WO2021238191A1 - 机器人的定位补偿方法及装置 - Google Patents

机器人的定位补偿方法及装置 Download PDF

Info

Publication number
WO2021238191A1
WO2021238191A1 PCT/CN2020/139939 CN2020139939W WO2021238191A1 WO 2021238191 A1 WO2021238191 A1 WO 2021238191A1 CN 2020139939 W CN2020139939 W CN 2020139939W WO 2021238191 A1 WO2021238191 A1 WO 2021238191A1
Authority
WO
WIPO (PCT)
Prior art keywords
robot
neurons
moving
error
compensation
Prior art date
Application number
PCT/CN2020/139939
Other languages
English (en)
French (fr)
Inventor
郭东生
钟文涛
高小云
张志波
万文洁
周家裕
张睿
王佳威
衷镇宇
Original Assignee
珠海格力智能装备有限公司
珠海格力电器股份有限公司
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 珠海格力智能装备有限公司, 珠海格力电器股份有限公司 filed Critical 珠海格力智能装备有限公司
Publication of WO2021238191A1 publication Critical patent/WO2021238191A1/zh

Links

Images

Classifications

    • BPERFORMING OPERATIONS; TRANSPORTING
    • B25HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25JMANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00Programme-controlled manipulators
    • B25J9/16Programme controls
    • BPERFORMING OPERATIONS; TRANSPORTING
    • B25HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25JMANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00Programme-controlled manipulators
    • B25J9/16Programme controls
    • B25J9/1656Programme controls characterised by programming, planning systems for manipulators
    • B25J9/1664Programme controls characterised by programming, planning systems for manipulators characterised by motion, path, trajectory planning
    • BPERFORMING OPERATIONS; TRANSPORTING
    • B25HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25JMANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00Programme-controlled manipulators
    • B25J9/16Programme controls
    • B25J9/1679Programme controls characterised by the tasks executed
    • B25J9/1692Calibration of manipulator

Definitions

  • This application relates to the technical field of robot control, and in particular to a method and device for positioning compensation of a robot.
  • the embodiments of the present application provide a robot positioning compensation method and device, so as to at least solve the technical problem that the robot positioning error compensation speed in the related art is slow, and the adaptability is low, and the positioning accuracy supplementary requirements of the robot cannot be met.
  • a method for positioning compensation of a robot including: collecting a set of movement points of the robot when the robot moves in a preset spatial grid, where the spatial grid refers to the working area of the robot A grid divided by a cube with a predetermined step length; an error compensation model is obtained by training using the moving point set, wherein the error compensation model is based on the error between the theoretical position and the actual position of the flange center point of the robot The model established by the compensation parameters; obtain the actual coordinates of the target position that the robot intends to reach; use the error compensation model to calculate the compensation parameters for the robot to move to the actual coordinates, wherein the compensation parameters are used to move the robot Compensation is performed to the positioning accuracy of the actual coordinates; based on the compensation parameters, the robot is controlled to reach the target position.
  • the step of collecting a set of moving points when the robot moves in a preset space grid includes: taking the base coordinate system of the robot as a reference, and taking the predetermined step length as a division rule, and dividing the work of the robot The area is divided into multiple space cubes; starting from the first base point of the robot's work area, the three-dimensional coordinates of the sampling points at the end of the robot are collected in multiple space cubes according to a preset collection rule; based on the three-dimensional coordinates , To determine the moving point set of the robot when it is moving.
  • the positioning compensation method further includes: dividing the plurality of spatial cubes into a plurality of sampling layers; collecting the robot in each The change parameters and change characteristics of each coordinate system direction in the sampling layer; based on the change parameters and the change characteristics, the error compensation model is established.
