Multi-robot cooperative flatting method, system and device based on optimal transmission theory

Abstract

The application discloses a kind of based on optimal transmission theory's multi-robot cooperative flat warehouse method, system and equipment, it is related to warehousing automation technical field, including: obtaining the current height distribution of material in work area and preset target height distribution;Obtain the single transport capacity parameter of each flat warehouse robot;Based on the material profit and loss state of each grid and the single transport capacity parameter of each robot, solve the optimal transmission model with the minimum total cost of material transfer as target;According to the global material transfer plan, the work path of each robot is planned, and the path task is issued;Reacquire the current height distribution of material, determine the material profit and loss state of each grid based on the current height distribution of material and the target height distribution of the reacquired material.The application makes the joint optimization of the task allocation, path selection and material distribution of multi-robot, and the grain surface shape is dynamically adapted in real time, so as to significantly improve the flatness precision and overall efficiency of flat warehouse operation.

Description

Technical Field

[0001] This invention relates to the field of warehouse automation technology, and in particular to a multi-robot collaborative warehouse leveling method, system and equipment based on optimal transport theory. Background Technology

[0002] Leveling is the process of flattening unevenly piled bulk materials in warehouses, granaries, or freight yards to a uniform height or target outline. It is a crucial step in grain storage and bulk material yard management. Traditional leveling operations mainly rely on manual labor using tools such as shovels and scrapers, or are completed by a single bulldozer or loader in conjunction with manual direction. However, manual methods are labor-intensive, inefficient, and the leveling accuracy depends entirely on the operator's experience and judgment, making it difficult to guarantee the overall uniformity of the grain surface. While single-machine operation reduces the manpower burden to some extent, it lacks a holistic perspective. Operators rely solely on visual observation of the grain surface level, making it impossible to accurately determine the material surplus or deficit in each area. They can only roughly allocate pushing tasks based on local impressions, leading to arbitrary work zoning, lengthy transfer routes, repeated pushing, and frequent omissions.

[0003] In recent years, grain leveling robots with autonomous navigation and pushing capabilities have begun to be applied in the field of warehouse automation. However, the operating strategies of existing grain leveling robots are still relatively primitive, generally lacking refined perception and quantitative analysis methods for the distribution of materials on the surface of grain piles. Robots typically only obtain the overall average height once before operation and push materials evenly according to fixed rules or preset trajectories, unable to identify the specific volume differences between surplus and shortage areas. This extensive task allocation method based on the overall average cannot determine the amount of material transferred between areas according to the actual surplus or shortage, causing the robot to repeatedly operate in areas with small surpluses while areas with large shortages are not filled in time. Task scheduling and path planning are disconnected, ultimately resulting in low leveling accuracy and low overall operational efficiency. Summary of the Invention

[0004] The main objective of this invention is to provide a multi-robot collaborative warehousing method, system, and equipment based on optimal transport theory. This aims to solve the problems of existing warehousing operations lacking precise means of quantifying material surplus and shortage, relying solely on overall height to roughly allocate tasks, resulting in unreasonable robot work zoning and task scheduling, poor warehousing effect, and low efficiency.

[0005] To achieve the above objectives, this invention proposes a multi-robot collaborative warehouse clearing method based on optimal transport theory, comprising the following steps: Step S10: Obtain the current height distribution and the preset target height distribution of materials within the work area; grid the work area and determine the material profit and loss status of each grid based on the current height distribution and the target height distribution; obtain the single transport capacity parameters of each flattening robot. Step S20: Based on the material surplus / deficit status of each grid and the single transport capacity parameters of each robot, solve the optimal transport model with the goal of minimizing the total cost of material transfer. The material transfer plan includes the amount of material transferred from surplus grids to empty grids. Step S30: Plan the operation path for each robot according to the global material transfer plan and issue the path task; Step S40: Reacquire the current height distribution of the material, determine the material surplus / deficit status of each grid based on the reacquired current height distribution and the target height distribution, and jump to step S20 until the updated material height distribution meets the preset flattening termination condition, and then end the operation.

[0006] Furthermore, obtaining the material surplus / deficit status of each grid cell specifically includes: The current material volume of each grid cell is calculated based on the current height distribution, and the target material volume of each grid cell is calculated based on the target height distribution. For each grid cell, calculate the difference between the current material volume and the target material volume; When the difference is greater than a preset positive threshold, the grid is identified as a surplus grid; when the difference is less than a preset negative threshold, the grid is identified as a vacant grid; otherwise, it is identified as a balanced grid.

