Automatic stacking method and device, terminal and storage medium

Abstract

The invention discloses an automatic stacking method and device, a terminal and a storage medium, and the method comprises the steps that package information of a to-be-stacked package is obtained according to an acquisition device, and an optimal stack type is matched for the to-be-stacked package in combination with historical stack type data of an automatic stacking system; carrying out point cloud segmentation on the tray with the stacked parcels through a preset segmentation model to obtain a segmentation result, calculating a tray gap in combination with the optimal stack type and a preset path planning algorithm, and generating a stacking path; the flexible clamping jaw device is controlled to grab the parcels to be stacked, the grabbing posture is corrected in real time in combination with feedback of the torque sensor, and the mechanical arm is controlled to execute stacking according to the stacking path; and when the segmentation result of the tray is matched with the optimal stack type, a stacking three-dimensional model is generated, the stacking three-dimensional model is bound with the AGV below the tray, and the AGV is controlled to be automatically put in storage. According to the method, stacking and warehousing are completed through stack type matching and path planning, and the adaptability and the space utilization rate of a stacking system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of warehouse logistics, in particular to an automatic stacking method and device, a terminal and a storage medium. BACKGROUND

[0002] With the rapid development of the logistics industry, the stacking operation of goods has become a crucial link in the logistics process. Efficient and accurate stacking operation has a profound impact on the smooth operation of the entire logistics supply chain. Traditional manual stacking methods have been difficult to meet the large-scale and efficient stacking demand.

[0003] In the past logistics stacking scenarios, there are various common means to solve the problem of stacking goods. On the one hand, manual stacking is a relatively basic way. Workers rely on their own experience and physical strength to carry and stack goods on pallets one by one. On the other hand, early automatic stacking often uses simple mechanical devices, such as fixed-program stacking machines. It follows a pre-set program to stack goods of specific specifications and shapes. Some semi-automatic stacking equipment requires manual assistance for part of the operation, such as adjusting the position and direction of the goods, to cooperate with the mechanical device to complete the stacking task.

[0004] However, the existing stacking methods have obvious defects. Manual stacking is inefficient and is easily affected by factors such as worker fatigue and mood, and cannot maintain a high and consistent stacking quality for a long time. Simple mechanical stacking devices lack flexibility and are difficult to adapt to goods of different shapes, sizes and materials. They also cannot dynamically adjust the stacking scheme according to real-time information of the goods and actual conditions of the warehouse. At the same time, they have limited ability to handle complex stacking types and respond to changes in goods placement, resulting in low utilization of warehouse space and high risk of collision and damage during the stacking process. SUMMARY

[0005] The purpose of the present application is to overcome the above technical problems, and provide an automatic stacking method, device, terminal and storage medium.

[0006] In a first aspect, the present application provides an automatic stacking method, which adopts the following technical means: An automatic stacking method based on an automatic stacking system, the automatic stacking system comprising a mechanical arm and a flexible gripper device at the end of the mechanical arm, the method comprising the following steps: Obtaining package information of a package to be stacked, and matching an optimal stacking type for the current package to be stacked in combination with the package information and historical stacking type data of the automatic stacking system; Performing point cloud segmentation on the pallet on which the package is stacked by a pre-set segmentation model to obtain a segmentation result, and calculating the pallet gap and generating a stacking path in combination with the optimal stacking type and a pre-set path planning algorithm; The flexible gripper device is controlled to grasp the to-be-stacked package, a feedback of a preset torque sensor is combined to correct a grasping posture in real time, and the mechanical arm is controlled to perform stacking according to the stacking path; When the segmentation result of the tray matches the optimal stacking type, a stacking three-dimensional model is generated, the stacking three-dimensional model is bound with an AGV vehicle under the tray, and the AGV vehicle is controlled to automatically enter a warehouse.

[0007] By adopting the technical scheme, the to-be-stacked package information can be accurately obtained and matched with the optimal stacking type, the tray gap can be accurately calculated to generate a reasonable stacking path, the flexible gripper device can be used to grasp the package and correct the grasping posture in real time, the stacking accuracy can be ensured, the stacking three-dimensional model and the AGV vehicle can be bound and automatically entered into the warehouse when the segmentation result of the tray matches the optimal stacking type, and the stacking efficiency and accuracy are improved.

[0008] Preferably, the package information of the to-be-stacked package is obtained according to a preset acquisition device, and a unique ID is generated for each to-be-stacked package, and the optimal stacking type is matched for the current to-be-stacked package in combination with historical stacking type data of the automatic stacking system, and the method specifically includes the following steps: The package information of the to-be-stacked package is obtained according to a preset acquisition device, and the acquisition device includes a code scanning device and a binocular depth camera, the face sheet information is obtained by scanning the to-be-stacked package according to the code scanning device, the package surface material and the package size of the to-be-stacked package are obtained according to the binocular depth camera, the unique ID of the to-be-stacked package is generated according to the face sheet information, and the package surface material and the package size are associated with the ID to form the package information; The position of the tray for stacking the to-be-stacked package is determined according to the face sheet information, and whether the tray is empty is determined; If the tray is empty, the historical successful stacking type of the automatic stacking system is obtained from the cloud according to the package surface material, the candidate stacking type is selected from the historical successful stacking type according to a cosine similarity algorithm, the optimal stacking type is selected from the candidate stacking type according to a preset optimal stacking type matching rule, and the package information and the optimal stacking type corresponding to the current to-be-stacked package are uploaded to the cloud; If the tray is not empty, the optimal stacking type of the current tray is obtained from the cloud.

[0009] By adopting the above technical solution, the face sheet information, the package surface material and the package size of the to-be-piled package can be comprehensively and accurately obtained by using the code scanning device and the binocular depth camera, thereby providing basic data for subsequent processing; the tray position can be accurately determined and it can be judged whether the tray is empty or not, so as to select a suitable mode to match the optimal pile type according to different situations; for the empty tray, the cosine similarity algorithm is used to screen the candidate pile type from the historical successful pile type, and then the optimal pile type is determined according to the preset rule and uploaded to the cloud, which helps to improve the pile efficiency and rationality; for the non-empty tray, the optimal pile type can be directly obtained from the cloud to simplify the process and save time.

