A dynamic positioning method for vertical warehouse pallets and an electric gripper device

Abstract

The present invention belongs to the field of vertical warehouse grasping and positioning technology, and specifically discloses a dynamic positioning method for vertical warehouse pallets, which includes the following steps: S1: obtaining the row and column information of the target pallet; S2: constructing an original coordinate system, and calculating the original coordinate data of the object pick-up and placement points through the row and column information of the pallet; S3: identifying the teaching points through QR codes, and using the least squares method to perform curve fitting on the teaching points to generate a dynamic coordinate system for the pallet, and further obtaining the original coordinate data of the object and correspondingly updating the coordinate data in the dynamic coordinate system to complete dynamic positioning. The present invention can achieve precise positioning of the object pick-up and placement points of vertical warehouse pallets that are non-rectangular and have curved long sides and parallelograms; the information collection method of the present invention only requires posting four QR codes to achieve precise positioning of the material pick-up and placement positions of the entire pallet.

Description

Technical Field

[0001] The present invention belongs to the technical field of vertical warehouse grasping and positioning, and in particular relates to a dynamic positioning method for a vertical warehouse pallet and an electric gripper device. Background Art

[0002] With rising labor costs and advancements in automation technology, the use of automated robotics is becoming increasingly widespread in industry. Grasping and placing objects is a typical warehousing task, requiring extremely high precision from automated equipment.

[0003] High-bay warehouses, commonly used in warehousing, are often used in conjunction with pallets. They combine the advantages of stacking and high-rise racking, utilizing space in a compact manner by increasing stacking height while protecting stored goods from damage. Compared to traditional warehousing models, high-bay warehouses offer efficient space utilization. By changing storage locations, items can be relocated, making them more convenient for access. However, accessing and placing items on each pallet requires significant labor costs, making the process from warehouse entrance to accessing items a pressing issue for the automation industry.

[0004] Given the high precision and continuous operation of robots, many solutions utilize industrial robotic workstations to perform object placement. However, when multiple intelligent vertical warehouses are installed, they cannot be arranged in parallel, and the relative positions of the pallets within the same warehouse vary. Therefore, when the robotic arm picks and places objects from different pallets in the same warehouse, locating the object placement point becomes a major challenge.

[0005] Current methods mostly use static teaching methods, manually marking the positions of all pallets and entering the position information into the robot system. When the robot grasps, the position is retrieved from the system. However, in a three-dimensional warehouse, the rigid connectors between pallets are prone to deformation. Due to the squeezing of the three-dimensional warehouse's internal handling devices on both sides, the entire pallet deforms, and the overall shape becomes a non-rectangular parallelogram. There is also a bend in the middle of the long side of the parallelogram, resulting in large irregularities and differences in the pick-and-place positions of this type of pallet and conventional rectangular pallets. This phenomenon introduces errors in the robot arm's grasping. Therefore, the existing standard equidistant rectangular pallet point calculation method cannot match the non-rectangular parallelogram vertical warehouse pallets with curved long sides. Summary of the Invention

[0006] The present invention provides a dynamic positioning method for a vertical warehouse pallet, which is used to solve the problem that the current three-dimensional warehouse will deform after long-term use and the existing algorithm cannot match the vertical warehouse pallet that is non-rectangular and has a parallelogram shape with curved long sides.

[0007] In order to solve the above technical problems, the technical solution of the present invention is: a dynamic positioning method for a vertical warehouse pallet, which includes the following steps:

[0008] S1: Get the row and column information of the target tray;

[0009] S2: Construct the original coordinate system and calculate the original coordinate data of the object pick-up and placement points based on the row and column information of the pallet;

[0010] S3: Identify the teaching points through the QR code, and use the least squares method to perform curve fitting on the teaching points to generate the dynamic coordinate system of the pallet. Further obtain the original coordinate data of the object and update the coordinate data accordingly in the dynamic coordinate system to complete dynamic positioning.

[0011] In a preferred embodiment of the present invention, step S3 includes:

[0012] S31: Select a teaching point in the original coordinate system, determine the actual position information of the teaching point in the original coordinate system by identifying the QR code corresponding to the teaching point, fit the coordinate point based on deep reinforcement learning, and construct a dynamic coordinate system;

[0013] S32: establishing a transformation matrix between the dynamic coordinate system and the original coordinate system;

[0014] S33: Select at least three teaching points in the dynamic coordinate system for fitting, and obtain the object pick-up and placement points based on the row offset and the point data of each column in the first row;

[0015] S34: Generate updated coordinates of the object placement location in the dynamic coordinate system through the transformation matrix, and obtain coordinate information of the updated coordinates in the original coordinate system.

