An automatic shelving method and system for logistics packaging boxes

Abstract

The application provides a logistics package box automatic shelving method and system, relates to the automatic logistics technical field, and is applied to a logistics robot, and comprises the following steps: when receiving a logistics package box shelving instruction, the current position of a target logistics package box and the current position of a target storage unit are acquired; path planning is performed in combination with the current position of the logistics robot to obtain a target movement path; the logistics robot is controlled to move to the target logistics package box according to the target movement path, and image data of the target logistics package box is acquired; external feature data is determined according to the image data, the grabbing mode of the logistics robot is further determined, and the grabbing strategy of the mechanical arm of the logistics robot is determined; the target logistics package box is grabbed according to the grabbing strategy, the package box is carried to the target storage unit, and the package box is subjected to shelving operation through a force-position hybrid control algorithm. The application reduces the damage risk of the logistics package box in the shelving process due to human factors and reduces the loss of goods.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics automation, in particular to a method and system for automatically shelving logistics packaging boxes. BACKGROUND

[0002] Due to the rapid development of intelligent logistics, the freight center generally adopts the mode of fixed conveying line combined with manual stacking to complete the shelving operation of logistics packaging boxes. The operator drives a forklift or places a box at the end of the conveying line according to the storage location information displayed on paper or a handheld terminal, and then places the packaging box on the designated shelf according to experience.

[0003] In related technologies, since manual operation is the main operation, human errors are prone to occur during the shelving of logistics packaging boxes, causing damage to the logistics packaging boxes or low placement accuracy, thereby increasing the loss of goods, causing the interruption of subsequent logistics links and the increase of additional costs, occupying channel resources and blocking subsequent outbound, causing serious delays in the logistics operation of shelving logistics packaging boxes, and being difficult to adapt to the high-efficiency operation demand of large-scale freight centers. SUMMARY

[0004] The problem solved by the present application is how to solve the problem of loss of logistics packaging boxes caused by manual transportation.

[0005] To solve the above problems, the present application provides a method and system for automatically shelving logistics packaging boxes.

[0006] In a first aspect, the present application provides a method for automatically shelving logistics packaging boxes, applied to a logistics robot, the method comprising:

[0007] After receiving the shelving instruction of the logistics packaging box, the current position of the target logistics packaging box and the current position of the target storage unit are obtained;

[0008] According to the current position of the target logistics packaging box and the current position of the target storage unit, the current position of the logistics robot is combined to plan a path, and a target movement path of the logistics robot is obtained;

[0009] According to the target movement path, the logistics robot is controlled to move to the target logistics packaging box, and image data of the target logistics packaging box is obtained;

[0010] According to the image data, the external feature data of the target logistics packaging box is determined, and according to the external feature data, the grabbing mode of the logistics robot is determined, and according to the grabbing mode, the grabbing strategy of the mechanical arm of the logistics robot is determined;

[0011] According to the grabbing strategy, the target logistics packaging box is grabbed, and then the target logistics packaging box is transported to the target storage unit, and a force-position hybrid control algorithm is used for shelving operation of the target logistics packaging box.

[0012] Optionally, after receiving the logistics packaging box shelving instruction, the current position of the target logistics packaging box and the current position of the target storage unit are obtained, comprising:

[0013] When the logistics packaging box shelving instruction is received, the current environment of the logistics robot is dynamically scanned to obtain point cloud data of the current environment;

[0014] According to the point cloud data combined with the image data of the current environment, a global environment model is generated;

[0015] According to the global environment model, the current position of the target logistics packaging box and the current position of the target storage unit are determined.

[0016] Optionally, according to the current position of the target logistics packaging box and the current position of the target storage unit, combined with the current position of the logistics robot, a path planning is performed to obtain a target movement path of the logistics robot, comprising:

[0017] According to the global environment model, the environment feature data from the current position of the logistics robot to the current position of the target logistics packaging box and from the current position of the target logistics packaging box to the current position of the target storage unit is determined;

[0018] According to the environment feature data, a multi-objective optimization function is used to search for a path in the global environment model to obtain a first path from the current position of the logistics robot to the current position of the target logistics packaging box and a second path from the current position of the target logistics packaging box to the current position of the target storage unit;

[0019] According to the first path and the second path, the target movement path is generated.

[0020] Optionally, according to the target movement path, the logistics robot is controlled to move to the target logistics packaging box, and image data of the target logistics packaging box is obtained, comprising:

[0021] According to the target movement path, the logistics robot is controlled to move to the target logistics packaging box according to preset speed and acceleration parameters;

[0022] When the logistics robot reaches the target logistics packaging box, initial image data of the target logistics packaging box is obtained;

[0023] Preprocess the initial image data, and determine whether the target logistics packaging box is complete according to the preprocessed initial image data;

[0024] If not, stop the logistics packaging box shelving task, and issue an alarm information;

[0025] If yes, the initial image data is used as the image data of the target logistics packaging box.

[0026] Optionally, the determining of the external feature data of the target logistics packaging box according to the image data comprises:

[0027] extracting features of the image data of the target logistics packaging box through a deep learning algorithm to obtain surface material features and volume shape features of the target logistics packaging box;

[0028] determining a surface material type of the target logistics packaging box according to the surface material features;

[0029] determining a size parameter of the target logistics packaging box according to the volume shape features;

[0030] using the surface material type and the size parameter as the external feature data of the target logistics packaging box.

[0031] Optionally, the determining of the grabbing mode of the logistics robot according to the external feature data comprises:

[0032] determining a grabbing tool of the logistics robot according to the surface material type and the size parameter of the target logistics packaging box;

[0033] querying a preset grabbing mode database according to the grabbing tool to determine the grabbing mode corresponding to the grabbing tool.

[0034] Optionally, the determining of the grabbing strategy of the mechanical arm of the logistics robot according to the grabbing mode comprises:

[0035] determining a grabbing point of the mechanical arm on the target logistics packaging box according to the grabbing mode in combination with the size parameter;

[0036] determining a grabbing force of the mechanical arm on the target logistics packaging box according to the grabbing point in combination with the surface material type;

[0037] generating a grabbing path of the mechanical arm according to a current position of the target logistics packaging box and the grabbing point;

[0038] obtaining the grabbing strategy of the mechanical arm according to the grabbing path and the grabbing force.

[0039] Optionally, the carrying the target logistics packaging box to the target storage unit is subjected to a shelving operation by a force-position hybrid control algorithm, comprising:

[0040] When the logistics robot reaches the target storage unit, pose data of the target storage unit and current pose data of the target logistics packaging box are acquired;

[0041] According to the pose data of the target storage unit and the current pose data of the target logistics packaging box, a pose deviation between the target logistics packaging box and the target storage unit is determined, the pose deviation comprising a position deviation and an attitude deviation;

[0042] According to the position deviation and the attitude deviation, a position strategy, an attitude strategy and a grasping force strategy of the mechanical arm are determined by a force-position hybrid control algorithm;

[0043] According to the position strategy, the attitude strategy and the grasping force strategy, a comprehensive control instruction of the mechanical arm is generated;

[0044] The target logistics packaging box is subjected to a shelving operation by the comprehensive control instruction.

[0045] Optionally, the method further comprises:

[0046] After the target logistics packaging box is placed, a placement position image of the target logistics packaging box in the target storage unit is acquired, and a shelving result of the target logistics packaging box is verified according to the placement position image;

[0047] If the shelving result is verified, an automatic shelving task of the logistics packaging box is completed;

[0048] If the shelving result is not verified, the position strategy, the attitude strategy and the grasping force strategy are adjusted according to the verification result, and the shelving operation is performed again according to the position strategy, the attitude strategy and the grasping force strategy until the shelving result of the target logistics packaging box is verified.

[0049] In a second aspect, the logistics packaging box automatic shelving system of the present application, the method is applied to a logistics robot, and the system comprises:

[0050] A position acquisition module is configured to acquire a current position of a target logistics packaging box and a current position of a target storage unit after receiving a logistics packaging box shelving instruction;

[0051] A path planning module is configured to plan a target moving path of the logistics robot according to a current position of the target logistics packaging box, a current position of the target storage unit, and a current position of the logistics robot.

[0052] An image acquisition module is configured to control the logistics robot to move to the target logistics packaging box according to the target moving path and acquire image data of the target logistics packaging box.

[0053] A feature extraction module is configured to determine external feature data of the target logistics packaging box according to the image data.

[0054] A grabbing mode determination module is configured to determine a grabbing mode of the logistics robot according to the external feature data.

[0055] A grabbing strategy generation module is configured to determine a grabbing strategy of a mechanical arm of the logistics robot according to the grabbing mode.

[0056] A shelving operation module is configured to grab the target logistics packaging box according to the grabbing strategy, carry the target logistics packaging box to the target storage unit, and perform shelving operation on the target logistics packaging box by using a force-position hybrid control algorithm.

[0057] The logistics packaging box automatic shelving method and system can obtain the best path of the logistics robot to the target position by acquiring the current positions of the logistics packaging box and the storage unit and combining the position of the logistics robot, thereby avoiding the inaccurate positioning caused by poor visibility and insufficient experience in manual operation, enabling the logistics packaging box to be accurately placed at the specified shelf position, and improving the shelving accuracy.

[0058] Meanwhile, the logistics robot can automatically complete various operations according to the preset program during the entire shelving process, thereby avoiding the errors that may occur in manual operation, such as misplacing the shelf and colliding with the packaging box, significantly reducing the damage risk of the logistics packaging box caused by human factors during the shelving process, and reducing the loss of goods. According to the external feature data of the logistics packaging box, the grabbing mode and the grabbing strategy of the mechanical arm can be determined, the most suitable grabbing mode can be adopted for the packaging boxes with different shapes, sizes, and weights, the protection of the packaging boxes during the grabbing and carrying process can be ensured, the damage of the packaging boxes caused by improper grabbing force or unreasonable grabbing position can be reduced, and the loss of goods can be effectively reduced.

[0059] The present application realizes the full-process automation of the shelving of logistics packaging boxes, without the need for manual driving of a forklift or manual stacking of the packaging boxes, greatly improving the efficiency of the shelving operation. The logistics robot can quickly and continuously complete the shelving of multiple packaging boxes according to preset instructions and paths, reducing the waiting time and repetitive labor in manual operation and improving the overall efficiency of the logistics operation. Through path planning, the logistics robot can select the optimal path to reach the target position, avoiding problems such as unreasonable path selection and channel congestion that may occur in manual operation, improving the operating efficiency of the robot and the utilization rate of the channel. At the same time, combined with the current position and task allocation of the logistics robot, the collaborative scheduling between multiple robots can be realized, further optimizing the logistics operation process and improving the logistics efficiency.

