WO2021223262A1 - 物体转移装箱过程策略生成方法、装置、计算机设备 - Google Patents

物体转移装箱过程策略生成方法、装置、计算机设备 Download PDF

Info

Publication number
WO2021223262A1
WO2021223262A1 PCT/CN2020/090524 CN2020090524W WO2021223262A1 WO 2021223262 A1 WO2021223262 A1 WO 2021223262A1 CN 2020090524 W CN2020090524 W CN 2020090524W WO 2021223262 A1 WO2021223262 A1 WO 2021223262A1
Authority
WO
WIPO (PCT)
Prior art keywords
boxed
objects
packed
packing
target
Prior art date
Application number
PCT/CN2020/090524
Other languages
English (en)
French (fr)
Inventor
胡瑞珍
黄惠
张皓
龚明伦
许聚展
陈滨
Original Assignee
深圳大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳大学 filed Critical 深圳大学
Publication of WO2021223262A1 publication Critical patent/WO2021223262A1/zh
Priority to US17/947,933 priority Critical patent/US20230011757A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B65—CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65B—MACHINES, APPARATUS OR DEVICES FOR, OR METHODS OF, PACKAGING ARTICLES OR MATERIALS; UNPACKING
    • B65B1/00—Packaging fluent solid material, e.g. powders, granular or loose fibrous material, loose masses of small articles, in individual containers or receptacles, e.g. bags, sacks, boxes, cartons, cans, or jars
    • B65B1/30—Devices or methods for controlling or determining the quantity or quality or the material fed or filled
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • G06Q10/083—Shipping
    • G06Q10/0832—Special goods or special handling procedures, e.g. handling of hazardous or fragile goods

