A size-aware strip-box layer case-packing path planning system and method

Abstract

The application discloses a kind of strip box layering case packing path planning system and method based on size perception, belong to the field of automated logistics, the system includes data acquisition module, virtual space modeling module, layering layout planning module, path generation and execution module and space model updating module;The method constructs and maintains a dynamic three-dimensional grid space model by obtaining known container and article size from database;When planning for each to-be-placed article, generate all feasible candidate placement points, and use a comprehensive evaluation function or a simplified strategy to calculate the score function of each candidate point, which comprehensively considers the placement height, compactness with surrounding articles and connectivity of remaining space;Finally, select the pose with the highest score to guide the robot arm to perform placement, and update the space model in real time, significantly improving the case packing space utilization and operation efficiency, especially suitable for pharmaceutical cold chain logistics and other scenarios.

Description

Technical Field

[0001] This invention relates to the fields of automated control and logistics warehousing technology, specifically to an automated packing path planning system and method for regular rectangular items such as vaccine boxes and strip packaging boxes, which is particularly suitable for scenarios with strict requirements for space utilization and cost control, such as pharmaceutical cold chain warehousing and outbound. Background Technology

[0002] In automated outbound operations of pharmaceutical cold chain logistics, various sizes of vaccine boxes or strip packaging boxes need to be removed from storage baskets and placed into unified turnover boxes for transportation. Currently, the mainstream automated solutions and their problems are as follows: ① Random placement without palletizing: This is the most common existing technology. The robotic arm grabs items according to instructions and places them in any empty space within the storage basket in a fixed posture or performs simple stacking. Due to a lack of space planning, many irregular gaps are generated between items of different sizes, resulting in extremely low space utilization, measured at only 40%-45%. This leads to a situation where one basket is not fully filled, requiring multiple baskets, significantly increasing the procurement and turnover costs of storage baskets, as well as the energy consumption and cost of subsequent cold chain transportation. ② 3D camera-guided palletizing solution: To improve utilization, this solution introduces a 3D vision sensor to scan the space inside the basket in real time, identifying empty areas and planning placement positions through point cloud processing. While this solution can improve utilization, it suffers from inherent drawbacks such as high hardware costs (the industrial-grade 3D cameras used cost tens of thousands to hundreds of thousands of yuan), poor environmental adaptability (easily leading to recognition failures in cold storage environments with low temperatures, frost, and changing light), slow processing speed, and complex system maintenance. These drawbacks make it difficult to apply on a large scale in the cost-sensitive and environmentally complex pharmaceutical distribution sector. Summary of the Invention

[0003] This invention aims to address the problems in existing automated packing technologies, which either result in low space utilization leading to resource waste or rely on high-cost, low-reliability 3D vision perception systems. It provides a purely software algorithm-driven solution that utilizes only known item and container size data. Through virtual space modeling and intelligent planning, it guides a robotic arm to achieve high-density, multi-level, and tightly stacked placement, thereby significantly improving container space utilization and reducing overall operating costs without increasing hardware costs.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] In a first aspect, the present invention provides a path planning system for stacked strip boxes based on size awareness.

[0006] The system is deployed at automated packing workstations and communicates with the warehouse management system. It includes:

[0007] Data acquisition module: used to acquire the internal three-dimensional dimension data of the container to be packed and the packaging dimension data of the strip box to be placed. That is, it queries and obtains the internal three-dimensional dimension data of the container to be packed and the outer packaging dimension data of the strip box to be placed from the database of the warehouse management system. The dimension data are all pre-stored known data, and no real-time measurement sensors are required when the system is running.

[0008] Virtual Space Modeling Module: This module is used to construct a three-dimensional mesh space model based on the internal dimensions of the container, and to create and maintain a three-dimensional array reflecting the occupancy status of the mesh. Specifically, based on the internal dimensions of the container, its internal space is discretized into a virtual space model composed of uniform three-dimensional mesh units; a three-dimensional array Space[x][y][z] is created and maintained, where each element corresponds to a mesh unit, and its value represents the occupancy status of that unit; during initialization, all mesh units are marked as idle.

[0009] Layered layout planning module: As the core algorithm engine of the system, it is used to calculate the optimal placement pose for the strip box to be placed; the layered layout planning module further includes:

[0010] (1) Candidate position generation unit: used to traverse the three-dimensional mesh space model and generate all feasible candidate placement points. Preferably, the virtual space model is traversed, and the current strip box is simulated, considering its optional rotation attitude, usually 0° or 90°, and placed in various possible positions with the bottom left corner of the strip box aligned with the mesh; for each simulated position, boundary checks are performed, such as whether it exceeds the container range, and collision detection is performed, such as whether it conflicts with the already occupied mesh, and all feasible candidate placement points are filtered out; the information of each candidate point includes three-dimensional coordinates [x, y, z] and rotation state.

[0011] (2) Evaluation function calculation unit: used to calculate a comprehensive evaluation score for each candidate placement point; the comprehensive evaluation function used by the evaluation function calculation unit is a function that comprehensively weighs multiple optimization objectives, and includes at least:

[0012] ① Height Priority Factor [f_height]: Prioritizes placing items at the lowest possible height to lower the overall center of gravity, enhance stability, and reserve space for subsequent placement on higher levels. Its value is usually negatively correlated with the placement height z. A preferred calculation method is: f_height[P] = 1 - [z_current / Z_max], where z_current is the z-coordinate of the candidate point, and Z_max is the total number of grid layers in the container.

