Multi-stage parallel physical simulation stability detection system and method for mixed palletizing
By constructing an adversarial physical simulation environment, combined with a GPU parallel simulation platform and a degradation factor, the efficiency and accuracy issues of stability assessment in hybrid palletizing were resolved, achieving efficient selection of palletizing candidate positions and meeting the high-speed requirements of industrial production.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING GREEN MFG IND INNOVATION RES INST
- Filing Date
- 2026-03-16
- Publication Date
- 2026-06-26
AI Technical Summary
Existing palletizing simulation algorithms struggle to accurately assess the physical stability of hybrid palletizing in dynamic environments. Traditional high-fidelity physical simulation calculations are time-consuming and cannot meet the high-cycle requirements of industrial palletizing robots. Furthermore, there are discrepancies between the real environment and the ideal simulation environment.
A multi-level parallel physical simulation stability detection system for hybrid palletizing is adopted. It constructs an adversarial physical simulation environment through an order delivery module, a detection and identification module, a stability detection module, and a palletizing decision module. It performs efficient stability detection based on a GPU parallel simulation platform, introduces a degradation factor for directional perturbation testing, and screens candidate palletizing positions.
It significantly improves the robustness and accuracy of the detection, enables high-precision dynamic stability verification to be completed in a short time, meets the high-cycle requirements of industrial applications, reduces simulation time and reduces the risk of collapse.
Smart Images
Figure CN122284362A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of physical simulation technology, specifically to a multi-level parallel physical simulation stability testing system and method for hybrid palletizing. Background Technology
[0002] The global manufacturing industry is rapidly evolving towards digitalization, networking, and intelligence, with flexible demands for diverse product types, small batches, and rapid delivery becoming the norm. In the final stages of logistics and manufacturing, the compact stacking of boxes of different sizes and weights onto the same pallet to form mixed pallets is crucial for automated outbound processing. However, production environments are characterized by frequent process changes and dynamic order fluctuations, and pallets after palletizing typically require long-distance transport via traversing vehicles. This makes it difficult for traditional packing algorithms to accurately assess the physical stability of pallets in dynamic environments, especially in industrial scenarios involving diverse product types and disordered mixed palletizing.
[0003] Specifically, existing palletizing algorithms mostly rely on static geometric rules to determine stability, ignoring the influence of inertial forces when the pallet is transferred at high speed by a traversing vehicle. This leads to the safety hazard of "seemingly stable when statically stacked, but collapsing as soon as it starts moving." Traditional high-fidelity physical simulation calculations are time-consuming and cannot meet the high cycle time requirement of industrial palletizing robots, which need to control a single decision within 5 seconds. Standard physics engine environments are usually based on overly idealized environments such as uniform friction and perfect contact surfaces, which often leads to optimistic simulation results that cannot cover the collapse risk caused by uncertain factors such as carton deformation and sensor errors in reality. Summary of the Invention
[0004] To address the technical problems of existing palletizing simulation algorithms, such as the disconnect between static planning and dynamic transportation, the contradiction between high-precision simulation and industrial production cycle time, and the discrepancy between the ideal simulation environment and the complex real environment, the present invention aims to provide a multi-level parallel physical simulation stability detection system for hybrid palletizing. The specific technical solution adopted is as follows:
[0005] The order delivery module is used to: acquire boxes to be palletized and transfer them to the conveyor line via a palletizing robot;
[0006] The detection and identification module is used to: detect boxes to be palletized on the conveyor line, determine the target box, and provide size information of the target box and the boxes in the current palletizing space.
[0007] The stability detection module is used to: analyze the size information of the target box, construct an adversarial physics simulation environment by combining the size information of the boxes in the current palletizing space, determine and feedback the stability based on the adversarial physics simulation environment, and obtain the palletizing candidate position;
[0008] The palletizing decision module is used to: evaluate candidate palletizing locations and determine the palletizing placement position;
[0009] The palletizing execution module is used to: place the target box into the palletizing position by the palletizing robot.
[0010] Preferably, the system detects boxes to be palletized on the conveyor line, identifies target boxes, and provides size information of the target box and the boxes in the current palletizing space, specifically:
[0011] A three-dimensional coordinate system is established based on the palletizing robot. The box currently being palletized on the conveyor line is defined as the target box. An array of box size information is established, including the box's length, width, height, coordinate position, and mass.
[0012] Preferably, the size information of the target box is analyzed, and an adversarial physics simulation environment is constructed in combination with the size information of the boxes in the current palletizing space. Based on the stability of the adversarial physics simulation environment, candidate palletizing positions are obtained, including:
[0013] Based on the size information of the target box and the boxes in the current palletizing space, a coarse screening is performed according to geometric rules to obtain the initial candidate positions and determine the initial candidate palletizing type;
[0014] A deterioration factor is introduced into the initial candidate stacking type, and the initial candidate stacking type is analyzed to determine the most vulnerable direction of the target box. Directional disturbance tests are carried out along the most vulnerable direction to construct an adversarial physical simulation environment.