[0016] In a preferred embodiment of the present invention, the method for fitting coordinate points through deep reinforcement learning in step S33 includes the following steps:

[0017] First, the position data of all teaching points are collected several times, and Gaussian noise is added to the position data;

[0018] Then, a neural network with two hidden layers is constructed based on the above data;

[0019] Finally, according to the neural network, the state space, action space and reward function are obtained.

[0020] In a preferred embodiment of the present invention, the state space, action space and reward function are obtained by the following formula:

[0021] S=[x i ,y i , z i ], i=(0, 1, 2, 3, 4) (1)

[0022] Among them, the state space S is a set that describes the robot state, including all possible positions and states of the robot. x, y, and z represent the coordinates of the three axes in the space respectively. i represents the serial number of the teaching point, which refers to a sequence of positions that the robot needs to reach. When i = 0, it indicates the starting position of the robot in the motion trajectory, that is, the initial state of the robot.

[0023] A=[x a ,y a , z a ] (2)

[0024] Where A represents a point in the three-dimensional space corresponding to each action in the action space. a represents the current spatial position of the point. The three components x, y, and z represent the x, y, and z coordinates of the robot in the Cartesian coordinate system respectively;

[0025] R=K*((x a -x i ) 2 +(y a -y i ) 2 +(z a -z i ) 2 ) (3)

[0026] Among them, R is the reward function and K is a constant.

[0027] In a preferred embodiment of the present invention, the step S31 further includes optimizing the reward function, wherein the optimization step is to sequentially calculate equations 4, 5, 6, and 7 to obtain an optimized reward function J;

[0028]

[0029] Among them, V represents the state value function, d represents the probability distribution of the state, Q represents the action state value function, and pi represents the state transition function.

[0030]

[0031] Among them, it means The result after derivation is the state transfer function Derivation and gradient calculation can realize iterative calculation of reward function.

[0032]

[0033] in Indicates the gradient of the state value function.

[0034]

[0035] Among them, to represent the gradient function.

[0036] In a preferred embodiment of the present invention, in step S33, the calculation method of the pick-up and drop-off point data of each column in the first row specifically includes:

[0037] When there are two teaching points in the first row, calculate the average column distance, and calculate the coordinate values of each column in the first row based on the average column distance to obtain the point data of each column in the first row;

[0038] When there are at least three teaching points in the first row, select the teaching points at both ends to establish a dynamic coordinate system, use a curve to fit all the teaching points in the first row, calculate the average column distance between adjacent teaching points, and calculate the coordinate values of each column in the first row based on the average column distance to obtain the point data of each column in the first row;

[0039] The calculation method of row offset includes:

[0040] The number of teaching points in the first column is two. The inter-row offset unit vector between the two teaching points is calculated, and the row offset vector is calculated based on the row spacing and the offset unit.

[0041] In a preferred embodiment of the present invention, step S34 specifically includes:

[0042] According to the coordinates of the teaching point, calculate the point data and row offset vector of each column in the first row

[0043] Calculate the coordinate values of all the pick-up and place positions of the pallet based on the row offset vector and the coordinates of the points in each column of the first row;

[0044] Convert the dynamic coordinate system to the original coordinate system coordinates.

[0045] In a preferred embodiment of the present invention, the specific method for calculating the coordinate values of all the pick-up and placement positions of the pallet is as follows:

[0046] Generate the coordinate information of all the pick-and-place points of the pallet in the dynamic coordinate system based on the position information of each pick-and-place point in the first row and the offset vector between rows;

[0047] Multiply the coordinate information by the transformation matrix to obtain the position information of the pick-up and drop points in the original coordinate system.

[0048] The present invention also discloses an electric gripper device for picking and placing in a vertical warehouse, which includes: a control system that adopts the dynamic positioning method of the vertical warehouse pallet, a material grabbing electric gripper controlled by the control system to pick and place items from the vertical warehouse, and a 3D camera for identifying the QR code set on the vertical warehouse.

[0049] The technical solution provided by the present invention has the following advantages compared with the prior art:

[0050] When the present invention is working, compared with the existing standard equally spaced rectangular pallet point calculation method, the present invention can achieve precise positioning of the object picking and placing points of non-rectangular vertical warehouse pallets with parallelograms and curved long sides; compared with the existing robot pallet teaching method, the information collection method in the present invention only requires posting four QR codes to achieve precise positioning of the entire pallet material picking and placing position. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, they can also obtain drawings of other embodiments based on these drawings.

