A hybrid palletizing method and palletizing system

Abstract

The application discloses a mixed palletizing method and a palletizing system. The mixed palletizing method comprises the following steps: receiving order data, generating a palletizing scheme, the palletizing scheme comprising material warehouse-out sequence, material sorting sequence and tray palletizing position; executing material warehouse-out and sorting according to the generated palletizing data; conveying the material to a palletizing robot; conveying an empty tray to the robot; a mechanical hand of the robot palletizing the material to the tray according to the tray palletizing position; wrapping the material after palletizing is completed; removing the tray after the material is wrapped from the robot; and the mechanical hand palletizing the material to the tray, which comprises the following steps: when the mechanical hand clamps the next layer of material to the tray, the tray is lowered synchronously. The mechanical hand only needs to move horizontally to grab the material and palletize the material to the upper layer of material, and does not need to perform height direction operation due to avoiding the height space after palletizing the layer of material, so that the operation path of the mechanical hand is shortened and simplified, the operation time of the mechanical hand is saved, and the palletizing efficiency is improved.

Description

Technical Field

[0001] This invention relates to the fields of intelligent manufacturing and automation technology, and in particular to a hybrid palletizing method and palletizing system in the field of logistics and warehousing. Background Technology

[0002] Palletizing is a common operation in logistics warehousing. A mixed-case palletizing system is an automated system that intelligently places cartons of different sizes, weights, specifications, and SKUs onto a single pallet. With the continuous growth of market demand and the increasing requirements for logistics efficiency, the innovation and development of warehousing technology has become a focus of industry attention.

[0003] Existing technology CN103764523A discloses a method and facility for storing and / or picking product units. By combining a transport shuttle with a lifting unit, it solves the problems of space waste and high workload in the storage and picking of product units in logistics systems, achieving rapid and automated handling and processing of product units, and improving the system's economic efficiency and space utilization. However, it does not address a method or system for intelligently stacking cartons of different sizes, weights, specifications, and SKUs onto a single pallet.

[0004] Existing technology CN119706148A discloses an online mixed palletizing method, apparatus, and equipment. By determining the free space of the carrier and calculating the matching degree in a logistics warehousing scenario, it selects the optimal placement position, solving the automation problem of online mixed palletizing, improving mixed palletizing efficiency and stability, and enhancing logistics turnover efficiency. However, it does not involve the optimization of the palletizing robot method, and the efficiency needs to be further improved.

[0005] Specifically, existing hybrid palletizing systems have the following shortcomings: First, regarding empty pallet replenishment, existing systems typically begin supplying the next empty pallet only after a pallet has been completed and moved out of the palletizing station. The entire process is executed sequentially, and the robotic arm cannot start the next round of palletizing operations before the empty pallet is in place, resulting in a significant downtime.

[0006] Secondly, regarding the wrapping process, in traditional solutions, the wrapping station is independent of the palletizing station. The wrapping process can only begin after all pallets have been palletized and transferred to the wrapping station. The wrapping process (which typically takes 30-60 seconds) is executed sequentially with the next round of palletizing operations, directly consuming cycle time and becoming a bottleneck for system efficiency.

[0007] Third, regarding palletizing efficiency, after the robotic arm picks up materials from the conveyor line, it needs to lift the materials to the highest stacking layer of the pallet for placement. As the number of stacking layers increases, the lifting stroke becomes increasingly longer, leading to a gradual extension of the single stacking cycle, and an overall decrease in palletizing efficiency as palletizing progresses. Summary of the Invention

[0008] The main objective of this invention is to overcome the shortcomings of the prior art and provide a hybrid palletizing method and system. Through further optimization of the palletizing method and system, the efficiency of hybrid palletizing is further improved.

[0009] To solve the above-mentioned technical problems, the technical solution provided by the present invention is as follows: A hybrid palletizing method, the method comprising: Receive order data and generate a palletizing plan, which includes the material outbound sequence and pallet palletizing position; Material sorting and outbound processing are performed according to the generated palletizing scheme; Materials are conveyed to the palletizing robot; Empty pallets are transported to the robot; The robot's robotic arm stacks materials onto the pallet according to the pallet's stacking position; After palletizing, the material is wrapped with film; The tray containing the wrapped material is removed from the robot; The robotic arm stacks materials onto the pallet, including: When the robotic arm picks up the material from the next layer and places it onto the pallet, the pallet descends synchronously. When the material is wrapped after palletizing, the robotic arm simultaneously palletizes the material to the next empty pallet; This includes replenishing empty pallets to the pallet buffer area in advance, and simultaneously moving empty pallets from the pallet buffer area to the material palletizing area when the materials are wrapped after palletizing is completed.

[0010] This invention also provides a hybrid palletizing system for use with the hybrid palletizing method, the system comprising: A processing control unit is configured to receive order data and generate a palletizing scheme, the palletizing scheme including the material outbound sequence and pallet palletizing position. A material sorting and output mechanism, configured to perform material sorting and outbound processing according to the generated palletizing scheme; A palletizing device configured to palletize materials onto the pallet according to the pallet palletizing position, wrap the palletized materials with film after palletizing, and remove the wrapped materials and the pallet. A material conveyor line configured to convey materials from the material sorting output mechanism to the palletizing device; in, The palletizing device includes a palletizing robot, a first-stage palletizing elevator, and a second-stage palletizing elevator; The robot includes a robotic arm, which is located above the first-stage palletizing elevator, which is located above the second-stage palletizing elevator. The palletizing section elevator is configured to support the pallet for the robotic arm to palletize materials. When the robotic arm picks up the next layer of material and places it on the pallet, the palletizing section elevator performs a descent, causing the pallet to descend synchronously.

[0011] Compared with the prior art, the present invention has the following beneficial effects: (1) When the robot completes the material stacking of one layer and moves on to the next layer, the pallet is lowered synchronously by the lifting machine of the stacking section. At this time, the robot only needs to move horizontally to grab the material and stack it on the material of the next layer. It does not need to perform vertical movement to avoid the height space after the material is stacked. This shortens and simplifies the robot's running path, saves the robot's running time, and improves the stacking efficiency.

[0012] (2) By setting up segmented elevators (a first-stage palletizing elevator and a second-stage palletizing elevator) at the palletizing station, a parallel operation mode of palletizing and wrapping is realized. After a full pallet is moved from the first-stage palletizing elevator to the second-stage palletizing elevator, the wrapping mechanism immediately wraps the material; at the same time, the next empty pallet is moved synchronously from the pallet buffer mechanism to the first-stage palletizing elevator, and the robot arm can start the palletizing operation of the new pallet without waiting, which significantly reduces the single pallet cycle time.

[0013] (3) By independently setting up a pallet lifting machine and a pallet buffer mechanism, the empty pallets can be replenished in advance and continuously supplied, eliminating the downtime of the robotic arm caused by waiting for the empty pallets to be in place, and further improving the overall capacity of the system.

