A Collaborative Task Allocation Method for Flying Box Stacking Robots Based on Digital Twins

Abstract

This invention relates to the field of flying box stacking robot technology, and particularly to a collaborative task allocation method for flying box stacking robots based on digital twins. The method includes the following steps: constructing a digital twin model based on flying box data, robot mechanism data, shelf data, and environmental data; building a stacking rule base within the digital twin model; receiving flying box stacking task instructions; generating multiple sets of candidate task allocation schemes using an improved genetic algorithm; simulating and scoring the multiple sets of candidate task allocation schemes based on the digital twin model and the stacking rule base; outputting the candidate task allocation scheme with the highest score as the optimal scheme; and sending the optimal scheme to the robot's control unit for real-time monitoring via the digital twin model. This invention eliminates the problems of heavy boxes stacked on top of light boxes and fragile boxes being stacked beyond their designated layers by constructing a multi-dimensional digital twin model, and improves the shortcomings of a single allocation logic through multi-objective optimization of the improved genetic algorithm.

Description

Technical Field

[0001] This invention relates to the field of flying box stacking robot technology, and in particular to a collaborative task allocation method for flying box stacking robots based on digital twins. Background Technology

[0002] In modern logistics warehousing, port freight, and industrial production, container stacking operations are a core link in cargo storage and transshipment, and their efficiency and safety directly determine the overall supply chain's turnover speed and operating costs. With the popularization of automation technology, stacking robots have gradually replaced traditional manual operations, becoming the mainstream equipment for completing high-frequency, high-intensity stacking tasks. However, the current task allocation and operation control system for container stacking robots still has significant technical shortcomings, making it difficult to meet the comprehensive requirements for stacking safety, stability, and efficiency in complex scenarios.

[0003] Existing task allocation logic for stacking robots generally relies on a single decision rule of distance priority, that is, prioritizing the allocation of tasks to the robot closest to the target storage location. This focuses solely on shortening the movement path of a single robot, without considering the global optimization of multi-robot collaboration. For example, when multiple robots respond to adjacent tasks simultaneously, local clustering or path conflicts can easily occur, leading to increased overall operation time and low equipment resource utilization.

[0004] A single allocation logic does not take into account the robot's actual load capacity and task characteristics. For example, assigning a heavy flying box to a robot with a weak load capacity may lead to equipment overload failure. Concentrating high-frequency light-load tasks on a single robot will cause it to run under high load continuously, while other robots are in a low-load standby state, resulting in uneven equipment wear and significant differences in robot lifespan.

[0005] Because the allocation logic does not correlate the weight attributes of the cargo boxes, heavier cargo boxes are often stacked on top of lighter ones during operation. Lighter cargo boxes, with their limited load-bearing capacity, are easily deformed or damaged under the pressure of heavier boxes, leading to damage to the contents. Simultaneously, the upward shift of the center of gravity in the stacking structure reduces overall stability, increases the probability of stack collapse, and consequently causes equipment failure or operational interruption.

[0006] For crates marked with "fragile," existing systems lack specific constraints on the number of stacking layers. During task execution, robots may stack fragile crates upwards without limit. The combined weight of the upper crates can easily exceed the compression resistance limit of the lower crates, resulting in a high rate of cargo breakage and increasing the company's cargo damage costs and customer complaint risks.

[0007] Therefore, it is necessary to design a collaborative task allocation method for flying box stacking robots based on digital twins. Summary of the Invention

[0008] To address the technical deficiencies in the background technology, this invention proposes a collaborative task allocation method for a flying box stacking robot based on digital twins, which solves the aforementioned technical problems and meets practical needs. The specific technical solution is as follows:

[0009] A collaborative task allocation method for a flying box stacking robot based on digital twins includes the following steps:

[0010] A digital twin model is constructed based on flying box data, robot mechanism data, shelf data, and environmental data. A stacking rule library containing placement constraints, shelf load-bearing constraints, and fragile box stacking layer constraints is built in the digital twin model.

[0011] Upon receiving the flying box stacking task instruction, based on the digital twin model and constrained by the stacking rule base, an improved genetic algorithm is used to generate multiple sets of candidate task allocation scheme data.

[0012] Based on the digital twin model and stacking rule base, multiple sets of candidate task allocation scheme data are simulated, verified and scored, and the candidate task allocation scheme data with the highest score is output as the optimal scheme data.

[0013] The optimal solution data is sent to the robot's control unit and monitored in real time through a digital twin model.

[0014] Furthermore, the specific steps for constructing a digital twin model are as follows:

[0015] Collect 3D data of the flying box to construct a 3D model of the flying box, bind a unique radio frequency identification tag to the 3D model of the flying box, and associate the flying box's weight parameters, outbound priority parameters, and fragility level parameters;

[0016] A robot mechanism model is constructed by collecting the robot's kinematic and load characteristics. The kinematic characteristics include the maximum lifting height and lifting speed of the lifting mechanism and the translation speed of the translation mechanism. The load characteristics include the maximum clamping force of the clamping mechanism.

[0017] Collect structural parameters, load-bearing parameters, and environmental data of stacking scenarios to construct a shelf model and an environmental model. The shelf model and environmental model collect real-time load-bearing data of the shelf and dynamic interference data in the stacking scenario.

[0018] The aforementioned 3D model of the flying box, robot mechanism model, shelf model, and environment model are integrated to form a digital twin model.