[0052] Figure 1 This is an overall flow chart of a dynamic positioning method for a vertical warehouse pallet according to one embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram of the teaching point position in a dynamic positioning method for a vertical warehouse pallet according to one embodiment of the present invention;

[0054] Figure 3 This is a flow chart for obtaining the position information of the pick-up and placement points in the original coordinate system in a dynamic positioning method for a vertical warehouse pallet described in one embodiment of the present invention. DETAILED DESCRIPTION

[0055] For ease of understanding, a vertical warehouse pallet dynamic positioning method is described below in conjunction with embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0056] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations and positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0057] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood broadly. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediary; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention.

[0058] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The accompanying drawings illustrate preferred embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present disclosure.

[0059] The equipment unit in the present invention mainly includes: an industrial serial robot, an electric material grabbing gripper, a 3D camera, and an intelligent vertical warehouse.

[0060] In this embodiment, the electric gripper for material grasping adopts parallel movement of fingers, and picks and places objects by moving the fingertips in parallel; the intelligent vertical warehouse contains multiple pallets, and pallets can be switched according to the selected pallet number; each pallet has a rectangular outer frame, and there are 49 rows and 10 columns of theoretically evenly spaced pick and place positions inside; because the actual situation is that the electric gripper is at the end of the robot, the robot is first raised to a safe height, moved horizontally into the vertical warehouse, positioned above the pick and place position, and then vertically descended to the object position, and the fingertips are moved in parallel to pick and place the object, and then it is lifted vertically upward to a safe position, and finally the robot arm is moved horizontally to leave the interior of the vertical warehouse. Therefore, when the robot picks and places objects, there is no need to consider the robot's motion interference, and it only needs to accurately locate the object position.

[0061] A dynamic positioning method for vertical warehouse pallets, wherein reference is made to Figure 1 As shown, the following steps are included.

[0062] 1. Generate and receive instructions

[0063] The intelligent vertical warehouse can send the pallet number and row and column instructions for entry and exit, and the controller receives the instructions.

[0064] 2. Calculate the coordinate data of the object pick-up and placement points through the pallet number and row and column information

[0065] 3. Get object placement information

[0066] A QR code is fixed at the teaching point, the teaching point is identified by the QR code, and a least squares method is used for curve fitting to generate a dynamic coordinate system to obtain the object placement position information. The specific steps include the following.

[0067] 3.1 Simplify the actual object placement situation into coordinate points in a two-dimensional plane, input model parameters and establish a simplified model.

[0068] 3.2 Establish the original coordinate system in the model according to the robot base coordinate position.

[0069] 3.3 Based on deep reinforcement learning DRL fitting coordinate points, build a dynamic coordinate system

[0070] Select teaching points P1, P2, P3, and P4 in the original coordinate system. In this embodiment, refer to Figure 2 As shown, the selected teaching points P1 (10th column, 1st row), P2 (5th column, 1st row), P3 (1st column, 1st row), and P4 (1st column, 49th row) are fixed at the teaching points. The original coordinate system is used as the reference coordinate system to identify the position of the QR code and obtain the spatial position information of the teaching points.

[0071] Based on deep reinforcement learning (DRL), fitting coordinate points includes the following steps.

[0072] (1) Collect 1000 sets of teaching point data and add Gaussian noise to the teaching data;

[0073] (2) Design a neural network with two hidden layers, where the number of neurons in the input layer is 15, the number of neurons in the hidden layer is 64, and the number of output neurons is 3;

[0074] (3) Based on the neural network, design the state space, action space and reward function.

[0075] The state space S is formula 1:

[0076] S=[x i ,y i , z i ], i=(0, 1, 2, 3, 4) (1)

[0077] Among them, represents the spatial position of the teaching point and the serial number of the currently selected point;

[0078] The action space A is formula 2:

[0079] A=[x a ,y a , z a ] (2)

[0080] Among them, a represents the current point spatial position;

[0081] The reward function R is formula 3:

[0082] r=K*((x a-x i ) 2 +(y a -y i ) 2 +(z a -z i ) 2 ) (3)

[0083] Where K is a constant

[0084] (4) Optimize the reward function value

[0085] The optimization method of the reward function value includes the following steps:

[0086] First, establish the gradient model, which is formula 4:

[0087]

[0088] Among them, V represents the state value function, d represents the probability distribution of the state, Q represents the action state value function, and pi represents the state transfer function

[0089] Secondly, the solution to obtain the gradient value through the gradient model

[0090]

[0091] Then, by deriving the state value function, we can get:

[0092]

[0093] Finally, we can update the strategy by approximating the change:

[0094]

[0095] (5) Update policy gradient

[0096] By deducing from the above, we can derive a method for optimizing the gradient function, measuring the teaching points from real samples, and using them to update our policy gradient. This method relies on a complete teaching process, which includes the following steps.