[0010] Preferably, the optimal pile type is obtained from the candidate pile type according to the preset optimal pile type matching rule, and specifically includes the following steps: A reinforcement learning environment including a state space, an action space and a reward function is constructed, and the reward function is calculated according to the feasible path of the to-be-piled package, the space score and the support score of the candidate pile type; The candidate pile type is quantitatively evaluated, and the quantitative evaluation includes feasible path quantitative analysis, space score calculation and support score calculation, to obtain the feasible path score, the space score and the support score; According to the PPO reinforcement learning algorithm, the policy network is trained and iteratively optimized in combination with the state space and the action space, based on the trained policy network, the candidate pile type is simulated to be executed, the comprehensive reward score is calculated according to the feasible path of the to-be-piled package, the space score and the support score, and the candidate pile type with the highest comprehensive reward score is output as the optimal pile type.

[0011] By adopting the above technical solution, a reinforcement learning environment including a state space, an action space and a reward function is constructed, so that the feasible path of the to-be-piled package, the space score and the support score of the candidate pile type are comprehensively considered, and the adaptability of each candidate pile type is more accurately measured; according to the PPO reinforcement learning algorithm, the policy network is trained and iteratively optimized, the comprehensive reward score is calculated after simulating the execution of the candidate pile type, and the optimal pile type is output, which can effectively improve the accuracy and reliability of the screening and provide a more reasonable pile type selection for subsequent pile.

[0012] Preferably, the tray on which the packages are stacked is segmented by a preset segmentation model, and the tray gap is calculated and the pile path is generated in combination with the optimal pile type and a preset path planning algorithm, specifically including the following steps: The tray point cloud data for stacking the to-be-piled package is obtained, the package contour on the tray is segmented according to the preset YOLOv8 segmentation model and the tray point cloud data, and the segmentation result is obtained, the segmentation result including the available gap; According to the optimal stacking type, the available spaces in the segmentation result are preliminarily screened, and the available spaces in conflict with the structure of the optimal stacking type are excluded; All the available spaces obtained after the preliminary screening are traversed, the minimum circumscribed rectangle of each space is calculated, all the minimum circumscribed rectangles not smaller than the package size are screened and collected into a space candidate set, the space utilization and the optimal placement pose of each candidate space in the space candidate set are calculated, and all the optimal placement poses are collected into a candidate placement pose list, which is arranged according to the space utilization.

[0013] By adopting the above technical scheme, the package contour is segmented by the YOLOv8 segmentation model, the distribution of the package on the tray and the space condition can be accurately identified, the available spaces are preliminarily screened according to the optimal stacking type, the conflict spaces are excluded, and it is ensured that the subsequent stacking conforms to the optimal stacking type, the minimum circumscribed rectangle is calculated by traversing the screened available spaces, the space candidate set is collected and the space utilization and the optimal placement pose are calculated, and the candidate placement pose list is arranged according to the space utilization, which can effectively utilize the tray space and provide a basis for selecting a suitable stacking position, thereby improving the stacking efficiency and the space utilization.

[0014] Preferably, the method further comprises the following steps: The current pose of the mechanical arm and the candidate spaces and the optimal placement poses thereof are obtained, a tray geometric model is constructed based on the tray point cloud data, and a working space model carrying an anti-collision region mark is obtained; A plurality of candidate paths satisfying preset constraint conditions are generated based on the improved RRT* algorithm, the constraint conditions include a space constraint condition and an acceleration constraint condition, the space constraint condition is that the candidate path does not collide with the tray geometric model or the anti-collision region, and the acceleration constraint condition is that the corner acceleration of the candidate path does not exceed a preset threshold; The candidate paths satisfying the constraint conditions are scored according to a preset path scoring rule, and the candidate path with the highest score obtained is taken as the stacking path.

[0015] By adopting the above technical scheme, the tray geometric model is constructed based on the tray point cloud data, and the working space model carrying the anti-collision region mark is obtained, which can more accurately reflect the stacking environment; the candidate paths satisfying the space and acceleration constraint conditions are generated by using the improved RRT* algorithm, which can avoid collision and control the corner acceleration, thereby ensuring the running safety and stability; the candidate path with the highest score is selected as the stacking path according to the preset path scoring rule, which can obtain the optimal stacking path and improve the stacking efficiency and accuracy.

[0016] Preferably, the control of the flexible gripper device to grab the to-be-palletized package is combined with the real-time correction of the grabbing posture based on the feedback of the preset torque sensor, the mechanical arm is controlled to perform palletizing according to the palletizing path, and the method specifically comprises the following steps: An initial clamping force is obtained according to the material of the package by calling a preset clamping force parameter, the flexible gripper device comprises a gripper and a pneumatic tendon for driving the gripper, a torque sensor is pre-installed on the flexible gripper device, and a rotary encoder and an inertial measurement unit are pre-installed on the mechanical arm; The pneumatic tendon is controlled to drive the gripper to grab the to-be-palletized package with the initial clamping force; In the process of palletizing performed by the mechanical arm, the real-time clamping force is monitored by the torque sensor, and when it is detected that the deformation amount of the to-be-palletized package exceeds a preset deformation threshold, the real-time clamping force is dynamically adjusted; The joint angle fed back by the rotary encoder and the end acceleration collected by the inertial measurement unit are used to estimate the actual grabbing posture deviation, a joint compensation instruction is generated according to the actual grabbing posture deviation and a PID controller, and the grabbing posture of the mechanical arm is corrected according to the joint compensation instruction; When the torque sensor detects a preset contact force condition, a gripper release instruction is triggered, and the to-be-palletized package is placed.

[0017] By adopting the above technical solution, the initial clamping force is determined according to the material of the package, which can adapt to the grabbing requirements of packages of different materials; the clamping force is monitored and dynamically adjusted in real time, which can prevent the package from being damaged due to excessive force or falling due to insufficient clamping force; the grabbing posture deviation is estimated and corrected, which can improve the palletizing accuracy; and the package is released when the contact force reaches the condition, which can ensure accurate placement of the package.

[0018] Preferably, after the placement of the to-be-palletized package is completed, the following steps are further included: The placement posture of the current package is obtained, the gravity center coordinates of the current package are calculated in combination with the package size, the gravity center coordinates of the remaining packages on the tray are obtained, and the overall gravity center coordinates of all the packages on the tray are calculated in combination with the gravity center coordinates of the current package; The stable center coordinates of the tray are obtained, the gravity center offset is calculated in combination with the overall gravity center coordinates, and the gravity center offset is compared with a preset warning range; When the gravity center offset falls within the warning range, a gravity center offset control strategy is executed in combination with the overall gravity center coordinates and the stable center coordinates.