[0019] Furthermore, the following steps are included before generating multiple sets of candidate task allocation scheme data:

[0020] The clamping mechanism data of the 3D model of the flying box and the robot mechanism model are extracted as the first collision group, and the dynamic interference data of the shelf model and the environment model are extracted as the second collision group. It is stipulated that only inter-group collision detection is performed, and collision group rule data is generated.

[0021] Differentiated bounding bodies are designed for different motion stages of the robot. The differentiated bounding bodies include a fine bounding body for the robot's lifting stage and a simplified fine bounding body for the robot's translation stage. Dynamically scaled bounding bodies are designed for temporary obstacles in dynamic interference data, and hierarchical bounding body parameter data is generated.

[0022] Furthermore, the improved genetic algorithm includes:

[0023] The encoding mechanism module is used to convert the matching relationship between the flying box stacking task, the robot, and the shelf into a gene sequence that the algorithm can recognize, including an outer gene representing the robot number and an inner gene representing the corresponding robot task sequence.

[0024] The initial population construction module is used to generate task allocation schemes. For each task allocation scheme to be generated, it is first filtered through a stacking rule base.

[0025] The fitness function module contains a built-in fitness calculation formula, which is used to calculate the fitness value of a task allocation scheme.

[0026] The genetic operation module is used to iteratively optimize the initial task allocation scheme;

[0027] The iteration termination module is used to control the number of iterations in the improved genetic algorithm.

[0028] Furthermore, the specific steps for generating multiple sets of candidate task allocation scheme data using the improved genetic algorithm are as follows:

[0029] Based on the flying box stacking task instructions, retrieve the corresponding flying box data, robot data and shelf data from the digital twin model, and output the shelf coordinate system, the flying box and compliant storage location mapping table and the robot state matrix;

[0030] Based on the shelf coordinate system, the mapping table between the flying box and the compliant storage location, and the robot state matrix, a task allocation is performed for each robot through the coding mechanism module. With the stacking rule base as a constraint, a preliminary task allocation scheme is generated through the initial population construction module.

[0031] The fitness function module calculates the fitness values ​​of all preliminary task allocation schemes and sorts them by value. The top few preliminary task allocation schemes are used as parents, and the genetic operation module performs iterative optimization. When the termination condition of the iteration termination module is met, the iteration stops and the task allocation scheme with the highest fitness value is output.

[0032] Based on collision group rule data and hierarchical bounding volume parameter data, the task allocation scheme to be verified is simulated and verified using a digital twin model, and multiple sets of candidate task allocation scheme data without collision risk are output.

[0033] Furthermore, the fitness calculation formula is as follows:

[0034] ,

[0035] Where F represents the fitness value, This represents the weighting coefficient for total stacking time, where T represents the total stacking time. This represents the load balancing weighting coefficient. Indicates the standard deviation of the task volume. This represents the weighting coefficient for high-risk operations, where H represents the number of high-risk operations. This represents the penalty coefficient for violations, and C represents the number of violations.

[0036] Furthermore, the specific steps for simulation verification and scoring include:

[0037] Based on the digital twin model, normal operation scenario, dynamic interference scenario and extreme abnormal scenario were built respectively, and collision group rule data and hierarchical bounding volume parameter data were loaded.

[0038] Substitute the candidate task allocation scheme data into normal operation scenarios, dynamic interference scenarios, and extreme abnormal scenarios respectively for offline simulation testing, and output simulation data;

[0039] The simulation data is substituted into the preset evaluation index system, and the score of each candidate task allocation scheme is calculated by the weighted scoring method. The candidate task allocation scheme with the highest score is output.

[0040] Furthermore, the evaluation index system includes primary and secondary indicators. Different weights are assigned to the primary and secondary indicators using a weighted scoring method. Before outputting the candidate task allocation scheme data with the highest score, candidate task allocation schemes that do not meet the stacking rule base and have a fault tolerance rate of less than the preset value in extreme abnormal scenarios are excluded first.

[0041] Furthermore, the real-time monitoring via the digital twin model specifically includes:

[0042] Real-time collection of flying box data, robot mechanism data, shelf data, and environmental data; real-time updates of the digital twin model;

[0043] The updated digital twin model is monitored for collision risk in real time, and violations of stacking rules and equipment anomalies are detected. When an anomaly is detected, a dynamic adjustment mechanism is activated to generate an adjustment plan and send it to the robot's control unit.

[0044] Furthermore, after the adjustment scheme is executed, the data of the adjustment process is archived in the exception handling database, and the parameters of the digital twin model are optimized by using the archived data in the exception handling database.

[0045] Compared with existing technologies, the collaborative task allocation method for flying box stacking robots based on digital twins provided by this invention has the following advantages:

[0046] This invention lays a solid foundation for safety and precision in stacking operations by constructing a multi-dimensional digital twin model. By binding a unique RFID tag to the 3D model of the crate, linking it with weight, outbound priority, and fragility level parameters, and combining this with the robot's kinematic and load characteristics to construct a mechanistic model, and simultaneously collecting real-time rack load data and environmental dynamic interference data to form a digital twin, it can accurately match crate attributes, robot capabilities, and rack load-bearing capacity. This eliminates the problem of heavy crates being stacked on top of lighter crates or fragile crates being stacked beyond their capacity at the data level, reducing the risk of cargo damage and stack collapse. Simultaneously, it ensures that robot operations always remain within their capability boundaries, reducing the probability of equipment failure.