[0007] Furthermore, the method for constructing the optimal transport model with the objective of minimizing the total cost of material transfer includes: By taking the surplus grid set as the source distribution and the empty grid set as the target distribution, the multi-robot collaborative warehouse clearing problem is modeled as an optimal transport problem with constraints such as robot carrying capacity and distance, and the optimization objective is to minimize the Wasserstein distance between the source distribution and the target distribution. The unit material transportation cost from the source grid to the target grid is determined by the path distance between the two grids and the single transportation capacity parameter. The global material transfer plan is obtained by solving the optimal transfer model.

[0008] Furthermore, the unit material transportation cost is obtained by dividing the path distance between two grids by the single transportation capacity parameter, or by multiplying the number of transfer trips required by the path distance between two grids, wherein the number of transfer trips is the amount of material to be transferred divided by the single transportation capacity parameter and rounded up.

[0009] Furthermore, the optimal transmission model is solved using any one of the following: the Sinkhorn entropy regularization algorithm, the distributed primal-dual algorithm based on Kantorovich duality, or a linear programming solver.

[0010] Furthermore, the algorithm used to plan the operation path for each robot is A-star algorithm, fast randomized tree (RRT) algorithm, artificial potential field method, velocity obstacle method, or model predictive control (MPC) algorithm.

[0011] Furthermore, the path task distribution includes: selecting a portion of the planned work paths according to a preset threshold for the number of paths per operation or a threshold for the duration of the operation, and using these as the path tasks to be distributed in this round.

[0012] Furthermore, the preset flattening termination condition is: traversing all grids, the absolute value of the difference between the current material height of each grid and the target height of the corresponding grid in the target height distribution is less than the preset maximum height difference threshold.

[0013] This invention also proposes a multi-robot collaborative warehouse clearing system based on optimal transport theory, comprising: The sensing module is used to collect the three-dimensional point cloud of materials within the work area and obtain the current height distribution of the materials; The control module is configured to execute a multi-robot collaborative warehouse clearing method based on optimal transport theory, in order to determine the material profit and loss status, construct and solve the optimal transport model, plan the path and issue tasks, and perform cyclic replanning and termination determination. The communication module is used to transmit data and instructions between the control module and each robot; The execution module, installed on each robot, is used to control the robot to complete the material pushing operation according to the received path task.

[0014] The present invention also proposes a multi-robot collaborative closing device based on the optimal transmission theory, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the multi-robot collaborative closing method based on the optimal transmission theory.

[0015] This invention quantifies the profit and loss of grain piles through gridding, constructs an optimal transmission model that integrates robot carrying capacity and transportation distance, solves the globally optimal material transfer plan using Wasserstein distance as a metric, and combines batch path execution and closed-loop replanning to achieve joint optimization of task allocation, path selection and material transportation of multiple robots. The grain surface shape is dynamically adapted in real time, thereby significantly improving the leveling accuracy and overall efficiency of the leveling operation. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the multi-robot collaborative warehouse clearing method based on optimal transmission theory of the present invention. Figure 2 This is a schematic diagram of the modular structure of the multi-robot collaborative warehouse leveling system based on the optimal transmission theory of the present invention.

[0019] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of the present invention and are not intended to limit the present invention.

[0021] To better understand the technical solution of the present invention, a detailed description will be provided below in conjunction with the accompanying drawings and specific embodiments.

[0022] like Figure 1 As shown, Figure 1 This is a flowchart illustrating the multi-robot collaborative warehouse clearing method based on optimal transmission theory of the present invention.

[0023] A multi-robot collaborative closing method based on optimal transport theory includes the following steps: Step S10: Obtain the current height distribution and the preset target height distribution of materials within the work area; grid the work area and determine the material profit and loss status of each grid based on the current height distribution and the target height distribution; obtain the single transport capacity parameters of each flattening robot. Step S20: Based on the material surplus / deficit status of each grid and the single transport capacity parameters of each robot, solve the optimal transport model with the goal of minimizing the total cost of material transfer. The material transfer plan includes the amount of material transferred from surplus grids to empty grids. Step S30: Plan the operation path for each robot according to the global material transfer plan and issue the path task; Step S40: Reacquire the current height distribution of the material, determine the material surplus / deficit status of each grid based on the reacquired current height distribution and the target height distribution, and jump to step S20 until the updated material height distribution meets the preset flattening termination condition, and then end the operation.