  • using the moving point set to train to obtain the error compensation model includes: expressing the corresponding actual coordinates of each sampling point in the moving point set with a first vector, and expressing each sample point in the moving point set with a second vector The theoretical coordinates corresponding to the sampling points; determine the feedback neural network for constructing the error compensation model, the feedback neural network including: an input layer, a hidden layer, and an output layer; a vector set composed of the first vector and the second vector Input to the input layer of the feedback neural network; determine the first connection weight between the neurons in the input layer and the neurons in the hidden layer, and determine the neurons in the hidden layer and the neurons in the hidden layer The second connection weight between neurons in the output layer; based on the first connection weight and the second connection weight, determine the theoretical coordinate set and the test coordinate set of the sampling point; determine based on the theoretical coordinate set
  • the error compensation model is used to test the error compensation model by using the test coordinate set.
  • the second connection weight includes: determining the number of neurons in the hidden layer; based on the number of neurons, generating the first value between the neurons in the input layer and the neurons in the hidden layer Connection weight; determining the activation function and hidden layer output matrix for constructing the error compensation model; based on the activation function, the hidden layer output matrix, the first connection weight and the number of neurons in the hidden layer
  • the threshold determines the second connection weight between the neurons in the hidden layer and the neurons in the output layer.
  • the positioning compensation method further includes: determining the error vector and absolute value corresponding to each sampling point based on the actual and theoretical coordinates of each sampling point Error value; based on the error vector and the absolute error value, determine the predicted deviation value corresponding to each sampling point; calculate the predicted deviation average based on the predicted deviation value of each sampling point; based on the predicted deviation average , Screening the target value interval, wherein the target value interval is set to optimize the calculation accuracy of the error compensation model.
  • an extreme learning algorithm ELM is used to construct the error compensation model.
  • a positioning compensation device for a robot, including: a collection unit configured to collect a set of movement points of the robot when the robot moves in a preset spatial grid, wherein the spatial grid Refers to the grid divided by cubes of predetermined step length in the working area of the robot; the training unit is set to use the moving point set to train to obtain an error compensation model, wherein the error compensation model is based on the flange of the robot The model established by the error compensation parameters between the theoretical position of the center point and the actual position; the acquisition unit is configured to acquire the actual coordinates of the target position that the robot intends to reach; the movement unit is configured to calculate the movement of the robot using the error compensation model The compensation parameter to the actual coordinate, wherein the compensation parameter is used to compensate the positioning accuracy of the robot moving to the actual coordinate; the control unit is set to control the robot to reach the location based on the compensation parameter. The target location.
  • the acquisition unit includes: a first division module configured to divide the robot's work area into a plurality of spaces based on the base coordinate system of the robot and the predetermined step length as a division rule A cube; a first acquisition module, set to start from the first base point of the working area of the robot, and collect the three-dimensional coordinates of the sampling point at the end of the robot in a plurality of space cubes according to a preset acquisition rule; the first determination module And set to determine a set of moving points of the robot when moving based on the three-dimensional coordinates.
  • the positioning compensation device further includes: a second division module configured to divide the plurality of spatial cubes into a plurality of sampling layers after collecting a set of movement points of the robot when the robot moves in a preset spatial grid;
  • the second acquisition module is configured to collect the change parameters and change characteristics of the coordinate system directions of the robot in each sampling layer;
  • the first establishment module is set to establish the error based on the change parameters and the change characteristics Compensation model.
  • the training unit includes: a second determining module configured to use a first vector to represent the corresponding actual coordinates of each sampling point in the moving point set, and use a second vector to represent each sample point in the moving point set The theoretical coordinates corresponding to the sampling points;
  • the third determining module is configured to determine the feedback neural network for constructing the error compensation model, and the feedback neural network includes: an input layer, a hidden layer, and an output layer;
  • the first input module is configured to The vector set consisting of the first vector and the second vector is input to the input layer of the feedback neural network;
  • the fourth determining module is configured to determine the difference between the neurons in the input layer and the neurons in the hidden layer And determine the second connection weight between the neurons in the hidden layer and the neurons in the output layer;
  • the fifth determining module is set to be based on the first connection weight And the second connection weight to determine the theoretical coordinate set and the test coordinate set of the sampling point;
  • the sixth determining module is configured to determine the error compensation model based on the theoretical coordinate set, and use the
  • the fourth determining module includes: a first determining sub-module configured to determine the number of neurons in the hidden layer; a first generating module configured to generate the number of neurons based on the number of neurons The first connection weight between the neurons in the input layer and the neurons in the hidden layer; the second determining sub-module is set to determine the activation function and the hidden layer output matrix for constructing the error compensation model; the third determining sub-module Module, set to determine the neurons of the hidden layer and the output matrix based on the activation function, the output matrix of the hidden layer, the first connection weight and the threshold of the number of neurons of the hidden layer The second connection weight between neurons of the layer.
