous mobile robot during obstacle avoidance operations and dynamically adjusting the boundary of the obstacle avoidance zone accordingly. First, the minimum distance is calculated based on the current speed, current acceleration, and preset acceleration / deceleration constraints. This step considers the robot's real-time dynamic performance, ensuring sufficient buffer space for deceleration or acceleration at the boundary to avoid untimely response due to inertia. Second, the minimum turning space is calculated based on the current turning radius and preset minimum turning radius constraints. This step, considering the robot's steering characteristics and combining real-time turning capabilities with system limitations, calculates the minimum operating space required for turning, ensuring the robot can complete the turning action within a limited area. Finally, the initial obstacle avoidance zone boundary is adjusted inward or outward based on the minimum distance and minimum turning space. This step integrates the above calculation results, dynamically modifying the boundary to make the obstacle avoidance zone more closely match the robot's actual motion constraints, thereby generating a safe and efficient obstacle avoidance path. This solution, combined with the aforementioned solution that determines the initial avoidance zone based on the operational intent, ensures that the determined avoidance zone not only considers the expected behavior of the target assembly but also fully takes into account the kinematic limitations of the autonomous mobile robot itself. This allows the subsequently planned avoidance path to effectively avoid potential conflicts while maximizing the robot's motion performance and avoiding unnecessary deceleration or detours. As a result, the overall efficiency of warehousing operations is significantly improved while ensuring safety.

[0103] Preferably, the current motion state information also includes the current position of the autonomous mobile robot;

[0104] Step A404 may include:

[0105] Based on the adjusted avoidance area, construct a local cost map that includes both passable and impassable areas;

[0106] Based on the current position, current speed, and preset kinematic constraints of the autonomous mobile robot, multiple candidate avoidance paths are generated in the local cost map to ensure that the candidate avoidance paths satisfy the kinematic constraints of the autonomous mobile robot.

[0107] For each candidate avoidance path, calculate its path length, rate of curvature change, and minimum distance to the impassable area. Based on the path length, rate of curvature change, and minimum distance, evaluate the energy consumption cost and motion smoothness score of each candidate avoidance path.

[0108] Based on energy consumption cost and motion smoothness scores, combined with preset weighting coefficients, the comprehensive optimization index of each candidate avoidance path is calculated.

[0109] The candidate avoidance path with the best comprehensive optimization index is selected as the avoidance path for the autonomous mobile robot.

[0110] The current position of an autonomous mobile robot can be obtained in real time through the robot's internal positioning system, such as a combination of simultaneous localization and mapping (SLAM) algorithm and sensor data from LiDAR, visual odometry, and inertial measurement unit (IMU); or it can be provided by an external positioning system, such as an ultra-wideband (UWB) positioning system or a positioning system based on QR codes / reflectors.

[0111] Secondly, based on the adjusted avoidance area, a local cost map is constructed, comprising passable and impassable areas. The local cost map is a data structure used in robot path planning to represent environmental obstacles and passable spaces. It divides the environment into grids or lattices, with each grid assigned a "cost" value representing its passability or degree of danger. The local cost map can be dynamically constructed and updated based on real-time local environmental information acquired by sensors (e.g., LiDAR, depth camera) combined with known static map information; alternatively, it can be constructed by directly mapping the adjusted avoidance area onto the local grid, marking lattices within the avoidance area as impassable and the remaining areas as passable.

[0112] Based on this, according to the current position and speed of the autonomous mobile robot, and the preset kinematic constraints, multiple candidate avoidance paths are generated in the local cost map, ensuring that the candidate avoidance paths satisfy the kinematic constraints of the autonomous mobile robot. The candidate avoidance paths are alternative trajectories that the robot may take to avoid obstacles. When generating these paths, the robot's kinematic constraints must be considered, i.e., physical limitations such as the robot's actual achievable maximum speed, maximum acceleration, and minimum turning radius, to ensure that the generated paths are executable. The candidate avoidance paths can be generated using sampling methods, such as random sampling or state-space based sampling (e.g., Rapid Exploration Random Tree (RRT) or Probabilistic Path Graph (PRM), to generate a sequence of path points that satisfy the kinematic constraints; alternatively, methods such as Model Predictive Control (MPC) or Dynamic Window Method (DWA) can be used to generate multiple feasible trajectories in the local search space under the constraints of the robot's kinematic model.