[0014] (4) By using reinforcement learning, spatial segmentation and dynamic constraint control algorithms to automatically generate the optimal stacking scheme, and combining the sorting buffer output mechanism and the position posture correction function, it is ensured that the mixed materials of multiple SKUs can be stacked accurately according to the planned order and target posture, thereby improving the stacking quality and system reliability. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a hybrid palletizing method according to the present invention; Figure 2 This is a comparative schematic diagram of the principle of synchronous pallet descent in a hybrid palletizing method according to the present invention; Figure 3 This is a schematic diagram of the overall structure of a hybrid palletizing system according to the present invention; Figure 4 This is a schematic diagram of the overall structure of a hybrid palletizing system palletizing device according to the present invention; Figure 5 This is a top view of a hybrid palletizing system palletizing device according to the present invention; Figure 6 This is a diagram showing the positional relationship between the first-stage and second-stage palletizing elevators in a hybrid palletizing system according to the present invention. Figure 7 This is a schematic diagram of the structure of a hybrid palletizing system palletizing device according to the present invention; Figure 8 for Figure 7 Enlarged view of point A in the middle; Figure 9 This is a side view of a palletizing device in a hybrid palletizing system according to the present invention; In the diagram: 1. Processing control unit; 2. Material sorting and output mechanism; 3. Palletizing device; 31. Frame; 32. Robot arm; 33. Palletizing stage 1 elevator; 34. Palletizing stage 2 elevator; 35. Pallet elevator; 36. Wrapping mechanism; 37. Palletizing pickup position; 371. Lifting mechanism; 38. Pallet conveyor belt; 39. Palletizing material conveyor belt; 4. Material transfer line; 5. Pallet. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0019] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the term "and / or" throughout the text includes three solutions; taking A and / or B as an example, it includes technical solution A, technical solution B, and a technical solution that simultaneously satisfies A and B. Furthermore, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of a person skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0020] Example 1 Reference Figures 1 to 2 This embodiment provides a hybrid palletizing method, the method including: Receive order data and generate a palletizing plan, which includes the material outbound sequence, material sorting sequence, and pallet 5 palletizing positions. Perform material outbound and sorting based on the generated palletizing data; Materials are conveyed to the palletizing robot; Empty pallet 5 is conveyed to the robot; The robot's robotic arm 32 stacks the materials onto the pallet 5 according to the pallet 5's stacking position; After palletizing, wrap the materials with film; The tray 5, after the material wrapping is completed, is removed from the robot; Robotic arm 32 stacks materials onto pallet 5, including: When the robotic arm 32 picks up the material from the next layer and places it onto the pallet 5, the pallet 5 descends synchronously.

[0021] In this embodiment, the pallet 5 is lowered simultaneously when the robotic arm 32 places a layer of material, which allows the robotic arm 32 to always be in the optimal working height range for its gripping and placing actions. This eliminates the need for the robotic arm 32 to adjust its vertical working height to adapt to the stacking structure that rises layer by layer. This simplifies the running path of the robotic arm 32, reduces the action time, effectively improves the overall palletizing efficiency, and reduces the control complexity of the robotic arm 32, as well as the unnecessary wear of mechanical parts. During the mixed palletizing process, materials of different specifications will be arranged sequentially to the corresponding positions on pallet 5 according to the palletizing scheme. The fixed-height robotic arm 32, in conjunction with the lifting pallet 5, can better adapt to the mixed operation requirements of materials of different heights, without the need for additional redundant design of the stroke of the robotic arm 32.

[0022] In one embodiment, when the robotic arm 32 picks up the material from the next layer and places it onto the pallet 5, the robotic arm 32 moves horizontally, and the pallet 5 descends at a height greater than or equal to the height of the material from the next layer.

[0023] In this embodiment, by moving the robot arm 32 horizontally, it is ensured that when the material is placed at the target position, the robot arm 32 does not need to adjust its height upward to avoid the already stacked upper layer of material. The horizontal movement can complete the connection action of picking up and placing materials. This will not cause the material to scrape and collide with the already stacked part during the downward movement due to insufficient descent of the pallet 5. This not only ensures the neatness of the stacking, but also avoids material collision damage. At the same time, it further reduces the action adjustment time of the robot arm 32 and maintains a stable stacking operation rhythm.

[0024] In one embodiment, when the robot arm 32 stacks materials onto the pallet 5 according to the pallet 5 stacking position, the pallet 5 is located on the palletizing section elevator 33; When wrapping the materials after palletizing is completed, pallet 5 is moved to the second-stage palletizing elevator 34; Among them, the first-stage palletizing elevator 33 is located above the second-stage palletizing elevator 34.

[0025] In this embodiment, the palletizing operation and the wrapping operation are respectively arranged on the running paths of the palletizing first-stage elevator 33 and the palletizing second-stage elevator 34, which are distributed vertically. There is no need to occupy additional space on the same horizontal plane to plan the transfer path, which can save workshop floor space. At the same time, after the palletizing is completed, the pallet 5 can be directly moved down to the palletizing second-stage elevator 34 to carry out the wrapping operation. By setting up segmented elevators (first-stage palletizing elevator 33 and second-stage palletizing elevator 34) at the palletizing station, a parallel operation mode of palletizing and wrapping is achieved. After a full pallet 5 is moved from the first-stage palletizing elevator 33 to the second-stage palletizing elevator 34, the wrapping mechanism 36 immediately wraps the material. At the same time, the next empty pallet 5 is moved synchronously from the pallet buffer mechanism to the first-stage palletizing elevator 33. The robot arm 32 can start the palletizing operation of the new pallet 5 without waiting, which significantly reduces the cycle time of a single pallet 5.

[0026] In one embodiment, when the material is wrapped after palletizing is completed, the robot arm 32 simultaneously palletizes the material to the next empty pallet 5; This includes replenishing empty pallets 5 to the pallet buffer area in advance, and when the materials are wrapped after palletizing, the empty pallets 5 are simultaneously moved from the pallet buffer area to the material palletizing area.

[0027] In this embodiment, while palletizing, empty pallets 5 are replenished to the buffer area in advance for standby. While the current pallet 5 carrying materials is moved out of the palletizing area for wrapping, the empty pallet 5 can be directly moved into the working position. The robot arm 32 does not need to wait for the empty pallet 5 to be replenished before starting the next round of palletizing, eliminating the process gap of waiting for the transfer of empty pallets 5, further increasing the palletizing workload per unit time, enhancing the continuity of the overall operation process, and improving the operating efficiency of the entire palletizing production line.

[0028] In one embodiment, the pallet elevator 35 replenishes the empty pallet 5 to the pallet buffer area in advance. When the material is wrapped after the palletizing is completed, the palletized material and pallet 5 are moved to the second-stage palletizing elevator 34, and the next empty pallet 5 is moved to the first-stage palletizing elevator 33 located in the material palletizing area. Among them, the first-stage palletizing elevator 33 and the second-stage palletizing elevator 34 are located on one side of the pallet elevator 35, and the first-stage palletizing elevator 33 is located above the second-stage palletizing elevator 34.

[0029] In this embodiment, two-stage elevators are arranged in upper and lower layers. Combined with the pre-replenished empty pallet 5 buffer scheme, the vertical layout saves the floor space of the workshop. The parallel operation of the process eliminates the gap of waiting for empty pallets 5, allowing the two core processes of palletizing and wrapping to be carried out continuously and synchronously. This balances the flexibility of the production line layout and the efficiency of the overall operation. At the same time, the structure of the whole system is simple, easy to install and maintain, and adaptable to the needs of production workshops of different sizes. Furthermore, this application achieves advance replenishment and continuous supply of empty pallets 5 through independently set pallet lifting machine 35 and pallet buffer mechanism, eliminating the downtime waiting time of robot arm 32 caused by waiting for empty pallets 5 to be in place, and further improving the overall capacity of the system.

[0030] In one embodiment, material outbound and sorting are performed based on generated palletizing data, including: performing material outbound according to the outbound order and conveying it to the sorting step, and performing material sorting according to the material sorting order; The material sorting process includes temporarily storing materials in order to the material buffer line according to the palletizing position of pallet 5. Once all materials on a pallet 5 have arrived and the sorting is completed, the materials are moved sequentially to the material conveyor line 4 and then transported to the palletizing robot.