[0015] The stability of the initial candidate positions is determined and fed back through an adversarial physical simulation environment, and the candidate positions for palletizing are then selected.
[0016] Preferably, based on the size information of the target box and the boxes in the current palletizing space, a coarse screening is performed according to geometric rules to obtain initial candidate positions and determine the initial candidate palletizing type, including:
[0017] The possible candidate locations of the target box are selected from the current palletizing space, and the contact surfaces between the possible candidate locations and the current palletizing space are analyzed to determine the effective support ratio.
[0018] Preset a screening threshold, compare the effective support ratio with the screening threshold, and select the initial stable position;
[0019] Based on the initial stable position, the center of gravity projection of the target box is detected to determine whether there is a risk of overturning. If not, the initial candidate position is obtained, and the initial candidate stacking type is determined in combination with the current stacking space.
[0020] Preferably, the contact surface between possible candidate locations and the current palletizing space is analyzed to determine the effective support ratio, specifically:
[0021] Based on the possible candidate locations and the current palletizing space, the supported areas are determined, and the corresponding areas are obtained. Combined with the size information of the target box, the effective support ratio is obtained.
[0022] Preferably, the deterioration factors include increasing the mass of the target box, applying a virtual counterweight above the suspended box in the current stacking space, and adding a protrusion at the center of the contact surface between the target box and the current stacking space.
[0023] Preferably, a deterioration factor is introduced into the initial candidate stacking type, and the initial candidate stacking type is analyzed to determine the most vulnerable direction of the target box to collapse. Directional perturbation tests are then conducted along the most vulnerable direction to construct an adversarial physical simulation environment, including:
[0024] Modify the physical properties of the target box in the initial candidate stacking type to increase its mass. Obtain the unsupported area based on the possible candidate positions and the current stacking space. Analyze the unsupported area, apply virtual counterweights to strengthen the center of gravity offset, and add protrusions to the center of the contact surface to simulate uneven contact.
[0025] The center of the unsupported area and the bottom of the target box are obtained respectively, and the direction most prone to collapse is determined by combining the results. Based on the current stacking space, a preset moving speed and acceleration are applied, and random perturbations are added to complete the construction of the adversarial physical simulation environment.
[0026] Preferably, the stability of the initial candidate positions is determined and fed back through an adversarial physical simulation environment to filter palletizing candidate positions, specifically as follows:
[0027] By monitoring the pose data of the target box in an adversarial physics simulation environment, and based on preset tilt angle and height thresholds in the vertical direction, the pose data is compared with the tilt angle and height thresholds respectively, and stable initial candidate positions are selected as palletizing candidate positions.
[0028] To address the aforementioned issues, this invention also provides a multi-level parallel physical simulation stability detection method for hybrid palletizing. This method utilizes the multi-level parallel physical simulation stability detection system for hybrid palletizing as described in any of the preceding claims to perform palletizing stability detection on the boxes to be palletized, determine candidate palletizing positions, and filter palletizing placement positions to achieve the placement of the boxes to be palletized.
[0029] The present invention has the following beneficial effects:
[0030] 1. By analyzing the size information of the target box and combining it with the size information of the boxes in the current palletizing space, an adversarial physical simulation environment is constructed. This environment includes various harsh working conditions such as mass weighting and bottom protrusion, which effectively reduces the deviation between the ideal simulation conditions and the actual physical environment. It also replaces the traditional inefficient random shaking test, significantly improving the robustness and accuracy of the detection. Furthermore, based on this adversarial physical simulation environment, stability can be efficiently determined and fed back, that is, high-precision dynamic stability verification can be completed in a short time, generating palletizing candidate positions. This not only greatly improves the detection speed, but also effectively takes into account the high cycle time requirements and high reliability requirements of industrial applications, making hybrid palletizing decision-making feasible and efficient in actual production.
[0031] 2. The multi-level parallel physical simulation stability detection method for hybrid palletizing provided by this invention has the same beneficial effects as the multi-level parallel physical simulation stability detection system for hybrid palletizing provided by this invention, and will not be described in detail here. Attached Figure Description
[0032] To more clearly illustrate the technical solutions and advantages 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 these drawings without creative effort.
[0033] Figure 1 This is a schematic block diagram of a multi-level parallel physical simulation stability detection system for hybrid palletizing, provided in one embodiment of the present invention.