Definitions

  • This application relates to the technical field of object packing, and in particular to a method, device, computer equipment and storage medium for generating an object transfer packing process strategy.
  • the packing problem is a well-known discrete optimization problem, which has a wide range of geometric applications in the field of computer graphics, such as texture map generation, artistic typesetting, two-dimensional panel manufacturing, and three-dimensional printing.
  • more constraints must usually be dealt with.
  • An important change in the physical packing problem is that at the beginning, the object is already in a certain spatial configuration (such as accumulated warehouse inventory, etc.). The movement of the object must follow some sequence, such as the first object pressed on the first object. Before the second object is transferred and boxed, the first object cannot be moved. Therefore, in the actual physical object boxing application, the boxing process is very important to the final boxing state of the object.
  • a method for generating object transfer and packing process strategy including:
  • the geometric information, dependency information, and the current height map of the target container of each object to be boxed in the set of boxed objects are processed by the neural network to perform convolution operations to determine the current object to be boxed and the corresponding handling state;
  • the object packing sequence the handling state of each object to be packed and the preset placement strategy, the packing process strategy of the set of objects to be packed is determined.
  • An object transfer packing process strategy generation device includes:
  • An obtaining module configured to obtain a priority map of a set of objects to be boxed in the initial container, where the priority map is used to describe dependency constraints between each object to be boxed in the set of objects to be boxed;
  • An encoding module configured to encode the priority map to obtain geometric information and dependency information of each object to be boxed in a preset state
  • the convolution module is used to perform convolution operations on the geometric information, dependency information, and the current height map of the target container of each object to be boxed in the set of objects to be boxed through the neural network to determine the object to be boxed and the corresponding handling State, generate the object packing sequence of the set of objects to be packed;
  • the determining module is used to determine the packing process strategy of the set of objects to be packed according to the packing target, the packing sequence of the objects, the handling state of each object to be packed and the preset placement strategy.
  • a computer device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of the method for generating the above object transfer and packing process strategy are realized.
  • a computer-readable storage medium has a computer program stored thereon, and when the computer program is executed by a processor, the steps of a strategy generation method for an object transfer and packing process are realized.
  • the above-mentioned object transfer packing process strategy generation method, device, computer equipment and storage medium obtain the priority map of the set of objects to be packed in the initial container, and encode the priority map to obtain that each object to be packed is in a preset state Under the geometric information and dependency information; the geometric information of each object to be boxed in the set of boxed objects, the dependent information and the current height map of the target container are convolved through the neural network to obtain the objects of the set of objects to be boxed
  • Packing sequence Determine the packing process strategy of the set of objects to be packed according to the packing target, the object packing sequence, the handling state of each packed object and the preset placement strategy set.
  • the packing process strategy Since the packing process strategy is obtained according to the packing sequence of the objects, the handling status of each packed object and the preset placement strategy, the packing process strategy takes into account the packing order of the objects, the handling status during the packing process, and After the transfer is placed, the packing work can be guided according to the packing process strategy, thereby solving the problem of object transfer and packing, and improving the efficiency of object transfer and packing.
  • FIG. 1 is a schematic flowchart of a method for generating a strategy for an object transfer and packing process in an embodiment
  • 2A-2D are corresponding schematic diagrams of translational and rotational transportation of objects to be boxed in an embodiment
  • Figure 3 is an encoding result of geometric information and dependency information of the object to be boxed in an embodiment
  • 4A-4D are schematic diagrams of packing corresponding to the method for generating a strategy for the object transfer packing process in an embodiment
  • Figure 5 is a schematic flow chart of establishing a priority map in an embodiment
  • Figure 6 is a schematic diagram of a process of establishing a priority map in an embodiment
  • FIG. 7 is a schematic flowchart of determining the current object to be boxed and the corresponding handling state in an embodiment
  • Fig. 8 is a schematic structural diagram of a TAP-Net neural network in an embodiment
  • FIG. 9 is a schematic flowchart of determining a target placement strategy in an embodiment
  • 10A-10D are schematic diagrams of the effects of the tightness value, the taper degree value, and the stability value in an embodiment
  • FIG. 11 is a schematic flowchart of a method for generating a strategy for an object transfer and packing process in another embodiment
  • FIG. 12 is a schematic diagram of an object transfer packing process strategy generation method applied to rolling packing in an embodiment
  • FIG. 13 is a schematic diagram of the transfer and packing of objects to different target containers in an embodiment
  • FIG. 14 is a schematic structural diagram of a device for generating a strategy for an object transfer and packing process in an embodiment
  • 15 is a schematic structural diagram of a device for generating a strategy for an object transfer and packing process in another embodiment
  • Fig. 16 is a schematic diagram of the internal structure of a computer device in an embodiment.
  • a method for generating a strategy for the object transfer and packing process is provided.
  • This embodiment uses the method applied to a terminal for illustration. It is understandable that the method can also be applied to a server. , Can also be applied to systems including terminals and servers, and can be realized through interaction between terminals and servers.
  • the method includes the following steps:
  • Step 102 Obtain a priority map of the set of objects to be boxed in the initial container, where the priority map is used to describe the dependency constraints between each object to be boxed in the set of objects to be boxed.
  • the object to be boxed is a two-dimensional object or a three-dimensional object.
  • the packing of a two-dimensional object is taken as an example for description.
  • Dependency constraint is the transfer restriction between each object to be packed in the transfer process; among them, the transfer restriction can be blocking, whether the object to be packed can be touched, etc.
  • the initial container includes a first object to be boxed and a second object to be boxed; the first object to be boxed is stacked on the second object to be boxed, and the second object to be boxed is covered by the first object to be boxed Block, so the second object to be boxed can only be boxed after the first object to be boxed is transferred.
  • the priority map graphically describes the dependency constraints between the objects to be boxed; the priority map is determined by analyzing the initial position information and initial geometric information of each object to be boxed in the initial container.
  • the initial position information refers to the relative coordinate position of each object to be boxed in the initial container; the initial geometric information is the width and height of each object to be boxed in the initial container.
  • the handling order of the objects to be packed in the initial container can be determined.
  • the terminal receives the initial position information and initial geometric information of each object to be boxed in the set of objects to be boxed in the initial container; analyzes the initial position information and initial geometric information of each object to be boxed, and determines each The dependency constraints between the objects to be boxed; the dependency constraints between each object to be boxed are described graphically, and the priority map of the set of objects to be boxed in the initial container is obtained.
  • Step 104 Encode the priority map to obtain geometric information and dependency information of each object to be boxed in a preset state.
  • the preset state refers to a preset object handling state
  • the handling state includes translational handling and rotational handling.
  • Figures 2A and 2B are examples of transfer and boxing in the case of translational transportation
  • Figures 2C and 2D are examples of transfer and boxing in the case of rotating transportation
  • Figures 2A and 2C illustrate the state of the object in the initial container
  • Figure 2B And Figure 2D illustrates the state of the object in the target container.
  • the terminal sends the priority map of the set of objects to be boxed to the neural network.
  • the neural network can be, but is not limited to, the transfer boxing neural network (TAP-Net).
  • TAP-Net includes an encoder and a Attention mechanism decoder; the encoder in TAP-Net extracts each object to be boxed in the set to be boxed from the priority map, and performs geometric information and dependent information of each object to be boxed in a preset state Encode, get the geometric information and dependency information after encoding.
  • the encoded dependency information can be represented by "0" and "1", "0" represents no restriction, and "1" represents restriction.
  • the encoded geometric information and dependent information may include geometric information and dependent information corresponding to the translational transportation of the object to be packed, and geometric information and dependent information corresponding to the rotational transportation.
  • FIG 3 there are objects A, B, C, D, and E in the set of objects to be boxed.
  • the geometric information of the translational transportation of the object A to be boxed is (w A , h A ), w A represents the width of the object A to be boxed, and h A represents the height of the object A to be boxed;
  • the top dependency information is expressed as "00011” , That is, there are objects D and E on the top when object A to be boxed is moved in translation;
  • the dependency information on the left is "00000", that is, there is no dependency constraint on the left when object A to be boxed is moved in translation;
  • the dependency information on the right is "00000", that is, there is no dependency constraint on the right side of the object A to be packed in translation.
  • the geometric information of the rotating transportation of the object A to be boxed is (h A , w A ), and the top dependency information is "00011", that is, there are boxed objects D and E on the top when the object A to be boxed is rotated and transported; the dependence on the left
  • the information is "10000”, that is, the left side of the object A to be packed is next to the wall of the target container when it is rotated and transported;
  • the dependency information on the right is "01000", that is, the object A to be packed is rotated and transported on the right, and there is an object B to be packed on the right.
  • Step 106 Perform a convolution operation on the geometric information, dependency information, and the current height map of the target container of each object to be boxed in the set of objects to be boxed through the neural network to determine the current object to be boxed and the corresponding handling state.
  • Step 108 Generate an object boxing sequence of the set of objects to be boxed.
  • the height map is used to describe the actual packing state of the target container, and there is a mapping relationship between the height map and the target container.
  • the height map is a multi-dimensional array.
  • the index value of each element in the array locates a different grid, and each element value is the height of the grid.
  • the neural network includes an encoder and a decoder.
  • the geometric information, dependency information, and current height map of the target container in the set of encoded objects to be boxed are used as the input information of the neural network, and the input information is processed by the encoder and decoder in the neural network. Perform convolution operation to determine and output the current object to be boxed and its corresponding handling status.
  • the neural network performs n times of convolution operation, determines the current objects to be boxed and their corresponding handling states n times, and determines the current objects to be boxed according to n times Box objects and their corresponding handling states, generate an object packing sequence for the collection of objects to be packed.
  • Step 110 Determine the packing process strategy of the set of objects to be packed according to the packing target, the object packing sequence, the handling state of each object to be packed and the preset placement strategy.
  • the packing target is used to judge the standard of packing quality, and the packing target includes at least one of a density target, a taper degree target, and a stability target.
  • the density target is judged by the ratio of the total area of all objects to be boxed, the projected area of all objects to be boxed in the bottom direction of the target container, and the rectangular area determined by the highest height and width of the target container;
  • the taper degree target passes all the objects to be boxed
  • the total area of the object is judged by the ratio of the projected area of all objects to be boxed in the bottom direction of the container;
  • the stability target is judged by the ratio of the number of stable objects in the target container to the total number of boxed objects.
  • the stability judgment of each object to be packed in the packing process includes: firstly, through geometric analysis, find out all the contact points between the lower surface of the object to be packed in the position and the edge of the packed object; find out this The center point of the lower surface of the object to be boxed; connect the contact points two by two to form a contour area, which is a line segment in the case of a two-dimensional object, and a polygon in the case of a three-dimensional object; judge whether the center point is in this In the contour area, if the center point is in the contour area, the object to be boxed is stable, otherwise the object to be boxed is unstable.
  • the preset placement strategy is a preset object placement strategy; the preset placement strategy is a packing strategy based on the maximum remaining space, including the leftmost and highest packing strategy and the multi-corner packing strategy, both of which are in the target container Find the largest remaining space in the remaining space. Among them, the leftmost and highest packing strategy only tests the lower left corner position of each maximum remaining space, and then selects the strategy with the highest packing target value. The multi-corner boxing strategy tests each corner of each maximum remaining space, and then selects the strategy with the highest boxing target value.
  • the object packing sequence and the handling status of each packed object are calculated in each pre-packed object.
  • Set the packing target value under the placement strategy obtain the maximum packing target value from the calculated packing target value, and use the preset placement strategy corresponding to the maximum packing target value as the set of objects to be packed Strategy of the packing process.
  • the packing process strategy of the set of objects to be packed is determined according to the packing sequence of the objects, the handling state of each packed object and the preset maximization of the reachable convex space placement strategy.
  • the current objects to be packed are in accordance with the corresponding handling state, Place them in the reachable convex space in the target container in turn, calculate the largest reachable convex space area in the remaining space after placing the object, and determine the largest reachable convex space area corresponding to the preset maximized reachable convex space placement
  • the strategy is used as the packing process strategy of the set of objects to be packed. This method can improve the utilization rate of the placement space of the target container.
  • the terminal obtains the priority map ( Figure 4B) of the set of objects to be boxed in the initial container ( Figure 4A), where the priority map is used to describe the dependency constraints between each object to be boxed in the set of objects to be boxed;
  • the encoder in TAP-Net encodes the priority map to obtain the geometric information and dependency information of each object to be boxed in the preset state; the geometric information and dependent information of each object to be boxed in the preset state and
  • the height map of the target container is input to TAP-Net ( Figure 4C).
  • the encoder and the decoder with attention mechanism in TAP-Net perform convolution operations on the input information in turn to determine the current object to be boxed and its corresponding Handling status; according to all the determined current objects to be packed and their corresponding handling status, generate the object packing sequence of the objects to be packed; according to the packing target, the object packing sequence, the handling status of each packed object and Preset placement strategy and determine the packing process strategy for the collection of objects to be packed.
  • the terminal communicates with the handling equipment, and generates a corresponding control instruction according to the determined packing process strategy.
  • the control instruction instructs the handling equipment to pack the objects to be packed in the initial container into the target container in turn according to the packing process strategy to complete the object Boxing, the boxing effect is shown in Figure 4D.
  • the priority map of the set of objects to be packed in the initial container is obtained, where the priority map is used to describe the dependency constraints between each object to be packed in the set of objects to be packed ,
  • the priority map is used to describe the dependency constraints between each object to be packed in the set of objects to be packed .
  • the geometric information and dependency information of each object to be boxed in the preset state are obtained; the geometric information and dependency information of each object to be boxed in the set of boxed objects are treated in turn through the neural network Perform convolution operation with the current height map of the target container to determine the current object to be boxed and its corresponding handling state, and obtain the object packing sequence of the object to be packed; according to the packing target, the object packing sequence, and each object to be packed.
  • the handling state of the box objects and the preset placement strategy determine the boxing process strategy for the collection of objects to be boxed.
  • the packing process strategy Since the packing process strategy is obtained according to the packing sequence of the objects, the handling status of each packed object and the preset placement strategy, the packing process strategy takes into account the packing order of the objects, the handling status during the packing process, and The position to be placed after the transfer. According to the packing process strategy, the packing work can be guided, thereby solving the problem of object transfer and packing, and further improving the efficiency of object transfer and packing.
  • a method for establishing a priority graph is provided.
  • the method is applied to a terminal for illustration. The method includes the following steps:
  • Step 502 Obtain initial position information and initial geometric information of each object to be boxed in the set of objects to be boxed in the initial container.
  • the initial position information refers to the relative coordinate position of each object to be boxed in the initial container; the initial geometric information is the width and height of each object to be boxed in the initial container.
  • the terminal receives the initial container information input from the interface, and obtains the initial position information and initial geometric information of each to-be-packed object in the set of to-be-packed objects in the initial container according to the initial container information.
  • Step 504 Perform geometric analysis on each object to be boxed in the set of objects to be boxed according to the initial position information and the initial geometric information, and determine a priority map of the set of objects to be boxed.
  • performing geometric analysis on each object to be boxed in the set of objects to be boxed according to the initial position information and initial geometric information, and determining the priority map of the set of objects to be boxed includes:
  • each direction of each object to be packed in the set of objects to be packed is traversed, and the restriction state of each direction is determined; according to the restriction state, the priority map of the set of objects to be packed is determined.
  • the initial position information and the initial geometric information are acquired, and the traversal algorithm is used to sequentially traverse each direction of each object to be boxed, and determine the restriction state of each direction. For example, when traversing the vertical direction, if there are other objects to be boxed, a corresponding top blocking mark is established between the current object to be boxed and each other object to be boxed, such as "TB"; traversing the current object to be boxed If there are other objects to be boxed in the space above the left middle position of the box, a corresponding left-hand touch blocking mark is established between the object to be boxed and each other object to be boxed, such as "LAB"; In the space above the middle position of the right side of the boxed object, if there are other objects to be boxed, establish a right-hand reach blocking mark, such as "RAB", between the front object to be boxed and each other object to be boxed.
  • a right-hand reach blocking mark such as "RAB
  • Figure 6 is a schematic diagram of the process of establishing a priority map in an embodiment.
  • the solid black line in the priority map represents the top blocking mark (eg, TB); , LAB); the dark dashed line indicates the right side access barrier mark (such as RAB).
  • RAB right side access barrier mark
  • each object to be boxed in the set of objects to be boxed is obtained according to the initial position information and initial geometric information.
  • the geometric analysis of the objects is performed to determine the priority map of the set of objects to be boxed; the change in the dependence information of the objects to be boxed in the initial container during the boxing process is obtained according to the priority map, and the objects in the target container are continuously optimized according to the changed dependence information. Deployment strategy to improve the packing effect and the utilization of the target container space.
  • a method for determining the packing sequence of objects includes the following steps:
  • step 702 the encoder in the neural network performs a convolution operation on the geometric information and dependency information of each object to be boxed, and maps the operation result to a high-dimensional space to obtain the corresponding first high-dimensional vector.
  • the input of the convolutional layer of the encoder is the geometric information of each object to be boxed in the preset state and the dependency information in the priority map; the geometric information is static, and the dependency information is dynamic.
  • the geometric information and dependent information of each object to be boxed are mapped to a high-dimensional space through convolution operations, and the first high-dimensional vector is obtained, denoted as e, and each input information is recorded.
  • Step 704 The decoder in the neural network performs a convolution operation on the geometric information of the last determined object to be boxed and the current height map of the target container, and maps the operation result to a high-dimensional space to obtain the corresponding second high-dimensional vector.
  • the decoder includes a recurrent neural network (Recurrent Neural Network, RNN), which accumulates the handling status, geometric information, and dependency information of the objects to be packed each time determined during the packing process.
  • RNN includes two inputs. The first input is the cumulative vector obtained by convolution of the geometric information of the last determined object to be boxed and the current height map of the target container. The second input is the RNN when the last object to be boxed is determined The second highest dimensional vector of the output.
  • the decoder uses the output of the last attention mechanism, that is, the geometric information of the last selected boxed object, and the current height map of the target container, through the convolutional layer to map to the high-dimensional space and merge, and then input
  • the output of the binning RNN is obtained by the output of the decoder, which is mapped to the high-dimensional space through the convolutional layer and merged to obtain the cumulative vector, and the operation result is mapped to the high-dimensional space to obtain the corresponding second high-dimensional vector.
  • the output vector of the decoder contains the transportation status and geometric information of all the previously determined objects to be packed.
  • Step 706 Determine the boxing probability value of each object to be boxed in a preset state according to the first high-dimensional vector and the second high-dimensional vector.
  • the first high-dimensional vector and the second high-dimensional vector are input into the attention mechanism, and the boxing probability value of each object to be boxed in a preset state is calculated.
  • Step 708 Determine the current object to be packed and the corresponding handling state according to the packing probability value.