[0031] In this embodiment, by pre-sorting and caching the materials, the scattered outbound materials are pre-integrated into the palletizing sequence of the corresponding pallet 5. The palletizing robot can continuously grab the entire group of materials in a preset order to complete the palletizing, without having to frequently switch and wait between different outbound materials. This avoids the idle gap of the robot waiting for materials to arrive, further improving the continuous operation efficiency of the palletizing operation. At the same time, it can accurately match the mixed coding and palletizing requirements of different materials, ensuring that the palletizing position is completely consistent with the preset scheme, improving the regularity and accuracy of mixed palletizing, and reducing the probability of mis-coding and garbled coding.

[0032] In one embodiment, receiving order data and generating a palletizing scheme includes automatically generating the optimal palletizing scheme using reinforcement learning, spatial segmentation, and dynamic constraint control algorithms. The order data includes the packaging box dimensions, weight, and material type for different materials.

[0033] In this embodiment, reinforcement learning algorithms are used to learn and optimize massive historical palletizing data. By combining the packaging box size, weight, and material type attribute information of different materials, the algorithm can quickly adapt to different material combinations under the comprehensive dynamic constraints of spatial segmentation rules. It outputs the optimal stacking scheme that balances space utilization and palletizing stability. Without the need for manual pre-setting of palletizing templates, it can flexibly cope with the palletizing needs of mixed orders with multiple categories and small batches, greatly improving the generation efficiency and adaptability of mixed palletizing schemes, reducing the workload of manual scheme adjustment, and ensuring that the generated palletizing scheme meets the requirements of placing heavy goods at the bottom and light goods at the top, and grouping and organizing similar materials, thereby improving the safety of storage and transportation.

[0034] In one embodiment, receiving order data and generating a palletizing scheme includes: S1. Obtain the attribute information, pallet parameters, robot operation parameters, and business rule information of the set of boxes to be palletized; S2. Based on attribute information, pallet 5 parameters, robot operation parameters and business rule information, establish a physical model of the box and a palletizing constraint model; S3. Use an integer scheme to perform local combination optimization on one or more boxes to be stacked, and construct a MiniBin. S4. Calculate the external dimensions, total weight, center of gravity, bottom support features, placement orientation, load-bearing features, and gripping features of each MiniBin. S5. Use a layer-by-layer stacking strategy to determine the current layer to be placed, and establish the current layer's placed state, the current layer's remaining space state, and the unplaced MiniBin state. S6. Input the current layer's placed state, the current layer's remaining space state, and the unplaced MiniBin state into the reinforcement learning decision model, and output a four-degree-of-freedom action decision. S7. Based on the four-degree-of-freedom action decision, candidate placements are made for the target MiniBin, and the candidate placement results are constrained and verified. S8. If the constraint check passes, the target MiniBin is placed and the palletizing status is updated; if the constraint check fails, the four-degree-of-freedom motion decision is masked, corrected, or re-determined. S9. When the current layer meets the layer switching condition, switch to the next layer to continue executing the placement decision until all palletized boxes are placed or the termination condition is met, and output the final palletizing scheme.

[0035] In this embodiment, by pre-combining multiple boxes into MiniBins, the complexity of the palletizing decision space is reduced, the computational efficiency is accelerated, and physical information can be pre-integrated through the feature calculation of MiniBins, providing a reliable input basis for subsequent reinforcement learning decisions. The use of a reinforcement learning model to output four-degree-of-freedom action decisions can quickly adapt to the dynamic changes of different box attributes and different operation parameters. Compared with traditional fixed rule palletizing, it can effectively improve the space utilization of pallet 5, while meeting the various constraints of robot operation and ensuring the stability and reliability of the palletizing process. Furthermore, by utilizing reinforcement learning, spatial segmentation, and dynamic constraint control algorithms to automatically generate the optimal stacking scheme, and combining it with the sorting buffer output mechanism and the cargo position posture correction function, it can also ensure that multi-SKU mixed incoming materials can be accurately stacked according to the planned order and target posture, thereby improving the stacking quality and system reliability. The Learning Environment World Model (LEWM) introduced in this embodiment is a neural network prediction model based on the attention mechanism. Its core function is to predict the environmental changes (including the change in fill rate Δfill, immediate reward r, task termination probability p_done, and placement success probability p_success) that will occur after performing a candidate action in a given stacking state.

[0036] In one embodiment, obtaining the attribute information of the set of boxes to be palletized includes the set of box attributes, the set of pallet parameters, the set of robot operation parameters, and the set of business rule constraints. The set of parameters is obtained by inputting the set of box attributes, the set of pallet parameters, the set of robot operation parameters, and the set of business rule constraints into the definition.

[0037] In this embodiment, Input is defined as {B, P, R, Ru}, where B is the set of box attributes, P is the set of pallet parameters, R is the set of robot operation parameters, and Ru is the set of business rule constraints. Among them, the box assembly: ; L b,i W b,i H b,i The original length, width, and height of the enclosure; Mb,i Total weight of the enclosure; C b,i =(C b,i,x C b,i,y C b,i,z ): Local center of gravity of the box; θ b,i : The set of rotation angles that the enclosure is allowed to rotate; N b,i : Maximum number of stacking layers in the vertical direction of the box.

[0038] Tray parameter set: ; L p W p Pallet's effective load-bearing length and width; Maximum height of the stack type; Pallet rated load capacity; d gap Minimum safety clearance between units.

[0039] Robot operation parameter set: ; p r =(p rx ,p ry ,p rz ): The three-dimensional coordinates of the robot base relative to the origin of the tray; : Length of connecting rods on each axis; R max Maximum horizontal working radius; Z r,min Z r,max : Upper and lower limits of vertical travel at the end; o tcp : Fixture TCP offset; d safe Safe distance during exercise Business rule constraint set: .

[0040] MixFlag: Whether multiple SKUs are allowed to be mixed in the same layer; min : Minimum overlap ratio of upper-level enclosure supports; OverHangFlag: Whether to prohibit the enclosure from being suspended and cantilevered; : Permissible offset threshold for stack center of gravity.

[0041] In one embodiment, the steps for establishing the physical model of the container and the palletizing constraint model include: S21. Based on the center coordinates and rotational attitude of the enclosure, establish a standardized AABB bounding box model to accurately represent the area occupied by the enclosure:

[0042] Where, x i y i , z i For spatial coordinates, ; : Equivalent dimensions of the box after rotation; S22. Construct a complete set of constraints C0, including boundaries, collisions, height, load, robot reachability, and stacking stability: .

[0043] In one embodiment, the steps of constructing a MiniBin include: employing a box local clustering and combination strategy to construct an indivisible MiniBin basic stacking unit, defining box subsets that are mutually exclusive and completely cover the set of boxes to be stacked:

[0044] Introduce 0-1 integer variables to represent box affiliation relationships:

[0045] The goal is to maximize the utilization of unit space, achieving a compact assembly of cabinets:

[0046] Solving for the optimal MiniBin unit set M={m k}

[0047] In one embodiment, the geometry, mass, center of gravity, attitude, support, load-bearing, and grasping features of each MiniBin unit are quantitatively calculated.

[0048] In this embodiment, geometric features (dimensions): ; Quality characteristics: ; Weighted central features: ; Attitude orientation features: Supports 0° / 90° adaptive rotation switching; Comprehensive state feature vector: .