[0034] Figure 2 This is a schematic diagram of the target box placed at any possible candidate position in the current palletizing space, according to an embodiment of the present invention, of a multi-level parallel physical simulation stability detection system for hybrid palletizing.
[0035] Figure 3 Figure 1 is a schematic diagram of an adversarial physical simulation environment for a multi-level parallel physical simulation stability detection system for hybrid palletizing, provided in an embodiment of the present invention. Figure 2(a) is a schematic diagram of the adversarial simulation environment in a 2D scene; Figure 3(b) is a schematic diagram of the simulation for determining the most vulnerable direction to collapse.
[0036] Figure 4 This is a schematic diagram illustrating the stability judgment of dynamic simulation in a multi-level parallel physical simulation stability detection system for hybrid palletizing, provided as an embodiment of the present invention.
[0037] Figure 5This is a schematic diagram of the execution of an adversarial physical simulation environment for a multi-level parallel physical simulation stability detection system for hybrid palletizing, provided in an embodiment of the present invention; wherein, Figure (a) is the current palletizing space; and Figure (b) is the result after dynamic simulation. Detailed Implementation
[0038] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a multi-level parallel physical simulation stability detection system and method for hybrid palletizing proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0040] The following description, in conjunction with the accompanying drawings, details the specific scheme of a multi-level parallel physical simulation stability detection system and method for hybrid palletizing provided by the present invention.
[0041] To better illustrate this, there are currently three main types of technologies for detecting the stability of mixed palletizing in industrial settings:
[0042] Firstly, there is the static judgment technique based on geometric rules. This type of method mainly assesses stability by calculating the proportion of the support area at the bottom of the box, for example, a support rate greater than 75%; or by using simple physical rules to predict changes in the center of gravity to assess stability. However, this method completely ignores the actual physical properties of the box, such as its center of gravity height, mass distribution, and the friction between layers. It cannot simulate the dynamic response of the box when subjected to external disturbances such as sudden stops or turns, making it extremely easy for the box to slip and collapse during subsequent transportation.
[0043] Secondly, simulation technology based on general-purpose physics engines. Some studies have attempted to introduce physics engines such as PyBullet (PyBullet Physics Server) or Gazebo (Gazebo Simulation Software) into the algorithm for simulation. However, this method is computationally inefficient. Each box may have dozens of candidate placement positions that need to be verified, and serial simulation often takes tens of seconds to complete one round of verification, seriously slowing down the production line cycle. Later, physical simulation steps that support parallel acceleration by graphics processing units (GPUs) such as Isaac Gym (parallel simulation platform) began to emerge, further improving the verification speed. However, the existing simulation environment settings are too idealized and lack modeling of unstructured factors in the real world, such as uneven surfaces of cardboard boxes and deformation due to moisture. This leads to false biases in the simulation results, making the boxes safe in the simulation environment but still collapsing in the actual physical environment.
[0044] Thirdly, there is the physical prediction technology based on neural networks. In recent years, the academic community has begun to try to use deep learning models to directly predict physical stability from the state information or point cloud data of stacked objects, attempting to fit physical laws by learning from massive samples, thereby replacing cumbersome dynamic calculations. Such data-driven models currently have significant shortcomings in industrial applications: due to their heavy reliance on training distribution, they are prone to prediction failure due to insufficient generalization when facing unknown categories or complex stacking types, making it difficult to adapt to the needs of flexible manufacturing; in addition, the "black box" probabilistic characteristics of neural networks lack rigorous mechanical support, failing to provide absolute safety guarantees for industrial applications, and even extremely low probability misjudgments can lead to serious accidents; moreover, such methods often ignore key details such as tiny overhangs or deformations at the contact surface, easily overlooking microscopic factors that lead to dynamic instability.
[0045] To address this, a multi-level parallel physical simulation stability detection system for hybrid palletizing is proposed. It establishes an adversarial physical simulation environment to perform stability detection on palletized boxes. Its GPU-based parallel simulation platform constructs an "adversarial" virtual physical environment that is more stringent than the real environment. In this environment, directional dynamic perturbations are applied to candidate pallet patterns, i.e., the pallet patterns formed by the possible placement of palletized boxes. It can accurately eliminate potentially unstable positions within the required time.
[0046] Please see Figure 1 The diagram illustrates a schematic block diagram of a multi-level parallel physical simulation stability detection system for hybrid palletizing, provided in the first embodiment of the present invention. The system includes:
[0047] The order delivery module is used to: acquire boxes to be palletized and transfer them to the conveyor line via a palletizing robot;
[0048] The detection and identification module is used to: detect boxes to be palletized on the conveyor line, determine the target box, and provide size information of the target box and the boxes in the current palletizing space.