[0060] At the same time, after adopting the automated shelving method, the dependence on manual labor is reduced, and the labor cost is lowered. In particular, in large-scale freight centers, the application of automation technology can reduce the demand for forklift drivers, stacking workers, and other positions, thereby saving a large amount of labor costs. As a result of reducing the damage to logistics packaging boxes caused by human operation errors, equipment failures and maintenance caused by damaged packaging boxes are avoided, and the wear and tear and maintenance costs of the equipment are reduced. At the same time, the operation of the automated equipment is more stable and reliable, and its maintenance cost is relatively low, which also helps to reduce the overall cost of the logistics operation. In addition, manual operation is easily affected by factors such as the fatigue, mood, and skill level of the operator, resulting in instability and uncertainty in the shelving operation. The logistics robot operates according to the established program and algorithm and is not disturbed by human factors, and can always maintain a stable operating state, ensuring the continuity and reliability of the shelving operation of the logistics packaging boxes. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 FIG. 1 is a flowchart of the logistics packaging box automatic shelving method according to an embodiment of the present application;

[0062] Figure 2 FIG. 2 is a structural diagram of the logistics packaging box automatic shelving system according to an embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings. Although some embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms, and should not be interpreted as being limited to the embodiments described herein, on the contrary, these embodiments are provided to make the present application more thorough and complete. It should be understood that the drawings and embodiments of the present application are only for illustrative purposes, and are not intended to limit the scope of protection of the present application.

[0064] It should be understood that each of the steps recited in the method embodiments of the present application can be performed in different orders and / or in parallel. In addition, the method embodiments can include additional steps and / or omit performing the steps shown. The scope of the present application is not limited in this regard.

[0065] The term "comprises" and variations thereof such as "comprising" and "comprises" as used herein are open-ended, that is, "comprising but not limited to," and allow for the inclusion of non- listed alternatives and / or insubstantial changes; the term "based on" is "based, at least in part, on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "optional embodiments." Related definitions are given throughout the detailed description. It is to be noted that the concepts mentioned in the present application are illustrative and not restrictive, and those skilled in the art should understand that "one" "multiple" modification is illustrative and not restrictive, and unless otherwise explicitly stated in the context, it should be understood as "one or more."

[0066] It should be noted that the "one" "multiple" modification mentioned in the present application is illustrative and not restrictive, and those skilled in the art should understand that "one" "multiple" modification is illustrative and not restrictive, and unless otherwise explicitly stated in the context, it should be understood as "one or more."

[0067] In conjunction Figure 1 As shown, the logistics packaging box automatic shelving method provided by the embodiments of the present application is applied to a logistics robot, and the logistics packaging box automatic shelving method comprises:

[0068] When receiving the logistics packaging box shelving instruction, the current position of the target logistics packaging box and the current position of the target storage unit are obtained.

[0069] Specifically, the logistics robot receives the shelving instruction through the control system, which is usually issued by the management terminal or the automation system of the logistics center. After receiving the instruction, the robot scans the surrounding environment using its built-in sensor system, such as laser radar or visual sensor, to determine the positions of the target logistics packaging box and the target storage unit. These sensors can provide accurate environmental data to help the robot identify the position and direction of the target object. After processing these data, the robot obtains the current position information of the target logistics packaging box and the target storage unit, which provides a basis for subsequent path planning.

[0070] According to the current position of the target logistics packaging box and the current position of the target storage unit, the current position of the logistics robot is combined to plan a path, and a target movement path of the logistics robot is obtained.

[0071] Specifically, according to the position information of the target logistics packaging box and the target storage unit, combined with the position data of the logistics robot itself, a path planning algorithm is called, in the embodiment of the present application, the path planning algorithm comprehensively considers the physical layout, obstacle position and dynamic situation of other robots or equipment in the logistics center. Through calculation, an optimal path from the current position of the robot to the position of the target logistics packaging box is generated, so as to ensure that the robot can safely and efficiently reach the target position; after the path planning is completed, the robot will move according to the path.

[0072] According to the target moving path, the logistics robot is controlled to move to the target logistics packaging box and obtain image data of the target logistics packaging box.

[0073] Specifically, the logistics robot starts to move according to the planned path, and continuously monitors the surrounding environment during the movement to ensure the smoothness and safety of the path. And after reaching the position of the target logistics packaging box, the robot uses its visual sensor, such as a camera, to take pictures of the target logistics packaging box to obtain corresponding image data. These image data are used for subsequent feature recognition and grasping strategy formulation, and the image data are obtained through the visual system of the robot, and the appearance features of the packaging box are captured to provide necessary information for the next step of processing.

[0074] According to the image data, the external feature data of the target logistics packaging box is determined, and according to the external feature data, the grasping mode of the logistics robot is determined, and according to the grasping mode, the grasping strategy of the mechanical arm of the logistics robot is determined.

[0075] Specifically, the obtained image data are analyzed and processed, such as through image recognition technology, to extract the external features of the target logistics packaging box, such as size, shape, color, etc. These feature data are used to guide the subsequent grasping operation, and the feature data determine the grasping mode and strategy that the robot needs to adopt. For example, if the size of the packaging box is large, the robot needs to adjust the range of its grasping device; if the shape of the packaging box is irregular, a special grasping angle and force need to be adopted. The determination of the feature data in this embodiment is based on image analysis algorithm analysis and acquisition, which can accurately identify and measure the object features in the image.

[0076] Based on the external feature data of the target logistics packaging box, a suitable grasping mode is determined through a preset decision logic. For example, if the surface of the packaging box is smooth, it is more suitable to adopt the suction cup type grasping; if the shape of the packaging box is regular and the weight is moderate, it is more suitable to adopt the jaw type grasping. The selection of the grasping mode is based on the comprehensive consideration of the size, shape, material and weight of the packaging box, and in this embodiment, the most suitable mode for the current task is selected from the preset grasping mode library according to these feature data.

[0077] After determining the grabbing mode, the grabbing strategy of the robotic arm is further refined, including adjusting the morphology and parameters of the end effector of the robotic arm to adapt to the specific characteristics of the target logistics packaging box. For example, if a gripper grabbing is selected, the opening and closing degree of the gripper is adjusted according to the size of the packaging box; if a suction cup grabbing is selected, the suction force of the suction cup is adjusted. In addition, the system will also adjust the grabbing force and angle according to the weight and material of the packaging box to ensure the stability and safety of the grabbing process. The above adjustments are automatically completed through the control system of the robot, ensuring that the robotic arm can accurately and efficiently complete the grabbing task.

[0078] According to the grabbing strategy, the target logistics packaging box is grabbed, and then the target logistics packaging box is transported to the target storage unit. A force-position hybrid control algorithm is used to perform the shelving operation on the target logistics packaging box.

[0079] Specifically, the robotic arm performs a grabbing action according to the determined grabbing strategy. During the grabbing process, the robot uses its force sensor to monitor the grabbing force in real time, ensuring the stability and safety of the grabbing process. After successful grabbing, the target logistics packaging box is transported to the target storage unit according to the path generated by the path planning algorithm. During the transportation process, the robot continuously monitors the surrounding environment to ensure the smoothness and safety of the path. Upon reaching the target storage unit, the robot uses a force-position hybrid control algorithm to accurately place the packaging box into the storage unit. At the same time, the force-position hybrid control algorithm can ensure that the robot adjusts the placement force and position according to the actual situation of the storage unit during the placement process, ensuring that the packaging box can be safely and accurately placed in place.

[0080] The logistics packaging box automatic shelving method and system of the present embodiment can obtain the best path for the logistics robot to reach the target position by obtaining the current positions of the logistics packaging box and the storage unit, combining the position of the logistics robot for path planning, avoiding the problem of inaccurate positioning caused by poor visibility and insufficient experience in manual operation, enabling the logistics packaging box to be accurately placed at the specified shelf position, and improving the shelving accuracy. Furthermore, the image data is used to determine the external characteristics of the logistics packaging box, such as size, shape, weight distribution, etc., providing accurate basis for subsequent grabbing operations. Compared with manually judging the grabbing method of the packaging box based on experience, this method is more scientific and accurate, which can effectively avoid the damage or unstable placement of the packaging box caused by improper grabbing, further improving the accuracy and stability of the shelving.

[0081] Meanwhile, during the entire shelving process, the logistics robot can automatically complete various operations according to the preset program, avoiding errors that may occur in manual operation, such as misplacing the shelves, colliding with the packaging boxes, etc., thereby significantly reducing the risk of damage to the logistics packaging boxes during the shelving process due to human factors and reducing the loss of goods. According to the external feature data of the logistics packaging box, the grabbing mode and the grabbing strategy of the mechanical arm are determined, the most suitable grabbing method can be adopted for packaging boxes of different shapes, sizes and weights, the protection of the packaging boxes during grabbing and carrying is ensured, the damage of the packaging boxes caused by improper grabbing force or unreasonable grabbing position is reduced, and the loss of goods is effectively reduced.

[0082] The embodiment realizes the full-process automation of the shelving of logistics packaging boxes, without the need for manual driving of a forklift or manual stacking of packaging boxes, greatly improving the efficiency of shelving operations. The logistics robot can quickly and continuously complete the shelving work of multiple packaging boxes according to the preset instructions and paths, reducing the waiting time and repetitive labor in manual operation and improving the overall efficiency of logistics operations. Through path planning, the logistics robot can choose the optimal path to reach the target position, avoiding problems such as unreasonable path selection and channel congestion that may occur in manual operation, improving the operating efficiency of the robot and the utilization rate of the channel. At the same time, combined with the current position of the logistics robot and task allocation, the collaborative scheduling between multiple robots can be realized, further optimizing the logistics operation process and improving the logistics efficiency.

[0083] At the same time, after adopting the automated shelving method, the dependence on human labor is reduced, and the labor cost is reduced. In particular, in large-scale freight centers, the application of automation technology can reduce the demand for forklift drivers, stacking workers and other positions, thereby saving a large amount of labor cost. Since the damage of logistics packaging boxes caused by human operation errors is reduced, equipment failures and maintenance caused by packaging box damage are avoided, and the wear and tear and maintenance cost of the equipment are reduced. At the same time, the operation of the automated equipment is more stable and reliable, and the maintenance cost is relatively low, which also helps to reduce the overall cost of logistics operation. In addition, manual operation is easily affected by factors such as the fatigue, mood and skill level of the operator, resulting in instability and uncertainty in shelving operations. The logistics robot operates according to the established program and algorithm and is not disturbed by human factors, can always maintain a stable operating state, and ensures the continuity and reliability of the shelving operation of the logistics packaging boxes.