  • the packing probability value with the largest numerical value from the calculated packing probability values, determine the object to be packed and the corresponding preset state corresponding to the packing probability value with the largest numerical value; and set the packing probability value with the largest numerical value
  • the corresponding object to be packed and the corresponding preset state are used as the current object to be packed and the corresponding handling state. According to all the determined objects currently to be boxed, the object boxing sequence of the set of objects to be boxed is obtained.
  • after determining the current object to be boxed and the corresponding handling state it further includes: removing the determined object to be boxed from the set of objects to be boxed; updating the priority map of the set of objects to be boxed And the current height map of the target container, jump to the step of obtaining the priority map of the set of objects to be boxed in the initial container.
  • the geometric information and dependency information of each object to be boxed are convolved by the encoder in the neural network, and the operation result is mapped to the high-dimensional space to obtain the corresponding first high-dimensional vector; in the neural network
  • the decoder performs a convolution operation on the geometric information of the last determined object to be boxed and the current height map of the target container, and maps the result of the operation to a high-dimensional space to obtain the corresponding second high-dimensional vector;
  • the dimensional vector and the second high-dimensional vector determine the packing probability value of each object to be packed in the preset state; determine the current object to be packed and the corresponding handling state according to the packing probability value, and determine the current to be packed
  • the box object is removed from the set of objects to be boxed; the priority map of the set of objects to be boxed and the current height map of the target container are updated, and the step of obtaining the priority map of the set of objects to be boxed in the initial container is jumped to.
  • the target position of the objects to be boxed in the target container is determined, thereby improving the rationality of placing the boxed objects and further improving The rationality of the space utilization of the target container and the efficiency of transfer and packing.
  • FIG. 8 is a schematic structural diagram of a TAP-Net neural network.
  • the TAP-Net neural network includes an encoder and a decoder with an attention mechanism.
  • the TAP-Net neural network obtains the object packing sequence of the object set to be boxed: each object to be boxed has two different states, namely the translation state and the rotation state, and there are n objects to be packed in the set of objects to be boxed.
  • the box object corresponds to 2n states, and the TAP-Net neural network only outputs n states.
  • TAP-Net handles the transfer and packing problem of n objects to be packed
  • the TAP-Net neural network repeats the steps of determining the current objects to be packed and the corresponding handling status n times.
  • the input of the decoder includes the state y t-1 selected when the step of determining the current to-be-packed object and the corresponding handling state is executed at the t-1 th time, the current height map h t of the target container, and the t-th
  • the output ⁇ t-1 of the cyclic neural network RNN when the step of determining the current to-be-packed object and the corresponding handling state is executed once.
  • the encoder will input Convert to high-dimensional vector
  • the decoder converts the input y t-1 and h t into high-dimensional vectors ⁇ t , and the attention mechanism is based on Calculate the 2n probability values corresponding to the 2n states with the data of these high-dimensional vectors and ⁇ t , and select the transportation state of the object to be packed with the highest probability from the 2n probability values as the current output y t , which will determine
  • the current objects to be boxed are deleted from the set of objects to be boxed, and the priority map of the set of objects to be boxed in the initial container is updated.
  • the value of y t-1 is the initial value, and the initial value can be zero.
  • the geometric information and dependency information of each object to be packed are convolved by the encoder in the neural network, and the operation result is mapped to the high-dimensional space to obtain the corresponding first high-dimensional vector;
  • the decoder in the neural network performs a convolution operation on the geometric information of the last determined object to be boxed and the current height map of the target container, and maps the operation result to a high-dimensional space to obtain the corresponding second high-dimensional vector;
  • the first high-dimensional vector and the second high-dimensional vector determine the boxing probability value of each object to be boxed in the preset state; determine the current object to be boxed and the corresponding handling state according to the boxing probability value; All the objects currently to be boxed generate the object boxing sequence of the set of objects to be boxed. According to the priority map of the objects to be boxed in the initial container and the current height map of the target container, the current objects to be boxed and the corresponding handling status are sequentially determined, ensuring the rational
  • a method for determining a target placement strategy is provided. Taking the method applied to a terminal as an example for description, the method includes the following steps:
  • Step 902 According to the packing target, the object packing sequence, and the handling status of each object to be packed, calculate the packing target value of each preset placement strategy set corresponding to the set of objects to be packed.
  • the packing target includes at least one of a density target, a cone degree target, or a stability target; the packing target value includes at least one of a tightness value, a cone degree value, or a stability value.
  • the packing target value R can be, but not limited to, expressed by the following functional relation:
  • the compactness value C is defined as the ratio between the total area A packed of all objects to be boxed and the current highest height of the target container and the rectangular area determined by the container width A rect , as shown in Figure 10A, The compactness measure tends to pack the objects to be boxed tightly, and the compactness value C is 1 if and only when all the objects to be boxed completely fill the rectangular area.
  • the tightness value C can have a value range of 0-1.
  • the cone degree value P is defined as the ratio between the total area A packed of all objects to be packed and the projected area A proj of all objects to be packed in the bottom direction of the container, as shown in Figure 10B, The area outside the projection area is allowed to be filled by future objects to be boxed. If the current state enables the subsequent objects to be boxed to be tightly boxed, the greater the taper degree value P is.
  • the value range of the taper degree value P may be 0-1.
  • the stability value S is defined as the ratio of the number of stable objects N stable to the number of all packed objects N packed ; optionally, the value range of the stability value S can be 0 to 1.
  • the stability judgment of each object includes: firstly, through geometric analysis, find out all the contact points between the bottom surface of the object and the edge of the packed object when the object is at that position; find the center of the bottom surface of the object Point; connect the contact points in pairs to form a contour area, which is a line segment in the case of a two-dimensional object, and a polygon in the case of a three-dimensional object; judge whether the center point is in the area, and if the center point is in the In this area, the object is considered stable, otherwise the object is considered unstable, as shown in Figure 10C and 10D, the stability value of Figure 10C Stability value greater than Figure 10D That is, the boxing effect shown in Figure 10C is better than the boxing effect of 10D.
  • Step 904 Determine the target placement strategy of the set of objects to be boxed from the preset placement strategy set according to the packing target value.
  • the packing target includes at least one of a density target, a cone degree target, or a stability target; according to the packing target, the object packing sequence, and the handling status of each object to be packed, the corresponding pre-set of the object set to be packed is calculated.
  • the packing target value of each preset placement strategy in the set placement strategy set includes: according to the object packing sequence, the handling state of each object to be packed and the preset placement strategy, each item to be packed in the set Packed objects are packed into the target container in turn; calculate the tightness value, taper degree value and stability value of the target container after packing; calculate the preset placement according to the tightness value, taper degree value and stability value
  • the packing target value corresponding to each preset placement strategy in the strategy set The preset placement strategy with the largest binning target value is obtained from the calculated binning target values, and the preset placement strategy with the largest binning target value is taken as the target placement strategy for the collection of objects to be binned.
  • the target placement strategy of the set of objects to be packed is determined from the preset placement strategy.
  • the target placement strategy of the set of objects to be packed is determined from the preset placement strategy.
  • a method for generating an object transfer boxing process strategy is provided. Taking the method applied to a terminal as an example, the method includes the following steps:
  • Step 1102 Obtain initial position information and initial geometric information of each object to be boxed in the set of objects to be boxed in the initial container.
  • Step 1104 Perform geometric analysis on each object to be boxed in the set of objects to be boxed according to the initial position information and initial geometric information, and determine the priority map of the set of objects to be boxed.
  • Step 1106 Encode the priority map to obtain geometric information and dependency information of each object to be boxed in a preset state.
  • Step 1108 The encoder in the neural network performs a convolution operation on the geometric information and dependency information of each object to be boxed, and maps the operation result to a high-dimensional space to obtain the corresponding first high-dimensional vector.
  • Step 1110 the decoder in the neural network performs a convolution operation on the geometric information of the last determined object to be boxed and the current height map of the target container, and maps the operation result to a high-dimensional space to obtain the corresponding second high-dimensional vector.
  • Step 1112 Determine the current object to be boxed and the corresponding handling state according to the packing probability value, remove the determined object to be boxed from the set of objects to be boxed, and update the priority map and target container of the set of objects to be boxed The current height map.
  • Step 1114 Generate a boxing sequence of objects to be boxed.
  • the capacity of the TAP-Net is determined according to the number of training objects N.
  • N objects can be selected
  • the set ( ⁇ ) of TAP-Net is used for transfer and boxing.
  • the object to be boxed is deleted from the initial container, and another object to be boxed
  • Add to ⁇ update the priority map of the object set to be boxed and the current height map of the target container; repeat the process until the packing order of all objects is determined, and get the object boxing sequence of the set of objects to be boxed.
  • TAP-Net uses 4 objects for training and needs to process 6 objects to be boxed, first select 4 objects into the initial set ⁇ , and then use TAP-Net selects the first object to be boxed, such as object C. After taking out the object C to be boxed, another object such as object F is added to ⁇ . Repeat this process until all objects are packed.
  • Step 1116 Calculate the boxing target value of each preset placement strategy set corresponding to the set of objects to be boxed according to the boxing target, the object boxing sequence and the handling state of each boxed object.
  • Step 1118 Determine the target placement strategy of the set of objects to be boxed from the preset placement strategy set according to the boxing target value.
  • each object to be boxed in the set of objects to be boxed needs to be boxed into a different target container, the target container number corresponding to each object to be boxed is obtained, and each object to be boxed
  • the target container number corresponding to the box object and the geometric information and dependency information of each object to be boxed are subjected to a convolution operation, the operation result is mapped to a high-dimensional space, and the corresponding first high-dimensional vector is obtained.
  • the geometric information of the last determined object to be boxed and the current height map of each target container number corresponding to the target container are convolved, and the result of the operation is mapped to the high-dimensional space to obtain the corresponding The second high-dimensional vector.
  • determine the current object to be packed, the corresponding handling status and the corresponding target container remove the determined object to be packed from the set of objects to be packed, and update the priority map of the set of objects to be packed and The current height map of the target container; generate a boxing sequence of objects to be boxed in different target containers.
  • the initial configuration of the initial container includes the objects to be boxed A-J. Perform the above steps to box the objects to be boxed into the corresponding target container 1 and target container 2, respectively.
  • the above object transfer packing process strategy generation method is to obtain the initial position information and initial geometric information of each object to be boxed in the set of objects to be boxed in the initial container; according to the initial position information and initial geometric information, the set of objects to be boxed is included Perform geometric analysis of each object to be boxed to determine the priority map of the set of objects to be boxed; encode the priority map to obtain the geometric information and dependency information of each object to be boxed in the preset state; through the neural network
  • the encoder in the encoder performs convolution operation on the geometric information and dependent information of each object to be boxed, and maps the result of the operation to the high-dimensional space to obtain the corresponding first high-dimensional vector;
  • the geometric information of the determined object to be packed and the current height map of the target container are subjected to convolution operation, and the operation result is mapped to the high-dimensional space to obtain the corresponding second high-dimensional vector; the current object to be packed is determined according to the packing probability value And the corresponding handling status, remove the determined
  • the object packing sequence and the handling status of each packed object calculate the packing target value of each preset placement strategy set corresponding to the set of objects to be packed; according to the packing target value, Determine the target placement strategy of the set of objects to be boxed from the preset placement strategy set.
  • the object boxing is three-dimensional object boxing, by obtaining the priority map of the set of objects to be boxed in the initial container; the priority map is used to describe the relationship between each object to be boxed in the set of objects to be boxed Dependency constraints; by encoding the priority map, the geometric information and dependency information of each object to be boxed in the preset state are obtained; the geometric information of each object to be boxed in the set of boxed objects is treated in turn through the neural network , Rely on the information and the current height map of the target container to perform convolution operations to determine the current object to be boxed and the corresponding handling status; generate the object boxing sequence of the object to be boxed collection; according to the boxing target, object boxing sequence, and each The handling state of the boxed objects and the preset placement strategy determine the boxing process strategy for the collection of objects to be boxed.
  • the efficiency of transfer and boxing is improved.
  • the specific definition of the device for generating the three-dimensional object transfer and packing process strategy can refer to the above definition of the two-dimensional object transfer and packing process strategy generation method, which will not be repeated here.
  • a device for generating an object transfer boxing process strategy includes: an acquisition module 1402, an encoding module 1404, a convolution module 1406, and a determination module 1408, in which:
  • the obtaining module 1402 is configured to obtain a priority map of the set of objects to be boxed in the initial container, where the priority map is used to describe the dependency constraints between each object to be boxed in the set of objects to be boxed.
  • the encoding module 1404 is used to encode the priority map to obtain geometric information and dependency information of each object to be boxed in a preset state.
  • the convolution module 1406 is used to perform convolution operations on the geometric information, dependency information and the current height map of the target container of each object to be boxed in the set of objects to be boxed through the neural network to determine the current object to be boxed and the corresponding In the handling state, the object packing sequence of the collection of objects to be packed is generated.
  • the determining module 1408 is used to determine the packing process strategy of the set of objects to be packed according to the packing target, the object packing sequence, the handling state of each object to be packed and the preset placement strategy.
  • the priority map of the set of objects to be boxed in the initial container is obtained, where the priority map is used to describe the dependency constraints between each object to be boxed in the set of objects to be boxed ;
  • the priority map is used to describe the dependency constraints between each object to be boxed in the set of objects to be boxed ;
  • the geometric information and dependency information of each object to be boxed in the preset state are obtained;
  • the geometric information and dependency information of each object to be boxed in the set of boxed objects are treated in turn through the neural network Perform convolution operation with the current height map of the target container to determine the current object to be boxed and the corresponding handling state, and obtain the object boxing sequence of the object to be boxed collection; according to the boxing target, object boxing sequence, and each boxed object To determine the packing process strategy of the collection of objects to be packed.
  • the acquiring module 1402 is further configured to acquire the initial position information and initial geometric information of each object to be boxed in the set of objects to be boxed in the initial container.
  • the device for generating a strategy for the object transfer and packing process further includes:
  • the geometric analysis module 1410 is configured to perform geometric analysis on each object to be boxed in the set of objects to be boxed according to the initial position information and the initial geometric information, and determine the priority map of the set of objects to be boxed;
  • the traversal module 1412 is used to traverse each direction of each object to be boxed in the set of objects to be boxed according to the initial position information and initial geometric information, and determine the restriction state of each direction; according to the restriction state, determine the object to be boxed The priority map of the collection of objects.
  • the convolution module 1406 is also used to perform convolution operations on the geometric information and dependency information of each object to be boxed through the encoder in the neural network, and map the result of the operation to a high-dimensional space to obtain the corresponding The first high-dimensional vector; through the decoder in the neural network, the geometric information of the last determined object to be boxed and the current height map of the target container are convolved, and the result of the operation is mapped to the high-dimensional space to obtain the corresponding The second high-dimensional vector;
  • the first high-dimensional vector and the second high-dimensional vector determine the packing probability value of each object to be packed in the preset state; according to the packing probability value, determine the current object to be packed and the corresponding handling state to obtain The object packing sequence of the boxed object collection.
  • the removing module 1414 is used to remove the determined objects to be boxed from the set of objects to be boxed.
  • the update module 1416 is used to update the priority map of the set of objects to be boxed and the current height map of the target container, and jump to the step of the priority map of the set of objects to be boxed in the initial container.
  • the calculation module 1418 is configured to calculate the boxing target value of each preset placement strategy set corresponding to the set of objects to be boxed according to the boxing target, the object boxing sequence and the handling state of each boxed object.
  • the calculation module 1418 is also used to pack each object to be boxed in the set of objects to be boxed to the target in turn according to the object packing sequence, the handling state of each boxed object, and the preset placement strategy.
  • the container Calculate the tightness value, taper degree value and stability value of the target container after packing.
  • the determining module 1408 is further configured to determine the target placement strategy of the set of objects to be packed from the preset placement strategy set according to the packing target value; wherein the packing target includes a density target, a cone degree target or At least one of the stability goals.
  • the determining module 1408 is further configured to calculate the boxing target value corresponding to each preset placement strategy in the preset placement strategy set based on the tightness value, the taper degree value, and the stability value.
  • the initial position information and initial geometric information of each object to be packed in the set of objects to be packed in the initial container are acquired; the set of objects to be packed is set according to the initial position information and the initial geometric information Perform geometric analysis of each object to be boxed in the box to determine the priority map of the set of objects to be boxed; by encoding the priority map, the geometric information and dependency information of each object to be boxed in the preset state are obtained;
  • the encoder in the neural network performs a convolution operation on the geometric information and dependent information of each object to be boxed, and maps the result of the operation to a high-dimensional space to obtain the corresponding first high-dimensional vector; the decoder in the neural network
  • the geometric information of the last determined object to be boxed and the current height map of the target container are subjected to convolution operation, and the operation result is mapped to the high-dimensional space to obtain the corresponding second high-dimensional vector; according to the first high-dimensional vector and the first high-dimensional vector.
  • Removal of the set of boxed objects update the priority map of the set of objects to be boxed and the current height map of the target container, jump to the step of obtaining the priority map of the set of objects to be boxed in the initial container; generate the set of objects to be boxed Object packing sequence.
  • each object to be packed in the set of objects to be packed is packed into the target container in turn; according to the packing target, the calculation after packing is calculated The tightness value, taper degree value and stability value corresponding to the target container; according to the tightness value, taper degree value and stability value, the packing target value corresponding to each preset placement strategy in the preset placement strategy set is calculated ; According to the packing target value, the target placement strategy of the set of objects to be packed is determined from the preset placement strategy.
  • Each module in the above-mentioned object transfer and packing process strategy generation device can be implemented in whole or in part by software, hardware and a combination thereof.
  • the above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.
  • a computer device is provided.
  • the computer device may be a terminal, and its internal structure is shown in FIG. 16.
  • the computer equipment includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus.
  • the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system and a computer program.
  • the internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium.
  • the communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, an operator's network, NFC (near field communication) or other technologies.
  • the computer program is executed by the processor to realize a strategy generation method of the object transfer and packing process.
  • the display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, or it can be a button, a trackball or a touchpad set on the housing of the computer equipment , It can also be an external keyboard, touchpad, or mouse.
  • FIG. 16 is only a block diagram of part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.
  • a computer device including a memory and a processor, and a computer program is stored in the memory, and the processor implements the steps in the foregoing method embodiments when the computer program is executed.
  • a computer-readable storage medium on which a computer program is stored, and the computer program is executed by a processor to implement the steps in the foregoing method embodiments.
  • Non-volatile memory may include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, or optical storage.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc.