[0049] The stability detection module is used to: analyze the size information of the target box, construct an adversarial physics simulation environment by combining the size information of the boxes in the current palletizing space, determine and feedback the stability based on the adversarial physics simulation environment, and obtain the palletizing candidate position;
[0050] The palletizing decision module is used to: evaluate candidate palletizing locations and determine the palletizing placement position;
[0051] The palletizing execution module is used to: place the target box into the palletizing position by the palletizing robot.
[0052] It can be explained that the order delivery module, detection and identification module, stability detection module, palletizing decision module, and palletizing execution module are integrated to form a stability detection system. In actual operation, it is deeply integrated into the palletizing decision software of industrial production. It also includes a processor, communication interface, memory, and communication bus. It can complete logical instructions and communication with each other through the communication bus to implement closed-loop stability detection based on palletizing planning, so as to ensure the stable placement of palletized boxes.
[0053] It should be added that palletizing decision-making software in industrial production refers to the planning and coordination software of the factory, namely the Manufacturing Execution System (MES), which is responsible for managing and tracking the production status of each production link in the factory, and for message transmission, information recording, etc.
[0054] The order placement and delivery module comprises two parts: order placement and delivery. In the palletizing process, boxes to be palletized are output from the warehouse according to customer needs. The boxes are then transported one by one to the palletizing robot via a conveyor line, facilitating the robot's handling of the boxes. This delivery component allows for the transfer of different boxes to the corresponding palletizing line as needed. The palletizing robot is an industrial robot system that uses automation technology to stack, organize, and place goods. It can replace traditional manual labor in performing repetitive and labor-intensive palletizing tasks. It typically consists of a robotic arm, end effectors such as grippers or claws, a control system, a vision recognition system, and a conveyor line, working together to complete the entire process from receiving, identifying, grasping, and handling goods to stacking them into a pallet according to a specified pattern.
[0055] Furthermore, in the detection and recognition module, specifically:
[0056] A three-dimensional coordinate system is established based on the palletizing robot. The box currently being palletized on the conveyor line is defined as the target box. An array of box size information is established, including the box's length, width, height, coordinate position, and mass.
[0057] The box that needs to be placed on the conveyor line for the current task is the target box.
[0058] Specifically, in industrial automated production, palletizing robots need to stack boxes of different shapes and sizes in an orderly manner. They require precise perception of the boxes' position and orientation in three-dimensional space to provide a data basis for subsequent operations. Therefore, a three-dimensional coordinate system is established based on the palletizing robot to provide a unified spatial reference frame. The length and width dimensions of the pallet corresponding to the current palletizing space are denoted as follows: and The current number of boxes in the palletizing space is For each box that has been placed, a corresponding information representation array is created, denoted as . , Indicates the first The dimensions of each box; , , They represent the first The length, width, and height of the box; , , Indicates the first The coordinates of the boxes; Indicates the first The mass of each box; where the coordinate position of the box determines the data of the lower left corner point of the box in the three-dimensional coordinate system; similarly, the size information of the target box is determined, denoted as... , Indicates the target box.
[0059] Furthermore, the stability detection module includes:
[0060] Step S31: Based on the size information of the target box and the boxes in the current stacking space, perform coarse screening according to geometric rules to obtain the initial candidate positions and determine the initial candidate stacking type.
[0061] It is explained that a coarse screening, or preliminary filtering, is performed using geometric rules with extremely low computational cost to eliminate obviously infeasible solutions and reduce the load on subsequent simulation calculations.
[0062] Further, step S31 includes:
[0063] Step S311: Select possible candidate locations for the target box from the current palletizing space, analyze the contact surfaces between the possible candidate locations and the current palletizing space, and determine the effective support ratio.
[0064] The following explanation is provided: Based on the current palletizing space, and considering factors such as the physical constraints of the palletizing space, the geometric characteristics of the target box, the layout of already stacked boxes, and stacking rules, the possible placement positions of the target box in the current palletizing space are given, i.e., possible candidate positions, denoted as [missing information]. , , Indicates the first One possible candidate position This represents the number of possible candidate locations, and thus the size information corresponding to each possible candidate location of the target box is... To ensure that the results of subsequent decisions are safe and executable, each possible candidate position is screened to determine whether it is conducive to the stability of the subsequent stack formation.
[0065] Please see Figure 2 Furthermore, in step S311, the contact surfaces between possible candidate locations and the current palletizing space are analyzed to determine the effective support ratio, specifically:
[0066] Based on the possible candidate locations and the current palletizing space, the supported areas are determined, and the corresponding areas are obtained. Combined with the size information of the target box, the effective support ratio is obtained.