[0084] Optionally, after receiving the logistics packaging box shelving instruction, the current position of the target logistics packaging box and the current position of the target storage unit are obtained, comprising:

[0085] When the logistics packaging box shelving instruction is received, the current environment of the logistics robot is dynamically scanned to obtain point cloud data of the current environment;

[0086] According to the point cloud data and image data of the current environment, a global environment model is generated;

[0087] According to the global environment model, the current position of the target logistics packaging box and the current position of the target storage unit are determined.

[0088] Specifically, after the logistics robot receives the shelving instruction, the 4D millimeter wave radar equipped on it immediately starts a 360° omnidirectional scan of the surrounding environment. For example, in a large logistics warehouse, the robot is located at the starting position of the shelf aisle, and the radar emits millimeter wave signals and receives reflected signals at a scanning frequency of multiple times per second, thereby obtaining distance, speed and angle information of the surrounding objects. These information is converted into point cloud data, each point representing a reflection point on the surface of an object in the environment, and the point cloud data can accurately depict the three-dimensional structure of the environment around the robot, including the positions and shapes of shelves, aisles, other robots and logistics packaging boxes.

[0089] Moreover, while obtaining the point cloud data, the hyperspectral camera carried by the robot also takes pictures of the current environment to obtain image data, which contains color and texture information in the environment, thereby supplementing the details lacking in the point cloud data. For example, the point cloud data can accurately locate the position and shape of a logistics packaging box, but cannot distinguish the color and material of the packaging box, while the images taken by the hyperspectral camera can provide such information to help the robot more accurately identify the target object. In this embodiment, the point cloud data and image data are fused and processed, and the information of the two is supplemented by a specific algorithm to generate a global environment model containing information such as three-dimensional structure, color and texture of the environment, thereby providing the robot with comprehensive environmental information to better plan the path and perform the task.

[0090] Using the generated global environment model, the system searches for the features of the target logistics packaging box and the target storage unit in the model through a preset recognition algorithm. For example, the system can match in the global environment model according to the preset size range, color features of the logistics packaging box, and the specific shape and identification of the storage unit. When the matching features are found, the system can determine the position coordinates of the target logistics packaging box and the target storage unit in the model; these position coordinates are based on the robot's own coordinate system and can accurately indicate the positions of the target objects relative to the robot. For example, the system determines that the target logistics packaging box is located 3 meters in front of the robot and 1 meter to the left, and the target storage unit is located 10 meters in front of the robot and 2 meters to the right; these position information provides accurate basis for the subsequent path planning and grabbing operation of the robot.

[0091] In a preferred embodiment of the present application, the preset recognition algorithm is mainly based on a Transformer multi-modal fusion algorithm, which can effectively combine point cloud data and image data to achieve accurate recognition and positioning of target objects.

[0092] Specifically, first, the point cloud data obtained by the 4D millimeter wave radar is down-sampled and filtered to remove noise points and extract key point features. For example, the point cloud data is simplified by Voxel Grid Filter to retain important geometric structure information. The image data obtained by the hyperspectral camera is normalized to adjust the brightness and contrast of the image, so as to better extract color and texture features. Then, the point cloud processing library (such as PCL) is used to extract the geometric features of the point cloud, such as normal, curvature, etc. These features can describe the surface shape and structure of the object. Convolutional neural network (CNN) is used to extract semantic features of the image, such as color, texture and shape. For example, pre-trained ResNet or VGG model can be used to extract image features. The point cloud features and image features are fused. For example, the self-attention mechanism in the Transformer architecture is used to weight the fusion of point cloud features and image features to generate comprehensive features. The self-attention mechanism can automatically learn the weight between different features to improve the representation ability of the features. The fused features are used for target recognition. For example, a trained classifier (such as support vector machine SVM or deep learning classifier) is used to identify the features of the target logistics packaging box and the target storage unit to determine their positions and poses.

[0093] In a preferred embodiment of the present application, the global environment model is used to represent the three-dimensional spatial information of the current working environment of the logistics robot, and by integrating the data obtained from multiple sensors, including point cloud data and image data, a comprehensive and accurate environment representation is provided. The global environment model of this embodiment not only contains the geometric structure of the environment, but also contains semantic information such as the category and position of different objects.

[0094] The specific composition of the global environment model includes geometric structure information such as point cloud data, which is obtained by 4D millimeter wave radar. The point cloud data provides three-dimensional position and shape information of objects in the environment, and can accurately describe the geometric structure of objects such as shelves, channels, logistics packaging boxes, etc. The point cloud data is then converted into a three-dimensional grid to more intuitively represent the geometric structure of the environment. For example, the Marching Cubes algorithm is used to convert the point cloud data into a grid model to generate a three-dimensional surface of the environment.

[0095] Also included is semantic information, such as image data, which is acquired by a hyperspectral camera that provides color and texture information of objects in the environment, which is used to distinguish different color and material of the logistics packaging boxes. Deep learning algorithms (such as semantic segmentation networks) are used to process the image data to generate semantic segmentation maps of the environment. For example, different objects in the image (such as shelves, logistics packaging boxes, ground, etc.) are labeled as different categories to provide more rich environment information for the robot.

[0096] In this embodiment, after the robot receives the shelving instruction, the 4D millimeter wave radar and the hyperspectral camera are started simultaneously to acquire point cloud data and image data respectively. The point cloud data is processed by downsampling and filtering to extract key point features. The image data is processed by normalization to extract color and texture features, and then the point cloud features and image features are fused using a Transformer multi-modal fusion algorithm to identify the features of the target logistics packaging box and the target storage unit through a trained classifier to determine their positions and poses. The processed point cloud data is converted into a three-dimensional mesh model to generate a geometric structure representation of the environment, and the image data is semantically segmented to generate a semantic information representation of the environment. The geometric structure information and semantic information are combined to generate a global environment model. In the global environment model, the position coordinates of the target logistics packaging box and the target storage unit are determined through a pre-set recognition algorithm, and based on these position information, a path is planned to generate an optimal path from the current position of the robot to the position of the target logistics packaging box.

[0097] Illustratively, assume that in a large logistics warehouse, the robot is located at the starting position of the shelf aisle. After the robot receives the shelving instruction, the 4D millimeter wave radar scans the surrounding environment 360° to acquire point cloud data; at the same time, the hyperspectral camera takes pictures of the environment to acquire image data. These data are processed and fused to generate a global environment model.

[0098] Through downsampling and filtering, key point features are extracted to generate a three-dimensional mesh model of the environment; through normalization processing, color and texture features are extracted to generate a semantic segmentation map; then the point cloud features and image features are fused using a Transformer algorithm to generate comprehensive features. Through a trained classifier, the features of the target logistics packaging box and the target storage unit are identified to determine their positions and poses. In the global environment model, the position coordinates of the target logistics packaging box and the target storage unit are determined to generate an optimal path from the current position of the robot to the position of the target logistics packaging box. Through the above steps, the robot can accurately acquire the current position information of the target logistics packaging box and the target storage unit to provide accurate basis for subsequent path planning and grasping operations.

[0099] In this optional embodiment, by combining the point cloud data of the 4D millimeter wave radar and the image data of the hyperspectral camera, the robot can obtain more comprehensive environmental information. The point cloud data provides accurate three-dimensional geometric structure, and the image data supplements detailed information such as color and texture. This embodiment significantly improves the recognition and positioning accuracy of target objects by using a multi-modal data fusion method. The self-attention mechanism in the Transformer architecture is used to weight the fused features, which can automatically learn the weights between different features, further improving the feature representation ability and recognition accuracy. For example, in a complex environment, even if the target object is partially obscured, the robot can still accurately identify its position and attitude.

[0100] At the same time, the dynamic scanning function of the 4D millimeter wave radar can obtain real-time dynamic information in the environment, such as moving personnel and other robots. Combined with the image data of the hyperspectral camera, the robot can update the global environment model in real time, thereby better adapting to the dynamically changing logistics environment. Through semantic segmentation of image data by deep learning algorithm, the robot can identify the category and location of different objects in the environment. This not only improves the recognition ability of target objects, but also enhances the adaptability of the robot to complex environments, enabling it to operate stably in different types of logistics warehouses. The generated global environment model contains geometric structure and semantic information of the environment, providing a comprehensive and accurate basis for path planning. According to the information in the model, the optimal path from the current position to the target position can be planned to avoid collisions and congestion, improving the efficiency and safety of path planning. During path planning, the robot can monitor environmental changes in real time and adjust the path according to the dynamic information in the global environment model. For example, when detecting obstacles in front, the robot can quickly re-plan the path to ensure the smooth progress of the task.

[0101] Through accurate target positioning and attitude recognition, the robot can accurately grasp the logistics packaging box, reducing the failure or damage caused by inaccurate positioning. This not only improves the success rate of the task, but also reduces the time wasted due to repeated operations. Fast processing and fusion of multi-modal data enable the robot to complete environmental perception and target recognition in a short time, thereby quickly responding to the on-shelf instruction and improving the efficiency of task execution. Through automated environmental perception and target recognition, the robot can autonomously complete the on-shelf task of the logistics packaging box, reducing the dependence on manual operation. This not only reduces labor costs, but also reduces the damage and additional costs caused by human operation errors. Optimized path planning and real-time adjustment function enable the robot to operate efficiently in complex logistics environments, reducing idle time and maintenance costs of equipment, and improving equipment utilization.

[0102] Optionally, the path planning is performed according to the current position of the target logistics package box, the current position of the target storage unit, and the current position of the logistics robot, to obtain a target moving path of the logistics robot, including:

[0103] According to the global environment model, environment feature data from the current position of the logistics robot to the current position of the target logistics package box and from the current position of the target logistics package box to the current position of the target storage unit are determined.

[0104] According to the environment feature data, a multi-objective optimization function is used to search for a path in the global environment model, to obtain a first path from the current position of the logistics robot to the current position of the target logistics package box and a second path from the current position of the target logistics package box to the current position of the target storage unit.

[0105] According to the first path and the second path, the target moving path is generated.