[0024] This invention quantifies the profit and loss of grain piles through gridding, constructs an optimal transmission model that integrates robot carrying capacity and transportation distance, solves the globally optimal material transfer plan using Wasserstein distance as a metric, and combines batch path execution and closed-loop replanning to achieve joint optimization of task allocation, path selection and material transportation of multiple robots. The grain surface shape is dynamically adapted in real time, thereby significantly improving the leveling accuracy and overall efficiency of the leveling operation.

[0025] In this embodiment, the specific implementation process of step S10 is as follows. First, the three-dimensional point cloud data of the material surface in the working area is collected using the sensing module mounted on each leveling robot or fixedly installed on the structure inside the warehouse. The sensing module can integrate one or more sensors such as LiDAR, depth camera, and ultrasonic sensor to perform a full-coverage scan of the grain pile surface and obtain a dense three-dimensional spatial point set describing the surface morphology of the material. The control module performs noise reduction, registration, and fusion processing on the collected three-dimensional point cloud to reconstruct a three-dimensional surface model reflecting the current material accumulation form, and then extracts the material surface height value corresponding to each plane position in the working area to form the current height distribution. At the same time, the control module reads the target flat distribution from the preset operating process parameters. This target flat distribution is usually a uniform height surface parallel to the bottom plane of the warehouse, or a slope distribution set according to a specific warehouse capacity curve, and extracts the target height distribution. Subsequently, the control module divides the entire working area into two-dimensional uniform grid units of uniform size on the horizontal projection plane. Each grid corresponds to a fixed plane area inside the warehouse, discretizing the continuous height field. For each grid obtained, the control module records its center plane coordinates and grid bottom area, and uses the average value of the current height distribution within the grid coverage area as the current material height of the grid.

[0026] When acquiring the material surplus / deficit status of each grid, the control module first generates a globally unified benchmark average height surface Havg based on a preset target flat distribution. Combining this with the bottom area of ​​each grid, it calculates the theoretically required standard grain storage volume Vavg for that grid. Simultaneously, based on the product of the current material height and the grid's bottom area, it calculates the actual grain storage volume Vi for each grid. For each grid, the control module performs a difference calculation: ΔVi = Vi - Vavg, to obtain the material volume deviation value for that grid. The system has a pre-set volume deviation threshold δv, which can be customized according to the grain flatness accuracy requirements. The control module compares the volume deviation value ΔVi of each grid cell with the volume deviation threshold δv to determine its status: when ΔVi is greater than δv, the grid cell is identified as a surplus grid cell, indicating that there is excess grain in the grid cell that needs to be transferred out; when ΔVi is less than -δv, the grid cell is identified as a vacant grid cell, indicating that the grid cell needs to transfer grain from surplus grid cells to fill the vacancy; when the absolute value of ΔVi falls within the range of δv, the grid cell is identified as a balanced grid cell, indicating that the grain volume deviation of the grid cell is within the allowable threshold range and no transfer is required. Through the above step-by-step determination, the control module finally outputs a surplus and vacancy distribution matrix covering all grid cells. This matrix clearly records the surplus / deficit type and specific material deviation volume of each grid cell, serving as discrete input prior information for subsequent optimal transport modeling. This process, which discretizes a continuous three-dimensional surface into a grid, calculates the volume deviation grid by grid with a reference mean surface, and determines the result based on a threshold, achieves precise quantification of the grain pile's surplus or shortage status. It replaces the traditional method of relying solely on the overall height for rough estimation, providing accurate discrete prior information for subsequent optimal transport modeling. Simultaneously, the control module acquires the pre-configured single-transport capacity parameter Qr of each leveling robot within the system. This parameter corresponds to the maximum rated single-transport volume Rr of each robot, determined by the geometric volume and driving performance of the robot's grain-pushing bucket.

[0027] In this embodiment, the specific implementation process of step S20 is as follows. Based on the global grid surplus and shortage distribution matrix output in step S10 and the single transport capacity parameter Qr of each robot, the control module formalizes the multi-robot collaborative warehousing problem into a constrained optimal transport problem. Specifically, the control module takes the set of all surplus grids as the source distribution, denoted as μ0, which describes the distribution of material surplus at each spatial location; and takes the set of all shortage grids as the target distribution, denoted as μT, which describes the distribution of material shortage at each spatial location. For each surplus grid x, its total amount to be transported is denoted as Mout(x), which is equal to the volume deviation value ΔVi of the grid, representing the total amount of material that the grid needs to be transported outward; for each shortage grid y, its total amount to be replenished is denoted as Min(y), which is equal to the absolute value of the volume deviation of the grid, representing the total amount of material that the grid needs to be replenished from the outside.