[0113] Furthermore, for each candidate avoidance path, its path length, rate of curvature change, and minimum distance to the impassable area are calculated. These are key indicators for evaluating path quality. The path length directly affects energy consumption and time; the rate of curvature change reflects the smoothness of the path, affecting motion comfort and mechanical wear; and the minimum distance to the impassable area measures the path's safety. The path length can be obtained by summing the Euclidean distances between consecutive points on the path; the rate of curvature change can be obtained by calculating the curvature of each point on the path and then taking its derivative or difference; and the minimum distance can be obtained by calculating the shortest distance from all points on the path to the boundary of the impassable area and taking the minimum value among them.

[0114] Subsequently, based on the path length, the rate of curvature change, and the minimum distance, the energy cost and motion smoothness score of each candidate avoidance path are evaluated. The energy consumption cost can be estimated based on factors such as path length, speed changes, and number of accelerations and decelerations, combined with the robot's dynamics model (for example, the energy consumption cost can be estimated by multiplying the path length by the robot's traction energy consumption per unit distance, and superimposing the acceleration / deceleration efficiency and motor power curves defined in the robot's dynamics model to estimate the total energy consumption of all speed changes and acceleration / deceleration events). The motion smoothness score can be calculated based on indicators such as the rate of curvature change and turning frequency using a preset scoring function (for example, the scoring function is Mss=w1*max(0,(s1_max-s1) / s1_max)+w2*max(0,(s2_max-s2) / s2_max), where Mss is the motion smoothness score, s1 is the rate of curvature change, s1_max is the preset maximum acceptable rate of curvature change, s2 is the turning frequency, s2_max is the preset maximum acceptable turning frequency, and w1 and w2 are weights).

[0115] Next, based on the energy consumption cost and the motion smoothness score, and combined with preset weighting coefficients, a comprehensive optimization index is calculated for each candidate avoidance path. The comprehensive optimization index is a weighted sum of the energy consumption cost and the motion smoothness score to obtain a single value that measures the overall quality of the path. The weighting coefficients reflect the system's preference for energy consumption and smoothness.

[0116] Finally, the candidate avoidance path with the best comprehensive optimization index is selected as the avoidance path for the autonomous mobile robot. This step involves directly comparing the comprehensive optimization index of all candidate paths and selecting the path with the smallest (or largest, depending on the index definition) value as the final avoidance path to be executed.

[0117] Through the above technical solution, this application effectively solves the problems in warehouse human-robot collaborative safety management, where autonomous mobile robot obstacle avoidance path planning may not fully consider energy consumption optimization and motion smoothness, leading to low obstacle avoidance efficiency, and the planned path may have excessive energy consumption or non-smooth motion due to ignoring real-time motion status and path quality indicators. Specifically, by incorporating the current position of the autonomous mobile robot into the current motion state information and constructing a local cost map based on the adjusted avoidance area, more accurate and real-time environmental information is provided for path planning. On this basis, multiple candidate avoidance paths that meet the robot's kinematic constraints are generated, ensuring the feasibility of the path and its adaptability to the robot's real-time dynamics. Furthermore, by calculating the path length, rate of curvature change, and minimum distance to impassable areas for each candidate avoidance path, and evaluating energy consumption cost and motion smoothness score accordingly, this solution can quantify the efficiency, comfort, and safety of the path. Combined with preset weight coefficients to calculate comprehensive optimization indicators, the system can balance and optimize between energy consumption and motion smoothness according to actual needs. Ultimately, the path with the optimal comprehensive optimization indicators was selected, ensuring that the autonomous mobile robot could not only safely avoid the target assembly during the obstacle avoidance process, but also complete the avoidance operation with the lowest energy consumption and the smoothest motion trajectory. This significantly improved the efficiency and smoothness of the autonomous mobile robot's obstacle avoidance operation, avoiding resource waste or operational disruptions caused by improper path selection. Thus, while ensuring the safety of human-robot collaboration, it maximized the overall efficiency and smoothness of warehouse operations.