[0106] Specifically, after generating the global environment model, first, environment feature data related to path planning are extracted from the model, including information such as the distribution of obstacles, the width of the passage, and the flatness of the ground between the current position of the logistics robot and the target logistics package box and between the target logistics package box and the target storage unit. For example, the system identifies the positions and shapes of obstacles such as shelves, other robots, or goods on the path, and also evaluates whether the width of the passage is sufficient for the robot to pass through and whether the ground is significantly uneven. These environment feature data will be important reference for path planning. Then, a multi-objective optimization function is used to search for a path in the global environment model in combination with the extracted environment feature data. The multi-objective optimization function comprehensively considers multiple factors such as the length, safety, and energy consumption of the path. For example, the system will preferentially select a route with a shorter path and no obstacles, and also consider the energy consumption on the path to avoid the robot running out of power during task execution. Through the optimization algorithm, the system searches for a first path from the current position of the logistics robot to the target logistics package box and a second path from the target logistics package box to the target storage unit. Both paths are optimal paths that meet multiple conditions such as safety and efficiency.

[0107] After obtaining the first path and the second path, the two paths are integrated to generate a complete logistics robot target movement path. In an embodiment of the present application, the integration process includes smoothing the connection points of the two paths to ensure that the robot can transition smoothly from one path to another without sudden turns or stops. For example, if there is a certain angle difference between the end point of the first path and the start point of the second path, the system will insert a transition path to smooth the angle, so that the robot can smoothly move from the position of grabbing the logistics packaging box to the position of the target storage unit. The final target movement path will guide the logistics robot to complete the entire shelving task.

[0108] For example, assume that in a large logistics warehouse, after receiving the shelving instruction, the logistics robot obtains point cloud data and image data of the environment through the 4D millimeter wave radar and hyperspectral camera, and generates a global environment model. In this model, the system identifies the obstacle distribution between the current position of the robot and the target logistics packaging box position, as well as the environmental feature data such as the channel width and ground conditions between the target logistics packaging box position and the target storage unit position. Then, using a multi-objective optimization function, the system searches for two paths in the global environment model: the first path is the optimal path from the current position of the robot to the target logistics packaging box position, and the second path is the optimal path from the target logistics packaging box position to the target storage unit position. Both paths take into account multiple factors such as path length, safety, and energy consumption. Finally, the system integrates the two paths into a smooth target movement path to guide the robot to complete the entire shelving task from the current position to the target storage unit.

[0109] In this optional embodiment, through the global environment model, the robot can obtain comprehensive environmental information from the current position to the target position, including obstacle distribution, channel width, and ground conditions, which makes the overall path planning more accurate, and the robot can avoid obstacles, select the optimal path, and reduce errors and uncertainties in path planning. At the same time, the multi-objective optimization function of this embodiment considers multiple factors such as path length, safety, and energy consumption to ensure that the planned path is not only the shortest, but also the safest and most energy-efficient. For example, even if a path is slightly longer, it will be preferred if it is safer and has lower energy consumption.

[0110] Moreover, through the 4D millimeter wave radar and hyperspectral camera, the robot can perceive real-time environmental changes such as moving personnel and other robots, and these data are updated in real time into the global environment model, enabling the robot to dynamically adjust the path planning to adapt to environmental changes. During path planning, the robot can adjust the path according to real-time environmental data. For example, when an obstacle is detected in front, the robot can quickly re-plan the path to ensure the smooth progress of the task. Through a multi-objective optimization function, the robot can select the optimal path, reducing unnecessary detours and waiting time. This not only improves the efficiency of task execution, but also reduces the energy consumption of the robot. When generating the target moving path, the system smoothes the path connection points to ensure smooth transition of the robot during path switching, without sudden turns or stops. This improves the efficiency of the robot and reduces mechanical wear and tear.

[0111] Through automated path planning and environmental perception, the robot can autonomously complete tasks, reducing dependence on human labor. This not only reduces labor costs, but also reduces damage to goods and additional costs caused by human operation errors. Optimized path planning and real-time adjustment functions enable the robot to efficiently operate in complex logistics environments, reducing idle time and maintenance costs of equipment, and improving equipment utilization. By combining data from 4D millimeter wave radar and hyperspectral camera, the robot can more comprehensively perceive the environment, reducing errors and uncertainties that may be caused by a single sensor. For example, in low-light or complex texture environments, multi-modal data fusion can provide more accurate environmental information. During path planning, the robot can adaptively adjust according to real-time environmental data to ensure the robustness of path planning. Even in emergency situations, the robot can quickly adjust the path to complete the task. In this embodiment, the robot can autonomously complete tasks, reducing dependence on human labor and reducing labor costs.

[0112] Optionally, the method further comprises:

[0113] controlling the logistics robot to move to the target logistics packaging box according to the target moving path at a preset speed and acceleration parameter;

[0114] when the logistics robot reaches the target logistics packaging box, obtaining initial image data of the target logistics packaging box;

[0115] preprocessing the initial image data, and determining whether the target logistics packaging box in the initial image data is complete according to the preprocessed initial image data;

[0116] If not, stop the task of shelving the logistics packaging box, and issue an alarm information;

[0117] If yes, the initial image data is taken as the image data of the target logistics packaging box.

[0118] Specifically, after generating the target movement path, the robot is controlled to move along the path according to the path information, combined with the preset speed and acceleration parameters. In this embodiment, these parameters are pre-set according to the performance of the robot and the task requirements, to ensure the stability and efficiency of the robot during movement. For example, the robot moves at a constant speed in the channel, and gradually slows down when approaching the target logistics packaging box, so as to accurately stop at the target position. The control system will monitor the speed and acceleration of the robot in real time, to ensure that it moves smoothly according to the pre-set parameters, avoiding instability or collision risks caused by excessive speed or acceleration. When the robot reaches the target logistics packaging box position, the hyperspectral camera carried by it starts immediately, to take pictures of the logistics packaging box and obtain initial image data. These image data contain information such as the color, texture and shape of the logistics packaging box. For example, the camera takes pictures of the packaging box from multiple angles at high resolution, to ensure that comprehensive appearance information is obtained. These initial image data will be used for subsequent image processing and analysis, to judge the integrity and status of the packaging box.

[0119] After obtaining the initial image data, the image is pre-processed, wherein the pre-processing steps include image denoising, contrast enhancement and edge detection, etc., to improve the image quality and facilitate subsequent analysis. For example, Gaussian filtering is used to remove noise in the image, and histogram equalization is used to enhance the contrast of the image, to make the outline of the logistics packaging box clearer. After pre-processing, the system uses image recognition algorithms to analyze the packaging box, to judge whether it is complete. For example, by detecting whether the edges of the packaging box are continuous, whether the surface is damaged or deformed, etc., to judge the integrity of the packaging box.

[0120] If the image analysis result shows that the logistics packaging box is not complete, for example, if it is found that the packaging box is damaged or deformed, the shelving task will be stopped immediately. To prevent damaged packaging boxes from entering the subsequent logistics process and avoid possible greater losses. At the same time, the system will issue an alarm information through the pre-set communication module, to notify the operator or management system. For example, the alarm information can be sent to the monitoring system of the logistics center through wireless network, and the operator can take timely measures to handle the damaged packaging box.

[0121] If the image analysis result shows that the logistics packaging box is complete, the pre-processed initial image data will be used as the image data of the target logistics packaging box for subsequent grabbing operations. These image data contains detailed appearance information of the packaging box, which will be used to determine the grabbing position and method. For example, the system will select the appropriate grabbing point according to the color and texture information in the image to ensure the stability and safety of the grabbing process.

[0122] For example, assume that in a large logistics warehouse, after receiving the shelving instruction, the logistics robot moves smoothly to the position of the target logistics packaging box according to the generated target movement path at a preset speed and acceleration parameter. After arriving, the hyperspectral camera carried by the robot takes a picture of the packaging box to obtain initial image data. The control system pre-processes the image, including denoising, contrast enhancement, and edge detection, etc. operations to improve the image quality. Then, the system uses image recognition algorithms to analyze the integrity of the packaging box and finds that the surface of the packaging box is not damaged or deformed, judging that it is complete. Therefore, the robot uses the pre-processed image data as the image data of the target logistics packaging box and continues to perform the grabbing operation. If it is found that the packaging box is damaged during the analysis process, the robot will stop the task and issue an alarm information to notify the operator to handle.

[0123] In this optional embodiment, by using a hyperspectral camera to capture rich spectral information, not only visible light but also near-infrared and other wavebands, these information enable the robot to more accurately identify the color, texture and shape of the logistics packaging box. For example, even in low light or complex background environment, the hyperspectral camera can provide clear and accurate image data to ensure the accuracy of identification.

[0124] Through image denoising, contrast enhancement and edge detection preprocessing steps, the robot can further improve the image quality and reduce the interference of environmental factors on identification. For example, Gaussian filtering can effectively remove noise in the image, and histogram equalization can enhance the contrast of the image to make the outline of the logistics packaging box clearer, thereby improving the accuracy of identification.

[0125] Moreover, after the robot reaches the target logistics packaging box location, the packaging box is checked for integrity through image recognition algorithms. If damage or deformation is found on the packaging box, the robot will immediately stop the task and send an alert message to notify the operator to handle it. This process avoids damaged packaging boxes entering the subsequent logistics process, reducing packaging box losses due to delayed or negligent manual inspection. At the same time, only after confirming the integrity of the packaging box, the robot will continue to perform the grabbing operation. Through precise image recognition and grabbing point selection, the robot can more stably and safely grab the logistics packaging box, avoiding damage to the packaging box due to improper grabbing. For example, the robot can select the best grabbing point based on color and texture information in the image to ensure the stability and safety of the grabbing process. The robot can automatically complete the entire process from path planning to packaging box recognition and grabbing, reducing manual intervention. This not only improves the efficiency of logistics operations, but also reduces the time wasted due to human operation errors. During the identification process, the robot can provide real-time feedback on the status of the packaging box and adjust the operation as needed. For example, if a slight deformation is found on the packaging box, the robot can adjust the grabbing force and angle to ensure successful grabbing and avoid repeated operations.

[0126] This embodiment reduces the dependence on manual inspection through automated integrity checking, reducing labor costs. At the same time, it reduces packaging box losses due to delayed or negligent manual inspection, further reducing operating costs. The robot can efficiently and stably complete tasks, reducing idle time and maintenance costs of equipment, and improving equipment utilization.

[0127] Optionally, the method further comprises:

[0128] The image data of the target logistics packaging box is subjected to feature extraction through a deep learning algorithm to obtain surface material characteristics and volume shape characteristics of the target logistics packaging box.

[0129] The surface material type of the target logistics packaging box is determined according to the surface material characteristics.

[0130] The size parameters of the target logistics packaging box are determined according to the volume shape characteristics.

[0131] The surface material type and the size parameters are taken as the external feature data of the target logistics packaging box.