[0019] By adopting the technical scheme, after the to-be-stacked package is placed, the overall gravity center coordinates and the gravity center offset of all the packages on the tray are calculated, the gravity center offset is compared with the preset early warning range, and when the offset falls within the early warning range, the gravity center offset control strategy is executed, so that the stability of the package stacking on the tray is ensured, problems such as package dumping caused by excessive gravity center offset are avoided, and the safety and reliability of the stacking operation are improved.

[0020] In a second aspect, the application provides an automatic stacking device, which adopts the following technical means: An automatic stacking device based on an automatic stacking system, the automatic stacking system comprising a mechanical arm and a flexible gripper device at the end of the mechanical arm, comprising the following modules: An optimal stacking type matching module for obtaining package information of a to-be-stacked package, and matching an optimal stacking type for the current to-be-stacked package in combination with the package information and historical stacking type data of the automatic stacking system; A stacking path planning module for performing point cloud segmentation on a tray on which packages are stacked by using a preset segmentation model to obtain a segmentation result, calculating a tray gap in combination with the optimal stacking type and a preset path planning algorithm, and generating a stacking path; A grabbing posture correction module for controlling the flexible gripper device to grab the to-be-stacked package, correcting a grabbing posture in real time in combination with feedback of a torque sensor set in advance, and controlling the mechanical arm to perform stacking according to the stacking path; An automatic warehousing module for generating a stacking three-dimensional model when the segmentation result of the tray matches the optimal stacking type, binding the stacking three-dimensional model with an AGV vehicle under the tray, and controlling the AGV vehicle to automatically enter a warehouse.

[0021] By adopting the above technical scheme, the optimal stacking type matching module can accurately obtain package information and generate a unique ID for each package, match an optimal stacking type in combination with historical stacking type data, and improve the adaptability of package stacking. The stacking path planning module can perform point cloud segmentation on a tray, calculate a tray gap in combination with an optimal stacking type and a path planning algorithm, and generate a stacking path to improve space utilization. The grabbing posture correction module corrects a grabbing posture in real time through torque sensor feedback, ensures the accuracy of the stacking process, and prevents package extrusion deformation. The automatic warehousing module generates a stacking three-dimensional model and binds it with an AGV vehicle when the segmentation result of the tray matches the optimal stacking type, realizes automatic warehousing, and improves logistics efficiency.

[0022] In a third aspect, the application provides an intelligent terminal, which adopts the following technical scheme: An intelligent terminal comprises a memory and a processor, the memory stores at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to realize the automatic stacking method as described above.

[0023] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium, the readable storage medium stores at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to realize the automatic stacking method as described above.

[0024] In summary, the present application at least includes the following beneficial effects: (1) The method of the present application matches the optimal stacking type by combining package information and historical stacking type data, can dynamically adjust the stacking scheme according to real-time conditions, thereby improving the utilization rate of warehouse space; (2) The method of the present application divides the tray point cloud by a preset segmentation model, generates an optimal stacking path by combining an optimal stacking type and a path planning algorithm, and reduces the collision and damage risk in the stacking process; (3) The method of the present application uses a flexible gripper device to grasp the package, combines a torque sensor to feedback real-time correction of the grasping posture, can adapt to goods of different shapes, sizes and materials, makes the stacking process flexible, can adapt to a variety of goods, and improves the grasping success rate. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is the overall flowchart of the automatic stacking method of the embodiment of the present application; Figure 2 is the flowchart of stacking type matching and path planning of the automatic stacking method of the embodiment of the present application; Figure 3 is the flowchart of center of gravity offset judgment of the automatic stacking method of the embodiment of the present application; Figure 4 is the architecture diagram of the automatic stacking device of the embodiment of the present application. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings. The described embodiments are only possible technical implementations of the present application, but are not limited thereto, and those skilled in the art can obtain other embodiments without creative labor under the circumstances of the embodiments of the present application.

[0027] This application mainly adopts a solution of obtaining package information, matching pallet type, and planning path to complete palletizing and warehousing, which achieves the effects of improving the adaptability of the palletizing system, space utilization, positioning accuracy, and reducing the failure rate. The following is a further detailed description of this application.

[0028] The automated palletizing method of this application is based on an automated palletizing system, which includes a robotic arm and a flexible gripper device at the end of the robotic arm, such as... Figure 1 As shown, the method includes the following steps: S1. Obtain the package information of the packages to be palletized, and match the optimal pallet type for the current packages to be palletized by combining the package information with the historical pallet type data of the automatic palletizing system. This includes the following steps: S11. Obtain package information of the package to be palletized according to the preset acquisition device, which includes a barcode scanning device and a binocular depth camera.

[0029] Based on the information obtained from the waybill by scanning the package to be palletized by the scanning device, in this embodiment, the scanning device is a P2 scanner, which includes an integrated laser ranging module. When the package to be palletized enters the detection area, the photoelectric sensor in the system triggers the P2 scanner to work.

[0030] The surface material and dimensions of the packages to be palletized are obtained based on a binocular depth camera. In this embodiment, the surface material of the packages includes cardboard boxes, woven bags, foam boxes, and hard plastics, and the dimensions of the packages include length (L), width (W), and height (H).

[0031] S12. Generate a unique ID for the parcel to be palletized based on the waybill information, summarize the parcel surface material and parcel size into parcel information and associate it with the generated ID, and upload the parcel information to the cloud.

[0032] S13, such as Figure 2 As shown, the position of the pallet for stacking packages is determined based on the waybill information. It is then determined whether the pallet for stacking packages is empty. If the pallet is empty, it means that the current package to be stacked is the first package, so the optimal stacking type needs to be matched.

[0033] S131. When the pallet is empty, retrieve all historical successful pallet types of the automatic palletizing system from the cloud based on the surface material of the package. The surface material of the package matched by the retrieved historical successful pallet types should be the same as the surface material of the current package.

[0034] S132. Extract features from the current packages to be palletized and construct a package feature vector for the packages to be palletized, i.e. ; in, Material coding, The historical average density of the same material packaged together is obtained from the cloud and is expressed in kg / m³. In the embodiment, the material code of the carton is 1, the material code of the woven bag is 2, the material code of the foam box is 3, and the material code of the hard plastic is 4.