Abstract

一种物体转移装箱过程策略生成方法、装置、计算机设备。所述方法包括：获取初始容器中待装箱物体集合的优先级图；对优先级图进行编码；神经网络卷积运算，确定当前待装箱物体；生成待装箱物体集合的物体装箱序列；根据装箱目标、物体装箱序列、各装箱物体的搬运状态和预设摆放策略，确定待装箱物体集合的装箱过程策略。

Description

物体转移装箱过程策略生成方法、装置、计算机设备
相关申请的交叉引用
本申请要求于2020年5月6日提交中国专利局，申请号为202010371670.X，申请名称为“物体转移装箱过程策略生成方法、装置、计算机设备”的中国专利申请的优先权,在此将其全文引入作为参考。
技术领域
本申请涉及物体装箱技术领域，特别涉及一种物体转移装箱过程策略生成方法、装置、计算机设备和存储介质。
背景技术
装箱问题是众所周知的离散优化问题，在计算机图形学领域中具有广泛的几何应用，例如纹理图生成、艺术排版、二维面板制造和三维打印等。然而，在其他涉及物理对象的实际应用中(如机器人装箱运输等)，通常还必须应对更多的约束条件。物理装箱问题的一个重要变化是，在初始时，物体就已经处于某种空间配置中(如积累的仓库库存等)，物体的移动必须遵循一些先后顺序，比如压在第一物体上面的第二物体被转移并装箱之前，无法移动第一物体。因此，在实际物理对象装箱应用中，装箱过程对物体最终的装箱状态至关重要。
然而，几何装箱问题中，仅需优化虚拟对象的装箱效果，以用于显示、存储或制造等场景中。在物理装箱问题中，初始状态对物品的装箱顺序和朝向有严格的约束；现有装箱技术通过假定待装箱物体没有任何额外的空间约束或依赖约束进行装箱，得到的装箱效果仅具有装箱结果显示，实际装箱过程中，需要花费时间根据装箱效果分析如何搬运，从而导致物体装箱效率低。
发明内容
基于此，有必要针对上述技术问题，提供一种能够提高物体转移装箱效率的物体转移装箱过程策略生成方法、装置、计算机设备和存储介质。
一种物体转移装箱过程策略生成方法，包括：
获取初始容器中待装箱物体集合的优先级图，其中所述优先级图用于描述所述待装箱物体集合中每个待装箱物体之间的依赖约束；
对所述优先级图进行编码，得到所述每个待装箱物体在预设状态下的几何信息和依赖信息；
通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态；
生成所述待装箱物体集合的物体装箱序列；
根据装箱目标、所述物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，确定所述 待装箱物体集合的装箱过程策略。
一种物体转移装箱过程策略生成装置，所述装置包括：
获取模块，用于获取初始容器中待装箱物体集合的优先级图，其中所述优先级图用于描述所述待装箱物体集合中每个待装箱物体之间的依赖约束；
编码模块，用于对所述优先级图进行编码，得到所述每个待装箱物体在预设状态下的几何信息和依赖信息；
卷积模块，用于通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态，生成待装箱物体集合的物体装箱序列；
确定模块，用于根据装箱目标、所述物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，确定所述待装箱物体集合的装箱过程策略。
一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，所述处理器执行所述计算机程序时实现上述物体转移装箱过程策略生成方法的步骤。
一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现物体转移装箱过程策略生成方法的步骤。上述物体转移装箱过程策略生成方法、装置、计算机设备和存储介质，通过获取初始容器中待装箱物体集合的优先级图，对优先级图进行编码得到每个待装箱物体在预设状态下的几何信息和依赖信息；通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，得到待装箱物体集合的物体装箱序列；根据装箱目标、物体装箱序列、各装箱物体的搬运状态和预设摆放策略集，确定待装箱物体集合的装箱过程策略。由于装箱过程策略是根据物体装箱序列、各装箱物体的搬运状态和预设摆放策略得到的，因此，装箱过程策略考虑了物体的装箱顺序，装箱过程中的搬运状态以及转移后被摆放的位置，根据装箱过程策略即可对装箱工作进行指导，从而解决了物体转移和装箱问题，提高了物体转移装箱效率。
附图说明
图1为一个实施例中物体转移装箱过程策略生成方法的流程示意图；
图2A-2D为一个实施例中待装箱物体的平移搬运和旋转搬运对应的示意图；
图3为一个实施例中待装箱物体几何信息和依赖信息的编码结果；
图4A-4D为一个实施例中物体转移装箱过程策略生成方法对应的装箱示意图；
图5为一个实施例中建立优先级图的流程示意图；
图6为一个实施例中建立优先级图的过程示意图；
图7为一个实施例中确定当前待装箱物体和对应的搬运状态的流程示意图；
图8为一个实施例中TAP-Net神经网络的结构示意图；
图9为一个实施例中确定目标摆放策略的流程示意图；
图10A-10D为一个实施例中紧密度值、锥形程度值和稳定性值的效果示意图；
图11为另一个实施例中物体转移装箱过程策略生成方法的流程示意图；
图12为一个实施例中物体转移装箱过程策略生成方法应用于滚动装箱的示意图；
图13为一个实施例中物体转移装箱到不同目标容器的示意图；
图14为一个实施例中物体转移装箱过程策略生成装置的结构示意图；
图15为另一个实施例中物体转移装箱过程策略生成装置的结构示意图；
图16为一个实施例中计算机设备的内部结构示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
在一个实施例中，如图1所示，提供了一种物体转移装箱过程策略生成方法，本实施例以该方法应用于终端进行举例说明，可以理解的是，该方法也可以应用于服务器，还可以应用于包括终端和服务器的系统，并通过终端和服务器的交互实现。本实施例中，该方法包括以下步骤：
步骤102，获取初始容器中待装箱物体集合的优先级图，其中优先级图用于描述待装箱物体集合中每个待装箱物体之间的依赖约束。
其中，待装箱物体为二维待物体或三维物体。本实施中以二维物体装箱为例进行说明。依赖约束是每个待装箱物体在转移过程中相互之间的转移限制；其中，转移限制可以是阻挡、待装箱物体是否能被触及等。例如，初始容器中包括第一待装箱物体和第二待装箱物体；第一待装箱物体叠放在第二待装箱物体上面，第二待装箱物体被第一待装箱物体阻挡，故第二待装箱物体只有在第一待装箱物体被转移之后才能被装箱。
优先级图以图形的方式描述待装箱物体之间的依赖约束；优先级图是通过分析初始容器中每个待装箱物体的初始位置信息和初始几何信息确定的。其中，初始位置信息是指每个待装箱物体在初始容器中的相对坐标位置；初始几何信息是初始容器中的每个待装箱物体的宽度和高度。根据优先级图可以确定初始容器中待装箱物体的搬运顺序。
具体地，终端接收初始容器中待装箱物体集合中每个待装箱物体的初始位置信息和初始几何信息；对每个待装箱物体的初始位置信息和初始几何信息进行分析，确定每个待装箱物体之间的依赖约束；以图形的方式描述每个待装箱物体之间的依赖约束，得到初始容器中待装箱物体集合的优先级图。
步骤104，对优先级图进行编码，得到每个待装箱物体在预设状态下的几何信息和依赖信息。
其中，预设状态是指预先设置的物体搬运状态，搬运状态包括平移搬运和旋转搬运。图2A和图2B为平移搬运情况下的转移和装箱示例，图2C和图2D为旋转搬运情况下的转移和装箱示例；图2A和图2C示意物体在初始容器中的状态，图2B和图2D示意物体在目标容器中的状态。
具体地，终端把待装箱物体集合的优先级图发送到神经网络中，神经网络可以是但不仅限于是转移装箱神经网络(TAP-Net)，TAP-Net包括一个编码器和一个带有注意力机制的解码器；TAP-Net中编码器从优先级图提取待装箱集合中的每个待装箱物体，对每个待装箱物体在预设状态下的几何信息和依赖信息进行编码，得到编码后的几何信息和依赖信息。其中，编码后的依赖信息可以用“0”和“1”来表示，“0”代表无约束限制，“1”代表约束限制。可选地，编码后的几何信息和依赖信息可包括待装箱物体平移搬运对应的几何信息和依赖信息，以及旋转搬运对应的几何信息和依赖信息。如图3所示，待装箱物体集合中有待装箱物体A、B、C、D、E。待装箱物体A的平移搬运的几何信息为(w A，h A)，w A代表待装箱物体A的宽度，h A代表待装箱物体A的高度；顶部依赖信息表示为“00011”，即待装箱物体A平移搬运时顶部有待装箱物体D、E；左侧的依赖信息为“00000”，即待装箱物体A平移搬运时左侧没有依赖约束；右侧的依赖信息为“00000”，即待装箱物体A平移搬运时右侧没有依赖约束。待装箱物体A的旋转搬运的几何信息为(h A，w A)，顶部依赖信息为“00011”，即待装箱物体A旋转搬运时顶部有装箱物体D、E；左侧的依赖信息为“10000”，即待装箱物体A旋转搬运时左侧紧挨着目标容器壁；右侧的依赖信息为“01000”，即待装箱物体A旋转搬运时右侧有待装箱物体B。
步骤106，通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态。
步骤108，生成待装箱物体集合的物体装箱序列。
其中，高度图用于描述目标容器的实际装箱状态，高度图与目标容器之间存在映射关系。高度图是一个多维数组，数组中的每个元素的索引值定位不同的网格，每个元素值是网格的高度。
具体地，所述神经网络包括编码器和解码器。将编码后的待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图作为神经网络的输入信息，通过神经网络中的编码器和解码器，对输入信息执行卷积运算，确定并输出当前待装箱物体和其对应的搬运状态。可选地，待装箱物体集合中有n个待装箱物体，神经网络执行n次卷积运算，n次确定当前待装箱物体和其对应的搬运状态，根据n次确定的当前待装箱物体和其对应的搬运状态，生成待装箱物体集合的物体装箱序列。
步骤110，根据装箱目标、物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，确定待装箱物体集合的装箱过程策略。
其中，装箱目标用于判断装箱质量的标准，装箱目标包括密度目标、锥形程度目标和稳定性目标中的至少一种。密度目标通过所有待装箱物体的总面积、所有待装箱物体在目标容器底部方向的投影面积和目标容器最高高度和宽度确定的矩形面积的比例来判断；锥形程度目标通过所有待装箱物体的总面积和所有待装箱物体在容器底部方向的投影面积的比例来判断；稳定性目标通过目标容器中稳定物体的数量与已装箱物体的总数量的比例值来判断。装箱过程中每个待装箱物体的稳定性判断包括：首先通过几何分析，找出该待装箱物体在该位置时其下表面与已装箱物体的边缘的所有接触点；找出该待装箱物体下表面的中心点；将接触点两两相连， 将组成一个轮廓区域，该区域在二维物体情况下是一条线段，在三维物体情况下是一个多边形；判断中心点是否在该轮廓区域中，若中心点在该轮廓区域中，则该待装箱物体是稳定的，否则该待装箱物体是不稳定的。
预设摆放策略是预先设置的物体摆放策略；预设摆放策略是基于最大剩余空间的装箱策略，包括最左最高装箱策略和多角落装箱策略，两者都在目标容器的剩余空间中找到最大剩余空间。其中，最左最高装箱策略仅测试每个最大剩余空间的左下角位置，然后选择装箱目标值最高的策略。多角落装箱策略测试每个最大剩余空间的每个角落，然后选择装箱目标值最高的策略。
具体地，根据神经网络输出的物体装箱序列和各待装箱物体的搬运状态，以及预设摆放策略和装箱目标，计算物体装箱序列、各装箱物体的搬运状态在每个预设摆放策略下的装箱目标值，从计算得到的装箱目标值中获取数值最大的装箱目标值，将数值最大的装箱目标值对应的预设摆放策略作为待装箱物体集合的装箱过程策略。
在一个实施例中，根据物体装箱序列、各装箱物体的搬运状态和预设最大化可触及凸空间摆放策略，确定待装箱物体集合的装箱过程策略。可选地，在装箱之前，确定目标容器中所有可触及的凸空间；对目标容器进行装箱时，根据最大化可触及凸空间摆放策略将当前待装箱物体按照对应的搬运状态，依次摆放至目标容器中可触及的凸空间，计算物体摆放后的剩余空间中的最大可触及凸空间面积，确定最大可触及凸空间面积最大对应的预设最大化可触及凸空间摆放策略作为待装箱物体集合的装箱过程策略。该方法可以提高目标容器的摆放空间的利用率。
以下为物体转移装箱过程策略生成的一个应用场景，如图4A-4D所示。
终端通过获取初始容器中待装箱物体集合(图4A)的优先级图(图4B)，其中，优先级图用于描述待装箱物体集合中每个待装箱物体之间的依赖约束；TAP-Net中编码器对优先级图进行编码，得到每个待装箱物体在预设状态下的几何信息和依赖信息；每个待装箱物体在预设状态下的几何信息和依赖信息和目标容器的高度图输入到TAP-Net(图4C)中，TAP-Net中的编码器和带注意力机制的解码器依次对输入信息执行卷积运算，确定当前待装箱物体和其对应的搬运状态；根据所有确定的各当前待装箱物体和其对应的搬运状态，生成待装箱物体集合的物体装箱序列；根据装箱目标、物体装箱序列、各装箱物体的搬运状态和预设摆放策略，确定待装箱物体集合的装箱过程策略。终端与搬运设备进行通讯，根据确定的装箱过程策略生成对应的控制指令，该控制指令指示搬运设备根据装箱过程策略将初始容器中的待装箱物体依次装箱到目标容器中，完成物体装箱，装箱效果如图4D所示。