[0132] Specifically, after obtaining the image data of the target logistics packaging box, a pre-trained deep learning model is used to extract features from the image. For example, a convolutional neural network (CNN) model such as ResNet or VGG is used to process the image data, which can automatically learn features in the image, including surface material features and volume shape features. Specifically, the model extracts texture, color distribution, edge information, etc. in the image to identify the surface material of the packaging box; at the same time, by analyzing the geometric shape and size information in the image, the volume shape features of the packaging box are extracted. For example, the model can identify the smoothness of the packaging box surface, the texture type (such as plastic, paperboard, etc.), and the length, width, and height dimensions of the packaging box. After extracting the surface material features, a pre-set classifier is used to classify these features to determine the surface material type of the target logistics packaging box. For example, a support vector machine (SVM) or a deep learning classifier is used to classify the extracted surface material features. These classifiers are trained based on a large amount of labeled data and can accurately identify the features of different materials. For example, by analyzing the texture and color distribution in the image, the classifier can determine whether the packaging box surface is made of plastic or paperboard. This process ensures that the robot can accurately identify the surface material of the packaging box, providing important basis for subsequent grabbing operations. After extracting the volume shape features, geometric analysis algorithms are used to determine the size parameters of the target logistics packaging box. For example, by analyzing the geometric shape and size information in the image, the length, width, and height dimensions of the packaging box are calculated. Specifically, an edge detection algorithm is used to find the outline of the packaging box, and then its size parameters are calculated through geometric calculation. For example, by detecting the four corner points of the packaging box, its length and width are calculated; by analyzing the height information of the packaging box, its height is calculated; this process ensures that the robot can accurately obtain the size parameters of the packaging box, providing accurate data support for subsequent grabbing and handling operations.

[0133] After determining the surface material type and size parameters, these information is integrated into the external feature data of the target logistics packaging box, which is used for subsequent grabbing mode selection and grabbing strategy formulation. For example, if the packaging box is made of plastic and has a large size, the robot may choose to use a suction cup type grabber; if the packaging box is made of paperboard and has a small size, the robot may choose to use a jaw type grabber. These external feature data provide comprehensive packaging box information for the robot, ensuring the stability and safety of the grabbing operation.

[0134] In a preferred embodiment of the present application, the pre-trained deep learning model is a convolutional neural network (CNN) model based on the ResNet architecture. The model includes an input layer: the size of the input image is 224x224x3, where 224 is the width and height of the image, and 3 represents the RGB three color channels. Convolutional layers and pooling layers: the first convolutional layer uses a 7x7 convolutional kernel with a step size of 2 and outputs 64 channels. Followed by a max pooling layer with a pooling window size of 3x3 and a step size of 2. ResNet is composed of multiple residual modules, each containing two convolutional layers and a skip connection. In this embodiment, ResNet-50 is used, which contains a 50-layer deep network structure, which can be divided into the following stages: First stage: 3 residual modules, each module contains two 3x3 convolutional layers, and the output channel numbers are 64, 64, and 256 respectively. Second stage: 4 residual modules, each module contains two 3x3 convolutional layers, and the output channel numbers are 128, 128, and 512 respectively. Third stage: 6 residual modules, each module contains two 3x3 convolutional layers, and the output channel numbers are 256, 256, and 1024 respectively. Stage: 3 residual modules, each module contains two 3x3 convolutional layers, and the output channel numbers are 512, 512, and 2048 respectively. Fully connected layer: After the last residual module, a global average pooling layer is used to reduce the size of the feature map to 1x1x2048, then a fully connected layer is connected, which outputs the feature vector of the target logistics packaging box.

[0135] The training process of the deep learning model includes: first, collect a large amount of logistics packaging box image data, including packaging boxes of different materials (such as plastic, paperboard, etc.) and different sizes, and each image is labeled, including surface material type and size parameters. Preprocess the image data, including cropping, scaling, normalization, etc. to ensure that the image size of the input model is consistent.

[0136] The ResNet model is trained using the labeled image data. During training, the cross-entropy loss function is used to measure the difference between the model's predicted values and the true values. The stochastic gradient descent (SGD) optimizer is used, with a learning rate of 0.001, a momentum of 0.9, and a weight decay of 0.0001. During training, data augmentation techniques such as random cropping and horizontal flipping are used to increase the model's generalization ability. The training process continues for multiple epochs (e.g., 100 epochs) until the model's performance on the validation set no longer improves. The model's performance is evaluated on the test set using accuracy, recall, and F1 score as indicators of recognition accuracy. If the model's performance is not satisfactory, adjustments can be made to the model structure or training parameters, such as changing the network depth or adding Dropout layers, or adjusting training parameters such as learning rate and optimizer, and enhancing data augmentation strategies. Cross-validation is used to ensure the model's performance is stable across different data subsets, and the model's performance is evaluated using accuracy, recall, and other indicators. When the model reaches the predetermined performance indicators on the validation and test sets, and has stable performance, good generalization ability, and acceptable resource consumption, the model is determined as the final model for actual logistics packaging box recognition and grasping tasks.

[0137] In practical applications, after the robot obtains the image data of the target logistics packaging box, it first performs preprocessing, including cropping, scaling, normalization, and other operations to ensure that the input model's image size is 224x224x3. The preprocessed image is input into the trained ResNet model, which outputs the target logistics packaging box's feature vector, including surface material features and volume shape features. A pre-set classifier (such as SVM or deep learning classifier) is used to classify the surface material features to determine the surface material type. Geometric analysis algorithms are used to process the volume shape features to calculate the size parameters of the target logistics packaging box. The surface material type and size parameters are integrated into the external feature data of the target logistics packaging box, which is used for subsequent grasping operations.

[0138] Assuming that in a large logistics warehouse, the logistics robot receives the instruction to put on the shelf, arrives at the target logistics packaging box location, and obtains its image data. The robot control system first preprocesses the image, including cropping, scaling, and normalization, to ensure that the image size of the input model is 224x224x3. Then, the preprocessed image is input into the trained ResNet model, and the model outputs the feature vector of the target logistics packaging box. The system uses a pre-set classifier to classify the surface material features and determine that the packaging box surface is plastic. Then, the system calculates the size parameters of the packaging box, such as 50 cm long, 30 cm wide, and 20 cm high, through a geometric analysis algorithm. Finally, the system integrates these information into the external feature data of the target logistics packaging box, which is used for subsequent grabbing operations. If the packaging box is plastic and large in size, the robot may choose to use a suction cup type grabber to ensure the stability and safety of the grabbing process.

[0139] In this optional embodiment, deep learning algorithms, especially convolutional neural networks (CNN), are used to automatically learn complex features in images, including surface material and volume shape. For example, through a CNN model with a ResNet architecture, high-dimensional feature representations can be extracted, which can more accurately describe the surface material and volume shape of the logistics packaging box. This automated feature extraction method is more accurate and efficient than traditional manual feature extraction methods. By extracting surface material features through a deep learning model, subtle differences between different materials (such as plastic, paperboard, etc.) can be distinguished. For example, the model can identify the smoothness and reflectivity of a plastic surface, as well as the texture and color distribution of a paperboard surface, to accurately determine the type of packaging box surface material. Deep learning models can extract volume shape features of packaging boxes, including length, width, and height dimensions. By analyzing the geometric shapes and size information in the image, the model can accurately calculate the size parameters of the packaging box, ensuring the accuracy of the grabbing operation. Deep learning algorithms can automatically extract features from image data, reducing the workload of manual annotation and feature extraction. For example, traditional feature extraction methods require manual design and selection of features, while deep learning models can directly extract the most useful features from images through automatic learning, improving work efficiency. The model not only automatically extracts features, but also automatically classifies surface material types and calculates size parameters based on these features. For example, through a trained classifier, the model can automatically identify the material type of the packaging box without manual inspection and labeling, further reducing human intervention. Deep learning models can complete feature extraction and classification tasks in a short time, improving the efficiency of logistics operations. For example, the model can process an image and output the result in a few milliseconds, greatly shortening the processing time compared to traditional methods.

[0140] By accurately determining the surface material type and size parameters of the target logistics package, the robot can choose the most appropriate grabbing method, reducing the failure or damage caused by improper grabbing. For example, for a plastic material packaging box, the robot may choose a suction cup type grabbing; for a paperboard material packaging box, it may choose a jaw type grabbing. This precise grabbing operation improves the success rate and efficiency of logistics operations. Through automatic feature extraction and classification, the dependence on manual inspection is reduced, and the labor cost is reduced. For example, in traditional logistics operations, manual inspection of the integrity and material of the packaging box is required, while the automatic system of the present invention can automatically complete these tasks, reducing the need for manual inspection. Through precise feature extraction and classification, the robot can avoid grabbing damaged packaging boxes, reducing the loss of packaging boxes caused by human negligence. For example, if the model detects that the packaging box is damaged, the robot will stop the task and issue an alarm to avoid damaged packaging boxes from entering the subsequent process, thereby reducing operating costs.

[0141] Optionally, the method further comprises:

[0142] According to the surface material type and size parameters of the target logistics package, determining the grabbing tool of the logistics robot;

[0143] According to the grabbing tool, querying a pre-set grabbing mode database to determine the grabbing mode corresponding to the grabbing tool.

[0144] Specifically, after determining the surface material type and size parameters of the target logistics package, the appropriate grabbing tool is selected according to these external feature data. For example, if the target logistics package is plastic and large in size, the system will select a suction cup type grabbing tool, because the plastic surface is usually smooth and the suction cup can provide stable suction force. If the packaging box is made of paperboard and is small in size, the system will select a jaw type grabbing tool, because the jaw can better adapt to irregularly shaped and small sized objects. After selecting the appropriate grabbing tool, the pre-set grabbing mode database is queried to determine the grabbing mode corresponding to the grabbing tool. The grabbing mode database contains the operation parameters and steps of different grabbing tools in different situations. For example, for a suction cup type grabbing tool, the database may record the suction pressure, suction time, release pressure, etc. For a jaw type grabbing tool, the database may record the opening and closing range of the jaw, grabbing force, grabbing angle, etc. According to the specific characteristics of the target logistics package, the most suitable grabbing mode is selected from the database to ensure the stability and safety of the grabbing operation.

[0145] In a preferred embodiment of the present invention, the pre-set grabbing mode database contains the following fields:

[0146] (1) Gripper type: Identifies the type of gripper used, such as suction cup, jaw, etc.

[0147] (2) Suitable material: Identifies the type of material the packaging box surface is made of, such as plastic, paperboard, etc.