[0049] S support : Bottom surface supporting mask features; S ratio : Percentage of the supported area; Maximum load capacity at the top; grasp : Grasp the center point; Dir grasp Allow grab direction.

[0050] In one embodiment, the step of modeling the environment state of layer-by-layer palletizing includes: constructing a single-layer complete environment state space by adopting a layer-by-layer palletizing strategy.

[0051] In this embodiment, the environmental state consists of three parts: the placed state, the remaining space state, and the unplaced unit state. The spatial expression of the environmental state is as follows: in: The current layer already contains a MiniBin collection; This is a set of free rectangular regions within the layer; This is the candidate set of MiniBin to be placed.

[0052] In one embodiment, the four-degree-of-freedom action decision construction steps include: constructing a Markov Decision Process (MDP) model, taking the layer environment state as input, and outputting four-degree-of-freedom intelligent placement actions to achieve adaptive decision-making.

[0053] In this embodiment, MDP is defined as follows: ; Design a four-DOF discrete-continuous hybrid motion space: Where: a1: MiniBin selection to be placed; a2: Free area selection; a3: Placement rotation posture; a 4x / a 4y Normalized placement coordinates within the region; Construct a multi-dimensional fusion reward function that integrates space utilization, stack compactness, stability, and constraint penalties: ; By optimizing the strategy gradient, the long-term cumulative discount return is maximized, and intelligent decision-making iterative optimization is achieved.

[0054] In one embodiment, the steps of candidate placement of the target MiniBin based on four-degree-of-freedom action decisions and constraint verification of the candidate placement results include: Based on the reinforcement learning action parsing, the candidate placement poses and bounding boxes of MiniBin are obtained, and the global constraint validity is verified through the indicator function: .

[0055] In this embodiment, the environment state is adaptively updated based on the verification results. If the placement is legal, the layer occupancy state and free space are updated. If the placement is illegal, the original state is maintained and a penalty is applied, forming a closed-loop iterative optimization mechanism.

[0056] In one embodiment, during the process of establishing the palletizing constraint model in steps S1 to S2, the mode cargo selection model is used to determine the baseline layer height and dominant cargo type of the palletizing. The mode cargo selection model includes area threshold screening, comprehensive optimization scoring and baseline height determination.

[0057] In this embodiment, the area threshold screening method is as follows: For each type of box to be stacked, the bottom area when its maximum face is down is calculated and multiplied by the remaining quantity to obtain the total unfolded area. If the total unfolded area is ≥ pallet area × area threshold (default is 70%), then the box is marked as the "modal candidate". The setting of the area threshold ensures that the selected modal goods have a sufficient quantity to cover the main area of ​​the pallet, forming a stable and efficient baseline layer; The comprehensive selection and evaluation method is as follows: for all the majority candidate goods, the following comprehensive evaluation function is used to rank and select the best: Score_selection = S_volume·0.30 + S_coverage·0.40 + S_count·0.15+ S_base·0.15 Where: S_volume: Volume score, normalized from box volume to 0-100 points; S_coverage: Coverage score, normalized from the area coverage of the box when it is optimally arranged on the pallet to 0-100 points (weight 40%, the most important dimension). S_count: The maximum number of items that can be placed per layer, with 10 points per item and a maximum of 50 points. S_base: Base area score, normalized from 0-50 points by the ratio of the box's base area to the pallet area.

[0058] The method for determining the baseline height is as follows: select the mode item with the highest comprehensive score as the baseline item, and use the height of this item under the optimal coverage orientation as the baseline height Hbase of the current palletizing layer. All subsequent MiniBin generation is performed with this baseline height as the target for height matching, ensuring that the height of each MiniBin within the same layer is consistent, forming a flat palletizing layer surface.

[0059] Furthermore, in the MiniBin construction process described in step S3, an integer combination optimization method is used to locally combine one or more boxes with the same orientation to be stacked to form a MiniBin unit.

[0060] Specifically, the integer combination optimization method includes height matching and stacking optimization: For a given baseline height Hbase (determined by the mode cargo selection model), height matching is performed within the height tolerance δ_h (default is 10mm). If the height h of a single box in this orientation satisfies |h-Hbase|≤δ_h, then a MiniBin is directly generated as a single piece (stack number stack=1); if the height of a single box does not match but stacking is allowed, then the optimal stack number k is searched within the stacking limit Nmax (default is 5 layers) to satisfy |k·hHbase|≤δ_h, generating a stacked MiniBin; Orientation Optimization Scoring: For each item, iterate through all its available orientations (including maximum face down, maximum face down rotated 90°, second largest face down, and second largest face down rotated 90°), and select the best MiniBin from the candidate MiniBins generated for each orientation according to the following scoring function: S_item(l, w, q) = N_per_layer·50 + C_ratio·200 + U_ratio·100, in: , which is the maximum number of MiniBins that can be placed in a single layer of the tray; C_ratio = (N_per_layer· l· w) / A_pallet, which is the ratio of the coverage area to the pallet area; U_ratio = (l·w·min(q, N_per_layer)) / A_pallet, which is the actual utilization rate after considering the available quantity; q represents the remaining quantity of the item; Descending sorting and expansion: After constructing the optimal MiniBin for all items, sort them in descending order of volume. Then, expand each MiniBin according to its stack count to obtain the actual usable quantity (usable quantity = remaining quantity of the item / / stack count), for use by the subsequent reinforcement learning decision model. Volume-priority sorting ensures that larger MiniBins are given priority in placement decisions, helping to improve overall space utilization.

[0061] In one embodiment, in the reinforcement learning decision model of step S6, an improved Dueling DQN (Dueling Deep Q-Network) deep reinforcement learning framework is used to output four-degree-of-freedom action decisions during the layer-by-layer stacking process.

[0062] In this embodiment, the Dueling DQN network includes: a feature extraction layer: employing a two-layer fully connected network (512-256 dimensional), using the ReLU activation function to map the 200-dimensional palletizing state vector into a high-dimensional feature representation; a value stream branch: outputting a single scalar V(s), representing the baseline value of the current palletizing state; a dominance stream branch: outputting an |A|-dimensional vector A(s,a), representing the relative dominance of each candidate action in state s; and an action masking mechanism: when calculating the Q-value, the action masking matrix sets the Q-value of infeasible actions (such as MiniBins whose size exceeds the remaining space, MiniBin types whose quantity has been exhausted, etc.) to negative infinity, ensuring that the network selects only from legal actions. The network structure is as follows: The Dueling DQN network decomposes the Q-value function into two independent flows: the state value function V(s) and the advantage function A(s,a). The final Q-value is calculated using the following formula: Q(s,a)=V(s)+A(s,a)(1 / |A|)*Σ_{a'}A(s,a') Where s represents the current palletizing state (including pallet-placed, remaining space, and unplaced MiniBin states), a represents the candidate action (i.e., selecting a MiniBin type and determining its placement location and orientation), and |A| represents the number of selectable actions. This decomposition structure allows the network to learn the value of the state itself and the advantage of each action relative to the average level, thus maintaining a stable value function estimate even with a large action space.