[0013] ② Compactness Factor [f_compactness]: Prioritizes locations with large contact areas with already placed items, including bottom support area and adjacent side areas, to increase placement stability and reduce useless gaps. It is calculated as: f_compactness[P] = [A_supported + A_adjacent] / A_total_possible, where A_supported is the effective bottom support area, A_adjacent is the adjacent area of ​​the four sides, and A_total_possible is the maximum possible contact area of ​​the current strip box.

[0014] ③ Spatial connectivity factor [f_connectivity]: Prioritizes placing items in areas that do not create isolated gaps that are difficult for subsequent items to utilize, and strives to maintain the regularity and connectivity of the remaining free space to create favorable conditions for subsequent placement. It can be further decomposed into a weighted sum of connectivity gain factor and connectivity maintenance factor.

[0015] Optimal pose determination unit: used to select the optimal placement pose based on the comprehensive evaluation score; that is, based on the score output by the evaluation function calculation unit, the point with the highest comprehensive score is selected from all candidate points as the final target placement pose of the current strip box.

[0016] Path generation and execution module: used to control the robotic arm to perform operations according to the optimal placement pose; that is, to receive the optimal placement pose and convert it into an instruction format that the robotic arm controller can recognize, such as PLACEX=125Y=85Z=32ROT=1, thereby generating a collision-free motion trajectory from the gripping point to the target placement point, and sending it to the robotic arm to perform precise gripping and placement operations.

[0017] Space model update module: used to update the occupancy status of the three-dimensional array after placement is completed; after confirming that the robotic arm has successfully completed the placement operation, it calculates the grid range occupied by the strip box according to its actual size and final placement pose, and updates the status of the corresponding area in the Space array to be occupied, so as to truly reflect the latest space occupancy status in the container for use in the next item planning.

[0018] Preferably, the virtual space modeling module discretizes the internal space of the container into uniform cubic grid cells with a side length of δ; the dimensions of the three-dimensional array are X_max×Y_max×Z_max, where X_max=floor[L_bin / δ], Y_max=floor[W_bin / δ], Z_max=floor[H_bin / δ], and L_bin, W_bin, and H_bin are the length, width, and height of the container, respectively.

[0019] Preferably, in the virtual space modeling module, the selection of the grid cell side length δ needs to balance accuracy and computational complexity, with a recommended value range of 2mm to 20mm, and 5mm being preferred. The number of grid cells is determined by dividing the container size by δ and rounding down.

[0020] Preferably, an optimal expression of the evaluation function is: Score[P]=w1·f_height[P]+w2·f_compactness[P]+w3·f_connectivity[P], where w1, w2, and w3 are configurable weight coefficients, and w1+w2+w3=1, which can be adjusted according to the characteristics of the scenario (such as large boxes as the main focus, small boxes as the main focus, stability as the priority, etc.).

[0021] Preferably, the height priority factor f_height[P] = 1 - [z_current / Z_max], where z_current is the height coordinate of the candidate point and Z_max is the total number of layers; the compactness factor f_compactness[P] is the ratio of the contact area between the strip box and the occupied grid at the candidate placement point to the maximum possible contact area; and the spatial connectivity factor f_connectivity[P] is used to evaluate the regularity of the remaining space after placement.

[0022] Preferably, when the candidate location generation unit is screening candidate placement points, it performs a support check rule: for a candidate placement point P[x0, y0, z0], it is required that for each grid cell [i, j, z0] in its corresponding bottom area, the grid cell [i, j, z0-1] directly below it must be in the state of being occupied or being the bottom surface of the container [z0=0].

[0023] Preferably, the evaluation function calculation unit can adopt a simplified evaluation strategy, namely the "lowest-leftmost-deepest priority" strategy: scanning the grid space in the order of z-coordinate from smallest to largest, then y-coordinate from smallest to largest, and finally x-coordinate from smallest to largest, and taking the first feasible position encountered as the optimal placement pose. This strategy has extremely high computational efficiency and is suitable for scenarios with extremely high real-time requirements.

[0024] Secondly, the present invention provides a size-aware strip box stacking path planning method, which is applied to the system described above.

[0025] The method includes the following steps:

[0026] S1: Obtain the internal dimensions of the container and the dimensions of the strip box to be placed;

[0027] S2: Construct and maintain a 3D mesh model representing the internal space occupancy status of the container;

[0028] S3: Plan the optimal placement pose for the current strip box, including generating candidate points, calculating evaluation scores, and selecting the optimal pose. Specifically, this includes the following steps:

[0029] S3-1: Traverse the 3D mesh model to generate all candidate placement points that satisfy the boundary constraints and do not conflict with the already occupied space;

[0030] S3-2: Calculate a comprehensive evaluation score for each candidate point. The evaluation comprehensively considers the placement height, the compactness with surrounding objects, and the impact on the connectivity of the remaining space.

[0031] S3-3: Select the candidate point with the highest comprehensive evaluation score as the optimal placement pose;

[0032] S4: Based on the optimal placement pose, generate the robotic arm motion path and control the robotic arm to perform grasping and placement operations;

[0033] S5: After placement is complete, update the space occupancy status in the three-dimensional mesh model;

[0034] S6: Repeat steps S1 to S5 until all the strip boxes have been placed.