[0067] Specifically, in this embodiment, based on the first The possible candidate locations are described in detail to determine the target box at the [number]th [location]. The contact polygon area between the bottom surface of the item placed at one of the possible candidate positions and the already stacked surface in the current stacking space, i.e., the contact surface, is denoted as . It is the convex hull formed by the multiple small circles in the diagram, and its geometric center is determined accordingly. and area Based on the length and width of the target box, calculate the effective support ratio, i.e. Similarly, determine the effective support ratio for all possible candidate positions.
[0068] Step S312: Set a preset screening threshold, compare the effective support ratio with the screening threshold, and screen the initial stable position.
[0069] Specifically, based on the effective support ratio in step S311, the basic support rules are executed. When the effective support ratio of the target box is less than the screening threshold, it is directly regarded as an unstable position and excluded. Conversely, if the effective support ratio of the target box is greater than or equal to the screening threshold, it indicates that the current candidate position has a certain degree of stability and is retained and recorded as the initial stable position. In this embodiment, the screening threshold is 40%, which can be adjusted according to the actual situation. In particular, in the actual palletizing process, when the target box is large, for example, a box whose width is greater than half the length of the pallet needs to be set with a higher effective support ratio, i.e., 70%, to ensure that the target boxes of different sizes can be screened accordingly.
[0070] Step S313: Based on the initial stable position, perform center of gravity projection detection on the target box to determine whether there is a risk of overturning. If not, obtain the initial candidate position and determine the initial candidate stacking type in combination with the current stacking space.
[0071] Specifically, a centroid projection detection is performed on the initial stable position, the projection point of the geometric centroid of the target box on the horizontal plane is analyzed, and this projection point is compared with the contact surface in step S311. By comparison, if the projected point falls outside the contact surface, it indicates that the target box is at risk of tipping over when stationary, and is therefore eliminated. Conversely, if the projected point falls within the contact surface, it indicates that the current initial stable position is relatively stable, and this position is used as the initial candidate position for subsequent analysis. The initial candidate stacking type is then determined in conjunction with the current stacking space. , , Indicates the first The stack type corresponding to the initial stable position, that is, assuming the aforementioned analysis of the first stable position. After a preliminary screening, the possible candidate locations were determined to be the initial stable locations, and there was no risk of overturning. This indicates the number of locations in the initial stable position where there is no risk of overturning.
[0072] Please see Figure 3 Step S32: Introduce a deterioration factor into the initial candidate stack type, analyze the initial candidate stack type to determine the most vulnerable direction of the target box, conduct directional disturbance tests along the most vulnerable direction, and construct an adversarial physical simulation environment.
[0073] It is explained that in traditional palletizing simulation environments, the simulation is overly idealized. Therefore, a degradation factor is introduced to simulate the worst-case working conditions in reality, ensuring that the simulated location has sufficient safety redundancy in reality.
[0074] Furthermore, the deterioration factors include increasing the mass of the target box, applying a virtual counterweight above the suspended box in the current palletizing space, and adding a protrusion at the center of the contact surface between the target box and the current palletizing space.
[0075] Understandably, in the process of constructing an adversarial physical simulation environment, the initial candidate stacking types determined through coarse screening will be completed. The virtual boxes are loaded into the physical simulation environment in parallel. Based on the actual size, position and mass of the boxes, corresponding cuboid models are generated. In the preset physical simulation environment, the virtual boxes will be affected by gravity, horizontal movement and other factors and shake. After a certain physical disturbance, it is judged whether the stack still maintains the initial state, so as to judge the stability of the position. Therefore, in order to solve the problem that the simulation is less likely to collapse than reality, adversarial simulation is adopted, which artificially worsens the stability conditions of the simulation environment to provide robustness guarantee.
[0076] Further, step S32 includes:
[0077] Step S321: Modify the physical properties of the target box in the initial candidate stacking type, increase the mass, obtain the unsupported area based on the possible candidate positions and the current stacking space, analyze the unsupported area, apply virtual counterweights to strengthen the center of gravity offset, and add protrusions to the center of the contact surface to simulate uneven contact.
[0078] Specifically, firstly, the physical properties of the target box are modified, that is, the simulated mass of the target box is increased by a factor of the actual physical mass, so as to increase the inertial force and overturning moment of the target box during the movement; preferably, in this embodiment, the simulated mass is set to twice the actual physical mass, which can be adjusted according to the actual situation.
[0079] Next, the center of gravity offset is strengthened. A preset offset threshold is used to offset the center of gravity of target boxes whose effective support ratio is less than the threshold. For target boxes that meet the constraints, the unsupported area in the current stacking space is calculated, i.e., the area where the current stacking space does not directly contact the target box, denoted as... The corresponding geometric center is denoted as An additional virtual top counterweight is placed on the top of the target box in the unsupported area. This counterweight has the same density as the target box, and its dimensions are as follows: The counterweight is positioned at the furthest point from the center of the target box in the unsupported area to offset the center of gravity of the overall structure and increase its imbalance.