上述物体转移装箱过程策略生成方法中，通过获取初始容器中待装箱物体集合的优先级图，其中优先级图用于描述待装箱物体集合中每个待装箱物体之间的依赖约束，通过对优先级图进行编码，得到每个待装箱物体在预设状态下的几何信息和依赖信息；通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和其对应的搬运状态，得到待装箱物体集合的物体装箱序列；根据装箱目标、物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，确定待装箱物体集合的装箱过 程策略。由于装箱过程策略是根据物体装箱序列、各装箱物体的搬运状态和预设摆放策略得到的，因此，装箱过程策略考虑了物体的装箱顺序，装箱过程中的搬运状态以及转移后被摆放的位置。根据装箱过程策略即可对装箱工作进行指导，从而解决了物体转移和装箱问题，进一步提高了物体转移装箱效率。
在一个实施例中，如图5所示，提供了一种建立优先级图的方法，本实施例以该方法应用于终端进行举例说明，该方法包括以下步骤：
步骤502，获取初始容器中待装箱物体集合中每个待装箱物体的初始位置信息和初始几何信息。
其中，初始位置信息是指每个待装箱物体在初始容器中的相对坐标位置；初始几何信息是每个待装箱物体在初始容器中摆放的宽度和高度。
具体地，终端接收界面输入的初始容器信息，根据初始容器信息获取初始容器中待装箱物体集合中每个待装箱物体的初始位置信息和初始几何信息。
步骤504，根据初始位置信息和初始几何信息对待装箱物体集合中每个待装箱物体进行几何分析，确定待装箱物体集合的优先级图。
在一个实施例中，根据初始位置信息和初始几何信息对待装箱物体集合中每个待装箱物体进行几何分析，确定待装箱物体集合的优先级图包括：
根据初始位置信息和初始几何信息，遍历待装箱物体集合中每个待装箱物体每个方向，确定每个方向的限制状态；根据限制状态，确定待装箱物体集合的优先级图。
具体地，获取初始位置信息和初始几何信息，采用遍历算法依次遍历每个待装箱物体每个方向，确定每个方向的限制状态。例如遍历竖直方向上方，若有其他待装箱物体，则在当前待装箱物体和每一其他待装箱物体之间建立对应的顶部阻挡标识，如“TB”；遍历当前待装箱物体的左边中间位置以上的空间，若有其他待装箱物体，则在当前待装箱物体和每一其他待装箱物体之间建立对应的左侧触及阻挡标识，如“LAB”；遍历当前待装箱物体的右边的中间位置以上的空间，若有其他待装箱物体，则在前待装箱物体和每一其他待装箱物体之间建立右侧触及阻挡标识，如“RAB”。图6为一个实施例中建立优先级图的过程示意图，如图6所示，在优先级图中黑色实线表示顶部阻挡标识(如，TB)；浅色虚线表示左侧触及阻挡标识(如，LAB)；深色虚线表示为右侧触及阻挡标识(如，RAB)。通过遍历分析待装箱物体每个方向，确定待装箱物体每个方向是否存在其他待装箱物体。根据遍历结果，确定每个待装箱物体的限制状态，得到待装箱物体集合的优先级图，明确初始容器中待装箱物体的状态，从而减少转移装箱时间和提高装箱效率。
本实施例中，通过获取初始容器中待装箱物体集合中每个待装箱物体的初始位置信息和初始几何信息，根据初始位置信息和初始几何信息对待装箱物体集合中每个待装箱物体进行几何分析，确定待装箱物体集合的优先级图；根据优先级图获取装箱过程中初始容器中待装箱物体的依赖信息的变化，根据变化的依赖信息不断优化目标容器中物体摆放策略，从而提高装箱效果以及目标容器空间的利用率。
在一个实施例中，如图7所示，提供了一种确定物体装箱序列方法，以该方法应用于终端为例进行说明，包括以下步骤：
步骤702，神经网络中的编码器对每个待装箱物体的几何信息和依赖信息进行卷积运算，将运算结果映射到高维空间，得到对应的第一高维向量。
具体地，编码器的卷积层的输入是每个待装箱物体在预设状态下的几何信息和优先级图中的依赖信息；几何信息是静态的，依赖信息是动态的。每个待装箱物体的几何信息和依赖信息通过卷积运算，将运算结果映射到高维空间，得到第一高维向量，表示为e，记录每个输入信息。
步骤704，神经网络中的解码器对上一个已确定的待装箱物体的几何信息、目标容器的当前高度图进行卷积运算，将运算结果映射到高维空间，得到对应的第二高维向量。
其中，解码器包括一个循环神经网络(Recurrent Neural Network，RNN)，将装箱过程中每次确定的待装箱物体的搬运状态、几何信息以及依赖信息进行累计。RNN包括两个输入，第一输入是上一个已确定的待装箱物体的几何信息和目标容器的当前高度图卷积得到的累计向量，第二个输入是确定上一个待装箱物体时RNN输出的第二高维向量。
具体地，解码器通过对上一次注意力机制的输出即上一个被选择装箱的物体的几何信息，和目标容器的当前高度图，通过卷积层映射到高维空间并合并，再输入上一次装箱RNN输出得到解码器的输出，通过卷积层映射到高维空间并合并得到累计向量，将运算结果映射到高维空间得到对应的第二高维向量。其中，解码器的输出向量包含着前面所有已确定的待装箱物体的搬运状态和几何信息。
步骤706，根据第一高维向量和第二高维向量，确定每个待装箱物体在预设状态下的装箱概率值。
具体地，将第一高维向量和第二高维向量输入注意力机制中，计算每个待装箱物体在预设状态下的装箱概率值。
步骤708，根据装箱概率值确定当前待装箱物体和对应的搬运状态。
具体地，从计算得到的装箱概率值中确定数值最大的装箱概率值，确定数值最大的装箱概率值对应的待装箱物体和对应的预设状态；将数值最大的装箱概率值对应的待装箱物体和对应的预设状态作为当前待装箱物体和对应的搬运状态。根据确定的所有当前待装箱物体，得到待装箱物体集合的物体装箱序列。
在一个实施例中，在确定当前待装箱物体和对应的搬运状态之后，还包括：将确定的当前待装箱物体从待装箱物体集合移除；更新待装箱物体集合的优先级图和目标容器的当前高度图，跳转到获取初始容器中待装箱物体集合的优先级图的步骤。
具体地，通过神经网络中的编码器对每个待装箱物体的几何信息和依赖信息进行卷积运算，将运算结果映射到高维空间，得到对应的第一高维向量；通过神经网络中的解码器对上一个已确定的待装箱物体的几何信息、目标容器的当前高度图进行卷积运算，将运算结果映射到高维空间，得到对应的第二高维向量；根据第一高维向量和第二高维向量，确定每个待装箱物体在 预设状态下的装箱概率值；根据装箱概率值确定当前待装箱物体和对应的搬运状态，将确定的当前待装箱物体从待装箱物体集合移除；更新待装箱物体集合的优先级图和目标容器的当前高度图，跳转到获取初始容器中待装箱物体集合的优先级图的步骤。通过更新初始容器中待装箱物体集合中的待装箱物体和优先级图，确定待装箱物体在目标容器中的目标位置，从而提高了装箱物体摆放的合理性，进一步地提高了目标容器的空间利用合理性以及转移装箱的效率。
在一个实施例中，图8为一TAP-Net神经网络的结构示意图，如图8所示，TAP-Net神经网络包括一个编码器和一个带有注意力机制的解码器。TAP-Net神经网络得到待装箱物体集合的物体装箱序列的过程为：每个待装箱物体有两个不同的状态，即平移状态和旋转状态，待装箱物体集合中n个待装箱物体对应2n个状态，TAP-Net神经网络只输出n个状态。当TAP-Net处理n个待装箱物体的转移装箱问题时，TAP-Net神经网络重复执行确定当前待装箱物体和对应的搬运状态的步骤n次。在第t次执行所述确定当前待装箱物体和对应的搬运状态的步骤时，编码器模块的输入表示为
Figure PCTCN2020090524-appb-000001
其中s i＝{w i，h i} i＝1.....n，s i表示第i个待装箱物体的静态的位置信息，包括第i个待装箱物体的宽度w i和高度h i，
Figure PCTCN2020090524-appb-000002
则表示第i个待装箱物体在第t次执行所述确定当前待装箱物体和对应的搬运状态的步骤时的动态的依赖信息。解码器的输入包括在第t-1次执行所述确定当前待装箱物体和对应的搬运状态的步骤时选择的状态y t-1、目标容器的当前高度图h t，以及在第t-1次执行所述确定当前待装箱物体和对应的搬运状态的步骤时循环神经网络RNN的输出ω t-1。
编码器将输入的
Figure PCTCN2020090524-appb-000003
转换为高维度向量
Figure PCTCN2020090524-appb-000004
解码器将输入的y t-1和h t转换为高维向量ω t，注意力机制根据
Figure PCTCN2020090524-appb-000005
和ω t这些高维向量的数据，计算2n个状态所对应的2n个概率值，从2n个概率值中选出对应概率最大的待装箱物体的搬运状态作为当次输出y t，将确定的当前待装箱物体从待装箱物体集合中删除，更新初始容器的待装箱物体集合的优先级图。该过程反复执行直到获取n个待装箱物体的搬运状态的序列，即得到待装箱物体集合的物体装箱序列{y t} t＝1....n.。在TAP-Net最开始进行计算的时候，y t-1数值为初始值，初始值可以是零。
上述确定物体装箱序列中，通过神经网络中编码器对每个待装箱物体的几何信息和依赖信息进行卷积运算，将运算结果映射到高维空间得到对应的第一高维向量；通过神经网络中的解码器对上一个已确定的待装箱物体的几何信息、目标容器的当前高度图进行卷积运算，将运算结果映射到高维空间，得到对应的第二高维向量；根据第一高维向量和第二高维向量，确定每个待装箱物体在预设状态下的装箱概率值；根据装箱概率值确定当前待装箱物体和对应的搬运状态；根据确定的所有当前待装箱物体，生成待装箱物体集合的物体装箱序列。根据初始容器中待装箱物体的优先级图和目标容器的当前高度图，依次确定当前待装箱物体和对应的搬运状态，确保了转移装箱的合理性和准确性。
在一个实施例中，如图9所示，提供了一种确定目标摆放策略方法，以该方法应用于终端为例进行说明，包括以下步骤：
步骤902，根据装箱目标、物体装箱序列，各待装箱物体的搬运状态，计算待装箱物体集 合对应预设摆放策略集中每个预设摆放策略的装箱目标值。
其中，装箱目标包括密度目标、锥形程度目标或稳定性目标中至少一种；装箱目标值包括紧密度值、锥形程度值或稳定性值中至少一种。装箱目标值R可以但不仅限于用以下函数关系式来表示：
R＝(C+P+S)/3
其中，紧密度值C定义为所有待装箱物体的总面积A packed和目标容器的当前最高高度和容器宽度A rect确定的矩形面积之间的比例，如图10A所示，
Figure PCTCN2020090524-appb-000006
紧密度的度量倾向于将待装箱物体紧密装箱，当且仅当所有待装箱物体完全填满矩形区域时，紧密度值C为1。可选地，紧密度值C的取值范围可为0～1。
锥状程度值P定义为所有待装箱物体的总面积A packed和所有待装箱物体在容器底部方向的投影面积A proj之间的比例，如图10B所示，
Figure PCTCN2020090524-appb-000007
投影区域之外的区域允许被将来的待装箱物体填充，若当前状态能使后续待装箱物体紧密装箱时，锥状程度值P越大。可选地，锥状程度值P的取值范围可为0～1。
稳定性值S定义为稳定物体的数量N stable与所有已装箱物体数量N packed的比；可选地，稳定性值S的取值范围可为0～1。装箱过程中，每个物体的稳定性判断包括：首先通过几何分析，找出该物体在该位置时其下表面与已装箱物体的边缘的所有接触点；找出该物体下表面的中心点；将接触点两两相连，将组成一个轮廓区域，该区域在二维物体情况下是一条线段，在三维物体情况下是一个多边形；判断中心点是否在该区域中，若中心点是否在该区域中，则认为该物体是稳定，否则认为该物体是不稳定，如图10C和10D所示，图10C的稳定性值
Figure PCTCN2020090524-appb-000008
大于图10D的稳定性值
Figure PCTCN2020090524-appb-000009
即图10C所示的装箱效果比10D的装箱效果好。
步骤904，根据装箱目标值，从预设摆放策略集中确定待装箱物体集合的目标摆放策略。
具体地，装箱目标包括密度目标、锥形程度目标或稳定性目标中至少一种；根据装箱目标、物体装箱序列，各待装箱物体的搬运状态，计算待装箱物体集合对应预设摆放策略集中每个预设摆放策略的装箱目标值包括：根据物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，将待装箱物体集合中每个待装箱物体依次装箱到目标容器中；计算装箱后目标容器对应的紧密度值、锥形程度值和稳定性值；根据紧密度值、锥形程度值和稳定性值计算预设摆放策略集中每个预设摆放策略对应的装箱目标值。从计算得到的装箱目标值中获取装箱目标值最大的预设摆放策略，将装箱目标值最大的预设摆放策略作为待装箱物体集合的目标摆放策略。
上述确定目标摆放策略中，根据装箱目标、物体装箱序列和各装箱物体的搬运状态，计算待装箱物体集合对应预设摆放策略集中每个预设摆放策略的装箱目标值；根据装箱目标值，从预设摆放策略集中确定待装箱物体集合的目标摆放策略。通过计算待装箱物体集合对应预设摆 放策略集中每个预设摆放策略的装箱目标值，比较装箱目标值的大小，确定最佳预设摆放策略，从而提高目标容器的空间利用率。