[0035] Preferably, in step S2, the physical dimensions [L, W, H] of the strip box are converted into grid occupancy numbers [L_grid, W_grid, H_grid] according to the preset grid precision δ, where L_grid=ceil[L / δ], W_grid=ceil[W / δ], and H_grid=ceil[H / δ]. In the candidate point generation in step S3, it is necessary to ensure that for candidate points [x0, y0, z0], the boundary conditions are met: x0+L_grid≤X_max, y0+W_grid≤Y_max, z0+H_grid≤Z_max, and the collision detection condition is: for all i∈[x0, x0+L_grid-1], j∈[y0, y0+W_grid-1], k∈[z0, z0+H_grid-1], Space[i][j][k] are all free.

[0036] Preferably, in step S3, the calculation of the evaluation score adopts a comprehensive evaluation function, and its weight coefficients [w1, w2, w3] are dynamically configured according to the packing scenario. The configuration strategy includes: the general mixed size scenario is configured as [0.35, 0.40, 0.25], the large box as the main scenario is configured as [0.45, 0.40, 0.15], and the small box as the main scenario is configured as [0.25, 0.35, 0.40].

[0037] Preferably, in step S3, the optimal placement pose is directly selected by adopting the "lowest-leftmost-deepest priority" strategy, that is, linear scanning is performed in ascending order of z, y, x coordinates to select the first pose that meets the placement conditions.

[0038] Compared with the prior art, the beneficial effects of the present invention are:

[0039] (1) Extremely low hardware cost and easy to modify: It completely abandons the expensive 3D vision sensor and can achieve high-density packing by simply upgrading the software algorithm, which is particularly conducive to the low-cost and rapid modification of existing automated production lines.

[0040] (2) Significantly improved space utilization: Through precise virtual space modeling and intelligent layout algorithms, the compact arrangement and multi-layer stacking of items are achieved. Actual test data shows that the volume utilization rate of containers can be increased from about 42.9% of traditional random stacking to more than 70.6%, a relative improvement of more than 64%. The number of containers required to complete the same task can be reduced by about 40%.

[0041] (3) Strong environmental robustness and stable operation: The algorithm relies entirely on reliable database size information and is not affected by environmental factors such as on-site lighting, temperature, fog, etc. It is particularly suitable for stable operation in complex industrial environments such as low-temperature cold storage.

[0042] (4) Fast operation cycle and high efficiency: The algorithm is a pure calculation process, with an average planning time of only about 141 milliseconds, which is much lower than the single grasping and releasing cycle of the robotic arm (5-6 seconds). This avoids the acquisition waiting time in the vision solution and ensures the continuous and efficient operation of the production line.

[0043] (5) Significant economic benefits: The improvement in space utilization directly translates into a decrease in container procurement costs and cold chain transportation costs. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the system hardware architecture of Embodiment 1 of the present invention.

[0045] Figure 2 This is a data table showing the dimensions and grid occupancy of the five vaccine placement boxes in Embodiment 1 of the present invention.

[0046] Figure 3 This is a statistical table of the average data of 100 test orders in Embodiment 2 of the present invention. Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. However, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the present invention.

[0048] Example 1: Detailed binning process using a comprehensive evaluation function

[0049] Please see Figures 1 to 2 This embodiment uses automated vaccine box warehousing and packing as an application scenario to explain in detail the specific implementation method of using a comprehensive evaluation function.

[0050] In the cascaded layout planning module, the comprehensive evaluation function Score[P]=w1·f_height[P]+w2·f_compactness[P]+w3·f_connectivity[P] used by the evaluation function calculation unit has the following specific definitions and calculation methods for each factor:

[0051] 1. Height priority factor [f_height]

[0052] Definition: Used to evaluate the merits of candidate placement point P, encouraging the placement of items at lower heights to reduce the overall center of gravity, improve stability, and reserve upper-level space for subsequent stacking.

[0053] Calculation method: A normalized linear function is used. Let the height coordinate of candidate point P be z_current (grid unit), and the maximum number of grid layers in the container along the height direction be Z_max. Then:

[0054] f_height[P]=1-[z_current / Z_max]

[0055] Value range: f_height[P]∈[0,1]. When the item is placed at the bottom of the basket (z_current=0), f_height[P]=1; when placed at the top (z_current=Z_max-item height), the value approaches 0.

[0056] 2. Compactness factor [f_compactness]

[0057] Definition: Used to evaluate the degree of contact between the item at candidate placement point P and the already placed items and the inner wall of the container, encouraging a larger contact area to enhance placement stability and space utilization.

[0058] Calculation method: This factor is the standardized contact area ratio, calculated as follows:

[0059] f_compactness[P]=[A_supported+A_adjacent] / A_total_possible

[0060] A_supported (bottom support area): This refers to the total number of grid cells directly below (layer z-1) the bottom surface of the candidate item that are either "occupied" or at the bottom of the basket (z=0). It represents the direct physical support area obtained by the item, measured in the number of grid cells.

[0061] A_adjacent (side adjacency area): This refers to the total number of grids on the outer surface of the four sides of the candidate item that are immediately adjacent to it (i.e., in the x-1, x+item length, y-1, y+item width directions) and are either "occupied" or are located on the inner wall of a container. It represents the degree to which the item fits into its surroundings.

[0062] A_total_possible (maximum possible contact area): refers to the maximum theoretical total area that the bottom surface and four sides of the object can contact in the current candidate posture, i.e., [L_grid*W_grid]+2*[L_grid*H_grid]+2*[W_grid*H_grid], where L_grid, W_grid, and H_grid are the grid dimensions of the object in the current rotation state.