[0063] The state space is designed as follows: the palletizing state vector s is a 200-dimensional feature vector, containing the following information dimensions: Global tray status: current layer fill rate, flatness, number of placed items, current step count; Height map statistical characteristics: The mean, standard deviation, and maximum value of the pallet height map reflect the flatness of the current layer surface; Statistical characteristics of remaining MiniBin: mean and standard deviation of height of various types of remaining MiniBin, proportion of remaining quantity, proportion of remaining total volume, and estimated number of remaining layers; Surface quality characteristics: usable area ratio, height difference range, fill area ratio; Interlayer characteristics: the difference between the average height of remaining MiniBins and the current surface height, and the proportion of higher MiniBins; Individual characteristics of each type of MiniBin: remaining quantity ratio, bottom area ratio of the tray, height normalization value, weight normalization value, and availability indicator (each type of MiniBin has 5 dimensions, and a maximum of 34 types of MiniBins are supported). The action space is designed as follows: The action space is defined as the set of indices of all MiniBin types available in the current layer. For each selected MiniBin, the system automatically determines its optimal placement coordinates (x, y, z) and placement orientation through the spatial candidate position scoring model, thereby transforming the complex continuous spatial placement decision into a discrete MiniBin selection decision, which greatly reduces the exploration difficulty of reinforcement learning. In one embodiment, the training mechanism of the Dueling DQN deep reinforcement learning decision model is as follows: Experience Replay: A Prioritized Experience Replay (PER) mechanism is adopted, using the absolute value of the Temporal Difference Error (TD) as the sampling priority. This ensures that high-learning-value experiences in the training samples are replayed more frequently, improving sample efficiency. The sampling probability of the i-th experience is: ; Where pi = |δ_i| + ε is the priority of the i-th experience (δ_i is the TD error, ε is a small constant to prevent zero probability), and α is the priority index (α=0.6 in this embodiment).

[0064] Importance sampling weights: To correct the bias introduced by priority sampling, importance sampling weights w_i are used to weight the gradient updates. ; Where N is the size of the experience pool, and β is the bias correction exponent, which increases linearly from an initial value of 0.4 to 1.0.

[0065] Target network soft update: Maintain a target network whose parameter θ_target is gradually approximated to the policy network parameter θ_policy through soft updates. θ_target = τ · θ_policy + (1 - τ) · θ_target; Here, τ = 0.005 is the soft update coefficient, which ensures the stability of the training process and avoids drastic fluctuations in Q-value estimation.

[0066] ε-Greedy Exploration Strategy: During training, legal actions are randomly selected with probability ε for exploration, and the current optimal action is selected with probability (1-ε). The exploration rate ε decreases exponentially from an initial value of 1.0 to 0.01. ε= max(ε_end, ε·ε_decay); Where ε_decay = 0.995 is the decay factor per step.

[0067] Gradient clipping: Clips the gradients of network parameters, limiting the upper limit of the gradient norm to 10.0 to prevent gradient explosion.

[0068] In one embodiment, the architecture and operational logic of the LEWM learning environment world model are as follows: Model Architecture: LEWM consists of the following core modules: MiniBin Encoder: Employs a two-layer fully connected network (input 6D → 128D → 128D) to encode the physical features of each MiniBin (including remaining quantity ratio, base area ratio, height ratio, volume ratio, aspect ratio, and weight) into a 128-dimensional embedding vector; State Encoder: Employs a two-layer fully connected network (input 20D → 128D → 128D) to encode the palletizing state features into a 128-dimensional embedding vector; Induced Set Attention Block (ISAB): Employs a multi-layer ISAB structure, introducing a set of learnable inducing points (32 in this embodiment) as the attention bottleneck, reducing the computational complexity of multi-element self-attention in the MiniBin set from O(N) to O(N). 2 The number of minibins is reduced to O(N·M) (where N is the number of minibins and M is the number of induction points), enabling the model to maintain efficient forward inference capabilities under different numbers of minibins. Each ISAB block contains two multi-head attention operations: first, the MiniBin set is cross-attention aggregated by the inducing point; second, the MiniBin set is cross-attentioned again on the aggregated result, thereby summarizing the overall features of the MiniBin set. The multi-head pooling module (PMA) uses four learnable seed vectors to perform attention pooling on the ISAB output, extracting the global representation of the MiniBin set. The action embedding module transforms the embedding vectors corresponding to the selected MiniBin into a 32-dimensional action representation through a linear mapping. The fusion prediction module concatenates the state embedding, the pooled MiniBin global representation, and the action embedding, then performs deep fusion through a multi-layer fully connected network (512-dimensional → 256-dimensional → 128-dimensional), finally outputting Δfill (tanh activation), reward (linear), termination probability (sigmoid activation), and success probability (sigmoid activation) through four independent prediction heads.

[0069] Zero-Shot Planning: During the inference phase, LOWM can be directly used to assist action selection without additional training. For each candidate legal action, LOWM predicts the immediate reward and fill rate improvement of that action, and, taking into account the predicted success probability, ranks the candidate actions according to the following scoring formula: Score(a) = (r_pred·w_reward +Δfill_pred·w_fill)·p_success; Where w_reward and w_fill are weighting coefficients, and p_success is the predicted placement success probability. The system selects the candidate action with the highest score as the final action to be executed.

[0070] Multi-step forward planning: When multi-step planning is enabled (planning step size horizon>1), LEWM further estimates the forward value of the current surface quality for future palletizing layers based on single-step predictions. The smoother the surface quality, the more abundant the remaining MiniBins, and the higher the height matching degree, the higher the future planning score. The total score for multi-step planning is calculated using the following formula: Score_total(a) = Score_immediate(a) +γ·Future_potential; Where γ = 0.85 is the discount factor, and Future_potential is calculated by weighting features such as surface availability, high matching degree, and remaining MiniBin volume ratio, and decays with the remaining planning steps.

[0071] In one embodiment, during the candidate placement process of the target MiniBin in step S7, a multi-strategy candidate position generation and multi-dimensional scoring mechanism are adopted to ensure that each placement achieves the optimal space utilization effect and stacking quality.

[0072] In this embodiment, the candidate position generation strategy is as follows: the system generates candidate placement coordinates (x, y) in descending order of priority; Specifically, Strategy 1 is close-fitting placement: Iterates through the boundaries of all placed items and generates candidate coordinates that are closely adjacent to the placed items in the X or Y direction, achieving close fit between items and reducing gaps; Strategy 2 is edge alignment: It collects the X and Y boundaries of all placed items and prioritizes aligning the generated candidate coordinates with the edges of existing items to maintain the regularity of the stack shape; Strategy 3 is corner priority: It prioritizes the four corner positions of the pallet and the feasible placement positions of the corner points under the current item size; Strategy 4 is wall-side placement: It generates candidate positions by uniformly sampling along the four edges of the pallet with a search step size (default is 5mm); Strategy 5 is grid search: It performs grid sampling in the remaining space of the pallet with a fine step size as a fallback solution when the above strategies cannot generate effective candidates. The multi-dimensional scoring mechanism is as follows: For each candidate position (x, y), the system automatically determines the Z-axis height (i.e., placement height) based on the support conditions of the current layer, and performs a comprehensive evaluation according to the following multi-dimensional scoring function; Score(x,y,z)=Score_corner+Score_wall+Score_alignment+Score_compactness+Score_rectangularity-Penalty_gap-Penalty_distance; in: Score_corner (Corner Fill Score): Detects whether a newly placed item simultaneously touches the adjacent edges of multiple placed items to form a "corner fill" effect. Each corner fill is awarded 150 points, and multiple corner fills are awarded additional incremental bonuses to encourage items to be tightly embedded in the gaps between existing items. Score_wall (wall contact score): Calculates the ratio of the contact length between the placed item and the four edges of the tray to the item's perimeter. When placing an item for the first time, a maximum reward of 500 points (2 sides against the wall) or 200 points (1 side against the wall) is given for the corner position. When there are already items, a maximum reward of 50 points is given according to the contact ratio. Score_alignment (edge ​​alignment score): Checks whether the boundaries of the new item are aligned with the corresponding boundaries of the placed items (tolerance 5mm). X / Y start alignment is scored 40 points each, and X / Y end alignment is scored 30 points each. Score_compactness: Evaluates whether the new item falls within or partially within the range of an existing item group (80 points) and whether it is adjacent to an expansion of an existing item group (60 points each). It also applies a distance penalty (maximum 100 points) by calculating the distance between the center of the new item and the center of the existing item group. Score_rectangularity: Calculates the bounding box fill rate of all items (including newly placed items). A fill rate greater than 80% is awarded 200 points, and greater than 60% is awarded 100 points. A maximum of 80 points is awarded for bounding boxes with an aspect ratio close to a square. At the same time, a penalty of 100 points is imposed for creating a concave shape (items protruding too far from the center of the bounding box). Penalty_gap: Detects whether a small gap of less than 50mm is formed between a new item and an existing item (such gaps are difficult to be effectively filled by subsequent items), and applies a penalty of 80 points to each small gap; at the same time, it detects whether a "dead space" (a gap that is blocked by items on both sides and cannot be used) is formed between existing items, and applies a penalty of 300 points. Penalty_distance (sparseness penalty): When multiple items are placed on a tray, calculate the overall compactness of the bounding box (sum of the areas of all items ÷ bounding box area), and apply a penalty of up to 200 points to non-compact areas proportionally.