在另一个实施例中，如图11所示，提供了一种物体转移装箱过程策略生成方法，以该方法应用于终端为例进行说明，包括以下步骤：
步骤1102，获取初始容器中待装箱物体集合中每个待装箱物体的初始位置信息和初始几何信息。
步骤1104，根据初始位置信息和初始几何信息，对待装箱物体集合中每个待装箱物体进行几何分析，确定待装箱物体集合的优先级图。
步骤1106，对优先级图进行编码，得到每个待装箱物体在预设状态下的几何信息和依赖信息。
步骤1108，神经网络中的编码器对每个待装箱物体的几何信息和依赖信息进行卷积运算，将运算结果映射到高维空间，得到对应的第一高维向量。
步骤1110，神经网络中的解码器对上一个已确定的待装箱物体的几何信息、目标容器的当前高度图进行卷积运算，将运算结果映射到高维空间，得到对应的第二高维向量。
步骤1112，根据装箱概率值确定当前待装箱物体和对应的搬运状态，将确定的当前待装箱物体从待装箱物体集合移除，更新待装箱物体集合的优先级图和目标容器的当前高度图。
步骤1114，生成待装箱物体集合的物体装箱序列。
可选地，在训练TAP-Net神经网络时，TAP-Net的容量根据训练对象的数量N来确定，当初始容器中有M(M>N)个待装箱物体时，可以选择N个物体的集合(Ω)用于TAP-Net进行转移和装箱，当确定当前待装箱物体和对应的搬运状态后，将当前待装箱物体从初始容器中删除，并将另一个待装箱物体添加到Ω中，更新待装箱物体集合的优先级图和目标容器的当前高度图；重复该过程，直至确定完所有物体的装箱顺序，得到待装箱物体集合的物体装箱序列。如图12所示，初始容器中有6个待装箱物体，当TAP-Net使用4个物体训练而要处理6个待装箱的物体时，首先选择4个物体进入初始集Ω，然后使用TAP-Net选择第一个待装箱的物体，如物体C。取出待装箱物体C后，另一个物体如物体F被添加到Ω中。重复该过程，直至装箱完所有物体。
步骤1116，根据装箱目标、物体装箱序列和各装箱物体的搬运状态，计算待装箱物体集合对应预设摆放策略集中每个预设摆放策略的装箱目标值。
步骤1118，根据装箱目标值，从预设摆放策略集中确定待装箱物体集合的目标摆放策略。
可选地，在一个实施例中，待装箱物体集合中每个待装箱物体需要装箱到不同的目标容器中，获取每个待装箱物体对应的目标容器编号，将每个待装箱物体对应的目标容器编号和每个待装箱物体的几何信息和依赖信息进行卷积运算，将运算结果映射到高维空间，得到对应的第一高维向量。通过神经网络中的解码器对上一个已确定的待装箱物体的几何信息、每个目标容器编号对应目标容器的当前高度图进行卷积运算，将运算结果映射到高维空间，得到对应的第二高维向量。根据装箱概率值确定当前待装箱物体、对应的搬运状态以及对应的目标容器，将 确定的当前待装箱物体从待装箱物体集合移除，更新待装箱物体集合的优先级图和目标容器的当前高度图；生成待装箱物体集合在不同目标容器中的物体装箱序列。执行根据装箱目标、物体装箱序列和各装箱物体的搬运状态，计算待装箱物体集合对应预设摆放策略集中每个预设摆放策略的装箱目标值步骤。如图13所示，初始容器中初始配置的中包括待装箱物体A-J，执行上述步骤，把待装箱物体分别装箱到对应的目标容器1和目标容器2中。
上述物体转移装箱过程策略生成方法，通过获取初始容器中待装箱物体集合中每个待装箱物体的初始位置信息和初始几何信息；根据初始位置信息和初始几何信息对待装箱物体集合中每个待装箱物体进行几何分析，确定待装箱物体集合的优先级图；对优先级图进行编码，得到每个待装箱物体在预设状态下的几何信息和依赖信息；通过神经网络中的编码器对每个待装箱物体的几何信息和依赖信息进行卷积运算，将运算结果映射到高维空间得到对应的第一高维向量；通过神经网络中的解码器对上一个已确定的待装箱物体的几何信息、目标容器的当前高度图进行卷积运算，将运算结果映射到高维空间，得到对应的第二高维向量；根据装箱概率值确定当前待装箱物体和对应的搬运状态，将确定的当前待装箱物体从待装箱物体集合移除，更新待装箱物体集合的优先级图和目标容器的当前高度图；生成待装箱物体集合的物体装箱序列。
根据装箱目标、物体装箱序列和各装箱物体的搬运状态，计算待装箱物体集合对应预设摆放策略集中每个预设摆放策略的装箱目标值；根据装箱目标值，从预设摆放策略集中确定待装箱物体集合的目标摆放策略。通过确定待装箱物体集合的优先级图，根据优先级图对待装箱物体之间的约束进行转移规划，提高了转移装箱效率以及目标容器的空间利用率。
在一个实施例中，物体装箱为三维物体装箱，通过获取初始容器中待装箱物体集合的优先级图；优先级图用于描述待装箱物体集合中每个待装箱物体之间的依赖约束；通过对优先级图进行编码，得到每个待装箱物体在预设状态下的几何信息和依赖信息；通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态；生成待装箱物体集合的物体装箱序列；根据装箱目标、物体装箱序列、各装箱物体的搬运状态和预设摆放策略，确定待装箱物体集合的装箱过程策略。通过确定待装箱物体集合的优先级图，根据优先级图对待装箱物体之间的约束进行转移规划，提高了转移装箱效率。可选地，三维物体转移装箱过程策略生成装置的具体限定可以参见上文中二维物体转移装箱过程策略生成方法的限定，在此不再赘述。
应该理解的是，虽然图1、图5、图7、图9、图11的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，图1、图5、图7、图9、图11中的至少一部分步骤可以包括多个步骤或者多个阶段，这些步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤中的步骤或者阶段的至少一部分轮流或者交替地执行。
在一个实施例中，如图14所示，提供了一种物体转移装箱过程策略生成装置，包括：获 取模块1402、编码模块1404、卷积模块1406和确定模块1408，其中：
获取模块1402，用于获取初始容器中待装箱物体集合的优先级图,其中优先级图用于描述待装箱物体集合中每个待装箱物体之间的依赖约束。
编码模块1404，用于对优先级图进行编码，得到每个待装箱物体在预设状态下的几何信息和依赖信息。
卷积模块1406，用于通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态，生成待装箱物体集合的物体装箱序列。
确定模块1408，用于根据装箱目标、物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，确定待装箱物体集合的装箱过程策略。
上述物体转移装箱过程策略生成装置中，通过获取初始容器中待装箱物体集合的优先级图，其中优先级图用于描述待装箱物体集合中每个待装箱物体之间的依赖约束；通过对优先级图进行编码，得到每个待装箱物体在预设状态下的几何信息和依赖信息；通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态，得到待装箱物体集合的物体装箱序列；根据装箱目标、物体装箱序列、各装箱物体的搬运状态和预设摆放策略，确定待装箱物体集合的装箱过程策略。通过确定待装箱物体集合的优先级图，根据优先级图对待装箱物体之间的约束进行转移规划，解决了物体转移和装箱问题，进一步提高了物体转移装箱效率。
在一个实施例中，如图15所示，所述获取模块1402，进一步用于获取初始容器中待装箱物体集合中每个待装箱物体的初始位置信息和初始几何信息。
所述物体转移装箱过程策略生成装置还包括：
几何分析模块1410，用于根据初始位置信息和初始几何信息对待装箱物体集合中每个待装箱物体进行几何分析，确定待装箱物体集合的优先级图；和
遍历模块1412，用于根据初始位置信息和初始几何信息，遍历待装箱物体集合中每个待装箱物体的每个方向进行，确定每个方向的限制状态；根据限制状态，确定待装箱物体集合的优先级图。
在一个实施例中，卷积模块1406还用于通过神经网络中的编码器对每个待装箱物体的几何信息和依赖信息进行卷积运算，将运算结果映射到高维空间，得到对应的第一高维向量；通过神经网络中的解码器对上一个已确定的待装箱物体的几何信息、目标容器的当前高度图进行卷积运算，将运算结果映射到高维空间，得到对应的第二高维向量；
根据第一高维向量和第二高维向量，确定每个待装箱物体在预设状态下的装箱概率值；根据装箱概率值确定当前待装箱物体和对应的搬运状态，得到待装箱物体集合的物体装箱序列。
移除模块1414，用于将确定的当前待装箱物体从待装箱物体集合移除。
更新模块1416，用于更新待装箱物体集合的优先级图和目标容器的当前高度图，跳转到初始容器中待装箱物体集合的优先级图步骤。
计算模块1418，用于根据装箱目标、物体装箱序列和各装箱物体的搬运状态，计算待装箱物体集合对应预设摆放策略集中每个预设摆放策略的装箱目标值。
在一个实施例中，计算模块1418还用于根据物体装箱序列、各装箱物体的搬运状态和预设摆放策略，将待装箱物体集合中每个待装箱物体依次装箱到目标容器中；计算装箱后目标容器对应的紧密度值、锥形程度值和稳定性值。
在一个实施例中，确定模块1408还用于根据装箱目标值，从预设摆放策略集中确定待装箱物体集合的目标摆放策略；其中装箱目标包括密度目标、锥形程度目标或稳定性目标中至少一种。
在一个实施例中，确定模块1408还用于根据紧密度值、锥形程度值和稳定性值，计算预设摆放策略集中每个预设摆放策略对应的装箱目标数值。
上述物体转移装箱过程策略生成装置中，通过获取初始容器中待装箱物体集合中每个待装箱物体的初始位置信息和初始几何信息；根据初始位置信息和初始几何信息对待装箱物体集合中每个待装箱物体进行几何分析，确定待装箱物体集合的优先级图；通过对优先级图进行编码，得到每个待装箱物体在预设状态下的几何信息和依赖信息；通过神经网络中的编码器对每个待装箱物体的几何信息和依赖信息进行卷积运算，将运算结果映射到高维空间，得到对应的第一高维向量；通过神经网络中的解码器对上一个已确定的待装箱物体的几何信息、目标容器的当前高度图进行卷积运算，将运算结果映射到高维空间，得到对应的第二高维向量；根据第一高维向量和第二高维向量，确定每个待装箱物体在预设状态下的装箱概率值；根据装箱概率值确定当前待装箱物体和对应的搬运状态，将确定的当前待装箱物体从待装箱物体集合移除；更新待装箱物体集合的优先级图和目标容器的当前高度图，跳转到获取初始容器中待装箱物体集合的优先级图步骤；生成待装箱物体集合的物体装箱序列。
根据物体装箱序列、各装箱物体的搬运状态和预设摆放策略，将待装箱物体集合中每个待装箱物体依次装箱到目标容器中；根据装箱目标，计算装箱后目标容器对应的紧密度值、锥形程度值和稳定性值；根据紧密度值、锥形程度值和稳定性值计算预设摆放策略集中每个预设摆放策略对应的装箱目标数值；根据装箱目标值，从预设摆放策略集中确定待装箱物体集合的目标摆放策略。通过确定待装箱物体集合的优先级图，根据优先级图对待装箱物体之间的约束进行转移规划，提高了转移装箱效率以及目标容器的空间利用率。
关于物体转移装箱过程策略生成装置的具体限定可以参见上文中对于物体转移装箱过程策略生成方法的限定，在此不再赘述。上述物体转移装箱过程策略生成装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是终端，其内部结构如图16所示。该计算机设备包括通过系统总线连接的处理器、存储器、通信接口、显示屏和输入装置。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性 存储介质、内存储器。该非易失性存储介质存储有操作系统和计算机程序。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的通信接口用于与外部的终端进行有线或无线方式的通信，无线方式可通过WIFI、运营商网络、NFC(近场通信)或其他技术实现。该计算机程序被处理器执行时以实现一种物体转移装箱过程策略生成方法。该计算机设备的显示屏可以是液晶显示屏或者电子墨水显示屏，该计算机设备的输入装置可以是显示屏上覆盖的触摸层，也可以是计算机设备外壳上设置的按键、轨迹球或触控板，还可以是外接的键盘、触控板或鼠标等。
本领域技术人员可以理解，图16中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，还提供了一种计算机设备，包括存储器和处理器，存储器中存储有计算机程序，该处理器执行计算机程序时实现上述各方法实施例中的步骤。
在一个实施例中，提供了一种计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现上述各方法实施例中的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和易失性存储器中的至少一种。非易失性存储器可包括只读存储器(Read-Only Memory，ROM)、磁带、软盘、闪存或光存储器等。易失性存储器可包括随机存取存储器(Random Access Memory，RAM)或外部高速缓冲存储器。作为说明而非局限，RAM可以是多种形式，比如静态随机存取存储器(Static Random Access Memory，SRAM)或动态随机存取存储器(Dynamic Random Access Memory，DRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (15)