[0063] Value range: f_compactness[P]∈[0,1]. A higher value indicates that the item is placed more stably and compactly at that location.

[0064] 3. Spatial connectivity factor [f_connectivity]

[0065] Definition: Used to evaluate the impact of placing an item at candidate point P on the shape of the remaining free space in the container. It encourages the avoidance of creating isolated, narrow gaps that are difficult for subsequent standard items to utilize after placement, and to maintain the regularity and accessibility of the remaining space.

[0066] Calculation method: This factor is quantified by analyzing changes in the clusters of vacant spaces before and after placement. A preferred implementation includes the following steps:

[0067] Spatial Cluster Marking: Before placement, all free grids (values ​​of 0) in the three-dimensional array Space are marked using a connected component analysis algorithm (such as breadth-first search based on a three-dimensional mesh) to identify all interconnected free spatial clusters and calculate the volume (number of grids) V_before_i of each cluster.

[0068] Simulated placement: Virtually place the item into candidate point P, and obtain a temporary spatial state Space_temp.

[0069] Cluster reanalysis: The free grids in Space_temp are re-clustered to obtain a new cluster set and its volume V_after_j.

[0070] Factor calculation: f_connectivity[P] is composed of a weighted sum of two sub-factors:

[0071] Connectivity maintenance factor [f_maintain]: Encourages the retention of large idle clusters. Calculated as the ratio of the largest idle cluster volume after placement to the largest idle cluster volume before placement: f_maintain = max[V_after_j] / max[V_before_i].

[0072] Connectivity gain factor [f_gain]: Penalizes the generation of too many small fragments. It is calculated as the ratio of the total volume of all free clusters after placement to the number of clusters after placement (i.e., the average cluster volume), then normalized and compared with an ideal average volume (such as the volume of a standard item). The formula can be: f_gain=[ΣV_after_j / N_after] / V_standard, where N_after is the number of clusters after placement, and V_standard is a preset standard item mesh volume, which is set to 1 if it exceeds this value.

[0073] Finally, f_connectivity[P] = α·f_maintain + β·f_gain, where α and β are sub-weights, and α + β = 1 (e.g., α = 0.6, β = 0.4).

[0074] Value range: f_connectivity[P] is usually normalized to around [0, 1]. The higher the value, the better the quality of the remaining space after placement.

[0075] 4. Configuration strategy for weight coefficients [w1, w2, w3]

[0076] Definition: Weighting coefficients are used to adjust the relative importance of the three optimization objectives—height, compactness, and connectivity—in the final decision.

[0077] Configurable strategies: The system comes pre-loaded with multiple weight configuration templates, which can be dynamically selected based on business scenarios or fine-tuned by the user.

[0078] General mixed-size scenario: [w1, w2, w3] = [0.35, 0.40, 0.25]. This balances various factors and is suitable for mixed orders with large size differences.

[0079] Large objects dominate the scene: [w1, w2, w3] = [0.45, 0.40, 0.15]. Emphasis is placed on a low center of gravity and stability (height and compactness), as large objects have a greater impact on stability and the space occupied at the bottom layer.

[0080] The main scene consists of small items: [w1, w2, w3] = [0.25, 0.35, 0.40]. Emphasis is placed on the regularity and connectivity of the space, as there are many small items, requiring a good remaining space for subsequent placement.

[0081] Custom configuration: Users can directly input or adjust the weight values ​​through the management interface, and the system verifies w1+w2+w3=1 in real time.

[0082] In this embodiment, the internal dimensions of the seedling storage basket are set as follows: L_bin = 600mm, W_bin = 400mm, H_bin = 300mm. The grid precision is δ = 5mm.

[0083] Initialize modeling: Calculate mesh dimensions: X_max=floor[600 / 5]=120, Y_max=floor[400 / 5]=80, Z_max=floor[300 / 5]=60. Initialize all elements of the 3D array Space

[60] to 0.

[0084] Evaluation function weight configuration: using the "General Mixed Size" preset template: [w1, w2, w3] = [0.35, 0.40, 0.25].

[0085] Task sequence: Place 5 vaccine boxes. Please refer to the following for their dimensions and the number of boxes they occupy after grid conversion. Figure 2 .

[0086] (1) Placement box A:

[0087] Candidate point generation: As the first item, the only feasible position of the bottom left corner of the bottom face is [0, 0, 0], with an orientation rot=0. It is denoted as candidate point P_A.

[0088] Evaluation calculation:

[0089] f_height[P_A] = 1 - [0 / 60] = 1.0

[0090] f_compactness[P_A]: The bottom support is the basket bottom, with an area of ​​40*30=1200; of the four sides, the left and front are basket walls, and the right and rear are empty. Calculations show A_supported=1200, A_adjacent=480+640=1120 [estimated contact between the left and front walls], and A_total_possible=2*[40*30+40*16+30*16]=4640. Therefore, f_compactness≈[1200+1120] / 4640≈0.500.

[0091] f_connectivity[P_A]: Before placement, the entire basket is empty, and the maximum cluster volume V_max_before = 120 * 80 * 60 = 576,000. After placement, the maximum free cluster volume is approximately the original volume minus the volume of box A. The remaining space is regularly shaped but divided into L-shaped regions, and the maximum connected cluster volume is approximately 48,000. f_maintain = 480,000 / 576,000 ≈ 0.833, f_gain ≈ 0.90, and after weighting, f_connectivity ≈ 0.86.