[0073] Support verification and Z-axis height determination: For non-bottom layer palletizing (i.e., the current layer's reference Z-coordinate is greater than 0), the system calculates the support conditions at candidate locations using a support checker. Collect all lower-level items located below the candidate area with a height gap within the allowable range (default is 5mm); Calculate the overlap area between the top surface of these lower-level items and the bottom surface of the candidate items; If the ratio of the overlapping area to the bottom area of ​​the candidate item (i.e., the support rate) is lower than the minimum support threshold (default is 60%), the position is determined to be unreliable and the candidate is excluded. If the support ratio meets the requirements, the highest surface of the supporting object below is taken as the Z-axis placement height of the candidate object.

[0074] In one embodiment, during reinforcement learning training, after each placement action is performed, the environment calculates an immediate reward value based on a multi-objective reward function model to guide the DDQN agent in learning the optimal palletizing strategy. The reward function R is composed of a weighted combination of the following components: R = R_step + R_fill + R_contact + R_flatness + R_gap + R_lookahead; The basic step reward R_step = 0.5: each successful placement of a MiniBin provides a positive incentive, encouraging the agent to place as many items as possible; Fill rate bonus R_fill: consists of two parts: the current fill rate improvement Δfill_rate and the current absolute fill rate curr_fill. R_fill =Δfill_rate·8.0 + curr_fill·1.0; For every 1% increase in fill rate, an additional reward of approximately 0.08 points is given. At the same time, the higher the current fill rate, the greater the overall reward, thus guiding the agent to pursue high space utilization. Close contact reward R_contact: Takes into account both wall-side contact and object-to-object contact. R_contact = N_wall·0.5 + N_item·2.0 + Bonus_multi_contact; In the formula, N_wall is the number of walls in contact, and N_item is the number of items already placed in contact; when the number of contact surfaces between items is ≥2, an additional 3.0 points are awarded, and when it is ≥3, an additional 1.5 points are awarded, to encourage items to form multiple close contacts and improve the structural integrity of the stack; Flatness bonus R_flatness = flatness·0.5: Flatness is defined as 1 / (1+σ / 50), where σ is the standard deviation of the height of the top surface of all placed items in the current layer. The smaller the standard deviation, the higher the flatness and the greater the bonus. This encourages each layer to form a flat top surface, providing a good foundation for the next layer of stacking. Gap penalty R_gap: When the distance between a newly placed item and an already placed item in the X or Y direction is less than 50mm, a penalty of -0.5 points is applied to each small gap to guide the agent to avoid leaving small gaps that cannot be filled. Cross-layer lookahead reward R_lookahead: When there is preview information for the next layer, the system evaluates the matching degree between the surface quality of the current layer after placement and the target height of the next layer. The higher the matching degree, the greater the lookahead reward (maximum 1.5 points). If the current layer consumes the minimum quantity of the modulo goods required for the next layer, a reservation penalty of 0.5 points is applied to guide the agent to reserve the necessary core goods resources for the palletizing of subsequent layers while satisfying the current layer's fill rate.

[0075] Reference Figure 3 This embodiment also provides a hybrid palletizing system, the system comprising: Processing control unit 1 is configured to receive order data and generate a palletizing scheme, which includes the material outbound sequence and pallet 5 palletizing position. Material sorting output mechanism 2 is configured to sort and release materials according to the generated palletizing scheme. The palletizing device 3 is configured to palletize the material to the pallet 5 according to the palletizing position of the pallet 5, wrap the material after palletizing, and remove the wrapped material and the pallet 5. Material conveying line 4 is configured to convey materials from material sorting and output mechanism 2 to palletizing device 3. The palletizing device 3 includes a frame 31, a palletizing robot, a first-stage palletizing elevator 33, and a second-stage palletizing elevator 34. The palletizing robot, the first-stage palletizing elevator 33, and the second-stage palletizing elevator 34 are all installed on the frame 31. The robot includes a robotic arm 32, which is located above the first-stage palletizing elevator 33, which is located above the second-stage palletizing elevator 34. The palletizing elevator 33 is configured to support the pallet 5 for the robot arm 32 to palletize materials. When the robot arm 32 picks up the next layer of material and places it on the pallet 5, the palletizing elevator 33 performs a descent, causing the pallet 5 to descend synchronously.