[0028] The control module uses minimizing the Wasserstein distance between the source distribution μ0 and the target distribution μT as the global optimization objective to construct an optimal transmission model. The mathematical expression of this model is: ; The following explains each symbol and function in the expression. "min" represents minimization, i.e., finding the transport scheme that minimizes the total cost. "π" represents a transport plan, a measure defined on Ω×Ω, where π(x,y) describes the planned volume of material transported from surplus grid x to empty grid y. "∏(μ0,μT)" represents the set of all transport plans distributed with μ0 and μT as edges, i.e., the set of all π satisfying the following conditions: for any surplus grid x, the total amount of material transported out of x must equal the surplus amount Mout(x); for any empty grid y, the total amount of material transported into y must equal the shortage amount Min(y). "∫_{Ω×Ω}" represents integrating over all possible combinations of surplus grid x and empty grid y within the entire working area Ω, corresponding to summing over all grid pairs in the case of discrete grids. "dπ(x,y)" represents integrating over the infinitesimal element of the transport plan π at (x,y). “c(x,y,R,Qr)” is the unit material transportation cost function, where R refers to the set of robots participating in the operation, and Qr is the single transportation capacity parameter of the r-th robot. This function represents the transportation cost required to transport a unit volume of material from grid x to grid y.

[0029] Combining the grid discretization of the work area in step S10, the above-described continuous-form optimal transmission problem is equivalent to a discrete summation optimization problem in this embodiment: among all transmission plans that satisfy the edge distribution, find the summation problem. The smallest π. In this discrete form, π(x,y) has a dual meaning: it represents the transport plan itself and directly gives the specific material transport volume from surplus grid x to empty grid y, while simultaneously satisfying two edge constraints: for any surplus grid x, the sum of the material transported from x to all empty grids Σ_yπ(x,y) equals the total surplus of that grid Mout(x); for any empty grid y, the sum of the material transported from all surplus grids into y Σ_xπ(x,y) equals the total shortage of that grid Min(y). This equivalent expression in the discrete form completely connects the grid discretization process with the optimal transport model, allowing the theoretical model to directly correspond to a computable numerical optimization problem.

[0030] The unit material transportation cost c(x,y,R,Qr) is determined by the path distance between two grid cells and the robot's single-trip transportation capacity parameter. In one implementation of this embodiment, c(x,y,R,Qr) is directly taken as the path distance d(x,y) between the center coordinates of the two grid cells divided by the single-trip transportation volume Qr, i.e., c=d / Qr. This form simplifies the transportation cost and transportation volume to a linear relationship, making the optimal transportation model a standard linear programming problem that can be efficiently calculated using mature solvers. In another implementation, to more accurately reflect the actual constraint that the robot must make a full round trip, let c(x,y,R,Qr)=[Nr(x,y)·d(x,y)] / π(x,y), where Nr(x,y)=ceil(π(x,y) / Qr) represents the minimum number of round trips required to transport π(x,y) of materials (rounded up). In this case, the unit cost is related to the transport volume itself, and the model becomes nonlinear. It can be solved by alternating iteration or piecewise linear approximation to more realistically simulate the stepped cost generated by the number of transport trips. Regardless of the form used, the cost function incorporates the robot's limited carrying capacity Qr and the spatial distance d(x,y) into the optimization framework. This allows the model to automatically tend to match grid pairs with closer distances for large-scale material transfer tasks during global optimization, achieving a joint optimal trade-off between the number of transport trips and the transport distance. This avoids the global efficiency loss caused by the disconnect between task allocation and path planning in traditional methods.