[0080] Preferably, the offset threshold is set accordingly for different sizes of the target boxes. In this embodiment, the offset threshold for large-sized target boxes is 0.8, and the offset threshold for small-sized target boxes is 0.75. These can be adjusted according to the actual situation. A larger offset threshold means that more initial candidate positions will be subjected to deterioration and instability processing.
[0081] Then, the uneven contact is simulated, and the geometric center position is determined based on the bottom support surface of the target box. To add a tiny bottom bump, the dimensions of which are... This means reducing the actual effective contact area of the initial candidate positions in the simulation environment to simulate the unstable contact surface phenomenon caused by moisture bulging or uneven sealing of cardboard boxes in reality.
[0082] Step S322: Obtain the center of the unsupported area and the bottom of the target box respectively, and comprehensively determine the direction most prone to collapse. Apply a preset moving speed and acceleration based on the current stacking space, and add random perturbation to complete the construction of the adversarial physical simulation environment.
[0083] It is explained that after constructing the harsh working environment in the aforementioned step S321, the acceleration and deceleration behavior of the palletizing robot during transportation is simulated. Therefore, an efficient directional perturbation strategy is proposed, which can complete the simulation test in a very short time and replace the inefficient random shaking.
[0084] Specifically, the support situation at the bottom of the target box is analyzed, and the geometric center of the unsupported area is determined based on the aforementioned steps. Then, the geometric center corresponding to the bottom surface of the target box is denoted as... Determine the direction most prone to collapse; the formula for calculating the direction vector is as follows: This refers to the direction of weakest support. Subsequently, a simulation is performed on the current palletizing space, applying preset moving speeds and accelerations. This means controlling the pallets in the current palletizing space to move along the most vulnerable direction, thus moving the entire palletizing space. For example, the speed of a palletizing robot is typically between 1 m / s and 2 m / s. Assuming the pallet transfer speed of the palletizing robot in the current factory is 2 m / s, then the moving speed applied to the pallets is 2 m / s. To enhance robustness, a small-amplitude random perturbation is added to the most vulnerable direction, for example, rotating the most vulnerable direction around the Z-axis of the three-dimensional coordinate system established based on the palletizing robot by -3 to 3°; thus completing the construction of the adversarial physics simulation environment.
[0085] It can be noted that by testing only the weakest link, many unnecessary and redundant simulations are avoided, and the time required for a single physical verification can be kept at a low level. Combined with the parallel computing capabilities of Isaac Gym, the efficiency of the palletizing planning decision-making process is improved, ensuring that it can be completed within 5 seconds.
[0086] Step S33: Through an adversarial physical simulation environment, determine and provide feedback on the stability of the initial candidate positions, and filter the palletizing candidate positions.
[0087] The explanation is that an adversarial physics simulation environment is used to analyze the stability of the initial candidate positions, that is, high-precision physics simulation based on GPU parallel computing is used for fine screening, maximizing the system response speed while ensuring detection accuracy.
[0088] Please see Figure 4 Furthermore, in step S33, the specific steps are as follows:
[0089] By monitoring the pose data of the target box in an adversarial physics simulation environment, and based on preset tilt angle and height thresholds in the vertical direction, the pose data is compared with the tilt angle and height thresholds respectively, and stable initial candidate positions are selected as palletizing candidate positions.
[0090] Preferably, the tilt angle threshold and the height threshold are respectively denoted as... and It can be adjusted according to the actual situation. For example, when the size of the pallet corresponding to the current palletizing space is... At that time, the tilt angle threshold Height threshold .
[0091] Specifically, during the construction of the adversarial physics simulation environment, the pose data of the target box is monitored in real time to observe its related changes. During the simulation shaking process, if the pose data of the target box at the corresponding initial candidate position changes by more than the tilt angle threshold in the vertical direction relative to the initial state, and / or the descent distance of the target box in the vertical height exceeds the height threshold, it indicates that the target box has fallen or tilted severely, and the current initial candidate position is determined to be unstable and is excluded. Conversely, if the pose data of the target box does not exceed the tilt angle threshold and / or the height threshold, it indicates that the initial candidate position is stable, and the stable initial candidate position is used as the palletizing candidate position. Similarly, all palletizing candidate positions are determined from the initial candidate positions.
[0092] The palletizing decision module evaluates candidate palletizing positions to determine a compact and space-efficient location as the target box's placement position, placing the target box in the most suitable position within the current palletizing space. Finally, in the palletizing execution module, the palletizing robot is controlled to perform the actual box picking and placement based on the palletizing placement position.