[0035] S133, a historical pile type feature vector is constructed, that is, wherein, is the average value of the length of all packages in the historical successful pile type i, is the arithmetic mean of the width of all packages in the historical successful pile type i, is the maximum height of a single package in the historical successful pile type i, is the main material code, is the average density of the package in the historical successful pile type i, which is calculated based on the ratio of weight to volume.

[0036] S134, the candidate pile type is obtained from the historical successful pile type according to the preset cosine similarity algorithm; In the embodiment, S>0.75, the top five candidate pile types are obtained according to the cosine similarity algorithm, and in other implementable specific manners, the value of S can also be adjusted.

[0037] S135, the optimal pile type is obtained from the candidate pile type according to the preset optimal pile type matching rule, and the package information and the optimal pile type corresponding to the current to-be-stacked package are uploaded to the cloud, which specifically includes the following steps.

[0038] The reinforcement learning environment including state space, action space and reward function is constructed.

[0039] The state space includes the geometric features of the candidate pile type, the attributes of the to-be-stacked package and the tray space data; the geometric features of the candidate pile type include the number of layers of the pile type and the arrangement mode of the package; the attributes of the to-be-stacked package include the size, weight and shape of the package; and the tray space data includes the shape and the remaining space size.

[0040] The action space is to select one from the candidate pile type set as the current stacking scheme; The reward function calculates the comprehensive reward score according to the feasible path of the to-be-stacked package, the space score of the candidate pile type and the support score.

[0041] That is, wherein, are the feasible path score, the space score and the support score, respectively, ​​​​​​​are corresponding weight coefficients, which can be adjusted.

[0042] Then the strategy network (Actor network) and the value network (Critic network) of the PPO algorithm are constructed, and the network parameters are randomly initialized.

[0043] The candidate stack types are quantitatively evaluated, which includes feasible path quantitative analysis, space score calculation and support score calculation, to obtain the feasible path score, the space score and the support score.

[0044] In a specific embodiment, the feasible path quantitative analysis includes the following steps: For each candidate stack type, the shortest collision-free path length of the robot arm from the starting position to the package target placement position is calculated, and the score is mapped in proportion to the path length. In a specific embodiment, the shortest path length is 1 point and the longest path length is 0 point. Determine whether the joint angle of the robot arm when performing an action at the target placement position is within the preset safety range. If it exceeds the range, the feasible path score is 0. Determine whether the flexible gripper device of the robot arm will collide with the preset package position during the grabbing and placing process. If there is a collision risk, the feasible path score is 0. The above scores are normalized to the [0, 1] interval to obtain the final feasible path score.

[0045] In a specific embodiment, a score is calculated according to the utilization rate of the tray space by the candidate stack type. The space utilization rate = occupied tray space volume / total tray volume. The space utilization rate is mapped to the [0, 1] interval as the space score.

[0046] In a specific embodiment, the support score is obtained according to the stability of the candidate stack type. The stability includes the proportion of the support area below the package and whether the gravity center projection is within the support surface. If it meets the requirements, it is 1 point, otherwise it is 0 point.

[0047] According to the PPO reinforcement learning algorithm, the strategy network is trained and iteratively optimized in combination with the state space and the action space.

[0048] In a specific embodiment, data sampling is performed based on the current strategy network, candidate stack types are selected from the action space, and new states, rewards and flags indicating whether the end are obtained after execution. The state-action-reward-next state sequence data is recorded.

[0049] According to the sequence data, the strategy network parameters are updated by the PPO algorithm to optimize the selection strategy of the candidate stack type. The value network parameters are updated according to the sequence data. The data sampling and network updating process are repeated multiple times to gradually optimize and iterate the strategy network and the value network.

[0050] When the reward change of the continuous multiple rounds of iterations is less than a threshold or a preset training round number is reached, the training is stopped, and a trained strategy network is obtained.

[0051] Based on the trained strategy network, a candidate stacking type is simulated to be executed, a comprehensive reward score is calculated according to a feasible path, a space score and a support score of the to-be-stacked package, and the candidate stacking type with the highest comprehensive reward score is output as the optimal stacking type. S14, if the tray is not empty, the optimal stacking type of the current tray is obtained from the cloud.

[0052] If the tray is not empty, that is, the current to-be-stacked package is not the first package, therefore the optimal stacking type has been matched when the first package is placed on the tray, so the optimal stacking type matched by the tray can be directly obtained.

[0053] S2, point cloud segmentation is performed on the tray on which the package is placed by using a preset segmentation model, and a tray gap is calculated and a stacking path is generated by using a preset path planning algorithm, specifically including the following steps: S21, tray point cloud data for stacking the to-be-stacked package is obtained by using a binocular depth camera, and the tray point cloud data includes the stacked package and the gap. The binocular camera only scans the corresponding area of the tray, thereby reducing the calculation amount.

[0054] S22, the package contour on the tray is segmented according to a preset YOLOv8 segmentation model and the tray point cloud data, and a segmentation result is obtained, wherein the segmentation result includes an available gap.

[0055] An inflation operation is performed on the segmentation result, and in this embodiment, the inflation operation is to expand the gap boundary by 3-5 mm, so as to ensure the safety margin of the mechanical arm operation.

[0056] Specifically, the segmentation result is a boundary box of the package and a segmentation mask image.

[0057] S23, according to the structural features and arrangement rules of the optimal stacking type, the available gaps in the segmentation result are preliminarily screened, and the available gaps in conflict with the structure of the optimal stacking type are excluded.

[0058] In one implementable specific manner, the judgment rules of the structural conflict include that the geometric size and shape of the available gap do not conform to the arrangement rules of the stacking type, or the position of the available gap cannot meet the support stability requirement of the stacking type. For example, if the optimal stacking type requires the package to be stacked in a transverse direction (the long side along the X axis), when the minimum circumscribed rectangle width of the available gap is less than the long side size of the package, it is considered as a conflict; if the optimal stacking type is designed as a pyramid shape, that is, the number of lower layer packages is more than the number of upper layer packages, if a certain available gap is located at the edge and cannot be completely supported by the lower layer packages, it is considered as a conflict.

[0059] S24, traverse all available voids obtained after preliminary screening, calculate the minimum circumscribed rectangle of each void, screen all minimum circumscribed rectangles not smaller than the size of the package and collect them into a void candidate set.

[0060] S25, calculate the space utilization and optimal placement pose of each candidate void in the void candidate set, collect all optimal placement poses into a candidate placement pose list, and arrange the candidate placement pose list in descending order of space utilization.