[0047] This invention replaces the traditional distance-first single logic with an improved genetic algorithm for multi-objective optimization. Combined with global scene mapping from a digital twin, it can dynamically match task allocation for multiple robots. In practical applications, this reduces local clustering and path conflicts, improves equipment resource utilization, and enhances overall stacking operation efficiency. The digital twin integrates robot mechanism data, cargo box data, and shelf data, improving the genetic algorithm to prioritize matching task characteristics with equipment capabilities and cargo attributes with shelf constraints during allocation. For example, heavy cargo boxes are automatically assigned to high-load robots, while light-load tasks are evenly distributed across multiple robots, reducing the standard deviation of equipment load, minimizing differences in equipment lifespan, and lowering maintenance costs.

[0048] The improved genetic algorithm in this invention transforms the matching relationship between tasks, robots, and shelves into gene sequences through an encoding mechanism. It then combines a stacking rule base to screen initial schemes and uses a fitness function that integrates total stacking time, load balancing, high-risk operation counts, and violation penalties for iterative optimization. This results in a task allocation scheme that balances efficiency, load balancing, and safety, preventing robot load imbalance and improving overall operational efficiency. Simultaneously, it constructs three scenarios: normal, dynamic interference, and extreme anomalies, loads collision group rules and hierarchical bounding volume parameters for simulation verification, and prioritizes the elimination of schemes with violations and insufficient fault tolerance to ensure that the output scheme is stable and feasible in complex scenarios. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating a collaborative task allocation method for a flying box stacking robot based on digital twins, as described in this invention.

[0050] Figure 2 This is a schematic diagram of the module of the improved genetic algorithm in this invention. Detailed Implementation

[0051] In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "middle," and "inner," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, it should be noted that unless otherwise explicitly specified and limited, the terms "installed," "connected," and "joined" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention through specific circumstances.

[0052] The embodiments of the present invention will be described below with reference to the accompanying drawings and related examples. The embodiments of the present invention are not limited to the following examples, and the present invention relates to the relevant necessary components in this technical field, which should be regarded as well-known technology in this technical field and can be known and mastered by those skilled in this technical field.

[0053] See Figure 1 A collaborative task allocation method for a flying box stacking robot based on digital twins includes the following steps:

[0054] Step S1: Construct a digital twin model based on flying box data, robot mechanism data, shelf data, and environmental data. In the digital twin model, construct a stacking rule library that includes placement constraints, shelf load-bearing constraints, and constraints on the number of stacking layers of fragile boxes.

[0055] The digital twin model enables interconnected linkage between sub-models through a data interaction interface. It can synchronously replicate the state, movement, and constraints of the physical scene through real-time data, achieving virtual-real mapping and two-way interaction. The stacking rule library is a set of security constraints embedded in the digital twin model, specifically designed for risk points in the flying box stacking scenario, used to verify the compliance of task allocation schemes. The placement constraints in the stacking rule library are as follows: the weight of the flying box to be stacked must be less than the weight of the flying boxes already stacked below, and the weight difference must be less than a preset value, with a default preset value of 200kg, which can be adjusted according to actual needs; the shelf load-bearing constraints are as follows: the real-time load-bearing data of a single-level shelf is less than the preset load-bearing limit, and the total weight of all flying boxes in the same level is less than the upper limit of that level; the fragile box stacking layer constraints are as follows: for flying boxes with the fragile rating, the number of stacking layers in the same shelf level must be less than the preset number of layers, and no flying boxes can be stacked on top of them. The default preset number of layers is 2, which can be adjusted according to actual needs.

[0056] Step S2: Receive the flying box stacking task instruction, and based on the digital twin model, with the stacking rule base as a constraint, use an improved genetic algorithm to generate multiple sets of candidate task allocation scheme data.

[0057] The flying crate stacking task instruction includes the task type, number of flying crates, corresponding flying crate identifiers, and task deadline. Task types include inbound, outbound, and relocation. These need to be parsed into parameters recognizable by the algorithm. The improved genetic algorithm accepts these parameters, adhering to the stacking rule base as a constraint, and optimizing for multiple objectives: minimizing total stacking time, balancing robot load, and minimizing high-risk operations. This generates multiple sets of candidate task allocation schemes. The improved genetic algorithm is an intelligent algorithm optimized for flying crate stacking task allocation scenarios based on traditional genetic algorithms. The candidate task allocation scheme data is a set of feasible task allocation schemes output by the improved genetic algorithm after iterative optimization. Each scheme includes structured data such as the binding relationship between the robot, task, and storage location, motion parameters, and estimated execution time.

[0058] Step S3: Based on the digital twin model and stacking rule base, simulate and score multiple sets of candidate task allocation scheme data, and output the candidate task allocation scheme data with the highest score as the optimal scheme data.

[0059] This step uses multi-scenario simulations with a digital twin model to ensure the output solution remains safe and efficient in complex scenarios. The optimal solution data, verified through multi-scenario simulations and weighted scoring, is selected from the candidate solutions based on its overall performance and serves as the final basis for execution by the physical robot.

[0060] Step S4: Send the optimal solution data to the robot's control unit for real-time monitoring via the digital twin model.