[0093] Understandably, by analyzing the size information of the target box and combining it with the size information of the boxes in the current palletizing space, an adversarial physical simulation environment is constructed. This environment includes various harsh working conditions such as mass weighting and bottom protrusion, effectively reducing the deviation between ideal simulation conditions and the actual physical environment. It also replaces the traditional inefficient random shaking test, significantly improving the robustness and accuracy of the detection. Furthermore, based on this adversarial physical simulation environment, stability can be efficiently determined and fed back, that is, high-precision dynamic stability verification can be completed in a short time, generating palletizing candidate positions. This not only greatly improves the detection speed but also effectively takes into account the high cycle time requirements and high reliability requirements of industrial applications, making hybrid palletizing decision-making feasible and efficient in actual production.
[0094] Please see Figure 5 To better illustrate this, an adversarial physical simulation environment is proposed in a multi-level parallel physical simulation stability detection system for hybrid palletizing to verify whether the stability of the stack formed after the boxes are stacked is reliable. The current stacking space is the initial state of the stack, and the result of the dynamic simulation is the output of the adversarial physical simulation environment. Each stack moves in its own direction without interfering with each other, and the stability judgment of multiple positions can be completed simultaneously.
[0095] It can be noted that the adversarial physical simulation environment proposed in this invention reduces the time overhead for each box from over 20 seconds to within 5 seconds compared to traditional physical simulation environments, and reduces the collapse rate from over 30% to 1%. Furthermore, it has been successfully deployed and is operating stably for a long time in the smart logistics warehouses of China Post Group and the smart logistics park of Deli Group. The entire multi-level parallel physical simulation stability detection system for mixed palletizing can handle the disordered mixed palletizing of hundreds of types of express delivery boxes, supports a stacking cycle of 10 seconds per piece, and has cumulatively supported the efficient delivery of over 1 million items.
[0096] Therefore, compared with existing physical simulation methods, the system proposed in this invention significantly improves the robustness and safety of the system. By introducing dynamic simulation, the risk of "static stability but collapse during transportation" is effectively avoided. Furthermore, based on the construction of an adversarial physical simulation environment, unstable factors are artificially amplified, providing sufficient safety redundancy for uncontrollable factors such as carton deformation, dimensional errors, and sensor noise that may occur in real production, ensuring an extremely high success rate when actually implemented.
[0097] Secondly, it breaks through the efficiency bottleneck of physical simulation in industrial real-time control. By utilizing the GPU parallel simulation environment (Isaac Gym) to simultaneously process the verification of hundreds of candidate positions, random perturbations are applied to the most vulnerable direction to implement a directional perturbation strategy, which avoids meaningless random testing, significantly compresses the computation time, and enables the entire process decision, including complex physical verification, to be stable within 5 seconds, fully meeting the high cycle time requirements of automated production lines.
[0098] Furthermore, it achieves accurate dynamic prediction of complex mixed stacking patterns. In the entire adversarial physics simulation environment, it not only considers the stability of individual boxes, but also simulates the frictional coupling between the target box and the existing stacking pattern below, i.e., the current stacking space. It uncovers potential sliding risks that cannot be detected by geometric rules alone, and is particularly suitable for logistics scenarios involving highly dynamic transportation such as sudden stops and starts of lateral moving vehicles. It eliminates the corresponding candidate positions to ensure that the stacking candidate positions retained by the system are stable and reliable.
[0099] The second embodiment of this invention provides a multi-level parallel physical simulation stability detection method for hybrid palletizing. It applies the multi-level parallel physical simulation stability detection system for hybrid palletizing described in any of the preceding embodiments to perform palletizing stability detection on the boxes to be palletized, determine candidate palletizing positions, and filter palletizing placement positions to achieve the placement of the boxes to be palletized. In operation, it requires a multi-level parallel physical simulation stability detection system for hybrid palletizing. Therefore, integrating the method and program data or configuring different hardware to produce functions similar to those achieved by this invention falls within the protection scope of this invention. This method has the same beneficial effects as the aforementioned multi-level parallel physical simulation stability detection system for hybrid palletizing, and will not be elaborated upon here.
[0100] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0101] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
Claims
1. A multi-level parallel physical simulation stability testing system for hybrid palletizing, characterized in that, The system includes: The order delivery module is used to: acquire boxes to be palletized and transfer them to the conveyor line via a palletizing robot; The detection and identification module is used to: detect boxes to be palletized on the conveyor line, determine the target box, and provide size information of the target box and the boxes in the current palletizing space. The stability detection module is used to: analyze the size information of the target box, construct an adversarial physics simulation environment by combining the size information of the boxes in the current palletizing space, determine and feedback the stability based on the adversarial physics simulation environment, and obtain the palletizing candidate position; The palletizing decision module is used to: evaluate candidate palletizing locations and determine the palletizing placement position; The palletizing execution module is used to: place the target box into the palletizing position by the palletizing robot.