[0092] Score[P_A]=0.35*1.0+0.40*0.500+0.25*0.86≈0.35+0.200+0.215=0.765.

[0093] Decision and Execution: Select P_A. The robotic arm performs placement. Update Space[0:39][0:29][0:15]=1.

[0094] (2) Placement box B:

[0095] Candidate point generation: Search at z=0 level. A significant candidate point is P_B1[0, 30, 0, rot=0], which is adjacent to the right side of box A.

[0096] Support check: The bottom of the basket is directly below the bottom surface of P_B1 [0:19, 30:49, 0], and the support is complete.

[0097] Evaluation calculation:

[0098] f_height[P_B1]=1.0

[0099] f_compactness[P_B1]: The bottom support is intact. The left side is completely adjacent to the right side of box A (area 20*10=200), and the front side is adjacent to the basket wall. Calculated, f_compactness = (400+600) / 1600 = 1000 / 1600 = 0.625.

[0100] f_connectivity[P_B1]: After placement, it forms an L-shaped occupied area together with box A, and the remaining space is still relatively regular. Estimated ≈ 0.87.

[0101] Score[P_B1]≈0.35*1.0+0.40*0.625+0.25*0.87=0.35+0.25+0.2175=0.8175.

[0102] Decision and Execution: P_B1 scores higher than other corner positions, selected. Update Space[0:19][30:49][0:9]=1.

[0103] (3) Placement box C:

[0104] Candidate point generation: Search at z=0 level. Try P_C1[40, 30, 0, rot=1] (rotated 90° to make the size 40×30×12), located to the right of box B.

[0105] Support check: The bottom of the basket is below the [40:79, 30:59, 0] surface, and the support is intact.

[0106] Evaluation calculation:

[0107] f_height[P_C1]=1.0

[0108] f_compactness[P_C1]: The bottom support is intact. The left side is completely adjacent to the right side of box B (area 20*12=240), and the front side is adjacent to the basket wall. A_total_possible=2×(40×30+40×12+30×12)=2×(1200+480+360)=2×2040=4080. Calculated, f_compactness =(1200+1680) / 4080=2880 / 4080≈0.706.

[0109] f_connectivity[P_C1]: After placement, it forms a compact U-shaped bottom layer layout with A and B, with the remaining space concentrated in the right rear area, resulting in good connectivity. Estimated value ≈ 0.88.

[0110] Score[P_C1]≈0.35*1.0+0.40*0.706+0.25*0.88=0.35+0.2824+0.22=0.8524.

[0111] Decision and Execution: P_C1 score is high and better than the non-rotational pose. Selected. Update Space[40:79][30:59][0:11]=1.

[0112] (4) Placement box D

[0113] Candidate point generation: The bottom layer is already quite full. The algorithm searches the upper space. Candidate point P_D1[16, 0, 16, rot=0] is found, located directly above box A.

[0114] Key support check: The area directly below the bottom surface of P_D1 [16:39, 0:15, 16] [16:39, 0:15, 15] is completely occupied by box A, with a support rate of 100%.

[0115] Evaluation calculation:

[0116] f_height[P_D1]=1-[16 / 60]≈0.733

[0117] f_compactness[P_D1]:

[0118] A_supported=24*16=384 [Fully supported].

[0119] A_adjacent: The left side [24*14] is close to the basket wall; the front side [16*14] is close to the basket wall; the right side is adjacent to the side of box A. A_adjacent=224+336+224=784.

[0120] A_total_possible=2×(24×16+24×14+16×14)=2×(384+336+224)=2×944=1888.

[0121] The calculated value is f_compactness = (384+784) / 1888 = 1168 / 1888 ≈ 0.619.

[0122] f_connectivity[P_D1]: When stacked on box A, it did not create new obstacles at the bottom layer and preserved a large free area on the right side of the bottom layer. The maximum free cluster size is approximately 80% of that before placement, and the average cluster size is good. Estimated f_connectivity≈0.85.

[0123] Score[P_D1]≈0.35*0.733+0.40*0.619+0.25*0.85≈0.2566+0.2476+0.2125=0.7167.

[0124] Decision and Execution: P_D1 scores highest among the upper-level positions that satisfy the support. Selected. Update Space[16:39][0:15][16:29]=1.

[0125] (5) Placement box E

[0126] Dimensions: 180×120×40mm, Grid area: 36×24×8

[0127] After successfully placing box D, the virtual space model is in the following state:

[0128] Occupied area:

[0129] Box A: Space[0:39][0:29][0:15]

[0130] Box B: Space[0:19][30:49][0:9]

[0131] Box C: Space [40:79][30:59][0:11][after rotating 90°]

[0132] Box D: Space [16:39][0:15][16:29]

[0133] Unused areas: mainly the right side and back of the basket, and the space above box C.

[0134] Step S3: Plan the optimal placement pose for box E

[0135] S3-1: Candidate Position Generation

[0136] The candidate location generation unit traverses the virtual space model, considering two orientations of box E (0° rotation and 90° rotation), and strictly enforces the physical support rule: for each grid on the bottom surface (z layer) of a candidate location, there must be a support directly below it (z-1 layer), either at the bottom of the box or in an already occupied grid. The algorithm generates multiple feasible candidate points, three of which are as follows:

[0137] Candidate point P5-1 [high-level tiling]: [80, 0, 0, rot=0]

[0138] Description: Try placing box E in the empty area on the far right of the bottom of the basket without rotating it.