[0061] S26, obtain the current pose of the mechanical arm and the candidate void and its optimal placement pose, construct a tray geometric model based on the tray point cloud data, and obtain a workspace model carrying an anti-collision region mark.

[0062] In this embodiment, the anti-collision region is mainly the gantry, and it is necessary to ensure that the mechanical arm does not hit the gantry during movement. The workspace model carrying the anti-collision region mark is the safety boundary of the gantry.

[0063] S27, generate multiple candidate paths satisfying the preset constraint conditions based on the improved RRT* algorithm.

[0064] The constraint conditions include spatial constraint conditions and acceleration constraint conditions. The spatial constraint condition is that the candidate path does not collide with the tray geometric model or the anti-collision region. A random tree is generated according to the current pose of the mechanical arm, a spherical region around the optimal placement pose of the package is generated, a sampling point is randomly generated in the workspace, the node closest to the sampling point is selected from the tree, and a new node is generated in the direction of the sampling point with a step length δ. In a specific embodiment, δ can be 80mm; check whether the new node pose collides with the gantry, and discard the node if it collides.

[0065] In a specific embodiment, a 80mm safety buffer is set around the gantry.

[0066] The acceleration constraint condition is that the corner acceleration of the candidate path does not exceed a preset threshold.

[0067] Estimate the joint angle acceleration at the corner of the path. In a specific embodiment, take three consecutive nodes, determine the joint angle corresponding to each node through the rotary encoder installed on the mechanical arm, calculate the rate of change of the joint angle by combining the displacement change between nodes and the time interval, and then obtain the corner acceleration.

[0068] Compare the calculated corner acceleration with the preset threshold. If the acceleration exceeds the threshold, it means that the mechanical arm moves too fast, which may cause the package to slide due to inertia. In this case, the path node or path segment will be discarded, and resampling or path planning will be performed again. If the acceleration is within the threshold range, the node or path segment is retained.

[0069] For the complete candidate path, the above acceleration calculation and judgment are performed segment by segment to ensure that the corner acceleration of the entire path meets the requirements. If there is a part that does not meet the requirements, the path is optimized and adjusted or directly excluded.

[0070] S28, score the candidate paths that meet the constraint conditions according to the preset path scoring rules. The scoring content includes space safety score and acceleration smoothness score. The space safety is judged according to the minimum distance between the path and the obstacle. The greater the distance, the higher the space safety score. The acceleration smoothness is judged according to the reciprocal of the maximum corner acceleration of the path. The smaller the acceleration, the higher the acceleration smoothness score.

[0071] The final score is obtained by weighted calculation according to the weight coefficients of the pre-set space safety score and acceleration smoothness score. The weight coefficients can be adjusted.

[0072] S29, the candidate path with the highest score obtained is taken as the stacking path, and is converted into a joint angle-time sequence of the mechanical arm.

[0073] S3, control the flexible gripper device to grab the package to be stacked, and combine the feedback of the pre-set torque sensor to correct the grabbing posture in real time. Control the mechanical arm to perform stacking according to the stacking path, which specifically includes the following steps: S31, call the preset clamping force parameter according to the material of the package to obtain the initial clamping force. The flexible gripper device includes a gripper and a pneumatic tendon driving the gripper. A torque sensor is pre-installed on the flexible gripper device, and a rotary encoder and an inertial measurement unit are pre-installed on the mechanical arm.

[0074] The torque sensor is used to detect the force condition at the end of the mechanical arm, and to monitor the three-dimensional force (Fx, Fy, Fz) and torque (Mx, My, Mz) of the contact point between the gripper and the package in real time. The rotary encoder is used to feedback the joint angle and read the real-time angle of each joint of the mechanical arm. The inertial measurement unit is used to detect the vibration state of the end, and to collect the three-axis acceleration and angular velocity of the mechanical arm.

[0075] S32, control the pneumatic tendon to drive the gripper to grab the package to be stacked with the initial clamping force.

[0076] In this embodiment, the initial clamping force of the woven bag is ; the initial clamping force of the foam box is ; and the initial clamping force of the hard box body is .

[0077] S33, during the execution of the stacking by the mechanical arm, monitor the real-time clamping force by the torque sensor, estimate the deformation amount of the package to be stacked by the torque change rate, and when it is detected that the deformation amount exceeds the pre-set deformation threshold, i.e. At this time, the real-time clamping force is dynamically adjusted.

[0078] The formula for dynamic adjustment is, ; wherein is a preset control parameter.

[0079] The rate of change of the holding pressure is maintained .

[0080] S34, the joint angle feedback by the rotary encoder and the end acceleration collected by the inertial measurement unit are acquired, the data are fused through Kalman filtering, and the actual grasping pose deviation between the actual pose and the target pose is estimated.

[0081] The actual grasping pose deviation includes a position deviation (x, y, z) , , ) and an angle deviation (a, b, g) , , .

[0082] S35, a joint compensation instruction is generated according to the actual grasping pose deviation and a PID controller, and the grasping pose of the robot arm is corrected according to the joint compensation instruction.

[0083] In this embodiment, the process of calculating the compensation amount according to the actual grasping pose deviation and the PID controller is to synthesize a control amount according to the PID controller and a preset control parameter, and to generate a joint compensation instruction according to the control amount.

[0084] In this embodiment, the control parameters are Kp=0.8, Ki=0.2, and Kd=0.1.

[0085] The joint motor of the robot arm receives the joint compensation instruction, adjusts the joint angle, and realizes the pose correction.

[0086] S36, when the torque sensor detects a preset contact force condition, a jaw release instruction is triggered, and the placement of the package to be stacked is completed.

[0087] In this embodiment, when the torque sensor detects that the Z-axis contact force Fz >5N, the jaw release instruction is triggered, and the package is placed. In this embodiment, the condition of Fz >5N is determined as the package contacting the surface of the tray.

[0088] S37, the placement pose of the current package is acquired, and the center of gravity coordinates of the current package are calculated in combination with the package size, as shown in Figure 3 .

[0089] The center of gravity coordinates are ; ; Placement pose: (x, y, ), is the placement rotation angle.

[0090] Obtain the center of gravity coordinates of the remaining packages on the current tray, and calculate the overall center of gravity coordinates of all packages on the tray in combination with the center of gravity coordinates of the current package, specifically, the center of gravity contribution of all packages is accumulated.