[0118] refer to Figure 2 This application provides a warehouse human-robot collaborative safety management system for controlling autonomous mobile robots. The system includes:

[0119] The combined object recognition module 1 is used to identify personnel features and mobile device features based on environmental perception data, and to determine the target combined object based on the spatial and motion correlation between personnel features and mobile device features (refer to step A1 above for details).

[0120] The positioning module 2 is used to obtain the real-time position of the target assembly in the work environment map and determine the functional area where the target assembly is currently located based on the real-time position (refer to step A2 above for details).

[0121] The task intent recognition module 3 is used to obtain the motion state parameters of the target assembly and determine the task intent of the target assembly based on the current functional area of ​​the target assembly and the motion state parameters (for details, refer to step A3 above).

[0122] The obstacle avoidance module 4 is used to determine the corresponding obstacle avoidance area according to the task intention, and to plan the obstacle avoidance path of the autonomous mobile robot based on the obstacle avoidance area (for details, refer to step A4 above).

[0123] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for human-machine collaboration safety management in a warehouse, used to control an autonomous mobile robot, characterized in that, The steps of this method include: A1. Based on environmental perception data, identify personnel characteristics and mobile device characteristics, and determine the target assembly based on the spatial and motion correlation between the personnel characteristics and the mobile device characteristics; A2. Obtain the real-time location of the target assembly in the work environment map, and determine the functional area where the target assembly is currently located based on the real-time location; A3. Obtain the motion state parameters of the target assembly, and determine the operational intent of the target assembly based on the current functional area of ​​the target assembly and the motion state parameters; A4. Determine the corresponding avoidance area based on the stated task intention, and plan the avoidance path for the autonomous mobile robot based on the avoidance area; Step A1 includes: A101. Acquire environmental perception data, and identify personnel characteristics and mobile device characteristics based on the environmental perception data to obtain the location data of the personnel characteristics and the mobile device characteristics; A102. Based on the location data of the personnel feature and the mobile device feature, calculate the physical distance between the personnel feature and the mobile device feature, the movement speed vector of the personnel feature, and the movement speed vector of the mobile device feature; A103. Compare the physical distance with a preset distance threshold. When the physical distance is less than the distance threshold, confirm that the spatial association condition is met. A104. When the directional and magnitude deviations between the movement speed vectors of the personnel feature and the mobile device feature are both within the corresponding preset consistency range, it is confirmed that the motion association condition is met. A105. When both spatial association conditions and motion association conditions are met, the personnel characteristics and the mobile device characteristics are marked as the target combination belonging to the same physical whole; The functional areas include at least a picking area and a main aisle area; Step A3 includes: A301. Obtain the motion state parameters of the target assembly within a preset time window; the motion state parameters include start / stop status and moving speed; the preset time window is a window of preset duration with the current moment as the end point; A302. Based on the motion state parameters of the target assembly within a preset time window, calculate the average moving speed, number of starts and stops, and moving speed fluctuation amplitude of the target assembly within the preset time window; A303. When the target assembly is located in the picking operation area, if the average moving speed is lower than a preset speed threshold or the number of starts and stops exceeds a preset number threshold, then the operation intention is determined to be a picking operation intention; otherwise, the operation intention is determined to be an uncertain intention. A303. When the target assembly is located in the main channel area, if the fluctuation range of the moving speed is less than a preset range threshold, the operation intention is determined to be a passage intention; otherwise, the operation intention is determined to be an uncertain intention.

2. The method for human-machine collaborative safety management in warehouses according to claim 1, characterized in that, Step A2 includes: A201. Obtain location data of personnel characteristics and mobile device characteristics in the target assembly; A202. Calculate the real-time position of the target assembly in the work environment map based on the location data of personnel characteristics and mobile device characteristics in the target assembly; A203. Determine the current functional area of ​​the target assembly based on the real-time location.

3. The method for human-machine collaboration safety management in warehouses according to claim 1, characterized in that, Step A4 includes: A401. Based on the stated operational intent, and in conjunction with the physical boundary information of the target assembly and preset safety distance parameters, determine the initial avoidance zone; A402. Obtain the current motion state information of the autonomous mobile robot; the current motion state information includes the current speed, current acceleration, and current turning radius; A403. Based on the current motion state information of the autonomous mobile robot and the preset kinematic constraints, the initial avoidance area is adjusted to obtain the adjusted avoidance area; A404. Based on the adjusted avoidance area, plan the avoidance path of the autonomous mobile robot so that the avoidance path meets the preset energy consumption optimization target and motion smoothness requirements.