[0139] Feasibility: Boundary check passed. Collision detection passed (the area is empty). Support check passed (z=0, directly supported by the bottom of the basket).

[0140] Features: Placed at the bottom, it scores highly in height, but is located on the edge, has no contact with existing items, has poor compactness, and may divide the remaining space.

[0141] Candidate point P5-2 [Middle layer side placement]: [0, 50, 12, rot=0]

[0142] Description: Try placing box E without rotating it, at a middle height to the right of box B, near the back wall of the basket.

[0143] Feasibility: Boundary check passed (y+24=74<80). Collision detection passed (no overlap with any occupied areas). Support check failed: The area directly below its bottom surface [0:35, 50:73, 12] at level z=11 is completely empty, with no continuous support extending to the bottom of the basket. Therefore, this point is filtered out and not included in the evaluation stage.

[0144] Candidate point P5-3 [overlapping on C]: [40, 30, 12, rot=1]

[0145] Description: Rotate box E by 90° (the grid size becomes 24×36×8) and place it directly above the top surface of box C.

[0146] Feasibility: Boundary check: x=40, x+23=63≤120; y=30, y+35=65≤80; z=12, z+7=19≤60. Passed.

[0147] Collision detection: The target area Space [40:63][30:65][12:19] is completely empty. It is vertically continuous but staggered with box C area [40:79][30:59][0:11] (box C to z=11, box E from z=12), with no collisions. It has no collisions with other boxes. Pass.

[0148] Key support check: Directly below the bottom surface [40:63, 30:65, 12] is [40:63, 30:65, 0:11]. This area entirely belongs to the top of box C (box C occupies up to z=11), and the support is continuous and complete. Pass.

[0149] S3-2: Evaluation Function Calculation

[0150] The evaluation function calculation unit calculates P5-1 and P5-3 (P5-2 has been filtered) using the general mixed size weights [w1=0.35, w2=0.40, w3=0.25].

[0151] For P5-1[80, 0, 0, rot=0]:

[0152] f_height[P5-1]=1-[0 / 60]=1.0. Height scores the highest.

[0153] • f_compactness [P5-1]: The bottom is supported by the basket bottom, with a large area; however, of the four sides, only the left side (x=79) may be adjacent to empty space, the right and front sides are the basket walls, and the back side is empty, so the contact area with the items already placed is almost 0. A_total_possible=2×(36×24+36×8+24×8)=2×(864+288+192)=2×1344=2688.

[0154] The calculated value is f_compactness = (864 + 480) / 2688 = 1344 / 2688 = 0.500.

[0155] f_connectivity[P5-1]: Placed in the front right corner of the basket, it cuts the remaining large continuous space (middle and rear of the basket), increasing the degree of space fragmentation. Score is low ≈0.20.

[0156] Score[P5-1]≈0.35*1.0+0.40*0.50+0.25*0.68=0.35+0.2+0.17=0.72.

[0157] For P5-3[40, 30, 12, rot=1]:

[0158] f_height[P5-3]=1-[12 / 60]=0.80. The height score is good.

[0159] f_compactness[P5-3]:

[0160] Bottom support area: 100% (completely supported by the top surface of box C).

[0161] • Adjacent area: The left side (x=39) is in close contact with the side of box C; the front side (y=29) is in close contact with the side of box C; the right side (x=64) and the back side (y=66) are currently empty or close to the basket wall. The contact area with existing items is large. A_total_possible=2×(24×36+24×8+36×8)=2×(864+192+288)=2×1344=2688.

[0162] The calculated value is f_compactness = (864+0) / 2688 = 864 / 2688 ≈ 0.321.

[0163] f_connectivity [P5-3]: After placement, it forms a stable stack with box C, occupies the "air" space above box C, does not create new obstacles on the ground layer, maintains the regularity and connectivity of other areas on the ground layer, and is beneficial for subsequent possible placement (if there are more boxes). A high score is f_connectivity≈0.94.

[0164] Score[P5-3]≈0.35*0.80+0.40*0.321+0.25*0.94=0.28+0.1284+0.235=0.6434.

[0165] S3-3: Determining the optimal pose

[0166] The optimal pose determination unit comparison score is: Score[P5-1]=0.72>Score[P5-2]=0.6434.

[0168] P5-1 is selected as the optimal placement pose of box E, i.e. [x=80, y=0, z=0, rotation=0].

[0169] In this batch of packing tasks, box E, being the last item, received a different decision from the expected one based on the comprehensive evaluation function. Although P5-3 (stacked on box C) performed well in terms of compactness and connectivity, its height score was low, and the compactness factor significantly lowered the overall score due to its zero side adjacency area. Conversely, P5-1 (laid flat at the bottom), while having lower compactness and connectivity scores, achieved a perfect height score, and its compactness factor reached 0.50 due to contact with the box walls, resulting in a higher overall score. This indicates that under the "general mixed size" weighting configuration, the height priority factor has a significant impact on the final decision. The algorithm tends to prioritize filling the bottom layer space, reserving upper layer space for possible subsequent items while maintaining the overall stability of the center of gravity.

[0170] Step S4: The path generation and execution module sends the pose command PLACE X=80 Y=0 Z=0 ROT=0 to the robotic arm, and the robotic arm accurately places box E in the right side area of ​​the bottom of the basket.

[0171] Step S5: The spatial model update module updates the status of the region Space[80:115][0:23][0:7] occupied by box E to occupied.