[0061] Before sending out the optimal solution data, this step requires parsing to generate executable instructions for the robot control unit, and then sending them to the corresponding robot control unit via the 5G communication module. The instructions include the target cargo location coordinates, motion parameters, and clamping force. The digital twin model actively monitors the data in the stacking process in real time, actively collects real-time data to update the model, and performs operations such as collision risk warning, stacking rule violation warning, and equipment anomaly warning.

[0062] This invention constructs a digital twin model integrating data from flying crates, robots, shelves, and the environment, and embeds a stacking rule library containing constraints on placement, shelf load-bearing, and fragile crate stacking. It then uses an improved genetic algorithm to generate multiple candidate task allocation schemes. After multi-scenario simulation verification and selection of the optimal scheme, execution is implemented, and real-time monitoring is achieved through the digital twin. This significantly reduces the risk of cargo damage and equipment failure in flying crate stacking operations, improves the efficiency and load balancing of multi-robot collaborative operations, enhances the adaptability of the scheme to dynamic interference and extreme abnormal scenarios, reduces manual intervention costs, and achieves a comprehensive improvement in the safety, efficiency, automation, and intelligence of flying crate stacking operations.

[0063] In one embodiment of the present invention, the specific steps for constructing a digital twin model are as follows:

[0064] Step S101: Collect 3D data of the flying box to construct a 3D model of the flying box, bind a unique radio frequency identification tag to the 3D model of the flying box, and associate the weight parameters, outbound priority parameters, and fragility level parameters of the flying box.

[0065] The 3D data of the flying box includes geometric 3D data and physical attribute data. Geometric 3D data includes the length, width, height, and material of the flying box. For irregularly shaped flying boxes, accurate contour data needs to be obtained through 3D scanning equipment. During data acquisition, for standardized flying boxes, a 3D model is directly constructed according to the design drawings using modeling software. For non-standardized flying boxes, accurate contour data needs to be obtained first through 3D scanning equipment, and then processed by software to generate a 3D model. A unique RFID tag is affixed to each physical flying box, and a virtual RFID tag is embedded in a designated location on the flying box's 3D model. The virtual tag ID is mapped one-to-one with the physical tag ID through a data interface, ensuring that the virtual model can be accurately associated with the corresponding physical flying box through the tag ID. An RFID reader reads the tag information of the physical flying box and synchronously associates the core parameters of the flying box with the virtual model. RFID tags are a non-contact automatic identification technology, consisting of tags, readers, and data transmission modules, capable of storing and transmitting the associated information of objects.

[0066] Step S102: Collect the kinematic and load characteristics of the robot to construct a robot mechanism model. The kinematic characteristics include the maximum lifting height and lifting speed of the lifting mechanism and the translation speed of the translation mechanism. The load characteristics include the maximum clamping force of the clamping mechanism.

[0067] Based on the collected kinematic and load characteristics, modeling is achieved through two steps: kinematic chain modeling and parameter calibration. Kinematic modeling uses the DH parameter method to construct the robot's kinematic chain model, decomposing the robot into lifting mechanism kinematic chains, translation mechanism kinematic chains, and clamping mechanism kinematic chains. The joint parameters of each kinematic chain are set according to the actual parameters of the physical robot. Parameter calibration corrects the model parameters through actual motion testing of the robot. Load characteristics are physical parameters describing the maximum load capacity that the robot's actuators can withstand.

[0068] Step S103: Collect the structural parameters, load-bearing parameters and environmental data of the stacking scenario of the shelf to construct the shelf model and the environmental model. The shelf model and the environmental model collect the real-time load-bearing data of the shelf and the dynamic interference data in the stacking scenario in real time.

[0069] The racking structure parameters include the total number of levels, the height of each level, the number of storage locations, and the dimensions of each storage location. Load-bearing parameters are collected by installing pressure sensors at the bottom of each level of the physical racking to collect real-time load-bearing data. A load-bearing monitoring module is embedded at the bottom of the corresponding level in the virtual racking model. The real-time load-bearing data collected by the physical sensors is written to the virtual module via an edge computing module, achieving real-time synchronization between the virtual and physical racking load-bearing data. Dynamic interference data includes temporarily parked forklifts and temporary replenishment containers. Dynamic interference data is collected using LiDAR and high-definition cameras to generate a dynamic obstacle model and update the environmental model in real time.

[0070] Step S104: Integrate the 3D model of the flying crate, the robot mechanism model, the shelf model, and the environment model to form a digital twin model. Using an edge computing module, a data interaction bus is established between the 3D model of the flying crate, the robot mechanism model, the shelf model, and the environment model to achieve state synchronization and data exchange between the sub-models. The integration is verified through virtual-real synchronization to ensure that the integrated digital twin model can accurately reproduce the full-element state of the physical stacking scene. The edge computing module is deployed locally within the stacking scene, enabling it to process the collected real-time data locally.

[0071] This invention achieves high-fidelity replication of physical stacking scenarios in virtual space by constructing a fully mapped digital twin model. Specifically, a unique RFID tag is attached to the 3D model of the flying crate to ensure accurate tracking and real-time synchronization of flying crate parameters, preventing improper stacking due to missing parameters. The robot mechanism model replicates kinematics and load characteristics based on DH parameters, ensuring a perfect match between the virtual model's movements and the physical robot's capabilities, eliminating equipment malfunctions caused by over-capacity allocation. Pressure sensors are embedded in the shelf model to achieve real-time load synchronization, and the environmental model captures dynamic interference data; both work together to reproduce the static structure and dynamic changes of the scene. The resulting multi-element collaborative mapping model lays a high-fidelity foundation for subsequent task allocation, simulation verification, and real-time monitoring, while significantly reducing manual intervention costs and improving the automation level of stacking operations.