[0097] First, randomly initialize the policy parameters;

[0098] Secondly, in the strategy $pi θ Generate a series of teaching points. S1, A1, R2, S2, A2, ..., S T $;

[0099] Finally, for t=1, 2, ..., T, estimate the return rate G t , the strategy parameters can be updated according to the estimated return rate:

[0100] The present invention returns G t A baseline value is subtracted from to reduce the variance of the gradient estimate while keeping the bias unchanged.

[0101] In each round, the difference between the action value and the true value is calculated to give a reward value. The larger the reward value, the closer the neural network calculation result is to the true value. The smaller the reward value, the greater the deviation of the neural network calculation result. This method is used to correct the evolution direction of the neural network so as to achieve the goal.

[0102] 3.4 Establish the transformation matrix T between the dynamic coordinate system and the original coordinate system

[0103] 3.5 Obtaining the location information of the pick-up and drop-off points in the original coordinate system

[0104] Specifically, refer to Figure 3 As shown, the steps to obtain the position information of the pick-up and drop-off points in the original coordinate system are:

[0105] (1) Use the robot to teach the pallet to point to a specific point

[0106] (2) Record the coordinate values of the robot's teaching points

[0107] (3) Establish a simplified model for picking and placing a single pallet

[0108] (4) Establish pallet pick-up and placement point information based on the original coordinate system

[0109] (5) Calculate the point data and row offset vector of each column in the first row

[0110] The steps to calculate the coordinates of all points in the first row are:

[0111] ① Use the two teaching points P1 and P3 in the first row to establish a dynamic coordinate system

[0112] ②Use curve fitting to fit the three teaching points P1, P2, and P3 in the first row

[0113] ③ Use P1 and P2 to calculate the average column spacing of columns 5 to 10, and use P2 and P3 to calculate the average column spacing of columns 1 to 5

[0114] ④Calculate the coordinate values of each column in the first row based on the column spacing

[0115] ⑤ Calculate all point data in each column of the first row based on the curve function and the coordinate values of each column

[0116] The steps to calculate the row offset vector are:

[0117] ① Calculate the unit vector of the inter-row offset based on the two teaching points P3 and P4 in the first column

[0118] ②Calculate the row offset vector based on the row spacing and offset unit

[0119] (6) Calculate the coordinate values of all the pick-up and placement positions of the pallet based on the row offset vector and the coordinates of all the points in the first row

[0120] (7) Convert the dynamic coordinate system to the original coordinate system

[0121] According to the pick-and-place point information of the first row of objects and the inter-row offset vector, the coordinate information of all the pick-and-place points on the pallet in the dynamic coordinate system is generated, and the position information of the pick-and-place points in the original coordinate system is obtained through the coordinate information and the transformation matrix T.

[0122] Repeat steps 3.1 to 3.5 to update the coordinate system.

[0123] 4. According to the object pick-up and placement information, the robotic arm performs the object pick-up and placement action

[0124] At this point, the entire process of robotic arm grasping is completed.

[0125] Compared to existing methods for calculating the position of standard equally spaced rectangular pallets, this invention can accurately locate the placement and retrieval points of objects on non-rectangular, parallelogram-shaped vertical warehouse pallets with curved long sides. Compared to existing robot pallet teaching methods, the information collection method in this invention only requires posting four QR codes to accurately locate the placement and retrieval points of materials on the entire pallet.

[0126] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art will appreciate that the technical solutions described in the above embodiments may be modified or some or all of the technical features thereof may be replaced with equivalents, and that such modifications or replacements do not deviate from the essence of the corresponding technical solutions within the scope of the various embodiments of the present invention.

Claims

1. A dynamic positioning method for a vertical warehouse pallet, characterized in that: The following steps are involved: S1: Get the row and column information of the target tray; S2: Construct the original coordinate system based on the robot's base coordinate position, and calculate the original coordinate data of the object pick-up and placement points through the row and column information of the pallet; S3: Identify the teaching points through the QR code, and use the least squares method to perform curve fitting on the teaching points to generate the dynamic coordinate system of the pallet. Further obtain the original coordinate data of the object and update the coordinate data accordingly in the dynamic coordinate system to complete dynamic positioning; S31: Select a teaching point in the original coordinate system, determine the actual position information of the teaching point in the original coordinate system by identifying the QR code corresponding to the teaching point, fit the coordinate point based on deep reinforcement learning, and construct a dynamic coordinate system; S32: establishing a transformation matrix between the dynamic coordinate system and the original coordinate system; S33: Select at least three teaching points in the dynamic coordinate system for fitting, and obtain the object pick-up and placement points based on the row offset and the point data of each column in the first row; S34: Generate updated coordinates of the object placement location in the dynamic coordinate system through the transformation matrix, and obtain coordinate information of the updated coordinates in the original coordinate system.