[0076] In this embodiment, the processing control unit 1 is signal-connected to the material sorting output mechanism 2, the material conveying line 4 and the palletizing device 3. The processing control unit 1 includes a processor and a storage unit. The storage unit stores a computer program. When the processor executes the computer program, it implements the hybrid palletizing method. Specifically, the processing control unit 1 receives order data from upstream systems (such as WMS / ERP). The order data includes the SKU number, packaging box dimensions (length x height x width), weight, and quantity of each material. Simultaneously, the processing control unit 1 utilizes reinforcement learning, spatial segmentation, and dynamic constraint control algorithms to automatically generate the optimal palletizing scheme by comprehensively considering constraints such as cargo weight, size, type, stability, space utilization, and equipment feasibility, thus forming palletizing data. The palletizing data includes the target placement coordinates (X, Y, Z) and placement angle of each material, as well as the material feeding sequence at the palletizing pick-up position 37. Material sorting output mechanism 2, based on the material outbound order in the palletizing data, rearranges randomly arriving materials into a feeding order that conforms to the palletizing sequence via a box conveyor line, and then outputs them sequentially to material conveyor line 4. Material sorting output mechanism 2 manages materials in groups of 20, issues feeding order instructions in batches, and counts the number of materials to be sorted, allocated, and executed in each group in real time, displaying the sorting progress through a data dashboard; The material conveyor line 4 transports the sorted materials from the material sorting output mechanism 2 to the palletizing pick-up position 37 of the palletizing device 3. Before the materials arrive at the palletizing pick-up position 37, the actual three-dimensional dimensions and attitude angles of the materials are detected by a 3D vision detection device located above the end of the conveyor line. The detection data is fed back to the processing control unit 1. The processing control unit 1 compares the actual attitude angle with the target attitude angle of the material in the palletizing data, calculates the attitude deviation, and performs a rotation compensation action synchronously during the movement stroke of the robot arm 32 after it grips the material, so that the material is accurately placed at the target coordinate position with the correct orientation angle. The first-stage palletizing elevator 33 is located above the second-stage palletizing elevator 34, forming an upper and lower segmented structure; During the palletizing operation, the empty pallet 5 is located on the palletizing elevator 33. The robot arm 32 places the materials onto the pallet 5 one by one according to the palletizing position of the pallet 5 in the palletizing data. When the robot arm 32 has completed the placement of all the materials in the current layer and begins to pick up the materials of the next layer, the palletizing elevator 33 performs a descent action. Its descent height is greater than or equal to the height of the materials in the completed layer, so that the top surface of the pallet 5 (i.e. the target placement surface of the next layer) is kept near the optimal placement height of the robot arm 32. At this time, the robot arm 32 only needs to move horizontally to place the materials to the target position of the next layer without having to lift to a higher height. This effectively shortens the running path of the robot arm 32 and keeps the palletizing efficiency from decreasing as the number of palletizing layers increases. After palletizing is completed, the second-stage palletizing elevator 34 rises, and the first-stage palletizing elevator 33 lowers the entire pallet 5 and the stacked materials to the docking position with the second-stage palletizing elevator 34, and places the entire pallet 5 and the stacked materials on the second-stage palletizing elevator 34. The second-stage palletizing elevator 34 then descends, carrying the pallet 5 and the stacked materials away from the first-stage palletizing elevator 33. Specifically, the first-stage palletizing elevator 33 lowers the pallet 5 to the docking position and stops. Then, the second-stage palletizing elevator 34 lifts the pallet 5 upwards, detaches it from the support surface of the first-stage palletizing elevator 33, and positions it. Once the pallet 5 is completely suspended in the air, the first-stage palletizing elevator 33 retracts its forks on both sides, completing the separation from the pallet 5. After separation, the second-stage palletizing elevator 34 lowers the pallet 5 smoothly and sends it to the next process.

[0077] Reference Figures 4 to 9 In one embodiment, the palletizing device 3 further includes a pallet lifter 35, a pallet buffer mechanism, and a wrapping mechanism 36. The first-stage pallet lifter 33 and the second-stage pallet lifter 34 are located on one side of the pallet lifter 35 and are used to lift the empty pallet 5 from the low position to the pallet buffer mechanism. The pallet lifter 35, the pallet buffer mechanism, and the wrapping mechanism 36 are mounted on the frame 31. The pallet lifter 35 is configured to lift the empty pallet 5 to the pallet buffer mechanism; The film wrapping mechanism 36 is configured to wrap the film onto the material that has been stacked; In this process, after the material and pallet 5 are palletized, they are moved from the first-stage palletizing elevator 33 to the second-stage palletizing elevator 34. The film wrapping mechanism 36 wraps the film around the material, and the next empty pallet 5 is moved from the pallet buffer mechanism to the first-stage palletizing elevator 33 simultaneously.

[0078] In this embodiment, the pallet buffer mechanism is located between the outlet of the pallet elevator 35 and the first-stage palletizing elevator 33 to temporarily store empty pallets 5 in a ready state. The wrapping mechanism 36 is located on the side of the second-stage palletizing elevator 34 and is equipped with an automatic cargo height sensing device and a variable frequency speed control system. After all the materials on a pallet 5 have been stacked, the first-stage palletizing elevator 33 moves the pallet 5 and the materials together to the second-stage palletizing elevator 34. At the same time, the wrapping mechanism 36 starts and wraps the materials while the second-stage palletizing elevator 34 is descending, realizing the parallel operation of wrapping while descending. The wrapping mechanism 36 automatically adjusts the wrapping height range according to the material height sensed by the automatic cargo height sensing device to ensure complete wrapping coverage. As the full pallet 5 moves from the first-stage palletizing elevator 33 to the second-stage palletizing elevator 34, the pallet buffer mechanism pushes the next empty pallet 5, which has been pre-lifted to the ready position by the pallet elevator 35, into the first-stage palletizing elevator 33. The robot arm 32 can then immediately begin the palletizing operation of the new pallet 5 without waiting for the empty pallet 5 to arrive. While the robot arm 32 is stacking the last few items on the current pallet 5, the pallet elevator 35 starts lifting the next empty pallet 5 to the pallet buffer mechanism, ensuring that the next empty pallet 5 is ready when the full pallet 5 is removed. After the wrapping is completed, the second-stage palletizing elevator 34 descends to the elevation where it connects with the ground conveyor line. The full pallet 5 flows directly to the downstream conveyor line, where the downstream equipment completes the outbound transfer.

[0079] In one embodiment, the palletizing device 3 further includes a pallet conveyor belt 38 and a palletizing material conveyor belt 39. The pallet conveyor belt 38 is used to transport the pallet 5 to the pallet lifter 35, and the palletizing material conveyor belt 39 is used to transport the material after it has been palletized and wrapped by the wrapping mechanism 36.

[0080] In this embodiment, the pallet conveyor belt 38 and the palletizing material conveyor belt 39 are set up, and the operation sequence of each mechanism is uniformly scheduled by the processing control unit 1. This enables the parallel advancement of the entire process of removing full pallets 5, replenishing empty pallets 5, wrapping operations, and palletizing new pallets 5, minimizing the waiting interval between each link and improving the overall palletizing operation efficiency. When the system is running, each mechanism performs corresponding actions synchronously according to the preset sequence, which not only ensures the integrity of the wrapping packaging, but also prevents the palletizing operation of the robot arm 32 from being interrupted due to the replenishment of empty pallets 5 or the transfer of full pallets 5, ultimately achieving continuous and stable mixed palletizing operations.

[0081] With a dual-arm configuration (two palletizing robots stacking 20 pieces on the left and 20 pieces on the right respectively), the complete operation cycle time for a single pallet of 5 is approximately 140 seconds, the equipment efficiency is approximately 25 pallets / hour, and the corresponding material handling capacity is approximately 1000 pieces / hour (based on 540 pieces per pallet).

[0082] In one embodiment, the palletizing device 3 further includes a lifting mechanism 371, which is used to lift the material at the palletizing pick-up position 37 upwards and remove it from the palletizing pick-up position 37.

[0083] In this embodiment, the lifting mechanism 371 is used to cooperate with the robot arm 32. When the robot arm 32 is ready to pick up the material from the palletizing pick-up position 37, the lifting mechanism 371 lifts the material located at the palletizing pick-up position 37 upwards to facilitate the robot arm 32 to pick it up.

[0084] Example 2 The difference between this embodiment and Embodiment 1 is that the system is configured in single-arm mode, using one palletizing robot to handle 530 pieces of material on a single pallet. The order contains mixed incoming materials of 3 SKUs (with significant size differences).