[0072] In one embodiment of the present invention, the following steps are included before generating multiple sets of candidate task allocation scheme data:

[0073] The clamping mechanism data from the 3D model of the flying crate and the robot's mechanics model are extracted as the first collision group, while the dynamic disturbance data from the shelf model and the environment model are extracted as the second collision group. Only inter-group collision detection is performed, generating collision group rule data. The 3D model of the flying crate and the clamping mechanism data are bound together as the first collision group because they are always rigidly connected during operation and must participate in collision detection as a whole to avoid relative positional deviations between the flying crate and the clamping mechanism caused by separate detection. The shelf model and dynamic disturbance data are grouped into the second collision group because both are objects that need to be avoided but are not actively controlled during flying crate stacking, and must be uniformly used as the detected end in collision detection. The collision group rule data is a structured data set that clarifies the scope of collision detection objects and the detection logic, serving as the basis for the scope of the collision detection algorithm.

[0074] Differentiated bounding bodies are designed for different motion stages of the robot. These include a fine bounding body for the robot's lifting phase and a simplified fine bounding body for the robot's translation phase. Dynamically scaled bounding bodies are designed for temporary obstacles in dynamic interference data, generating hierarchical bounding body parameter data. Considering the motion characteristics of temporary obstacles in dynamic interference data, bounding bodies with adaptively adjustable sizes based on speed are designed to avoid missed detections or false positives due to motion inertia. When temporary obstacles are moving, they exhibit motion inertia, requiring enlargement of the bounding body to compensate for this inertia; when stationary, no enlargement is needed to avoid false positives due to excessively large bounding bodies. The hierarchical bounding body parameter data is a structured data set integrating all parameters of the fine bounding body, simplified fine bounding body, and dynamically scaled bounding body, serving as the parameter basis for the collision detection algorithm to call the bounding bodies.

[0075] This invention optimizes the collision detection system through collision group division and hierarchical bounding body design, effectively solving the problems of redundant calculations and the imbalance between accuracy and real-time performance in traditional detection methods. By dividing the system into first and second collision groups, only inter-group detection is performed, reducing unnecessary calculations and improving system response efficiency. A refined bounding body is designed for the robot's lifting phase, while a simplified bounding body is designed for the translation phase, balancing safety and real-time performance. For temporary obstacles, a dynamically scaling bounding body is designed, adaptively adjusting its size with speed to reduce the false positive and false negative rates of dynamic interference. This system provides a standardized detection foundation for verifying the collision risk of candidate solutions, ensuring the feasibility of the solutions while reducing the cost of manual parameter adjustment and improving the automation and scene adaptability of collision detection.

[0076] See Figure 2 In one embodiment of the present invention, the improved genetic algorithm includes:

[0077] The encoding mechanism module is used to convert the matching relationship between the flying crate stacking task, the robot, and the shelf into a gene sequence that the algorithm can recognize. This includes an outer gene representing the robot number and an inner gene representing the corresponding robot task sequence. The outer gene represents the unique number of the flying crate stacking robot in integer form, with each outer gene corresponding to one robot, clearly defining the executor of the task. Nested under the outer gene, it represents the task sequence of the corresponding robot in integer form. Each inner gene element is a unique identifier for the flying crate. At the same time, the target shelf location of the flying crate is implicitly contained through the pre-matching relationship between the flying crate and the shelf.

[0078] The initial population construction module is used to generate task allocation schemes. For each task allocation scheme to be generated, it is first filtered through a stacked rule base. The pre-screening through the stacked rule base reduces invalid and non-compliant individuals in the initial scheme and improves the algorithm iteration efficiency.

[0079] The fitness function module contains a built-in fitness calculation formula to calculate the fitness value of a task allocation scheme. Through the built-in fitness calculation formula, the compliance, efficiency, balance and security of each task allocation scheme are quantitatively evaluated, providing direction for iterative optimization.

[0080] The genetic operation module is used to iteratively optimize the initial task allocation scheme. Through three steps of selection operator, crossover operator and mutation operator, the initial scheme is iteratively evolved to generate a better task allocation scheme.

[0081] The iteration termination module is used to control the number of iterations in the improved genetic algorithm.

[0082] It should be noted that the specific steps for generating multiple sets of candidate task allocation scheme data using the improved genetic algorithm are as follows:

[0083] Step S201: Based on the flying box stacking task instruction, retrieve the corresponding flying box data, robot data, and shelf data from the digital twin model, and output the shelf coordinate system, the flying box-compliant storage location mapping table, and the robot state matrix. The shelf coordinate system is a three-dimensional Cartesian coordinate system established based on the physical space of the stacking scene, which serves as the spatial reference for unifying all location data. The flying box-compliant storage location mapping table is based on the stacking rule base, which filters out a list of shelf storage locations that meet the constraints for each flying box to be assigned, forming a structured data association between flying boxes and storage locations. The robot state matrix is ​​a structured table that integrates the key state parameters of the currently available flying box stacking robot, serving as the basis for the robot's ability to match tasks.

[0084] Step S202: Based on the shelf coordinate system, the mapping table of flying boxes and compliant storage locations, and the robot state matrix, tasks are assigned to each robot through the coding mechanism module. With the stacking rule base as a constraint, a preliminary task assignment scheme is generated through the initial population construction module.