[0031] The control module flexibly selects one of the following algorithms to solve the constructed optimal transport model, based on the actual warehouse size and hardware computing power: the Sinkhorn entropy regularization algorithm, the distributed primordial-dual algorithm based on Kantorovich duality, or a linear programming solver. The Sinkhorn entropy regularization algorithm, by introducing an entropy regularization term, transforms the original optimal transport problem into a form that can be solved iteratively and efficiently, making it particularly suitable for scenarios with a large number of grids and high requirements for solution speed. The distributed primordial-dual algorithm based on Kantorovich duality can fully utilize the multi-core parallel computing capabilities of edge computing nodes, making it suitable for distributed computing deployments in medium to large-scale warehouses. The linear programming solver is suitable for small-scale scenarios with high accuracy requirements. Regardless of the solution algorithm used, a globally optimal transport plan, denoted as π, can eventually converge stably. This transport plan, in the form of a matrix or mapping, explicitly specifies the volume of material to be transferred from each surplus grid x to each empty grid y, i.e., the amount of material transferred from surplus grids to empty grids. Since the model incorporates the robot's carrying capacity Qr into the transportation cost function c(x,y,R,Qr) during model construction, the globally optimal transportation plan π obtained by solving has implicitly considered the impact of each robot's single transport volume on the overall scheduling scheme. This ensures the global optimality of task allocation from the source when it is subsequently decomposed into the work paths that each robot can execute.

[0032] In this embodiment, the specific implementation process of step S30 is as follows: the control module takes the global material transfer plan obtained in step S20 as the core input, and combines the planar coordinate position of each grid in the work area and the single transport capacity parameters of each robot to plan a specific driving operation path for each robot. The control module can use any of the following path planning algorithms, such as the Fast Random Exploration Tree (RRT) algorithm, the artificial potential field method, the velocity obstacle method, or the Model Predictive Control (MPC) algorithm, and is not limited to any specific algorithm. The RRT algorithm can quickly explore complex work areas with non-convex structures and obstacles by randomly sampling in the workspace and incrementally building a search tree, generating a passable trajectory connecting surplus grids and empty grids; the artificial potential field method drives the robot from the position of surplus grid to the position of the target empty grid while avoiding other robots and obstacles on the warehouse wall by constructing virtual gravitational and repulsive fields; the velocity obstacle method generates a collision-free velocity command by judging the collision cone region in the velocity space in a multi-robot dynamic scenario; the Model Predictive Control (MPC) algorithm solves the optimization problem in the finite time domain in each control cycle, taking into account both path smoothness and robot dynamic constraints. The generated set of operational paths includes the entire trajectory of each robot from its current position, to the designated surplus grid material collection point, to the corresponding empty grid unloading point after loading materials, and multiple round trips between the grids. After obtaining all operational paths, the control module does not distribute all paths to each robot at once. Instead, it selects the highest priority paths from all paths based on preset thresholds for the number of paths per operation or operation duration. This phased distribution strategy ensures the controllability of segmented operations and avoids the system executing outdated paths after grain surface deformation due to distributing too many tasks at once. After selection, the control module distributes the path tasks to the corresponding leveling robots via wireless communication methods such as Wi-Fi. Upon receiving the path task, each robot's execution module controls its walking mechanism to travel along the planned trajectory according to the received path instructions and manipulates the grain-pushing bucket to complete the loading, transfer, and unloading of materials, realizing the material transfer from surplus grids to empty grids.

[0033] In this embodiment, step S40 is implemented as follows: After each robot completes all path tasks assigned in this round, the material in the work area undergoes significant surface deformation due to local shoveling and filling. The material quantity of some grids that were originally marked as surplus or empty has changed, and some grids may even have reached a balanced state. At this time, the control module triggers a new round of material distribution perception process. The perception module re-collects three-dimensional point cloud data of the material surface in the work area, and the control module processes the new point cloud data to obtain the updated current height distribution. Based on the updated current height distribution and the preset target height distribution, the control module recalculates the current material volume, target material volume, and volume difference for each grid, and updates the material surplus / deficit status of each grid according to the same volume deviation threshold judgment rule. This re-perception and recalculation process enables the system to track the dynamic changes in the grain pile morphology in real time, avoiding incorrect scheduling decisions based on outdated grain surface information. The control module then compares the updated material height distribution with the preset leveling termination condition. The specific termination condition for leveling is as follows: Traverse all grids within the work area and determine whether the absolute value of the difference between the current material height of each grid and the target height corresponding to that grid in the target height distribution is less than a preset maximum height difference threshold. The maximum height difference threshold is an allowable deviation value preset according to the accuracy requirements of the grain leveling process. When the height deviation of all grids meets the above condition, it indicates that the overall material surface in the work area has reached the target flatness requirement. The control module determines that the leveling operation is complete, sends a stop command to each robot, and ends the entire operation process.