[0172] Example 2: Comparative Test of Random Stacking, the Comprehensive Evaluation Function Scheme of the Present Invention, and the Simplified Strategy Scheme

[0173] Please see Figure 1 and Figure 3 This embodiment aims to quantitatively verify, through systematic comparative testing, the significant advantages of the technical solution described in this invention over traditional solutions in terms of space utilization, operational efficiency, and economic benefits. The tests rigorously simulate real-world pharmaceutical cold chain warehousing and outbound scenarios.

[0174] 1. Test Objective

[0175] Verify and quantify the actual performance of the following three packing schemes:

[0176] Comparative example: Traditional random stacking scheme (no space planning).

[0177] Experimental Group 1: The optimization scheme of this invention is based on the comprehensive evaluation function.

[0178] Experimental Group 2: The present invention is a simplified scheme based on the "lowest-leftmost-deepest" strategy.

[0179] 2. Test Environment and Configuration

[0180] Hardware platform:

[0181] Industrial computer (Intel i7, 16GB RAM).

[0182] Six-axis collaborative robotic arm (repeat positioning accuracy ±0.1mm).

[0183] Standard seedling storage basket (internal dimensions: 600mm×400mm×300mm).

[0184] Software environment:

[0185] Comparative example: Selecting an idle location based on a random number generator.

[0186] Experimental Group 1 and Experimental Group 2: Deploy the system described in this invention, with a virtual mesh accuracy of δ=5mm.

[0187] Test dataset:

[0188] The inventory includes 15 real vaccine box sizes, ranging from 50×50×30mm to 250×200×100mm.

[0189] Test Orders: Using a random number generator, 100 independent test orders are generated by randomly selecting items from the item library. The total volume of items in each order is 1.5 to 2 times the volume of the target container to ensure that all scenarios require the use of multiple containers for packing. The item sequence in the orders is randomized.

[0190] 3. Testing Methods

[0191] For each test order, three different automated packing simulations were performed, and the following metrics were recorded:

[0192] Volume utilization rate: The ratio of the total volume of all items placed in a single container to the internal volume of the container.

[0193] Average number of items per container: The average number of items per container across all containers required to complete this order.

[0194] Total number of containers required to complete the order.

[0195] Average single-item planning time: The average time it takes for the system to calculate the placement pose of a single item.

[0196] Stacking stability assessment: Record the number of times "virtual collapse" or improper placement occurs during the simulation due to insufficient support.

[0197] 4. Test Results

[0198] The statistical results of 100 test orders were averaged. For core data, please refer to [link / reference]. Figure 3 As shown.

[0199] 5. Results Analysis and Discussion

[0200] Space utilization: The space utilization rates of experimental groups 1 and 2 (~70%) were significantly higher than those of the control group (42.9%), with a relative improvement of over 60%. This directly proves that the virtual space modeling and intelligent planning algorithm of this invention can effectively guide the close arrangement of items and eliminate useless gaps. Experimental group 1, due to its multi-objective optimization, performed slightly better than experimental group 2.

[0201] Loading efficiency: Doubling the loading capacity of a single container means that the number of containers required to complete the same outbound task can be reduced by about 40%. This will directly reduce the costs of container procurement, cleaning, management, and transportation.

[0202] Algorithm efficiency: The comprehensive evaluation function calculation time of Experiment Group 1 is relatively high (141ms), but it is still much lower than the single grasping and releasing cycle of the robotic arm (usually 5-6 seconds), and does not affect the overall operation cycle. The simplified strategy of Experiment Group 2 has extremely high real-time performance (<5ms), and is suitable for production lines with extremely high cycle time requirements.

[0203] Stability: The comparative example showed multiple instances of improper placement due to a lack of support checks. Both solutions in this invention incorporate physical support rules, achieving near 100% stability and ensuring the reliability of automated operations.

[0204] 6. Economic benefit verification and calculation

[0205] Based on the test results, calculations were performed using a typical scenario where a municipal-level CDC produces 800,000 boxes of vaccines annually:

[0206] Cost parameter assumptions:

[0207] Price of seedling storage basket: 80 yuan / each.

[0208] Average cost per crate for cold chain transportation: 8 yuan / trip.

[0209] The cost of upgrading the software system of this invention is 200,000 yuan (one-time investment).

[0210] Calculation process:

[0211] Traditional solution: Annual crates required = 800,000 boxes ÷ 50 boxes / crab = 16,000 crates. Annual cost = container cost [16,000 × 80 = 1.28 million] + transportation cost [16,000 × 8 = 128,000] = 1.408 million yuan.

[0212] The present invention scheme (using data from experimental group 1):

[0213] Annual basket requirement = 800,000 boxes ÷ 100 boxes / basket = 8,000 baskets.

[0214] Annual cost = Container cost [8000 × 80 = 640,000] + Transportation cost [8000 × 8 = 64,000] = 704,000 yuan.

[0215] Verification conclusion:

[0216] Annual direct cost savings: 140.8 - 70.4 = 70.4 million yuan.

[0217] Software investment payback period: 200,000 yuan ÷ 704,000 yuan / year ≈ 0.28 years (approximately 3.4 months).

[0218] Long-term benefits: The 40% reduction in the number of containers also means reduced warehouse storage space, cleaning energy consumption, and management complexity.

[0219] 7. Test Conclusion

[0220] This effectiveness verification example uses objective data to demonstrate the following through simulation testing with a large number of random orders and economic benefit analysis:

[0221] Both technical solutions proposed in this invention can increase the space utilization rate of automated packing from about 42.9% in the traditional method to about 70% without increasing any hardware costs, resulting in a significant performance improvement.