[0085] In this embodiment, the three SKUs have significant size differences. The largest material (780mm long x 440mm wide x 180mm high) should be placed at the bottom of the pallet first. However, the arrival order of upstream materials is not fixed due to the influence of the conveying path. The processing control unit 1 reverses the feeding sequence based on the stacking planning results, and arranges large-sized and heavy materials at the beginning of the feeding sequence. The material sorting output mechanism 2 buffers the incoming materials to the corresponding track through the branching route and releases them according to the calculated sequence, ensuring that large materials arrive at the palletizing pick-up position 37 first. Regarding the correction of the material position, the 3D vision inspection device detects that the long material (780mm) is 90 degrees horizontal before the material arrives at the picking position (the actual long side is perpendicular to the conveying direction, while the target posture requires the long side to be parallel to the conveying direction). The processing control unit 1 calculates the correction angle as +90 degrees. The robot arm 32 performs rotation compensation in the movement stroke after grabbing the material, so that the material is accurately placed in the target coordinate in the correct direction, with a deviation from the planned coordinate not exceeding + / -5mm. The control logic for synchronous descent of pallet 5, parallel wrapping, and continuous supply of empty pallet 5 is the same as in Example 1. With a single-arm configuration of 30 pieces / pallet, the cycle time for a single pallet 5 is approximately 110 seconds, and the equipment efficiency is approximately 32.7 pallets / hour.

[0086] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A hybrid palletizing method, the method comprising: Receive order data and generate a palletizing plan, which includes the material outbound sequence, material sorting sequence and pallet palletizing position; Based on the generated palletizing data, material outbound and sorting are performed; Materials are conveyed to the palletizing robot; Empty pallets are transported to the robot; The robot's robotic arm stacks materials onto the pallet according to the pallet's stacking position; After palletizing, the material is wrapped with film; The tray containing the wrapped material is removed from the robot; The feature is that the robotic arm stacks materials onto the pallet, comprising: When the robotic arm picks up the material from the next layer and places it onto the pallet, the pallet descends synchronously. When the material is wrapped with film after the palletizing is completed, the robotic arm simultaneously palletizes the material to the next empty pallet; This includes replenishing empty pallets to the pallet buffer area in advance, and when the material is wrapped after palletizing, the empty pallets are simultaneously moved from the pallet buffer area to the material palletizing area.

2. The hybrid palletizing method according to claim 1, characterized in that, When the robotic arm picks up the material from the next layer and places it onto the pallet, the robotic arm moves horizontally, and the pallet descends synchronously to a height greater than or equal to the height of the material from the next layer.

3. The hybrid palletizing method according to claim 1, characterized in that, When the robotic arm stacks materials onto the pallet according to the pallet stacking position, the pallet is located on the first stacking section elevator; When the material is wrapped with film after the palletizing is completed, the pallet is moved to the second-stage palletizing elevator; The first-stage palletizing elevator is located above the second-stage palletizing elevator; The pallet elevator replenishes empty pallets to the pallet buffer area in advance. When the material is wrapped after the palletizing is completed, the material and pallet after the palletizing are moved to the second-stage palletizing elevator. The next empty pallet is simultaneously moved to the first-stage palletizing elevator located in the material palletizing area. The palletizing stage 1 elevator and the palletizing stage 2 elevator are located on one side of the pallet elevator, and the palletizing stage 1 elevator is located above the palletizing stage 2 elevator.

4. The hybrid palletizing method according to claim 1, characterized in that, The process of generating palletizing data and performing material outbound and sorting includes: performing material outbound according to the outbound order and conveying it to the sorting step; and performing material sorting according to the material sorting order. The material sorting process includes temporarily storing materials in order to the material buffer line according to the pallet pallet position. Once all materials on a pallet have arrived and the sorting is completed, the materials are moved sequentially to the material conveyor line and then transported to the palletizing robot.

5. The hybrid palletizing method according to claim 1, characterized in that, The process of receiving order data and generating a palletizing scheme includes: S1. Obtain the attribute information, pallet parameters, robot operation parameters, and business rule information of the set of boxes to be palletized; S2. Based on the attribute information, pallet parameters, robot operation parameters, and business rule information, establish a physical model of the box and a palletizing constraint model; S3. Use an integer scheme to perform local combination optimization on one or more boxes to be stacked, and construct a MiniBin; S4. Calculate the external dimensions, total weight, center of gravity, bottom support features, placement orientation, load-bearing features, and gripping features of each MiniBin. S5. Use a layer-by-layer stacking strategy to determine the current layer to be placed, and establish the current layer's placed state, the current layer's remaining space state, and the unplaced MiniBin state. S6. Input the current layer placement state, the current layer remaining space state, and the unplaced MiniBin state into the reinforcement learning decision model, and output a four-degree-of-freedom action decision. S7. Based on the four-degree-of-freedom action decision, candidate placements are made for the target MiniBin, and constraint verification is performed on the candidate placement results. S8. If the constraint verification passes, the target MiniBin is placed and the palletizing status is updated; if the constraint verification fails, the four-degree-of-freedom action decision is masked, corrected, or re-determined. S9. When the current layer meets the layer switching condition, switch to the next layer to continue executing the placement decision until all palletized boxes are placed or the termination condition is met, and output the final palletizing scheme.

6. The hybrid palletizing method according to claim 5, characterized in that, The steps for establishing the physical model of the box and the palletizing constraint model include: S21. Based on the center coordinates and rotational attitude of the enclosure, establish a standardized AABB bounding box model to accurately represent the area occupied by the enclosure: Where, x i y i , z i For spatial coordinates, ; : Equivalent dimensions of the box after rotation; S22. Construct a complete set of constraints C0, including boundaries, collisions, height, load, robot reachability, and stacking stability: 。 7. The hybrid palletizing method according to claim 5, characterized in that, The steps for constructing a MiniBin include: employing a box local clustering and combination strategy to construct an indivisible MiniBin basic stacking unit, defining box subsets that are mutually exclusive and completely cover the set of boxes to be stacked: Introduce 0-1 integer variables to represent box affiliation relationships: The goal is to maximize the utilization of unit space, achieving a compact assembly of cabinets: Solving for the optimal MiniBin unit set M={m k } 8. A hybrid palletizing system, characterized in that, The system includes: A processing control unit is configured to receive order data and generate a palletizing scheme, the palletizing scheme including the material outbound sequence and pallet palletizing position. A material sorting and output mechanism, configured to perform material sorting and outbound processing according to the generated palletizing scheme; A palletizing device configured to palletize materials onto the pallet according to the pallet palletizing position, wrap the palletized materials with film after palletizing, and remove the wrapped materials and the pallet. A material conveyor line configured to convey materials from the material sorting output mechanism to the palletizing device; in, The palletizing device includes a palletizing robot, a first-stage palletizing elevator, and a second-stage palletizing elevator; The robot includes a robotic arm, which is located above the first-stage palletizing elevator, which is located above the second-stage palletizing elevator. The palletizing section elevator is configured to support the pallet for the robotic arm to palletize materials. When the robotic arm picks up the next layer of materials and places it on the pallet, the palletizing section elevator performs a descent, causing the pallet to descend synchronously.

9. The hybrid palletizing system according to claim 8, characterized in that, The palletizing device also includes a pallet lifter, a pallet buffer mechanism, and a wrapping mechanism, with the first-stage pallet lifter and the second-stage pallet lifter located on one side of the pallet lifter; The pallet lifter is configured to lift empty pallets to the pallet buffer mechanism; The film wrapping mechanism is configured to wrap film onto the material that has been stacked; In this process, after the palletizing is completed, the material and pallet are moved from the first-stage palletizing elevator to the second-stage palletizing elevator, the film wrapping mechanism wraps the film around the material, and the next empty pallet is simultaneously moved from the pallet buffer mechanism to the first-stage palletizing elevator.

10. The hybrid palletizing system according to claim 8, characterized in that, The processing control unit is signal-connected to the material sorting and output mechanism, the material conveying line, and the palletizing device. The processing control unit includes a processor and a storage unit, the storage unit storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1-6.

Images