  • the positioning compensation device further includes: a seventh determining module, configured to determine the error compensation model based on the actual coordinates and theoretical coordinates of each sampling point based on the actual coordinates and theoretical coordinates of each sampling point after the error compensation model is obtained through the training of the moving point set.
  • the error vector and the absolute error value corresponding to the point The error vector and the absolute error value corresponding to the point; the eighth determining module is set to determine the predicted deviation value corresponding to each sampling point based on the error vector and the absolute error value; the calculation module is set to be based on each sampling Point prediction deviation value, calculate the average prediction deviation; the filtering module is set to filter the target value interval based on the prediction deviation average value, wherein the target value interval is set to optimize the calculation accuracy of the error compensation model Spend.
  • an extreme learning algorithm ELM is used to construct the error compensation model.
  • an industrial robot including: a processor; and a memory, configured to store executable instructions of the processor; wherein the processor is configured to execute the The instructions can be executed to execute any one of the robot positioning compensation methods described above.
  • a computer storage medium includes a stored program, wherein when the program is running, the device where the computer storage medium is located is controlled to execute any one of the foregoing items.
  • the described robot positioning compensation method is also provided.
  • the error compensation model when compensating the positioning accuracy of the robot, first collect the moving point set of the robot when it moves on the preset spatial grid, and then use the moving point set to train to obtain the error compensation model, and then obtain the target that the robot intends to reach The actual coordinates of the position, the error compensation model is used to calculate the compensation parameters for the robot to move to the actual coordinates, and based on the compensation parameters, the robot is controlled to reach the target position.
  • moving point sets with different sampling intervals can be used as training sets for error compensation model training and compensation testing, and an error compensation model between the theoretical position and actual position of the robot flange center point can be established, and the error compensation model It can compensate the absolute positioning accuracy of any point in the working area of the robot, with fast compensation speed, high absolute positioning accuracy and good generalization performance, which can meet the positioning accuracy requirements of the robot, so as to solve the problem of robot positioning error compensation speed in related technologies. It is slow, and has low adaptability, and cannot meet the technical problems of supplementary requirements for robot positioning accuracy.
  • Fig. 1 is a flowchart of an optional robot positioning compensation method according to an embodiment of the present application
  • Fig. 2 is a schematic diagram of an optional spatial grid sampling point plan according to an embodiment of the present application
  • Fig. 3 is a schematic structural diagram of an optional feedback neural network according to an embodiment of the present application.
  • Fig. 4 is a schematic diagram of an optional construction error compensation model according to an embodiment of the present application.
  • Fig. 5 is a schematic diagram of an optional conversion relationship between coordinate systems of an industrial robot according to an embodiment of the present application
  • Fig. 6 is a flowchart of an optional method for transforming various coordinate systems of a robot according to an embodiment of the present application
  • Fig. 7 is a schematic diagram of an optional robot positioning compensation device according to an embodiment of the present application.
  • ELM Extreme Learing Machine
  • extreme learning machine is a machine learning system or method based on a feedforward neural network.
  • ELM is applied to the positioning accuracy compensation of the robot, and an error compensation model based on the ELM algorithm is established, which can compensate the absolute positioning accuracy of any point in the working area.