[0222] The solution of this invention has excellent stability and good real-time performance, and can meet the reliable operation requirements in complex industrial environments.

[0223] This invention can bring direct and huge economic benefits, with a short investment recovery period and extremely high commercial promotion value.

[0224] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A size-aware, stacked packing path planning system for strip boxes, characterized in that, include: The data acquisition module is used to acquire the internal three-dimensional dimension data of the container to be packed and the packaging dimension data of the strip box to be placed. The virtual space modeling module is used to construct a three-dimensional mesh space model based on the internal dimensions of the container, and to create and maintain a three-dimensional array that reflects the mesh occupancy status. The stacked layout planning module is used to calculate the optimal placement pose for the strip box to be placed, and includes: The candidate location generation unit is used to traverse the three-dimensional mesh space model and generate all feasible candidate placement points; An evaluation function calculation unit is used to calculate a comprehensive evaluation score for each of the candidate placement points; The optimal pose determination unit is used to select the optimal placement pose based on the comprehensive evaluation score. The path generation and execution module is used to control the robotic arm to perform operations based on the optimal placement posture. The spatial model update module is used to update the occupancy status of the three-dimensional array after placement is completed.

2. The system according to claim 1, characterized in that, The virtual space modeling module discretizes the internal space of the container into uniform cubic grid cells with a side length of δ, where the value of δ ranges from 2mm to 20mm; the dimensions of the three-dimensional array are X_max×Y_max×Z_max, where X_max=floor[L_bin / δ], Y_max=floor[W_bin / δ], Z_max=floor[H_bin / δ], and L_bin, W_bin, and H_bin are the length, width, and height of the container, respectively.

3. The system according to claim 1 or 2, characterized in that, The comprehensive evaluation function used by the evaluation function calculation unit integrates at least the height priority factor, compactness factor, and spatial connectivity factor; the comprehensive evaluation function is expressed as: Score[P]=w1·f_height[P]+w2·f_compactness[P]+w3·f_connectivity[P], where w1, w2, and w3 are configured weight coefficients, and w1+w2+w3=1.

4. The system according to claim 3, characterized in that, The height priority factor f_height[P] = 1 - [z_current / Z_max], where z_current is the height coordinate of the candidate point and Z_max is the total number of layers; the compactness factor f_compactness[P] is the ratio of the contact area between the strip box and the occupied grid at the candidate placement point to the maximum possible contact area; the spatial connectivity factor f_connectivity[P] is used to evaluate the regularity of the remaining space after placement.

5. The system according to claim 1, characterized in that, When the candidate location generation unit is screening candidate placement points, it performs a support check rule: for a candidate placement point P[x0, y0, z0], it requires that for each grid cell [i, j, z0] in its corresponding bottom area, the grid cell [i, j, z0-1] directly below it must be in the state of being occupied or being the bottom surface of the container [z0=0].

6. The system according to claim 1, characterized in that, The evaluation function calculation unit adopts a simplified strategy of "lowest-leftmost-deepest priority": it scans the candidate points in order of priority, from smallest to largest height coordinate z, from smallest to largest depth coordinate y, and from smallest to largest width coordinate x, and determines the first feasible candidate point encountered as the optimal placement pose.

7. A size-aware, stacked packing path planning method for strip boxes, applied to the system described in any one of claims 1-5, characterized in that, Includes the following steps: S1: Obtain the internal dimensions of the container and the dimensions of the strip box to be placed; S2: Construct and maintain a 3D mesh model representing the internal space occupancy status of the container; S3: Plan the optimal placement pose for the current strip box, including generating candidate points, calculating evaluation scores, and selecting the optimal pose; S4: Control the robotic arm to perform grasping and placement operations according to the optimal placement posture; S5: Update the space occupancy status in the three-dimensional mesh model; S6: Repeat steps S1 to S5 until all the strip boxes have been placed.

8. The method according to claim 7, characterized in that, In step S2, the physical dimensions [L, W, H] of the strip box are converted into grid occupancy numbers [L_grid, W_grid, H_grid] according to the preset grid precision δ, where L_grid=ceil[L / δ], W_grid=ceil[W / δ], and H_grid=ceil[H / δ]. In the candidate point generation in step S3, it is necessary to ensure that for candidate points [x0, y0, z0], the boundary conditions are met: x0+L_grid≤X_max, y0+W_grid≤Y_max, z0+H_grid≤Z_max, and the collision detection condition is: for all i∈[x0, x0+L_grid-1], j∈[y0, y0+W_grid-1], k∈[z0, z0+H_grid-1], Space[i][j][k] are all free.

9. The method according to claim 7 or 8, characterized in that, In step S3, the evaluation score is calculated using a comprehensive evaluation function, and its weight coefficients [w1, w2, w3] are dynamically configured according to the packing scenario. The configuration strategy includes: [0.35, 0.40, 0.25] for general mixed-size scenarios, [0.45, 0.40, 0.15] for large boxes as the main scenario, and [0.25, 0.35, 0.40] for small boxes as the main scenario.

10. The method according to claim 7, characterized in that, In step S3, the optimal placement pose is directly selected using the "lowest-leftmost-deepest priority" strategy, that is, linear scanning is performed in ascending order of z, y, x coordinates to select the first pose that meets the placement conditions.

Images