[0091] Obtain the stable center coordinates of the tray, and calculate the center of gravity offset in combination with the overall center of gravity coordinates, the offset is the Euclidean distance between the overall center of gravity coordinates and the stable center coordinates of the tray.

[0092] S38, compare the center of gravity offset with the preset warning range.

[0093] In one specific embodiment, three threshold ranges are set, including a safe range, a warning range and a dangerous range, and the center of gravity offset is compared with the three threshold ranges respectively.

[0094] Where the center of gravity offset ≤ 50mm belongs to the safe range, the package can continue to be placed, and the state of the package placed on the tray is updated in real time; The center of gravity offset is between 50mm and 80mm, which belongs to the warning range, when the center of gravity offset falls within the warning range, the center of gravity offset control strategy is executed in combination with the overall center of gravity coordinates and the stable center coordinates.

[0095] The center of gravity offset control strategy is to place the subsequent package in the anti-center of gravity direction area preferentially, and increase the lateral pressure by 10%-20% according to the clamping force, to suppress the dumping trend.

[0096] The center of gravity offset > 80mm belongs to the dangerous range, the system immediately triggers emergency handling, stops the current tray stacking, and controls the AGV under the tray to be dispatched to the manual inspection area.

[0097] S4, when the segmentation result of the tray matches the optimal stacking type, a stacking three-dimensional model is generated, the stacking three-dimensional model is bound with the AGV car under the tray, and the AGV car is controlled to automatically enter the warehouse, specifically including the following steps: S41, the segmentation result is analyzed in real time to determine whether the package layout in the segmentation result conforms to the optimal stacking type, specifically, the system determines whether the available space, package layout, etc. on the current tray matches the required conditions of the optimal stacking type.

[0098] For example, check whether the size, shape and position of the available gap can place the remaining packages in the optimal stacking type, and whether the position and attitude of the placed packages meet the specifications of the optimal stacking type.

[0099] If it is consistent, the segmentation result matches the optimal stacking type.

[0100] By analyzing the segmentation result in real time and judging whether it matches the optimal stacking type, it can be ensured that the stacking meets the expectation.

[0101] S42, obtain the package information and placement pose of all packages on the tray, and generate a stacking three-dimensional model of the current tray and all packages through a WebGL engine.

[0102] Specifically, the system collects the position coordinates, size, and attitude information of all packages on the current tray, as well as the relevant data of the tray itself, and converts these data into a format suitable for WebGL processing, including converting spatial coordinate data into three-dimensional vectors and converting the geometric shape information of the packages and the tray into polygon mesh data.

[0103] A three-dimensional scene is created using a WebGL engine, and three-dimensional models of the packages and the tray are created according to the converted data and drawn into the scene to form a complete stacking three-dimensional model, which displays the stacking state of the current tray in real time. The WebGL engine is used to generate the stacking three-dimensional model, realizing the visual display of the stacking condition.

[0104] S43, associate identification information with the stacking three-dimensional model, and obtain information of an AGV car below the tray, including the number, position, and running state of the AGV car.

[0105] A corresponding relationship is established in the system database to bind the identification information with the information of the AGV car, so that the system can correspond the stacking three-dimensional model with a specific AGV car, providing a basis for subsequent control of the AGV car.

[0106] According to the package information, navigation information is generated for the AGV car, including the position coordinates of the target warehouse and the travel path planning instructions. After receiving these information, the AGV car travels autonomously according to the planned path.

[0107] The system controls the AGV car to transport the tray carrying the stacked packages to the target warehouse according to the navigation information, completes the automatic warehousing operation, facilitates subsequent cargo management and transportation, and improves the logistics efficiency.

[0108] Based on the same inventive concept as above, the embodiments of the present application also disclose an automatic stacking device based on an automatic stacking system, which includes a mechanical arm and a flexible gripper device at the end of the mechanical arm, as shown in Figure 4 The automatic stacking device includes the following modules: An optimal stacking type matching module is configured to obtain package information of packages to be stacked, and match an optimal stacking type for the current packages to be stacked in combination with the package information and historical stacking type data of the automatic stacking system. The palletizing path planning module is configured to perform point cloud segmentation on the pallet of the stacked package by using a preset segmentation model to obtain a segmentation result, calculate a pallet gap in combination with an optimal stacking type and a preset path planning algorithm, and generate a palletizing path; The grabbing posture correction module is configured to control the flexible gripper device to grab the package to be stacked, correct the grabbing posture in real time in combination with feedback of a preset torque sensor, and control the robot arm to perform palletizing according to the palletizing path. The automatic warehousing module is configured to generate a palletizing three-dimensional model when the segmentation result of the pallet matches the optimal stacking type, bind the palletizing three-dimensional model with an AGV vehicle under the pallet, and control the AGV vehicle to automatically enter a warehouse.

[0109] In a specific implementation, the optimal stacking type matching module includes the following units: The first optimal stacking type matching unit is configured to obtain package information of the package to be stacked by using a preset acquisition device, the acquisition device including a code scanning device and a binocular depth camera, obtain face sheet information by scanning the package to be stacked by using the code scanning device, obtain a package surface material and a package size of the package to be stacked by using the binocular depth camera, generate a unique ID of the package to be stacked according to the face sheet information, and associate the package surface material and the package size with the ID to form the package information. The second optimal stacking type matching unit is configured to determine a position of a pallet for stacking the package to be stacked according to the face sheet information, and determine whether the pallet is empty. If the pallet is empty, the historical successful stacking type of the automatic palletizing system is obtained from the cloud according to the package surface material, a candidate stacking type is selected from the historical successful stacking type according to a cosine similarity algorithm, an optimal stacking type is selected from the candidate stacking type according to a preset optimal stacking type matching rule, and the package information and the optimal stacking type corresponding to the current package to be stacked are uploaded to the cloud. If the pallet is not empty, the optimal stacking type of the current pallet is obtained from the cloud.

[0110] In a specific implementation, the second optimal stacking type matching unit includes the following sub-units: The optimal stacking type matching sub-unit is configured to construct a reinforcement learning environment including a state space, an action space, and a reward function, and calculate the reward function according to a feasible path of the package to be stacked, a space score of a candidate stacking type, and a support score. The candidate stacking type is quantitatively evaluated, the quantitative evaluation including feasible path quantitative analysis, space score calculation, and support score calculation, to obtain a feasible path score, a space score, and a support score. According to the PPO reinforcement learning algorithm, the policy network is trained and iteratively optimized in combination with the state space and action space. Based on the trained policy network, the candidate stacking type is simulated to be executed, and the comprehensive reward score is calculated according to the feasible path of the to-be-stacked package, the space score and the support score, and the candidate stacking type with the highest comprehensive reward score is output as the optimal stacking type.