  • the error compensation model involved in the embodiment of the present application can optimize the value of the hidden layer neuron and improve the positioning accuracy.
  • an embodiment of a method for positioning compensation of a robot is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and, Although a logical sequence is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than here.
  • Fig. 1 is a flowchart of an optional robot positioning compensation method according to an embodiment of the present application. As shown in Fig. 1, the method includes the following steps:
  • Step S102 Collect a set of moving points when the robot moves on a preset spatial grid, where the spatial grid refers to a grid divided by a cube with a predetermined step length in the working area of the robot;
  • step S104 an error compensation model is obtained by training using the moving point set, where the error compensation model is a model established based on the error compensation parameters between the theoretical position and the actual position of the flange center point of the robot;
  • Step S106 obtaining the actual coordinates of the target position that the robot intends to reach
  • Step S108 using the error compensation model to calculate compensation parameters for the robot moving to the actual coordinates, where the compensation parameters are used to compensate the positioning accuracy of the robot moving to the actual coordinates;
  • Step S110 based on the compensation parameters, control the robot to reach the target position.
  • the robot when compensating the positioning accuracy of the robot, first collect the moving point set of the robot when it moves on the preset spatial grid, and then use the moving point set to train to obtain the error compensation model, and then obtain the target position that the robot intends to reach
  • the error compensation model is used to calculate the compensation parameters for the robot to move to the actual coordinates, and based on the compensation parameters, the robot is controlled to reach the target position.
  • moving point sets with different sampling intervals can be used as training sets for error compensation model training and compensation testing, and an error compensation model between the theoretical position and actual position of the robot flange center point can be established, and the error compensation model It can compensate the absolute positioning accuracy of any point in the working area of the robot, with fast compensation speed, high absolute positioning accuracy and good generalization performance, which can meet the positioning accuracy requirements of the robot, so as to solve the problem of robot positioning error compensation speed in related technologies. It is slow, and has low adaptability, and cannot meet the technical problems of supplementary requirements for robot positioning accuracy.
  • the robots involved in the embodiments of this application include, but are not limited to: industrial robots (such as six-axis robots) and educational robots.
  • the robots can be used to achieve applications in welding, drilling and riveting, polishing, painting and other scenarios, and improve the robot’s performance in automated parts and components. Absolute positioning accuracy.
  • Step S102 Collect a set of moving points when the robot moves on a preset spatial grid, where the spatial grid refers to a grid divided by a cube with a predetermined step length in the working area of the robot.
  • the step of collecting a set of moving points when the robot moves in a preset space grid includes: dividing the robot's working area into multiple spaces based on the base coordinate system of the robot and using a predetermined step length as a dividing rule Cube: Starting from the first base point of the robot's work area, collect the three-dimensional coordinates of the sampling points at the end of the robot in multiple spatial cubes according to the preset collection rules; determine the set of moving points when the robot is moving based on the three-dimensional coordinates.
  • Fig. 2 is a schematic diagram of an optional spatial grid sampling point planning according to an embodiment of the present application.
  • the actual working area is first planned in the robot workspace, namely Cube P1P2P3P4P5P6P7P8, the distance d in the working area is a predetermined step (fixed sampling step), the sampling path can be: zigzag sampling on the plane P1P2P3P4, that is, from the starting point P1 to P2 and back to P1P3 from P1
  • the second line is collected at the position of the predetermined step length, until the sampling reaches P4, and then the second layer is collected in the same way, until the sampling end point P8 is collected.
  • Each sampling takes the zero position of the robot as the starting position and arrives at the sampling point in the same posture, and then uses the measuring device to measure the actual position of the robot end in three-dimensional coordinates to obtain a set of moving points.
  • the positioning compensation method further includes: dividing the multiple spatial cubes into multiple sampling layers; and collecting the robot in each sampling layer in each coordinate system Directional change parameters and change characteristics; based on the change parameters and change characteristics, an error compensation model is established.
  • each sampling layer includes: X direction, Y direction, and Z direction.