[0148] (3) Suitable size range: Identifies the size range of the packaging box that the gripper is suitable for, such as length, width, and height.

[0149] (4) Operating parameters: Identifies the specific parameters of the gripper when performing the gripping task, such as suction pressure, gripping force, suction time, release pressure, etc.

[0150] (5) Operating steps: Identifies the specific steps of the gripper when performing the gripping task, such as approaching the packaging box, suction / gripping, lifting, releasing, etc.

[0151] Table 1: Database query table for logistics packaging box gripping mode

[0152]

[0153] With the database shown in Table 1, first, determine the gripper, according to the surface material type and size parameters of the target logistics packaging box, select the most suitable gripper from the database. For example, if the target logistics packaging box is made of plastic, the size is 50x30x20 cm, the system will select a suction cup gripper. According to the selected gripper, query the corresponding gripping mode in the database. For the suction cup gripper, the system will find the following operating parameters and steps: operating parameters: suction pressure: 0.5 bar, suction time: 2 seconds, release pressure: 0.2 bar, operating steps include: move above the packaging box, suction the packaging box, lift the packaging box, release the packaging box, then perform the gripping operation, the robot performs the gripping operation according to the queried gripping mode. For example, the robot moves above the packaging box, starts the suction cup, suctions the packaging box at a pressure of 0.5 bar, holds for 2 seconds, lifts the packaging box, and releases the packaging box at a pressure of 0.2 bar after reaching the target position.

[0154] Suppose in a large logistics warehouse, the logistics robot receives the instruction to put on the shelf, reaches the target logistics packaging box position, and obtains its image data. The robot control system extracts the surface material characteristics and volume shape characteristics of the packaging box through deep learning algorithms, determines that the packaging box surface is made of plastic, and the size is 50x30x20 cm. According to these external characteristic data, the system queries the pre-set gripping mode database and selects a suction cup gripper. The system finds the gripping mode corresponding to the suction cup gripper, including suction pressure of 0.5 bar, suction time of 2 seconds, and release pressure of 0.2 bar. The robot performs the gripping operation according to these parameters to ensure the stability and safety of the gripping process.

[0155] In this optional embodiment, by analyzing the surface material type and size parameters of the target logistics packaging box, the robot can accurately select the most suitable grabbing tool. For example, for a plastic packaging box, a suction cup grabbing tool can be selected to take advantage of its smooth surface suction characteristics; for a paperboard packaging box, a jaw grabbing tool can be selected to better adapt to its irregular shape and smaller size. This precise matching significantly improves the success rate of grabbing. By querying the pre-set grabbing mode database, the robot can obtain the optimal grabbing mode corresponding to the selected grabbing tool. These modes contain verified operation parameters and steps to ensure the stability and reliability of the grabbing operation. For example, the suction pressure and time of the suction cup, the opening and closing range of the jaw, and the grabbing force of the jaw are optimized to reduce the risk of grabbing failure. At the same time, different packaging box materials and sizes require different grabbing strategies. Through accurate identification and matching, the robot can select the most suitable grabbing tool and mode according to the specific characteristics of the packaging box, reducing damage to the packaging box caused by improper grabbing. For example, for fragile plastic packaging boxes, using a suction cup grabbing tool can avoid the extrusion damage caused by a jaw. During the grabbing process, the robot can adjust the grabbing parameters according to real-time feedback to ensure the smoothness and safety of the grabbing process. For example, if a slight deformation of the packaging box surface is detected during the grabbing process, the robot can adjust the suction pressure of the suction cup or the grabbing force of the jaw to avoid damaging the packaging box.

[0156] By automatically selecting the grabbing tool and grabbing mode, the robot can quickly and accurately complete the grabbing task, reducing the time for manual intervention and adjustment. For example, traditional grabbing operations may require manual selection of grabbing tools and adjustment of parameters based on the characteristics of the packaging box, while the automatic system of the present invention can complete these operations in a few seconds, significantly improving logistics efficiency. Due to the improvement in the success rate of grabbing, repeated operations caused by grabbing failure are reduced. For example, if the grabbing fails, the robot needs to re-adjust the grabbing tool and parameters, which not only wastes time but also may further damage the packaging box. By optimizing the grabbing mode, this repeated operation is reduced, improving overall logistics efficiency.

[0157] By automatically selecting the grabbing tool and grabbing mode, the dependence on human labor is reduced, and the labor cost is lowered. For example, in traditional logistics operations, manual inspection of packaging box characteristics and selection of grabbing tools are required, while the automatic system of the present invention can automatically complete these tasks, reducing the need for manual inspection and adjustment. Through precise grabbing operation, damage to the packaging box caused by improper grabbing is reduced, and additional costs caused by damaged packaging boxes are reduced. For example, damaged packaging boxes may need to be re-packaged or processed, increasing operating costs. By optimizing the grabbing mode, this loss is reduced, improving operational efficiency.

[0158] Optionally, the determining the grasping strategy of the mechanical arm of the logistics robot according to the grasping mode comprises:

[0159] determining a grasping point of the mechanical arm on the target logistics packaging box according to the grasping mode and the size parameter;

[0160] determining a grasping force of the mechanical arm on the target logistics packaging box according to the grasping point and the surface material type;

[0161] generating a grasping path of the mechanical arm according to the current position of the target logistics packaging box and the grasping point;

[0162] obtaining the grasping strategy of the mechanical arm according to the grasping path and the grasping force.

[0163] Specifically, since the size parameters such as length, width, and height will directly affect the selection of the grasping point, after determining the grasping mode, the control system of the logistics robot determines the grasping point of the mechanical arm according to the size parameters of the target logistics packaging box to adapt to packaging boxes of different sizes. For example, if a suction cup type grasping tool is used, the control system will calculate the optimal suction position in the central region of the packaging box to ensure that the suction cup can cover a large enough area to provide stable grasping force. For a gripper type grasping tool, the control system will determine the edge position where the gripper should grip the packaging box to maintain stability during lifting,

[0164] The grasping force of the mechanical arm is adjusted according to the selected grasping point and the surface material type of the packaging box. For example, for smooth plastic material, the suction cup may need higher suction pressure to ensure that it does not slip; while for rough paperboard material, lower suction pressure may be sufficient. The control system will refer to the operation parameters in the pre-set grasping mode database, such as suction pressure or grasping force, to set the grasping force of the mechanical arm to adapt to packaging boxes of different materials. After determining the grasping point, the optimal path of the mechanical arm from the current position to the grasping point is calculated by a path planning algorithm, which will give priority to the kinematic constraints of the mechanical arm and obstacles in the working environment. For example, if the mechanical arm needs to bypass shelves or other obstacles to reach the grasping point, the path planning algorithm will generate a path that avoids these obstacles, thus ensuring that the path is smooth and continuous to avoid damage to the packaging box during grasping.

[0165] After the grasping path is generated and the grasping force is determined, the control system integrates these information to form a grasping strategy for the robotic arm, which includes how the robotic arm should move to the grasping point, how to apply the grasping force, and how to lift and move the package after grasping. For example, the strategy can instruct the robotic arm to move to the grasping point at a specific speed and acceleration, to grasp or clamp the package at a specific pressure, and to lift the package to the designated location at a steady speed; this strategy ensures the accuracy and safety of the entire grasping process.

[0166] In a preferred embodiment of the present application, the path planning algorithm is an artificial intelligence-based optimization algorithm that finds an optimal path for the logistics robot to move from the current location to the target logistics package location and from the target logistics package location to the target storage unit location. The path planning algorithm of this embodiment combines graph search algorithms and heuristic methods, such as A* search algorithm or Dijkstra algorithm.

[0167] For example, the specific implementation steps of the path planning algorithm include:

[0168] First, an environmental model is constructed based on the layout and obstacle information of the logistics warehouse, which is usually represented as a graph, where nodes represent possible robot positions and edges represent paths for the robot to move from one position to another.

[0169] An evaluation function is then defined to evaluate the path from the starting position to the target position, taking into account factors such as path length, estimated movement time, ease of avoiding obstacles, and whether special actions such as turning or climbing are required. Graph search algorithms are then applied to search for the shortest or optimal path from the starting point to the ending point. For example: A* search algorithm, which combines Dijkstra algorithm and heuristic search algorithm, uses heuristic function to estimate the cost from current node to target node, while ensuring that the path found is the shortest. Dijkstra algorithm is an algorithm that ensures the shortest path is found by constantly selecting the nearest unvisited node and updating the distance of its neighbor nodes.

[0170] At the same time, in order to improve the efficiency of the algorithm, this embodiment introduces heuristic method, which estimates the distance from any node to the target node based on specific features of the environment, such as straight-line distance or known obstacle paths. After finding the initial path, further optimization is performed to ensure smooth and safe movement of the robot. This may include path smoothing techniques, such as interpolation methods to make the path smoother and reduce sharp turns and unnecessary actions, thereby improving the efficiency and safety of the robot's movement. During the movement of the robot, if changes in the environment are detected (such as dynamic obstacles), the path planning algorithm can adjust the path in real time to avoid collisions and ensure task completion.

[0171] In this embodiment, the path planning algorithm utilizes information obtained from the global environment model, combined with the robot's current position and target position, to generate an efficient and safe path. This algorithm not only considers the constraints of robot motion but also the dynamics and complexity of logistics operations, thereby ensuring that the robot can efficiently perform tasks in complex environments.

[0172] For example, suppose a logistics robot needs to grasp a plastic packaging box measuring 50x30x20 cm. The control system first selects a suction cup gripping tool and corresponding operating parameters from a pre-set gripping pattern database based on the box's size and material. Next, the system determines that the robotic arm's gripping point is located in the center of the packaging box and sets the suction pressure of the suction cup to 0.5 bar. Then, the system generates a gripping path from the robotic arm's current position to the center of the packaging box, ensuring the path avoids surrounding obstacles. Finally, based on this path and suction pressure, the control system formulates the robotic arm's gripping strategy, including moving to the gripping point at a smooth speed, activating the suction cup, adsorbing the packaging box, lifting it, and moving it to the designated position. In this way, the robot can safely and accurately grasp the packaging box, improving the efficiency and reliability of logistics operations.