[0085] Step S203: Calculate the fitness values ​​of all preliminary task allocation schemes through the fitness function module, sort them by value, and use the top few preliminary task allocation schemes as parents. Iterate and optimize them through the genetic operation module. When the termination condition of the iteration termination module is met, stop the iteration and output the task allocation scheme to be verified with the highest fitness value. The termination condition of the iteration termination module specifically includes the fluctuation of the optimal fitness value being less than the preset value for several consecutive generations, the optimal fitness value reaching the preset value, or reaching the set number of iterations.

[0086] Step S204: Based on collision group rule data and hierarchical bounding volume parameter data, the task allocation scheme to be verified is simulated and verified using a digital twin model, outputting multiple sets of candidate task allocation schemes without collision risk. The digital twin model includes a collision detection module. It retrieves the collision group rule data and hierarchical bounding volume parameter data, integrates them into the collision detection module, simulates and verifies the task allocation scheme to be verified, eliminates task allocation schemes with collision risk, and outputs multiple sets of candidate task allocation schemes without collision risk.

[0087] It should be noted that the fitness calculation formula is as follows:

[0088] ,

[0089] Wherein, F represents the fitness value, which is a quantitative indicator that measures the overall quality of a single task allocation scheme and is the core basis for the algorithm to select the best scheme and eliminate the worst scheme. The total stacking time weight coefficient represents the importance of minimizing the total stacking time in the comprehensive evaluation. The higher the weight, the more the algorithm prioritizes efficiency. T represents the total stacking time, which is the total time from the start of operation to the completion of all assigned tasks for all flying box stacking robots in a single task allocation scheme. The time taken by the last robot to complete the task is taken. This represents the load balancing weight coefficient, reflecting the importance of the robot load balancing optimization goal. The higher the weight, the more the algorithm prioritizes the even distribution of the workload among the robots. The standard deviation of the task volume is a statistical indicator that quantifies the differences in task volume distribution among all flying box stacking robots. The smaller the standard deviation, the more balanced the task volume of each robot. The high-risk operation weight coefficient reflects the importance of minimizing the number of high-risk operations as a safety objective. The higher the weight, the more preferentially the algorithm avoids dangerous operations. H represents the number of high-risk operations, which is the total number of high-risk operations performed by the flying box stacking robot in a single task allocation scheme. F represents the constraint violation penalty coefficient, which is the penalty coefficient for schemes that violate the stacking rules. The larger the coefficient, the more severe the violation and the more severe the penalty. C represents the number of violations, the total number of times a single task allocation scheme violates the constraints of the stacking rule base. Each violation counts as one violation. This formula is essentially a combination of multi-objective positive optimization terms and violation penalty terms: the first three terms transform the smaller the value, the better the indicator, into a larger value, which contributes more positively. The last term deducts the penalty for violating schemes by multiplying the penalty coefficient by the number of violations. Ultimately, it achieves synergistic optimization of efficiency, balance, safety, and compliance. The larger the F value, the better the overall performance of the task allocation scheme.

[0090] In one embodiment of the present invention, the specific steps of performing simulation verification and scoring include:

[0091] Step S301: Based on the digital twin model, construct normal operation scenario, dynamic interference scenario, and extreme anomaly scenario respectively, and load collision group rule data and hierarchical bounding volume parameter data. The normal operation scenario refers to dynamic interference of objects and simulation verification under ideal equipment conditions. The dynamic interference scenario adds high-frequency dynamic interference from the stacking scenario to the normal operation scenario. Dynamic interference includes temporary obstacle passage and replenishment flying box deployment. It is used to verify the anti-interference ability of the candidate task allocation scheme under dynamic interference environment and avoid the situation where the scheme is feasible in the ideal scenario but fails in the actual scenario. The extreme anomaly scenario simulates low-probability but high-risk extreme situations in the physical scenario. It adds low equipment power and critical anomaly of shelf load capacity to the normal operation scenario to verify the fault tolerance ability of the candidate task allocation scheme under extreme risk environment and ensure that the scheme has the ability to cope with sudden risks.

[0092] Step S302: Substitute the candidate task allocation scheme data into the normal operation scenario, dynamic interference scenario, and extreme anomaly scenario respectively for offline simulation testing, and output simulation data; collect compliance data, efficiency data, and load balancing data for each candidate task allocation scheme in the simulation verification of the normal operation scenario; collect anti-interference data, collision risk data, and timing stability data in the simulation verification of the dynamic interference scenario; and collect fault tolerance data, safety redundancy data, and scheme feasibility data in the simulation verification of the extreme anomaly scenario.

[0093] Step S303: Substitute the simulation data into the preset evaluation index system, calculate the score of each candidate task allocation scheme using the weighted scoring method, and output the candidate task allocation scheme data with the highest score.

[0094] The compliance data includes the number of times stacking rules were violated, such as heavy boxes pressing on light boxes, exceeding shelf limits, and exceeding the shelf height for fragile boxes; efficiency data includes total stacking time and average execution time per task; load balancing data includes the workload of each robot and the standard deviation of the workload; anti-interference data includes the number of task interruptions and the interruption recovery time; collision risk data includes the number of collision warnings and the collision avoidance success rate; timing stability data is the deviation rate between the total estimated time of the solution and the actual simulation time; fault tolerance data includes the time for task redistribution and the number of anomaly responses; safety redundancy data includes the maximum load capacity of the shelf and the minimum remaining battery power of the robot; and solution feasibility data is the fault tolerance rate, obtained by comparing the number of tasks that could not be executed due to anomalies with the total number of tasks.