  1. 一种物体转移装箱过程策略生成方法，包括：
    获取初始容器中待装箱物体集合的优先级图，其中所述优先级图用于描述所述待装箱物体集合中每个待装箱物体之间的依赖约束；
    对所述优先级图进行编码，得到所述每个待装箱物体在预设状态下的几何信息和依赖信息；
    通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态；
    生成所述待装箱物体集合的物体装箱序列；
    根据装箱目标、所述物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，确定所述待装箱物体集合的装箱过程策略。
  2. 根据权利要求1所述的方法，其中，所述获取初始容器中待装箱物体集合的优先级图包括：
    获取初始容器中待装箱物体集合中每个待装箱物体的初始位置信息和初始几何信息；
    根据所述初始位置信息和所述始几何信息对所述待装箱物体集合中每个待装箱物体进行几何分析，确定所述待装箱物体集合的优先级图。
  3. 根据权利要求2所述的方法，其中，所述根据所述初始位置信息和所述始几何信息对所述待装箱物体集合中每个待装箱物体进行几何分析，确定所述待装箱物体集合的优先级图包括：
    根据所述初始位置信息和所述始几何信息，遍历所述待装箱物体集合中每个待装箱物体的每个方向，确定每个方向的限制状态；
    根据所述限制状态，确定所述待装箱物体集合的优先级图。
  4. 根据权利要求1所述的方法，其中，所述通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、动态信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态，包括：
    神经网络中的编码器对所述每个待装箱物体的几何信息和依赖信息进行卷积运算，将运算结果映射到高维空间，得到对应的第一高维向量；
    神经网络中的解码器对上一个已确定的待装箱物体的几何信息、目标容器的当前高度图进行卷积运算，将运算结果映射到高维空间，得到对应的第二高维向量；
    根据所述第一高维向量和所述第二高维向量，确定所述每个待装箱物体在预设状态下的装箱概率值；
    根据所述装箱概率值确定当前待装箱物体和对应的搬运状态。
  5. 根据权利要求1或4所述的方法，其中，在所述确定当前待装箱物体和对应的搬运状态之后，和所述生成所述待装箱物体集合的物体装箱序列之前，所述方法还包括：
    将确定的当前待装箱物体从所述待装箱物体集合移除；
    更新所述待装箱物体集合的优先级图和所述目标容器的当前高度图；
    跳转到所述获取初始容器中待装箱物体集合的优先级图的步骤。
  6. 根据权利要求1所述的方法，其中，所述根据装箱目标、所述物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，确定所述当前装箱过程策略包括：
    根据装箱目标、所述物体装箱序列和各待装箱物体的搬运状态，计算所述待装箱物体集合对应预设摆放策略集中每个预设摆放策略的装箱目标值；
    根据所述装箱目标值，从所述预设摆放策略集中确定所述待装箱物体集合的目标摆放策略。
  7. 根据权利要求6所述的方法，其中，所述装箱目标包括密度目标、锥形程度目标或稳定性目标中的至少一种；所述根据装箱目标、所述物体装箱序列和各待装箱物体的搬运状态，计算所述待装箱物体集合对应预设摆放策略集中每个预设摆放策略的装箱目标值包括:
    根据所述物体装箱序列、各待装箱物体的搬运状态和所述预设摆放策略，将所述待装箱物体集合中每个待装箱物体依次装箱到所述目标容器中；
    根据所述装箱目标，计算装箱后所述目标容器对应的紧密度值、锥形程度值或稳定性值的至少一种；
    根据所述紧密度值、锥形程度值或稳定性值的至少一种，计算预设摆放策略集中每个预设摆放策略对应的装箱目标值。
  8. 根据权利要求1所述的方法，其中，在所述确定当前待装箱物体和对应的搬运状态之后，和所述生成所述待装箱物体集合的物体装箱序列之前，所述方法进一步包括：
    判断所述待装箱物体集合中的所有待装箱物体的装箱顺序和对应的搬运状态是否全部都已确定；
    如果所述待装箱物体集合中的所有待装箱物体的装箱顺序和对应的搬运状态没有全都确定，则跳转到获取初始容器中待装箱物体集合的优先级图的步骤。
  9. 根据权利要求1或4所述的方法，其中，在所述确定当前待装箱物体和对应的搬运状态之后，和所述生成所述待装箱物体集合的物体装箱序列之前，所述方法还包括：
    将确定的当前待装箱物体从所述待装箱物体集合移除；
    判断所述待装箱物体集合中的所有待装箱物体的装箱顺序和对应的搬运状态是否全部都已确定；
    如果所述待装箱物体集合中的所有待装箱物体的装箱顺序和对应的搬运状态没有全都确定，则更新所述待装箱物体集合的优先级图和所述目标容器的当前高度图；
    跳转到所述获取初始容器中待装箱物体集合的优先级图的步骤。
  10. 根据权利要求7所述的方法，其中，每个待装箱物体的稳定性判断包括：
    通过几何分析，找出所述待装箱物体的下表面与已装箱物体的边缘的所有接触点，和所述待装箱物体下表面的中心点；
    将接触点两两相连，组成一个轮廓区域；
    判断所述中心点是否位于所述轮廓区域中；
    若所述中心点位于所述轮廓区域中，则所述待装箱物体是稳定的，否则所述待装箱物体是不稳定的。
  11. 根据权利要求1所述的方法，其中，所述待装箱物体为二维待装箱物体或三维待装箱物体。
  12. 根据权利要求1所述的方法，其中，每个所述待装箱物体在预设状态下的依赖信息用0和1表示，其中0代表无约束限制，1代表有约束限制。
  13. 一种物体转移装箱过程策略生成装置，包括：
    获取模块，用于获取初始容器中待装箱物体集合的优先级图，其中所述优先级图用于描述所述待装箱物体集合中每个待装箱物体之间的依赖约束；
    编码模块，用于对所述优先级图进行编码，得到所述每个待装箱物体在预设状态下的几何信息和依赖信息；
    卷积模块，用于通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态，生成待装箱物体集合的物体装箱序列；
    确定模块，用于根据装箱目标、所述物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，确定所述待装箱物体集合的装箱过程策略。
  14. 一种计算机设备，包括存储器和处理器，所述存储器存储计算机程序，其中，所述处理器执行所述计算机程序时，执行以下步骤：
    获取初始容器中待装箱物体集合的优先级图，其中所述优先级图用于描述所述待装箱物体集合中每个待装箱物体之间的依赖约束；
    对所述优先级图进行编码，得到所述每个待装箱物体在预设状态下的几何信息和依赖信息；
    通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态；
    生成待装箱物体集合的物体装箱序列；
    根据装箱目标、所述物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，确定所述待装箱物体集合的装箱过程策略。
  15. 一种计算机可读存储介质，其上存储有计算机程序，其中，所述计算机程序被处理器执行时，执行以下步骤：
    获取初始容器中待装箱物体集合的优先级图，其中所述优先级图用于描述所述待装箱物体集合中每个待装箱物体之间的依赖约束；
    对所述优先级图进行编码，得到所述每个待装箱物体在预设状态下的几何信息和依赖信息；
    通过神经网络依次对待装箱物体集合中每个待装箱物体的几何信息、依赖信息和目标容器的当前高度图进行卷积运算，确定当前待装箱物体和对应的搬运状态；
    生成待装箱物体集合的物体装箱序列；
    根据装箱目标、所述物体装箱序列、各待装箱物体的搬运状态和预设摆放策略，确定所述待装箱物体集合的装箱过程策略。
PCT/CN2020/090524 2020-05-06 2020-05-15 物体转移装箱过程策略生成方法、装置、计算机设备 WO2021223262A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/947,933 US20230011757A1 (en) 2020-05-06 2022-09-19 Method and apparatus for generating strategy of object transport-and-pack process, and computer device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010371670.X 2020-05-06
CN202010371670.XA CN111598316B (zh) 2020-05-06 2020-05-06 物体转移装箱过程策略生成方法、装置、计算机设备