4. The method for human-machine collaboration safety management in warehouses according to claim 3, characterized in that, Step A401 includes: B1. Based on the stated task intention, retrieve the expected behavior pattern corresponding to the task intention from a preset intention behavior library; the expected behavior pattern includes the expected movement direction, the expected movement speed range, and the expected stopping area; B2. Based on the expected behavior pattern and combined with the physical boundary information of the target assembly, determine the expected space occupied by the target assembly; B3. Based on the expected space to be occupied and combined with the preset safety distance parameters, the expected space to be occupied is expanded to obtain the initial avoidance area.

5. A method for human-machine collaboration safety management in warehouses according to claim 4, characterized in that, Step B2 includes: B201. Based on the expected direction of movement and the expected range of movement speed, and combined with the current position and physical boundary information of the target assembly, calculate the expected displacement vector and expected rotation angle of the target assembly within a preset time step; B202. Based on the expected displacement vector and the expected rotation angle, and combined with the physical boundary information of the target assembly, generate the expected trajectory segment of the target assembly within the preset time step, and determine the expected dynamic occupied space covered by the expected trajectory segment; B203. Based on the geometry and spatial extent of the expected dwelling area, and in conjunction with the physical boundary information of the target assembly, determine the expected static occupied space of the target assembly within the expected dwelling area; B204. Merge the expected dynamic occupied space with the expected static occupied space to obtain the expected occupied space of the target assembly.

6. The method for human-machine collaboration safety management in warehouses according to claim 3, characterized in that, Step A403 includes: Based on the current speed, the current acceleration, and the preset acceleration / deceleration capability constraints, calculate the minimum distance required for the autonomous mobile robot to decelerate or accelerate at the boundary of the initial avoidance zone. Based on the current turning radius and the preset minimum turning radius constraint, calculate the minimum turning space required for the autonomous mobile robot to perform a turning operation at the boundary of the initial avoidance zone; Based on the minimum distance and the minimum turning space, the boundary of the initial avoidance area is adjusted by shrinking inward or expanding outward to obtain the adjusted avoidance area.

7. A method for human-machine collaboration safety management in warehouses according to claim 3, characterized in that, The current motion state information also includes the current position of the autonomous mobile robot; Step A404 includes: Based on the adjusted avoidance area, a local cost map containing passable and impassable areas is constructed. Based on the current position, current speed, and preset kinematic constraints of the autonomous mobile robot, multiple candidate avoidance paths are generated in the local cost map, such that the candidate avoidance paths satisfy the kinematic constraints of the autonomous mobile robot. For each candidate avoidance path, calculate its path length, rate of curvature change, and minimum distance to the impassable area, and evaluate the energy consumption cost and motion smoothness score of each candidate avoidance path based on the path length, the rate of curvature change, and the minimum distance. Based on the energy consumption cost and the motion smoothness score, and combined with preset weighting coefficients, a comprehensive optimization index is calculated for each candidate avoidance path. The candidate avoidance path with the best comprehensive optimization index is selected as the avoidance path for the autonomous mobile robot.

8. A warehouse human-machine collaborative safety management system, used to control an autonomous mobile robot based on the warehouse human-machine collaborative safety management method according to any one of claims 1-7, characterized in that, The system includes: The assembly recognition module is used to identify personnel features and mobile device features based on environmental perception data, and to determine the target assembly based on the spatial and motion correlation between the personnel features and the mobile device features. The positioning module is used to obtain the real-time position of the target assembly in the work environment map, and determine the functional area where the target assembly is currently located based on the real-time position; The task intent recognition module is used to acquire the motion state parameters of the target assembly, and determine the task intent of the target assembly based on the current functional area of ​​the target assembly and the motion state parameters. The obstacle avoidance module is used to determine the corresponding obstacle avoidance area according to the operation intention, and to plan the obstacle avoidance path of the autonomous mobile robot based on the obstacle avoidance area.

Images