[0095] It should be noted that the evaluation index system includes primary and secondary indicators. A weighted scoring method is used to assign different weights to the primary and secondary indicators. Before outputting the candidate task allocation scheme with the highest score, candidate task allocation schemes that do not meet the stacking rule base or have a fault tolerance rate less than a preset value in extreme abnormal scenarios are prioritized for elimination. The primary indicators include stacking rule compliance, task execution efficiency, robot load balancing, and collision and disturbance rejection safety. Secondary indicators under stacking rule compliance include the number of violations and the fault tolerance rate in extreme scenarios. Secondary indicators under task execution efficiency include total stacking time and additional time spent in abnormal scenarios. Secondary indicators under robot load balancing include the standard deviation of task load and the number of task handovers. Secondary indicators under collision and disturbance rejection safety include the number of collision warnings and the dynamic disturbance time deviation rate.

[0096] This invention constructs three scenarios—normal operation, dynamic interference, and extreme anomaly—based on a digital twin model and loads collision detection data to conduct offline simulation tests on candidate solutions. Then, it combines a multi-dimensional evaluation index system and a weighted scoring method to select the optimal solution. This can cover most of the risk states in physical stacking scenarios, reduce the interruption rate of candidate solutions in dynamic interference scenarios, improve the fault tolerance rate in extreme anomaly scenarios, and significantly improve the adaptability of the solution.

[0097] In one embodiment of the present invention, the real-time monitoring via a digital twin model specifically includes:

[0098] Step S401: Collect real-time data on the flying box, robot mechanism, shelf, and environment, and update the digital twin model in real time. For the four core elements of flying box, robot, shelf, and environment, deploy multiple types of sensors to achieve full-dimensional and high-frequency data collection, ensuring that the physical state is complete and without lag. Data transmission is carried out through the edge computing module deployed locally in the stacking scenario to achieve synchronous updates of the digital twin model.

[0099] Step S402: Perform real-time collision risk monitoring, stacking rule violation detection, and equipment anomaly detection on the updated digital twin model. When an anomaly is detected, a dynamic adjustment mechanism is activated to generate an adjustment plan and send it to the robot's control unit. Real-time collision risk monitoring calls upon the stacking scenario-specific collision group and hierarchical bounding body data, performing differentiated detection according to the robot's movement stage to avoid collision accidents. Stacking rule violation detection verifies in real time whether physical operations violate the stacking rule library constraints to prevent violations from causing cargo damage or collapse. Equipment anomaly detection monitors the operating status of the robot and the data acquisition equipment to prevent operation interruptions or safety accidents due to equipment failure. The dynamic adjustment mechanism is a closed-loop mechanism that automatically generates and executes an adjustment plan by calling an improved genetic algorithm when collision risk, stacking violation, or equipment anomaly is detected.

[0100] It should be noted that after the adjustment scheme is executed, the data from the adjustment process is archived in the anomaly handling database. The parameters of the digital twin model are optimized using the archived data in the anomaly handling database. The size parameters of the hierarchical bounding volume are optimized based on the number of collision warnings archived in the anomaly handling database, and the warning deviation is minimized using the gradient descent algorithm. The constraint thresholds of the stacking rule base are adjusted based on the number of violations.

[0101] This invention collects multi-source data on flying boxes, robot mechanisms, shelves, and the environment in real time, and updates the digital twin model through an edge computing module, reducing the discrepancy between virtual and real states and eliminating risk misjudgments caused by monitoring lag. At the same time, based on collision group rule data, hierarchical bounding bodies, and stacking rule bases, it conducts multi-dimensional detection of collision risks, stacking violations, and equipment anomalies, improving the detection rate of hidden risks. When an anomaly occurs, it automatically matches and adjusts strategies, calls an improved genetic algorithm to generate a scheme, and issues it after verification through micro-scene simulation. This significantly reduces the time for manual intervention, ensuring the safety and compliance of stacking operations while adapting to unmanned needs and improving the continuity and automation level of operations.

[0102] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A collaborative task allocation method for a flying box stacking robot based on digital twins, characterized in that, Includes the following steps: A digital twin model is constructed based on flying box data, robot mechanism data, shelf data, and environmental data. A stacking rule library containing placement constraints, shelf load-bearing constraints, and fragile box stacking layer constraints is built in the digital twin model. Upon receiving the flying box stacking task instruction, based on the digital twin model and constrained by the stacking rule base, an improved genetic algorithm is used to generate multiple sets of candidate task allocation scheme data. Based on the digital twin model and stacking rule base, multiple sets of candidate task allocation scheme data are simulated, verified and scored, and the candidate task allocation scheme data with the highest score is output as the optimal scheme data. The optimal solution data is sent to the robot's control unit and monitored in real time through a digital twin model; The specific steps for constructing a digital twin model are as follows: Collect 3D data of the flying box to construct a 3D model of the flying box, bind a unique radio frequency identification tag to the 3D model of the flying box, and associate the flying box's weight parameters, outbound priority parameters, and fragility level parameters; A robot mechanism model is constructed by collecting the robot's kinematic and load characteristics. The kinematic characteristics include the maximum lifting height and lifting speed of the lifting mechanism and the translation speed of the translation mechanism. The load characteristics include the maximum clamping force of the clamping mechanism. Collect structural parameters, load-bearing parameters, and environmental data of stacking scenarios to construct a shelf model and an environmental model. The shelf model and environmental model collect real-time load-bearing data of the shelf and dynamic interference data in the stacking scenario. The aforementioned 3D model of the flying box, robot mechanism model, shelf model, and environment model are integrated to form a digital twin model; The real-time monitoring via digital twin model specifically includes: Real-time collection of flying box data, robot mechanism data, shelf data, and environmental data; real-time updates of the digital twin model; The updated digital twin model is monitored for collision risk in real time, and violations of stacking rules and equipment anomalies are detected. When an anomaly is detected, a dynamic adjustment mechanism is activated to generate an adjustment plan and send it to the robot's control unit.