[0034] Furthermore, when the height deviation of at least one grid exceeds the maximum height difference threshold, it indicates that the material surface has not yet reached the flatness target. The control module will jump to step S20 based on the updated material surplus and deficit status of each grid, reconstruct the optimal transmission model, solve the new global material transfer plan, plan a new round of operation path and issue it for execution, forming a closed-loop cycle operation process. This closed-loop cycle replanning mechanism enables the system to have the ability to respond to environmental changes in real time. It can adaptively adjust the subsequent scheduling plan according to the actual grain surface shape after each round of operation, fundamentally overcoming the problem of the open-loop execution mode being unable to cope with the deformation accumulation during the operation process, which leads to a decrease in flatness accuracy.

[0035] like Figure 2 As shown, Figure 2 This is a schematic diagram of the modular structure of the multi-robot collaborative warehouse leveling system based on the optimal transmission theory of the present invention.

[0036] refer to Figure 2In this embodiment, the multi-robot collaborative warehouse leveling method based on optimal transmission theory is implemented through a multi-robot collaborative warehouse leveling system based on optimal transmission theory. This system consists of four collaborative parts: a perception module 10, a control module 20, a communication module 30, and an execution module 40. The perception module integrates at least one sensor from LiDAR, a depth camera, and an ultrasonic sensor, mounted on each robot body or installed on a fixed structure inside the warehouse. It is responsible for collecting three-dimensional point cloud data of the material surface in the working area and transmitting the raw sensing data to the control module through the communication module, allowing the control module to reconstruct the current height distribution of the material. The control module is deployed on an edge computing node and runs a computer program that stores all the steps of the above method. It sequentially executes core computational tasks such as determining the material profit and loss status, constructing and solving the optimal transmission model, path planning and task batching, and closed-loop replanning and termination determination. The communication module, built on a wireless local area network, is responsible for bidirectional data and command transmission between the control module and each robot. This includes uploading sensor data collected by the perception module to the control module, and issuing path tasks and control commands generated by the control module to the execution modules of each robot, while simultaneously synchronizing the work progress and status feedback information of each robot. The execution module, installed on each leveling robot, includes a walking mechanism and a grain-pushing bucket. After receiving the path task from the control module, it controls the walking mechanism to drive the robot along the planned trajectory to the designated location, and uses the grain-pushing bucket to perform the scooping, transfer, and leveling operations of materials, completing the transfer of materials from surplus grids to vacant grids. The perception, control, communication, and execution modules work collaboratively in the above closed-loop process, jointly realizing automated and intelligent leveling operations from three-dimensional perception of the grain surface to multi-robot task allocation and path planning.

[0037] In this embodiment, the above method can also be implemented using a multi-robot collaborative leveling device based on optimal transport theory. This device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following operations: acquiring the current height distribution of materials within the work area and a preset target height distribution; gridding the work area and determining the material surplus / deficit status of each grid; simultaneously acquiring the single-transport capacity parameters of each leveling robot; constructing an optimal transport model based on the material surplus / deficit status and single-transport capacity parameters, aiming to minimize the total cost of material transfer, and solving for the global material transfer plan; planning work paths for each robot according to the global material transfer plan and issuing path tasks; after each robot executes its path task, re-acquiring the current height distribution of materials and updating the material surplus / deficit status of each grid; determining whether a preset leveling termination condition is met; if not, returning to the optimal transport model construction step to continue the loop. This device can be integrated as an independent control unit deployed in the grain warehouse control room, or embedded in the onboard computing unit of each leveling robot to collaboratively run the above steps in a distributed manner.

[0038] Furthermore, this invention transforms grain leveling operations into a dynamic optimal transport problem. First, through grid-based discretization, the three-dimensional shape of the grain pile is mapped to precisely calculable grid surplus / deficit quantities, achieving a refined quantification of the material surplus / deficit state. Then, in the modeling phase, the robot's carrying capacity constraints and spatial transport distance are jointly incorporated into the cost function, using Wasserstein distance as a measure of distribution difference. By solving the optimal transport model, a globally optimal material transfer plan is obtained, achieving global optimization of task allocation from the source. In the path generation phase, multiple path planning algorithms are supported for flexible adaptation and are deployed in batches, ensuring the controllability and safety of the operation process. Finally, through a closed-loop cyclical replanning mechanism, the grain surface shape is re-perceived and re-modeled after each round of segmented operations, enabling the scheduling scheme to dynamically adapt to the continuous changes in the material surface in real time. These technical features work together to significantly improve the leveling accuracy and overall efficiency of grain leveling operations, making it suitable for various bulk material leveling scenarios such as grain silos, bulk cargo yards, and bulk cargo ship holds.