[0111] In a specific implementable embodiment, the stacking path planning module comprises the following units: The first stacking path planning unit is configured to obtain tray point cloud data for stacking the to-be-stacked package, segment the package contour on the tray according to a preset YOLOv8 segmentation model and the tray point cloud data, and obtain a segmentation result, the segmentation result comprising available gaps; The second stacking path planning unit is configured to preliminarily screen the available gaps in the segmentation result according to the optimal stacking type, and exclude available gaps in conflict with the structure of the optimal stacking type; The third stacking path planning unit is configured to traverse all available gaps obtained after preliminary screening, calculate the minimum bounding rectangle of each gap, screen all minimum bounding rectangles not smaller than the package size and aggregate them into a gap candidate set, calculate the space utilization and the optimal placement pose of each candidate gap in the gap candidate set, and aggregate all optimal placement poses into a candidate placement pose list, the candidate placement pose list being arranged according to the space utilization.

[0112] The fourth stacking path planning unit is configured to obtain the current pose of the mechanical arm and the candidate gap and its optimal placement pose, construct a tray geometric model based on the tray point cloud data, and obtain a workspace model carrying anti-collision region markers; The fifth stacking path planning unit is configured to generate multiple candidate paths satisfying preset constraint conditions based on an improved RRT* algorithm, the constraint conditions comprising a space constraint condition and an acceleration constraint condition, the space constraint condition being that the candidate path does not collide with the tray geometric model or the anti-collision region, and the acceleration constraint condition being that the corner acceleration of the candidate path does not exceed a preset threshold; The sixth stacking path planning unit is configured to score the candidate paths satisfying the constraint conditions according to a preset path scoring rule, and obtain the candidate path with the highest score as the stacking path.

[0113] In a specific implementable embodiment, the grasping pose correction module comprises the following units: The first grasping pose correction unit is configured to call a preset clamping force parameter according to the package material to obtain an initial clamping force, the flexible gripper device comprising a gripper and a pneumatic tendon driving the gripper, a torque sensor being pre-installed on the flexible gripper device, and a rotary encoder and an inertial measurement unit being pre-installed on the mechanical arm; The second grasping posture correction unit is configured to control the pneumatic muscle tendon to drive the gripper to grasp the package to be stacked at an initial clamping force. The third grasping posture correction unit is configured to monitor the real-time clamping force by the torque sensor during the stacking process of the robot arm, and dynamically adjust the real-time clamping force when the deformation amount of the package to be stacked is detected to exceed the preset deformation threshold. The fourth grasping posture correction unit is configured to obtain the joint angle fed back by the rotary encoder and the end acceleration collected by the inertial measurement unit to estimate the actual grasping posture deviation, generate a joint compensation instruction according to the actual grasping posture deviation and the PID controller, and correct the grasping posture of the robot arm according to the joint compensation instruction. The fifth grasping posture correction unit is configured to trigger a gripper release instruction when the torque sensor detects a preset contact force condition, and complete the placement of the package to be stacked.

[0114] In a specific embodiment, the automatic stacking device further comprises the following modules: The center of gravity deviation warning module is configured to obtain the placement posture of the current package, calculate the center of gravity coordinates of the current package in combination with the package size, obtain the center of gravity coordinates of the remaining packages on the tray, and calculate the overall center of gravity coordinates of all the packages on the tray in combination with the center of gravity coordinates of the current package. The stable center coordinates of the tray are obtained, and the center of gravity deviation is calculated in combination with the overall center of gravity coordinates. When the center of gravity deviation falls within the warning range, the center of gravity deviation control strategy is executed in combination with the overall center of gravity coordinates and the stable center coordinates.

[0115] Based on the same inventive concept, the embodiments of the present application also disclose a computer readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set can be loaded and executed by a processor to realize the automatic stacking method provided by the above method embodiments.

[0116] Similarly, based on the same inventive concept, the embodiments of the present application also disclose a computer readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set can be loaded and executed by a processor to realize the automatic stacking method provided by the above method embodiments.

[0117] Those skilled in the art can understand that all or part of the steps of the above-mentioned embodiments can be completed by hardware, or can be instructed by programs to complete the related hardware, and the programs can be stored in a computer readable storage medium, such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes.

[0118] The above are only optional embodiments of the present application and are not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An automatic palletizing method, characterized by, Based on an automatic stacking system, the automatic stacking system includes a mechanical arm and a flexible gripper device at the end of the mechanical arm, the method comprises the following steps: Obtain the package information of the package to be stacked, and match the optimal stacking type for the current package to be stacked according to the package information and the historical stacking type data of the automatic stacking system; Through a preset segmentation model, the point cloud segmentation of the tray on which the package is stacked is performed to obtain a segmentation result, and the tray gap is calculated and the stacking path is generated according to the optimal stacking type and a preset path planning algorithm; Control the flexible gripper device to grab the package to be stacked, and combine the feedback of the pre-set torque sensor to correct the grabbing posture in real time, and control the mechanical arm to execute stacking according to the stacking path; When the segmentation result of the tray matches the optimal stacking type, a three-dimensional stacking model is generated, the three-dimensional stacking model is bound with the AGV car under the tray, and the AGV car is controlled to automatically enter the warehouse.

2. The automatic palletizing method according to claim 1, characterized in that, The package information of the package to be stacked is obtained, and the optimal stacking type for the current package to be stacked is matched according to the package information and the historical stacking type data of the automatic stacking system, which specifically comprises the following steps: According to a preset acquisition device, the package information of the package to be stacked is obtained, the acquisition device includes a code scanning device and a binocular depth camera, the face sheet information of the package to be stacked is obtained according to the code scanning device, the package surface material and the package size of the package to be stacked are obtained according to the binocular depth camera, the unique ID of the package to be stacked is generated according to the face sheet information, and the package surface material and the package size are associated with the ID to form the package information; According to the face sheet information, the position of the tray for stacking the package to be stacked is determined, and whether the tray is empty is judged; If the tray is empty, the historical successful stacking type of the automatic stacking system is obtained from the cloud according to the package surface material, the candidate stacking type is selected from the historical successful stacking type according to the cosine similarity algorithm, the optimal stacking type is selected from the candidate stacking type according to the preset optimal stacking type matching rule, and the package information and the optimal stacking type corresponding to the current package to be stacked are uploaded to the cloud; If the tray is not empty, the optimal stacking type of the current tray is obtained from the cloud.