  • the positioning compensation method further includes: determining the error vector and absolute error corresponding to each sampling point based on the actual and theoretical coordinates of each sampling point Value; Based on the error vector and absolute error value, determine the prediction deviation value corresponding to each sampling point; Based on the prediction deviation value of each sampling point, calculate the average prediction deviation; Based on the average prediction deviation, filter the target value interval, Among them, the target value interval is used to optimize the calculation accuracy of the error compensation model.
  • the third formula is:
  • Combining the acquisition path of each sampling point, comparing the absolute error value of each sampling layer (parallel to the plane P1P2P3P4) can reflect the change of the positioning error along the Z direction of the robot base coordinate system, and compare each sampling line on a separate sampling layer (and The absolute error value between the line segments P1P2 (parallel) can reflect the change of the positioning error along the X direction of the robot base coordinate system. Comparing the absolute value of the error between each sampling point on a separate sampling line can reflect the positioning error along the robot base coordinate system Y Changes in direction.
  • the direction of the positioning error vector is compared to obtain the change characteristics of the positioning error direction along each direction of the robot base coordinate system, and the size and direction of the robot positioning error
  • the change characteristics may include:
  • an error model can be further established to compensate the robot positioning accuracy.
  • step S104 an error compensation model is obtained by training using the moving point set, where the error compensation model is a model established based on the error compensation parameters between the theoretical position and the actual position of the flange center point of the robot.
  • the extreme learning algorithm ELM can be used to construct an error compensation model, and the structure of the ELM can be understood as a feedback neural network with a single hidden layer.
  • training to obtain the error compensation model using the moving point set includes: expressing the corresponding actual coordinates of each sampling point in the moving point set with a first vector, and expressing the theoretical coordinates corresponding to each sampling point in the moving point set with a second vector ;
  • Determine the feedback neural network for constructing the error compensation model the feedback neural network includes: input layer, hidden layer, output layer; input the vector set composed of the first vector and the second vector to the input layer of the feedback neural network; determine the input layer The first connection weight between the neurons in the hidden layer and the neurons in the hidden layer, and the second connection weight between the neurons in the hidden layer and the neurons in the output layer is determined; based on the first connection weight and the first connection weight Second, connect the weights, determine the theoretical coordinate set and the test coordinate set of the sampling point; determine the error compensation model based on the theoretical coordinate set, and use the test coordinate set to test the error compensation model.
  • Fig. 3 is a schematic structural diagram of an optional feedback neural network according to an embodiment of the present application. As shown in Fig. 3, it includes an input layer, a hidden layer, and an output layer.
  • Input layer and hidden layer, hidden layer and output layer neurons are fully connected.
  • I 1 , I 2 , I 3 ??I N represents the input layer neuron
  • O 1 , O 2 , O 3 ??O M represents the output layer neuron
  • q 1 , q 2 , q 3 ??q L represents the hidden layer neuron
  • the actual coordinates are determined by the fourth formula, and the theoretical coordinates are determined by the fifth formula.
  • Another option is to determine the first connection weight between the neurons in the input layer and the neurons in the hidden layer, and determine the second connection weight between the neurons in the hidden layer and the neurons in the output layer Values include: determining the number of neurons in the hidden layer; based on the number of neurons, generating the first connection weight between neurons in the input layer and neurons in the hidden layer; determining the activation function and Hidden layer output matrix; based on the activation function, the hidden layer output matrix, the first connection weight and the hidden layer neuron number threshold, determine the second connection between the neurons in the hidden layer and the neurons in the output layer Weight.
  • the first connection weight w is expressed by the eighth formula
  • the neuron threshold is expressed by the ninth formula, where the eighth formula is:
  • the ninth formula is:
  • H and T are known quantities
  • H is the hidden layer output matrix in the feedback neural network, represented by four known quantities: input P, activation function f(x), first connection weight w and threshold b .