2. A dynamic positioning method for a vertical warehouse pallet according to claim 1, characterized in that: The method for deep reinforcement learning fitting coordinate points in step S31 includes the following steps: First, the position data of all teaching points are collected several times, and Gaussian noise is added to the position data; Then, a neural network with two hidden layers is constructed based on the above data; Finally, according to the neural network, the state space, action space and reward function are obtained.

3. A dynamic positioning method for a vertical warehouse pallet according to claim 2, characterized in that: The state space, action space and reward function are obtained by the following formula: S=[x i ,y i ,z i ],i=(0,1,2,3,4) (1) The state space S is a set describing the robot's state, including all possible positions and states of the robot. x, y, and z represent the coordinates of the three axes in the space, respectively. i represents the serial number of the teaching point, which refers to a sequence of positions that the robot needs to reach. When i = 0, it represents the starting position of the robot in the motion trajectory, that is, the initial state of the robot. A=[x a ,y a ,z a ] (2) Where A represents a point in the three-dimensional space corresponding to each action in the action space; a represents the current point spatial position; the three components x, y, and z represent the x, y, and z coordinates of the robot in the Cartesian coordinate system respectively; R=K*((x a -x i ) 2 +(y a -y i ) 2 +(z a -z i ) 2 ) (3) Among them, R is the reward function and K is a constant.

4. A dynamic positioning method for a vertical warehouse pallet according to claim 3, characterized in that: The step S31 further includes optimizing the reward function, wherein the optimization direction is to calculate the optimized reward function J by using equations 4, 5, 6 and 7 in sequence; Among them, V represents the state value function, d represents the probability distribution of the state, Q represents the action state value function, and pi represents the state transition function; in, Represents the result after derivation, after the state transfer function Derivative calculation of gradients to achieve iterative calculation of reward function; in Indicates the gradient of the state value function; Among them, to represent the gradient function.

5. A dynamic positioning method for vertical warehouse pallets according to claim 4, characterized in that: The step S31 also includes updating the strategy parameters after the reward function is optimized. The updating method includes the following steps: First, randomly initialize the policy parameters; Then, a series of teaching points are generated on the pallet, S1, A1, R2, S2, A2, ..., S T $; Finally, the estimated reward value R t ,t=(1,2,...,T), calculate the updated policy parameters: Among them, θ represents the policy parameter.

6. A dynamic positioning method for a vertical warehouse pallet according to claim 1, characterized in that: In step S33, the calculation method of the pick-up and drop-off point data of each column in the first row specifically includes: When there are two teaching points in the first row, calculate the average column distance, and calculate the coordinate values of each column in the first row based on the average column distance to obtain the point data of each column in the first row; When there are at least three teaching points in the first row, select the teaching points at both ends to establish a dynamic coordinate system, use a curve to fit all the teaching points in the first row, calculate the average column distance between adjacent teaching points, and calculate the coordinate values of each column in the first row based on the average column distance to obtain the point data of each column in the first row; The calculation method of row offset includes: The number of teaching points in the first column is two. The inter-row offset unit vector between the two teaching points is calculated, and the row offset vector is calculated based on the row spacing and the offset unit.

7. A dynamic positioning method for a vertical warehouse pallet according to claim 6, characterized in that: Step S34 specifically includes: According to the coordinates of the teaching point, calculate the point data of each column in the first row and the row offset vector; Calculate the coordinate values of all the pick-up and place positions of the pallet based on the row offset vector and the coordinates of the points in each column of the first row; Convert the dynamic coordinate system to the original coordinate system coordinates.

8. A vertical warehouse pallet dynamic positioning method according to claim 7, characterized in that: The specific method to calculate the coordinate values of all the pick-up and placement positions of the pallet is as follows: Generate the coordinate information of all the pick-and-place points of the pallet in the dynamic coordinate system based on the position information of each pick-and-place point in the first row and the offset vector between rows; Multiply the coordinate information by the transformation matrix to obtain the position information of the pick-up and drop points in the original coordinate system.

9. An electric gripper device for picking and placing in a vertical warehouse, characterized in that: include: A control system for the dynamic positioning method of a vertical warehouse pallet according to any one of claims 1 to 8 is adopted, and the control system controls a material grabbing electric gripper for picking up and placing items from the vertical warehouse and a 3D camera for identifying a QR code set on the vertical warehouse.

Images