3. The automatic palletizing method according to claim 2, characterized in that, The optimal stacking type is selected from the candidate stacking type according to the preset optimal stacking type matching rule, which specifically comprises the following steps: A reinforcement learning environment including state space, action space and reward function is constructed, and the reward function is calculated according to the feasible path of the package to be stacked, the space score of the candidate stacking type and the support score; The candidate stacking type is quantitatively evaluated, which includes feasible path quantitative analysis, space score calculation and support score calculation, to obtain the feasible path score, the space score and the support score; According to the PPO reinforcement learning algorithm, the policy network is trained and iteratively optimized in combination with the state space and the action space, based on the trained policy network, a candidate stacking type is simulated to be executed, a comprehensive reward score is calculated according to the feasible path of the to-be-stacked package, the space score and the support score, and the candidate stacking type with the highest comprehensive reward score is output as the optimal stacking type.

4. The automatic palletizing method according to claim 2, characterized in that, The tray on which the packages are stacked is segmented by a preset segmentation model, the optimal stacking type and a preset path planning algorithm are combined to calculate the tray gap and generate a stacking path, and the method specifically comprises the following steps: Obtain the tray point cloud data for stacking the to-be-stacked packages, segment the package contour on the tray according to a preset YOLOv8 segmentation model and the tray point cloud data, and obtain a segmentation result, wherein the segmentation result includes available gaps; Preliminarily screen the available gaps in the segmentation result according to the optimal stacking type, and exclude the available gaps that conflict with the structure of the optimal stacking type; Iterate through all the available gaps obtained after preliminary screening, calculate the minimum bounding rectangle of each gap, screen all the minimum bounding rectangles that are not smaller than the package size, and aggregate them into a gap candidate set, calculate the space utilization and the optimal placement pose of each candidate gap in the gap candidate set, aggregate all the optimal placement poses into a candidate placement pose list, and arrange the candidate placement pose list according to the space utilization.

5. The automatic palletizing method according to claim 4, characterized in that, Further comprising the following steps: Obtain the current pose of the mechanical arm and the candidate gap and its optimal placement pose, construct a tray geometric model based on the tray point cloud data, and obtain a workspace model carrying anti-collision region markers; Generate multiple candidate paths that meet preset constraint conditions based on an improved RRT* algorithm, wherein the constraint conditions include a space constraint condition and an acceleration constraint condition, the space constraint condition is that the candidate path does not collide with the tray geometric model or the anti-collision region, and the acceleration constraint condition is that the corner acceleration of the candidate path does not exceed a preset threshold value; Score the candidate paths that meet the constraint conditions according to a preset path scoring rule, and take the candidate path with the highest score as the stacking path.

6. The automatic palletizing method according to claim 2, characterized in that, The flexible gripper device is controlled to grasp the to-be-stacked package, the grasping pose is corrected in real time in combination with the feedback of a pre-set torque sensor, and the mechanical arm is controlled to execute stacking according to the stacking path, and the method specifically comprises the following steps: Obtain an initial clamping force according to a preset clamping force parameter based on the package material, the flexible gripper device includes a gripper and a pneumatic tendon that drives the gripper, a torque sensor is pre-installed on the flexible gripper device, and a rotary encoder and an inertial measurement unit are pre-installed on the mechanical arm; Control the pneumatic tendon to drive the gripper to grasp the to-be-stacked package at the initial clamping force; During the execution of stacking by the mechanical arm, the real-time clamping force is monitored through the torque sensor, and the real-time clamping force is dynamically adjusted when it is detected that the deformation amount of the to-be-stacked package exceeds a preset deformation threshold. The joint angle feedback of the rotary encoder and the end acceleration collected by the inertial measurement unit are used to estimate the actual grabbing pose deviation, a joint compensation instruction is generated according to the actual grabbing pose deviation and a PID controller, and the grabbing pose of the mechanical arm is corrected according to the joint compensation instruction; When the torque sensor detects a preset contact force condition, a gripper release instruction is triggered to complete the placement of the to-be-stacked package.

7. The automatic palletizing method according to claim 6, characterized in that, After the placement of the to-be-stacked package is completed, the following steps are further included: The placement pose of the current package is obtained, the gravity center coordinates of the current package are calculated in combination with the package size, the gravity center coordinates of the remaining packages on the tray are obtained, and the overall gravity center coordinates of all the packages on the tray are calculated in combination with the gravity center coordinates of the current package; The stable center coordinates of the tray are obtained, the gravity center offset is calculated in combination with the overall gravity center coordinates, and the gravity center offset is compared with a preset warning range; When the gravity center offset falls within the warning range, the gravity center offset control strategy is executed in combination with the overall gravity center coordinates and the stable center coordinates.

8. An automatic palletizing device, characterized by Based on an automatic stacking system, the automatic stacking system includes a mechanical arm and a flexible gripper device at the end of the mechanical arm, and includes the following modules: An optimal stacking type matching module is used to obtain package information of a to-be-stacked package, and the optimal stacking type is matched for the current to-be-stacked package in combination with the package information and historical stacking type data of the automatic stacking system; A stacking path planning module is used to perform point cloud segmentation on a tray on which packages are placed by using a preset segmentation model to obtain a segmentation result, calculate a tray gap in combination with the optimal stacking type and a preset path planning algorithm, and generate a stacking path; A grabbing pose correction module is used to control the flexible gripper device to grab the to-be-stacked package, correct the grabbing pose in real time in combination with feedback of a torque sensor set in advance, and control the mechanical arm to perform stacking according to the stacking path; An automatic warehousing module is used to generate a stacking three-dimensional model when the segmentation result of the tray matches the optimal stacking type, bind the stacking three-dimensional model with an AGV vehicle under the tray, and control the AGV vehicle to automatically enter a warehouse.

9. A smart terminal, characterized by A memory and a processor are included, and the memory stores at least one instruction, at least one program, a code set, or an instruction set, which is loaded and executed by the processor to implement the automatic stacking method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The readable storage medium stores at least one instruction, at least one program, a code set, or an instruction set, which is loaded and executed by the processor to implement the automatic stacking method of any one of claims 1 to 7.

Images