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/947,933 Continuation US20230011757A1 (en) 2020-05-06 2022-09-19 Method and apparatus for generating strategy of object transport-and-pack process, and computer device

Publications (1)

Publication Number Publication Date
WO2021223262A1 true WO2021223262A1 (zh) 2021-11-11

Family

ID=72185621

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/090524 WO2021223262A1 (zh) 2020-05-06 2020-05-15 物体转移装箱过程策略生成方法、装置、计算机设备

Country Status (3)

Country Link
US (1) US20230011757A1 (zh)
CN (1) CN111598316B (zh)
WO (1) WO2021223262A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112407725B (zh) * 2020-11-20 2022-11-22 深圳市海柔创新科技有限公司 货物取出方法、装置、机器人、仓储系统及存储介质
CN112884410B (zh) * 2021-02-26 2023-01-13 深圳蓝胖子机器智能有限公司 装箱方法、电子设备以及存储介质
CN113516300B (zh) * 2021-06-18 2024-03-29 北京控制工程研究所 一种多任务多约束高可信实时自主决策优化方法及系统
CN113705860B (zh) * 2021-08-05 2023-10-17 北京航空航天大学 一种实时智能强鲁棒性的多形状制造件排样优化方法及系统
CN116228838B (zh) * 2023-05-10 2024-03-08 深圳大学 一种基于视觉检测的物体装箱强化学习方法及相关装置

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108009675A (zh) * 2017-11-28 2018-05-08 上海量科电子科技有限公司 货物装箱方法、装置及系统
CN108268878A (zh) * 2016-12-30 2018-07-10 乐视汽车（北京）有限公司 三维全卷积网络实现设备
CN109685278A (zh) * 2018-12-28 2019-04-26 深圳蓝胖子机器人有限公司 装箱方法、设备和计算机可读存储介质
WO2019152337A1 (en) * 2018-02-01 2019-08-08 Oracle International Corporation Energy program communication control system and method based on load shape analysis
CN110210685A (zh) * 2019-06-12 2019-09-06 广东三维家信息科技有限公司 物流装箱方法及装置
CN110288287A (zh) * 2019-05-23 2019-09-27 深圳市启海仓储有限公司 一种仓储货物的分理方法及系统
CN110929792A (zh) * 2019-11-27 2020-03-27 深圳市商汤科技有限公司 图像标注方法、装置、电子设备及存储介质
CN111099363A (zh) * 2020-01-09 2020-05-05 湖南视比特机器人有限公司 码垛方法、码垛系统和存储介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102415508B1 (ko) * 2017-03-28 2022-07-01 삼성전자주식회사 컨볼루션 신경망 처리 방법 및 장치
CN109345017A (zh) * 2018-10-08 2019-02-15 南京航空航天大学 一种考虑装箱约束的车间物料配送优化方法
CN110852324A (zh) * 2019-08-23 2020-02-28 上海撬动网络科技有限公司 一种基于深度神经网络集装箱箱号检测方法
CN110555424A (zh) * 2019-09-10 2019-12-10 深圳前海微众银行股份有限公司 港口集装箱布局检测方法、装置、设备及可读存储介质

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108268878A (zh) * 2016-12-30 2018-07-10 乐视汽车（北京）有限公司 三维全卷积网络实现设备
CN108009675A (zh) * 2017-11-28 2018-05-08 上海量科电子科技有限公司 货物装箱方法、装置及系统
WO2019152337A1 (en) * 2018-02-01 2019-08-08 Oracle International Corporation Energy program communication control system and method based on load shape analysis
CN109685278A (zh) * 2018-12-28 2019-04-26 深圳蓝胖子机器人有限公司 装箱方法、设备和计算机可读存储介质
CN110288287A (zh) * 2019-05-23 2019-09-27 深圳市启海仓储有限公司 一种仓储货物的分理方法及系统
CN110210685A (zh) * 2019-06-12 2019-09-06 广东三维家信息科技有限公司 物流装箱方法及装置
CN110929792A (zh) * 2019-11-27 2020-03-27 深圳市商汤科技有限公司 图像标注方法、装置、电子设备及存储介质
CN111099363A (zh) * 2020-01-09 2020-05-05 湖南视比特机器人有限公司 码垛方法、码垛系统和存储介质

Also Published As

Publication number Publication date
CN111598316A (zh) 2020-08-28
CN111598316B (zh) 2023-03-24
US20230011757A1 (en) 2023-01-12

Similar Documents

Publication Publication Date Title
WO2021223262A1 (zh) 物体转移装箱过程策略生成方法、装置、计算机设备
US20210110599A1 (en) Depth camera-based three-dimensional reconstruction method and apparatus, device, and storage medium
US10417821B2 (en) Method of simplifying a geometry model
US10346996B2 (en) Image depth inference from semantic labels
US9959670B2 (en) Method for rendering terrain
US20210166464A1 (en) Density coordinate hashing for volumetric data
US9412074B2 (en) Optimized trotterization via multi-resolution analysis
TW201805843A (zh) 用於在三維虛擬製造環境中電氣特性模型化的系統及方法
CN110276829A (zh) 通过多尺度体素哈希处理的三维表示
KR20140060454A (ko) 포인트 클라우드에서의 포인트 백필링
AU2018253470B2 (en) Device and method for extracting terrain boundary
US11367263B1 (en) Image-guided three dimensional modeling
JP6380051B2 (ja) 有限要素演算プログラム、有限要素演算装置および有限要素演算方法
KR20120122957A (ko) 네비게이션 장치, 고도 좌표를 결정하는 방법 및 데이터베이스를 생성하는 방법
CN104952086B (zh) 用于管理结构数据的设备及方法
WO2021218037A1 (zh) 目标检测方法、装置、计算机设备和存储介质
CN116227209A (zh) 一种点云数据多维线性差值方法、终端设备及存储介质
WO2022048468A1 (zh) 平面轮廓识别方法、装置、计算机设备和存储介质
CN114596423A (zh) 基于虚拟场景网格化的模型渲染方法、装置和计算机设备
KR101642598B1 (ko) 공간 분할 방법, 공간 분할 장치 및 기억 매체
JP7352706B2 (ja) 貨物計数方法及び装置、コンピュータ装置、並びに記憶媒体
CN116521782A (zh) 一种数据资产地图获取方法、装置、设备、介质及产品
CN117609412B (zh) 一种基于网络结构信息的空间对象关联方法及装置
Behar et al. Dynamic Minkowski sums under scaling
US20240104790A1 (en) Optimizing grid-based compute graphs

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20934649

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 13/03/2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20934649

Country of ref document: EP

Kind code of ref document: A1