[0173] In this optional embodiment, by precisely determining the robotic arm's grasping strategy, efficient and stable grasping of logistics packaging boxes of different sizes and materials is achieved, significantly improving the operational accuracy and efficiency of logistics automation. Utilizing surface material and size parameters extracted through deep learning algorithms, combined with a pre-set grasping pattern database, the robot can intelligently select the most suitable grasping tool and pattern, ensuring a high degree of match between the grasping strategy and the characteristics of the packaging box. Furthermore, by accurately calculating the grasping point and grasping force, the robot can adapt to packaging boxes of various shapes and sizes, reducing potential damage to goods during grasping and thus lowering the loss rate. The generated grasping path takes into account the kinematic limitations of the robotic arm and obstacles in the working environment, making the grasping process smoother and safer, avoiding equipment collisions or goods falling due to improper paths. Overall, this invention not only improves the level of automation in logistics operations but also reduces logistics operating costs by minimizing human intervention and optimizing resource utilization, enhancing the reliability and adaptability of the entire logistics system.

[0174] Optionally, the step of transporting the target logistics packaging box to the target storage unit and performing a shelving operation on the target logistics packaging box using a force-position hybrid control algorithm includes:

[0175] When the logistics robot arrives at the target storage unit, it acquires the pose data of the target storage unit and the current pose data of the target logistics packaging box.

[0176] According to the pose data of the target storage unit and the current pose data of the target logistics packaging box, a pose deviation between the target logistics packaging box and the target storage unit is determined, the pose deviation including a position deviation and an attitude deviation;

[0177] According to the position deviation and the attitude deviation, a position strategy, an attitude strategy, and a grasping force strategy of the mechanical arm are determined through a force-position hybrid control algorithm;

[0178] According to the position strategy, the attitude strategy, and the grasping force strategy, a comprehensive control instruction of the mechanical arm is generated;

[0179] The target logistics packaging box is subjected to a shelving operation through the comprehensive control instruction.

[0180] Specifically, when the logistics robot carrying the target logistics packaging box arrives at the target storage unit, first, the pose data of the target storage unit, including its position and orientation, is obtained by using the sensors carried by the robot, such as a camera or a laser radar. At the same time, the current pose data of the target logistics packaging box is also obtained, which can be achieved by the internal sensors of the robot, such as an encoder or an IMU (Inertial Measurement Unit), providing a basis for subsequent pose deviation calculation through the acquisition of these data.

[0181] According to the obtained pose data, the pose deviation between the target logistics packaging box and the target storage unit is calculated through a geometric transformation algorithm, including a position deviation, i.e., the straight-line distance and direction between the center points of the packaging box and the storage unit, and an attitude deviation, i.e., the difference between the orientation of the packaging box and the orientation of the storage unit. Ensuring that the robot can accurately adjust the position and attitude of the packaging box to match the requirements of the storage unit.

[0182] After determining the pose deviation, a force-position hybrid control algorithm is used to determine the position strategy, pose strategy, and grasp force strategy of the robotic arm based on the robot's dynamics model and control algorithms such as PID control or model predictive control. The position strategy guides the robotic arm on how to move to reduce the positional deviation, the pose strategy guides how to adjust the orientation of the packaging box to match the storage unit, and the grasp force strategy ensures that the packaging box is safely and stably grasped during the adjustment process. According to the determined position strategy, pose strategy, and grasp force strategy, the control system generates comprehensive control instructions, which detail how the robotic arm should move during the shelving operation, including speed, acceleration, rotation angle, and grasping force, etc. These instructions aim to ensure that the robotic arm can smoothly and accurately place the packaging box into the designated storage unit. By sending the comprehensive control instructions to the driver of the robotic arm, the shelving operation is executed. The robotic arm moves to the correct position according to the instructions, adjusts the pose of the packaging box, and then grasps and places the packaging box with appropriate force. Throughout the process, the control system continuously monitors the state of the robotic arm and the pose of the packaging box to ensure the accuracy and safety of the operation.

[0183] In a preferred embodiment of the present application, the pose data of the target logistics packaging box and the target storage unit is obtained from the robot sensors (such as vision sensors or position sensors), including position coordinates (x, y, z) and attitude (usually represented by quaternions or Euler angles). Using geometric transformation algorithms, the pose deviation between the target logistics packaging box and the target storage unit is calculated, including positional deviation (Δx, Δy, Δz) and attitude deviation (Δθ_x, Δθ_y, Δθ_z). Based on the robot's dynamics model, which describes the kinematics and dynamics characteristics of the robotic arm joints, including joint stiffness, damping, mass, etc. The pose deviation is input into the control algorithm, such as PID control or model predictive control (MPC). The PID controller calculates the required joint torque in real time by adjusting the proportional (P), integral (I), and derivative (D) parameters to reduce the pose deviation. MPC optimizes the control input by predicting future pose deviations to generate the optimal joint trajectory. The position strategy output by the control algorithm includes the target position and movement speed of each joint of the robotic arm. The kinematics model of the robotic arm calculates the precise motion path of the end effector based on these strategies, ensuring that the target logistics packaging box can be accurately moved to the position of the target storage unit.

[0184] The pose data of the target logistics packaging box and the target storage unit are obtained from the sensors. Using a geometric transformation algorithm, the pose deviation between the two is calculated. Meanwhile, this embodiment considers the pose adjustment capability of the robot end effector, including the kinematics and dynamics characteristics of the rotary joint. The pose deviation is input into the control algorithm, and the PID controller also calculates the torque required to be applied to the rotary joint in real time by adjusting the P, I, and D parameters, thereby achieving pose adjustment. The MPC optimizes the control input by predicting future pose changes, generating a smooth pose adjustment trajectory. The pose strategy output by the control algorithm includes the target angle and rotation speed of the rotary joint. The kinematics model of the robot calculates the pose adjustment path of the end effector based on these strategies, ensuring that the pose of the target logistics packaging box can match the target storage unit.

[0185] The physical characteristic data of the target logistics packaging box, such as weight, surface friction coefficient, etc., are obtained. Based on the dynamics model of the grabbing mechanism, which describes the relationship between the grabbing force and the force on the packaging box, including the stiffness, damping, etc. of the grabbing mechanism. The physical characteristic data is input into the control algorithm, and the PID controller also calculates the force required to be applied to the grabbing mechanism in real time by adjusting the P, I, and D parameters, thereby maintaining stable grabbing. The MPC optimizes the distribution and size of the grabbing force by predicting the force changes on the packaging box during grabbing and adjustment. The grabbing force strategy output by the control algorithm includes the target grabbing force of the grabbing mechanism and the dynamic adjustment parameters of the grabbing force. The grabbing force is monitored in real time through the force sensor of the grabbing mechanism, and closed-loop control is performed in combination with the control algorithm, ensuring that the grabbing force is stable at the set value while being able to dynamically adjust to adapt to different operating states.

[0186] For example, assume that the robot needs to place a 5kg logistics package box onto a target storage unit: the robot obtains the positional deviation between the package box and the storage unit as (0.1m, 0.05m, 0.15m) through the vision sensor. Based on the dynamic model and PID control algorithm, the angles that each joint of the robotic arm needs to move are calculated, so that the end effector moves along a straight path to the target position. The PID parameters are pre-tuned according to experiments to ensure that the robotic arm can quickly and smoothly reach the target position. The vision sensor also obtains the attitude deviation as (5°, 3°, 8°), and the MPC algorithm is used to predict the attitude change in the next few seconds, optimize the motion trajectory of the rotating joints, and gradually align the attitude of the package box with the storage unit. The MPC algorithm considers the dynamic characteristics and motion constraints of the robotic arm to ensure the smoothness and stability of the attitude adjustment. According to the weight and surface friction coefficient of the package box, the minimum required gripping force is calculated as 50N. Through the PID control algorithm, the gripping force is monitored in real time and closed-loop controlled to ensure that the gripping force is stable at 60N (including a certain safety margin). If the gripping force is detected to decrease during the attitude adjustment process, the PID controller will immediately increase the gripping force to prevent the package box from slipping. Through the above process, the robot can accurately adjust the position, attitude, and gripping force of the robotic arm, ensuring that the logistics package box is safely and stably placed into the target storage unit.

[0187] Suppose a logistics robot needs to place a plastic package box with dimensions 50x30x20 cm into a specific unit of a shelf. After the robot reaches the shelf, its control system first obtains the pose data of the shelf unit and the package box. Then, the system calculates the pose deviation between the two, including position and attitude. Based on these deviations, the control system determines how to adjust the position, attitude, and gripping force of the robotic arm through the force-position hybrid control algorithm. Then, the system generates a set of detailed control instructions to guide the robotic arm to accurately move to the position of the shelf unit, adjust the attitude of the package box, and then smoothly place the package box. In this way, the robot can efficiently and accurately complete the shelving task, reducing the need for manual operation and improving the automation level of logistics operations.

[0188] In this optional embodiment, by accurately obtaining the pose data of the target storage unit and the logistics package box, and calculating the pose deviation between them, the robot can very accurately adjust the motion of its robotic arm, thereby ensuring that the robot can accurately place the logistics package box on the specified storage location, reducing the errors that may occur due to human operation, and improving the overall accuracy of the shelving operation.

[0189] Meanwhile, the force-position hybrid control algorithm takes into account both position and attitude control, as well as the required grasping force during operation, providing a smooth and controllable operation mode for the robot. During the shelving process, the robot can adapt to the weight and shape changes of the packaging boxes, maintaining a stable grasping force to avoid damage to the goods due to improper force. Moreover, the integrated control instructions make the robot's shelving operation more smooth and efficient. The robot can quickly adjust its strategy based on real-time feedback, reducing unnecessary movement and waiting time, thereby speeding up the shelving speed of individual packaging boxes and improving the processing capacity of the entire logistics center. The application of the algorithm enables the robot to adapt to different sizes and weights of logistics packaging boxes, as well as different types and layouts of storage units. This flexibility and adaptability allows the robot to reliably complete the shelving task in a variable logistics environment. Automated shelving operations reduce the dependence on manual labor and also reduce the error rate caused by human factors. This not only reduces the labor costs of enterprises, but also improves the reliability and customer satisfaction of logistics operations.

[0190] Optionally, the method further comprises:

[0191] After the target logistics packaging box is placed, an image of the placement position of the target logistics packaging box in the target storage unit is obtained, and the shelving result of the target logistics packaging box is verified according to the image of the placement position;

[0192] If the shelving result verification is passed, the automatic shelving task of the logistics packaging box is completed;

[0193] If the shelving result verification is not passed, the position strategy, the attitude strategy, and the grasping force strategy are adjusted according to the verification result, and the shelving operation is performed again according to the position strategy, the attitude strategy, and the grasping force strategy until the shelving result verification of the target logistics packaging box is passed.