2. The multi-level parallel physical simulation stability detection system for hybrid palletizing according to claim 1, characterized in that, The system detects boxes to be palletized on the conveyor line, identifies target boxes, and provides dimensional information about the target box and the boxes currently in the palletizing space. Specifically: A three-dimensional coordinate system is established based on the palletizing robot. The box currently being palletized on the conveyor line is defined as the target box. An array of box size information is established, including the box's length, width, height, coordinate position, and mass.
3. The multi-level parallel physical simulation stability detection system for hybrid palletizing according to claim 1, characterized in that, Analyze the size information of the target box, and construct an adversarial physics simulation environment based on the size information of the boxes in the current palletizing space. Determine and feedback stability based on the adversarial physics simulation environment to obtain candidate palletizing positions, including: Based on the size information of the target box and the boxes in the current palletizing space, a coarse screening is performed according to geometric rules to obtain the initial candidate positions and determine the initial candidate palletizing type; A deterioration factor is introduced into the initial candidate stacking type, and the initial candidate stacking type is analyzed to determine the most vulnerable direction of the target box. Directional disturbance tests are carried out along the most vulnerable direction to construct an adversarial physical simulation environment. The stability of the initial candidate positions is determined and fed back through an adversarial physical simulation environment, and the candidate positions for palletizing are then selected.
4. The multi-level parallel physical simulation stability detection system for hybrid palletizing according to claim 3, characterized in that, Based on the size information of the target box and the boxes in the current palletizing space, a coarse screening is performed according to geometric rules to obtain initial candidate positions and determine the initial candidate palletizing types, including: The possible candidate locations of the target box are selected from the current palletizing space, and the contact surfaces between the possible candidate locations and the current palletizing space are analyzed to determine the effective support ratio. Preset a screening threshold, compare the effective support ratio with the screening threshold, and select the initial stable position; Based on the initial stable position, the center of gravity projection of the target box is detected to determine whether there is a risk of overturning. If not, the initial candidate position is obtained, and the initial candidate stacking type is determined in combination with the current stacking space.
5. The multi-level parallel physical simulation stability detection system for hybrid palletizing according to claim 4, characterized in that, Analyze the contact surfaces between potential candidate locations and the current palletizing space to determine the effective support ratio, specifically: Based on the possible candidate locations and the current palletizing space, the supported areas are determined, and the corresponding areas are obtained. Combined with the size information of the target box, the effective support ratio is obtained.
6. The multi-level parallel physical simulation stability detection system for hybrid palletizing according to claim 4, characterized in that, The deterioration factors include increasing the mass of the target box, applying a virtual counterweight above the suspended box in the current palletizing space, and adding a protrusion at the center of the contact surface between the target box and the current palletizing space.
7. The multi-level parallel physical simulation stability detection system for hybrid palletizing according to claim 6, characterized in that, A deterioration factor is introduced into the initial candidate stacking type, and the initial candidate stacking type is analyzed to determine the most vulnerable direction of the target box. Directional perturbation tests are conducted along the most vulnerable direction to construct an adversarial physics simulation environment, including: Modify the physical properties of the target box in the initial candidate stacking type to increase its mass. Obtain the unsupported area based on the possible candidate positions and the current stacking space. Analyze the unsupported area, apply virtual counterweights to strengthen the center of gravity offset, and add protrusions to the center of the contact surface to simulate uneven contact. The center of the unsupported area and the bottom of the target box are obtained respectively, and the direction most prone to collapse is determined by combining the results. Based on the current stacking space, a preset moving speed and acceleration are applied, and random perturbations are added to complete the construction of the adversarial physical simulation environment.
8. The multi-level parallel physical simulation stability detection system for hybrid palletizing according to claim 3, characterized in that, Through an adversarial physics simulation environment, the stability of the initial candidate positions is determined and feedback is provided to filter palletizing candidate positions, specifically as follows: By monitoring the pose data of the target box in an adversarial physics simulation environment, and based on preset tilt angle and height thresholds in the vertical direction, the pose data is compared with the tilt angle and height thresholds respectively, and stable initial candidate positions are selected as palletizing candidate positions.
9. A multi-level parallel physical simulation stability testing method for hybrid palletizing, characterized in that, The multi-level parallel physical simulation stability detection system for hybrid palletizing as described in any one of claims 1 to 8 is used to perform palletizing stability detection on the boxes to be palletized, determine the candidate palletizing positions and filter the palletizing placement positions, so as to realize the placement of the boxes to be palletized.