[0039] The above description is only a part of the embodiments of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made under the technical concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A multi-robot cooperative warehouse clearing method based on optimal transport theory, characterized in that, Includes the following steps: Step S10: Obtain the current height distribution and the preset target height distribution of materials within the work area; grid the work area and determine the material profit and loss status of each grid based on the current height distribution and the target height distribution; obtain the single transport capacity parameters of each flattening robot. Step S20: Based on the material surplus / deficit status of each grid and the single transport capacity parameters of each robot, solve the optimal transport model with the goal of minimizing the total cost of material transfer to obtain the global material transfer plan. The global material transfer plan includes the amount of material transferred from surplus grids to empty grids. Step S30: Plan the operation path for each robot according to the global material transfer plan and issue the path task; Step S40: Reacquire the current height distribution of the material, determine the material surplus / deficit status of each grid based on the reacquired current height distribution and the target height distribution, and jump to step S20 until the updated material height distribution meets the preset flattening termination condition, and then end the operation.

2. The method according to claim 1, characterized in that, The acquisition of the material surplus / deficit status of each grid cell specifically includes: The current material volume of each grid cell is calculated based on the current height distribution, and the target material volume of each grid cell is calculated based on the target height distribution. For each grid cell, calculate the difference between the current material volume and the target material volume; When the difference is greater than a preset positive threshold, the grid is identified as a surplus grid; when the difference is less than a preset negative threshold, the grid is identified as a vacant grid; otherwise, it is identified as a balanced grid.

3. The method according to claim 1, characterized in that, The method for constructing the optimal transport model with the objective of minimizing the total cost of material transfer includes: By taking the surplus grid set as the source distribution and the empty grid set as the target distribution, the multi-robot collaborative warehouse clearing problem is modeled as an optimal transport problem with robot carrying capacity and distance constraints, and the optimization objective is to minimize the Wasserstein distance between the source distribution and the target distribution. The unit material transportation cost from the source grid to the target grid is determined by the path distance between the two grids and the single transportation capacity parameter. The global material transfer plan is obtained by solving the optimal transfer model.

4. The method according to claim 3, characterized in that, The unit material transportation cost is obtained by dividing the path distance between two grids by the single transportation capacity parameter, or by multiplying the number of transfer trips required by the path distance between two grids. The number of transfer trips is the amount of material to be transferred divided by the single transportation capacity parameter and rounded up.

5. The method according to claim 3, characterized in that, The optimal transmission model is solved using any one of the following: the Sinkhorn entropy regularization algorithm, the distributed primal-dual algorithm based on Kantorovich duality, or a linear programming solver.

6. The method according to claim 1, characterized in that, The algorithm used to plan the operation path for each robot is A-star algorithm, fast randomized tree (RRT) algorithm, artificial potential field method, velocity obstacle method, or model predictive control (MPC) algorithm.

7. The method according to claim 1, characterized in that, The path task distribution includes: selecting a portion of the planned operation paths according to a preset threshold for the number of paths per operation or a threshold for operation duration, and using these as the path tasks for this round of distribution.

8. The method according to claim 1, characterized in that, The preset flattening termination condition is: traversing all grids, the absolute value of the difference between the current material height of each grid and the target height of the corresponding grid in the target height distribution is less than the preset maximum height difference threshold.

9. A multi-robot collaborative warehouse clearing system based on optimal transport theory, characterized in that, include: The sensing module is used to collect the three-dimensional point cloud of materials within the work area and obtain the current height distribution of the materials; The control module is configured to perform the method as described in any one of claims 1 to 8 to achieve material profit and loss status determination, optimal transmission model construction and solution, path planning and task assignment, and cyclic replanning and termination determination. The communication module is used to transmit data and instructions between the control module and each robot; The execution module, installed on each robot, is used to control the robot to complete the material pushing operation according to the received path task.

10. A multi-robot collaborative warehouse leveling device based on optimal transport theory, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multi-robot collaborative closing method based on optimal transmission theory as described in any one of claims 1 to 8.

Images