[0194] Specifically, after the logistics packaging box is placed in the target storage unit, the robot control system activates its mounted image acquisition device, such as a high-resolution camera, to obtain the latest placement position image of the packaging box in the storage unit. The image data of these placement position images provides visual information of the packaging box position, attitude, and stability for the subsequent shelving result verification process. For example, the camera can take pictures of the packaging box from different angles to ensure that the complete placement situation is obtained. The control system transmits the image data of the obtained placement position image to the built-in image processing module, which uses image recognition and machine learning algorithms to analyze whether the packaging box is correctly placed in the storage unit. The verification process includes checking whether the position of the packaging box is accurate, stable, and whether there is possible damage. For example, the system may compare the distance between the edge of the packaging box and the boundary of the storage unit to confirm whether it is in the predetermined position.

[0195] If the analysis result of the image processing module shows that the placement of the packaging box meets all the predetermined conditions, such as accurate position, correct and stable posture, the control system will determine that the shelving result is verified. At this time, the robot will mark the automatic shelving task of the logistics packaging box as completed, and can continue to execute the next task or enter the standby state. For example, the system may send a confirmation signal to the management system of the logistics center, indicating that the packaging box has been successfully shelved.

[0196] If the shelving result fails to pass the verification, the control system will analyze the reasons for the failure according to the verification result. For example, whether it involves position deviation, incorrect posture, or insufficient gripping force. Then, the system will adjust the corresponding position strategy, posture strategy and gripping force strategy. For example, if the position deviation is large, the system may increase the movement accuracy of the robot arm; if the posture is incorrect, the rotation angle of the robot arm may be adjusted. After adjustment, the robot will re-execute the shelving operation according to the new strategy until the shelving result is verified.

[0197] Suppose a logistics robot places a plastic packaging box into a specific unit of the shelf. After placement, the robot uses a camera to obtain an image of the packaging box's position on the shelf. The control system analyzes the image to confirm the position and posture of the packaging box. If it finds that the packaging box deviates slightly from the predetermined position, the system will adjust the position strategy of the robot arm according to this verification result, such as by reducing the movement amplitude of the robot arm or increasing the position correction steps to correct. After adjustment, the robot re-executes the shelving operation until the control system confirms through image data analysis that the packaging box is correctly placed, the position is accurate and stable, thus completing the shelving task. This process not only ensures that each packaging box is correctly shelved, but also improves the overall accuracy and reliability of the logistics operation.

[0198] In this optional embodiment, the shelving result is verified by the placement position image, ensuring that the logistics packaging box is accurately placed in the target storage unit. This automated verification process reduces human error and improves placement accuracy; if the shelving result fails to pass the verification, the system can automatically adjust the strategy and re-execute the shelving operation. This self-adjusting and correcting ability enhances the reliability of the system, ensuring that tasks can be successfully completed even in the face of complex or changing environmental conditions. The automated shelving and verification process reduces the need for human intervention, speeding up the processing of logistics packaging boxes. The robot can work continuously, improving the overall operational efficiency of the logistics center. At the same time, by reducing repeated work and potential damage to goods caused by placement errors, it helps to reduce the cost of logistics operations; the automated shelving process also reduces the dependence on manual labor, and accurate shelving operations reduce the risk of damage to goods and equipment.

[0199] In combination Figure 2As shown, the automatic shelving system for the logistics packaging box of the present application, the method is applied to a logistics robot, and the system comprises:

[0200] A position acquisition module is configured to acquire the current position of the target logistics packaging box and the current position of the target storage unit when receiving the shelving instruction of the logistics packaging box.

[0201] A path planning module is configured to plan a target movement path of the logistics robot according to the current position of the target logistics packaging box and the current position of the target storage unit in combination with the current position of the logistics robot.

[0202] An image acquisition module is configured to control the logistics robot to move to the target logistics packaging box according to the target movement path and acquire image data of the target logistics packaging box.

[0203] A grabbing strategy generation module is configured to determine external feature data of the target logistics packaging box according to the image data, determine a grabbing mode of the logistics robot according to the external feature data, and determine a grabbing strategy of a mechanical arm of the logistics robot according to the grabbing mode.

[0204] A shelving operation module is configured to grab the target logistics packaging box according to the grabbing strategy, carry the target logistics packaging box to the target storage unit, and perform shelving operation on the target logistics packaging box by using a force-position hybrid control algorithm.

[0205] The automatic shelving system for the logistics packaging box of the present application has the same advantages as the above-mentioned automatic shelving method for the logistics packaging box compared with the prior art, and thus will not be described here again.

[0206] Although the present application is disclosed as above, the protection scope of the present application is not limited to this. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application, and these changes and modifications will fall within the protection scope of the present application.

Claims

1. A method for automatically shelving logistics packaging boxes, characterized in that, The automatic shelving method for logistics packaging boxes, applied to logistics robots, includes: Upon receiving the instruction to place the logistics packaging box on the shelf, the current location of the target logistics packaging box and the current location of the target storage unit are obtained; specifically, upon receiving the instruction to place the logistics packaging box on the shelf, the current environment of the logistics robot is dynamically scanned to obtain point cloud data of the current environment; A global environment model is generated based on the point cloud data and the image data of the current environment; Based on the global environment model, determine the current location of the target logistics packaging box and the current location of the target storage unit; Based on the current position of the target logistics packaging box and the current position of the target storage unit, and combined with the current position of the logistics robot, path planning is performed to obtain the target movement path of the logistics robot; specifically, this includes: determining the environmental feature data from the current position of the logistics robot to the current position of the target logistics packaging box and from the current position of the target logistics packaging box to the current position of the target storage unit based on the global environment model; Based on the environmental feature data, a multi-objective optimization function is used to perform path search in the global environment model to obtain a first path from the current position of the logistics robot to the current position of the target logistics packaging box and a second path from the current position of the target logistics packaging box to the current position of the target storage unit. The target movement path is generated based on the first path and the second path; According to the target movement path, control the logistics robot to move to the target logistics packaging box and acquire image data of the target logistics packaging box; specifically, according to the target movement path, control the logistics robot to move to the target logistics packaging box at a preset speed and acceleration parameters; When the logistics robot arrives at the target logistics packaging box, it acquires the initial image data of the target logistics packaging box; The initial image data is preprocessed, and the integrity of the target logistics packaging box in the initial image data is determined based on the preprocessed initial image data. If not, stop the task of putting the logistics packaging boxes on the shelf and issue an alarm message; If so, the initial image data shall be used as the image data of the target logistics packaging box; Based on the image data of the target logistics packaging box, the external feature data of the target logistics packaging box is determined, specifically including: using a deep learning algorithm to extract features from the image data of the target logistics packaging box to obtain the surface material features and volume and shape features of the target logistics packaging box; Based on the surface material characteristics, determine the surface material type of the target logistics packaging box; Based on the volume and shape characteristics, determine the size parameters of the target logistics packaging box; The surface material type and the size parameters are used as the external feature data of the target logistics packaging box; and the grasping mode of the logistics robot is determined based on the external feature data, and the grasping strategy of the robotic arm of the logistics robot is determined based on the grasping mode. The target logistics packaging box is grasped according to the grasping strategy, and then the target logistics packaging box is transported to the target storage unit. The target logistics packaging box is then put on the shelf using a force-position hybrid control algorithm.

2. The automatic shelving method for logistics packaging boxes according to claim 1, characterized in that, Determining the grasping mode of the logistics robot based on the external feature data includes: The gripping tool of the logistics robot is determined based on the surface material type and size parameters of the target logistics packaging box. Based on the crawling tool, a preset crawling mode database is queried to determine the crawling mode corresponding to the crawling tool.

3. The automatic shelving method for logistics packaging boxes according to claim 2, characterized in that, The step of determining the grasping strategy of the robotic arm of the logistics robot based on the grasping mode includes: Based on the grasping mode and the size parameters, the grasping point of the robotic arm on the target logistics packaging box is determined; Based on the gripping point and the surface material type, the gripping force of the robotic arm on the target logistics packaging box is determined. Based on the current position of the target logistics packaging box and the gripping point, the gripping path of the robotic arm is generated; The grasping strategy of the robotic arm is obtained based on the grasping path and the grasping force.

4. The automatic shelving method for logistics packaging boxes according to claim 3, characterized in that, The step of moving the target logistics packaging box to the target storage unit and performing a shelving operation on the target logistics packaging box using a force-position hybrid control algorithm includes: When the logistics robot arrives at the target storage unit, it acquires the pose data of the target storage unit and the current pose data of the target logistics packaging box. Based on the pose data of the target storage unit and the current pose data of the target logistics packaging box, the pose deviation between the target logistics packaging box and the target storage unit is determined, and the pose deviation includes position deviation and attitude deviation. Based on the position deviation and the attitude deviation, the position strategy, attitude strategy and gripping force strategy of the robotic arm are determined by the force-position hybrid control algorithm. The robotic arm generates integrated control commands based on the position strategy, the posture strategy, and the gripping force strategy. The target logistics packaging box is put on the shelf using the integrated control commands.

5. The automatic shelving method for logistics packaging boxes according to claim 4, characterized in that, Also includes: After the target logistics packaging box is placed, an image of the placement position of the target logistics packaging box in the target storage unit is obtained, and the shelving result of the target logistics packaging box is verified based on the placement position image; If the listing result is verified, the automatic listing task of the logistics packaging boxes is completed. If the shelf placement result verification fails, the position strategy, posture strategy, and gripping force strategy are adjusted according to the verification result, and the shelf placement operation is re-executed according to the position strategy, posture strategy, and gripping force strategy until the shelf placement result verification of the target logistics packaging box passes.

6. An automatic shelving system for logistics packaging boxes, characterized in that, The automatic shelving system for logistics packaging boxes, as described in any one of claims 1 to 5, comprises: The location acquisition module is used to obtain the current location of the target logistics packaging box and the current location of the target storage unit after receiving the logistics packaging box shelving instruction; The path planning module is used to perform path planning based on the current position of the target logistics packaging box and the current position of the target storage unit, combined with the current position of the logistics robot, to obtain the target movement path of the logistics robot. The image acquisition module is used to control the logistics robot to move to the target logistics packaging box according to the target movement path, and to acquire image data of the target logistics packaging box; The grasping strategy generation module is used to determine the external feature data of the target logistics packaging box based on the image data of the target logistics packaging box, determine the grasping mode of the logistics robot based on the external feature data, and then determine the grasping strategy of the robotic arm of the logistics robot based on the grasping mode. The shelving operation module is used to grab the target logistics packaging box according to the grabbing strategy, then transport the target logistics packaging box to the target storage unit, and perform the shelving operation on the target logistics packaging box through a force-position hybrid control algorithm.

Images