2. The collaborative task allocation method for a flying box stacking robot based on digital twins according to claim 1, characterized in that, The following steps are included before generating multiple sets of candidate task allocation scheme data: The clamping mechanism data of the 3D model of the flying box and the robot mechanism model are extracted as the first collision group, and the dynamic interference data of the shelf model and the environment model are extracted as the second collision group. It is stipulated that only inter-group collision detection is performed, and collision group rule data is generated. Differentiated bounding bodies are designed for different motion stages of the robot. The differentiated bounding bodies include a fine bounding body for the robot's lifting stage and a simplified fine bounding body for the robot's translation stage. Dynamically scaled bounding bodies are designed for temporary obstacles in dynamic interference data, and hierarchical bounding body parameter data is generated.

3. The collaborative task allocation method for a flying box stacking robot based on digital twins according to claim 2, characterized in that, The improved genetic algorithm includes: The encoding mechanism module is used to convert the matching relationship between the flying box stacking task, the robot, and the shelf into a gene sequence that the algorithm can recognize, including an outer gene representing the robot number and an inner gene representing the corresponding robot task sequence. The initial population construction module is used to generate task allocation schemes. For each task allocation scheme to be generated, it is first filtered through a stacking rule base. The fitness function module contains a built-in fitness calculation formula, which is used to calculate the fitness value of a task allocation scheme. The genetic operation module is used to iteratively optimize the initial task allocation scheme; The iteration termination module is used to control the number of iterations in the improved genetic algorithm.

4. The collaborative task allocation method for a flying box stacking robot based on digital twins according to claim 3, characterized in that, The specific steps for generating multiple sets of candidate task allocation scheme data using the improved genetic algorithm are as follows: Based on the flying box stacking task instructions, retrieve the corresponding flying box data, robot data and shelf data from the digital twin model, and output the shelf coordinate system, the flying box and compliant storage location mapping table and the robot state matrix; Based on the shelf coordinate system, the mapping table between the flying box and the compliant storage location, and the robot state matrix, a task allocation is performed for each robot through the coding mechanism module. With the stacking rule base as a constraint, a preliminary task allocation scheme is generated through the initial population construction module. The fitness function module calculates the fitness values ​​of all preliminary task allocation schemes and sorts them by value. The top few preliminary task allocation schemes are used as parents, and the genetic operation module performs iterative optimization. When the termination condition of the iteration termination module is met, the iteration stops and the task allocation scheme with the highest fitness value is output. Based on collision group rule data and hierarchical bounding volume parameter data, the task allocation scheme to be verified is simulated and verified using a digital twin model, and multiple sets of candidate task allocation scheme data without collision risk are output.

5. A collaborative task allocation method for a flying box stacking robot based on digital twins according to claim 3, characterized in that, The fitness calculation formula is as follows: ， Where F represents the fitness value, This represents the weighting coefficient for total stacking time, where T represents the total stacking time. This represents the load balancing weighting coefficient. Indicates the standard deviation of the task volume. This represents the weighting coefficient for high-risk operations, where H represents the number of high-risk operations. This represents the penalty coefficient for violations, and C represents the number of violations.

6. The collaborative task allocation method for a flying box stacking robot based on digital twins according to claim 2, characterized in that, The specific steps for simulation verification and scoring include: Based on the digital twin model, normal operation scenario, dynamic interference scenario and extreme abnormal scenario were built respectively, and collision group rule data and hierarchical bounding volume parameter data were loaded. Substitute the candidate task allocation scheme data into normal operation scenarios, dynamic interference scenarios, and extreme abnormal scenarios respectively for offline simulation testing, and output simulation data; The simulation data is substituted into the preset evaluation index system, and the score of each candidate task allocation scheme is calculated by the weighted scoring method. The candidate task allocation scheme with the highest score is output.

7. A collaborative task allocation method for a flying box stacking robot based on digital twins according to claim 6, characterized in that, The evaluation index system includes primary and secondary indicators. Different weights are assigned to the primary and secondary indicators using a weighted scoring method. Before outputting the candidate task allocation scheme data with the highest score, candidate task allocation schemes that do not meet the stacking rule base and have a fault tolerance rate of less than the preset value in extreme abnormal scenarios are excluded first.

8. The collaborative task allocation method for a flying box stacking robot based on digital twins according to claim 1, characterized in that, After the adjustment plan is completed, the data from the adjustment process will be archived in the exception handling database, and the parameters of the digital twin model will be optimized using the archived data in the exception handling database.

Images