  • the value of the number of neurons in the hidden layer L is less than the number of training samples n, so the second connection weight ⁇ between the hidden layer and the output layer is solved by equation (12) :
  • H + is the generalized inverse matrix of the output matrix.
  • the value of the number of hidden layer neurons as the parameter of the ELM algorithm has a greater impact on the generalization performance of the error compensation model, too few hidden layer neurons will lead to larger errors in the fitting model, and too many hidden layers. Layered neurons may cause overfitting of the error model and reduce the efficiency of model training. Therefore, in the embodiment of the present application, the value of L is further adjusted to optimize the compensation model.
  • Figure 4 is a schematic diagram of an optional error compensation model constructed according to an embodiment of the present application.
  • the ELM algorithm when used for training, the theoretical coordinates of the sampling point and the actual coordinates of the sampling point are combined with the selected activation Function and the set number of hidden layer neurons are trained to obtain the error compensation model.
  • the predicted coordinates of the test point after prediction are obtained, and the prediction coordinates are based on the test point.
  • the theoretical coordinates are used to determine whether the positioning accuracy meets the requirements. If it does, the error compensation model can be further optimized. If it does not, the number of neurons in the hidden layer needs to be re-adjusted to further optimize the compensation model.
  • m represents the number of samples in the test point set.
  • the prediction deviation E of the error compensation model is the difference between the theoretical coordinate X t of the test point set and the theoretical coordinate X f predicted by the error model, expressed by equation (16):
  • the average forecast deviation E A is expressed by formula (18):
  • the prediction deviation E A is a function of the number of neurons L.
  • Step S106 Obtain the actual coordinates of the target position that the robot intends to reach.
  • the theoretical coordinates of the robot when it is moving are calculated through the error compensation model, and then the compensation parameters are used to control the robot to accurately reach the target position.
  • step S108 the error compensation model is used to calculate the compensation parameters of the robot moving to the actual coordinates, where the compensation parameters are set to compensate the positioning accuracy of the robot moving to the actual coordinates.
  • Step S110 based on the compensation parameters, control the robot to reach the target position.
  • the embodiment of the present application takes the GR20A 6-DOF industrial robot as an example for description to analyze the robot positioning error law and complete the error compensation experiment based on the ELM algorithm.
  • the measuring equipment uses a 3-dimensional three-optical measuring instrument.
  • FIG. 5 is a schematic diagram of an optional conversion relationship between coordinate systems of an industrial robot according to an embodiment of the present application, as shown in FIG. Display, including the establishment of the base coordinate system based on the center point of the robot base, the flange center coordinate system, the world coordinate system, and the measurement system coordinate system.
  • Fig. 6 is a flow chart of an optional conversion method of the coordinate systems of the robot according to an embodiment of the present application.
  • the flange coordinate system, the world coordinate system, and the measurement system coordinate system can be established by the measurement system first The relationship between the three, and then rotate the robot A1 and A6 axis respectively, and measure the coordinates of the marked points in the world coordinate system at each rotation position, measure the initial position coordinates of the marked points in the world coordinate system, and measure the theoretical positions at the same time The coordinates of the marker point are in the world coordinate system.
  • the marker points measured when the A1 and A6 axes are rotated can be fitted to circles to obtain the x-axis and z-axis of the robot in the world coordinate system.
  • the conversion relationship between the world coordinate system and the robot base coordinate system can be determined, and the translation relationship between the world coordinate system and the robot base coordinate system can be determined.
  • the absolute error is calculated by formulas (1)-(3), and the error range and error value before compensation are obtained.
  • the absolute error after compensation is obtained and compared with the absolute error value before error compensation.
  • FIG. 7 is a schematic diagram of an optional positioning compensation device for a robot according to an embodiment of the present application.
  • the positioning compensation device may include: an acquisition unit 71, a training unit 73, an acquisition unit 75, and a movement unit 77 , Control unit 79, in which,
  • the collection unit 71 is configured to collect a set of moving points of the robot when it moves on a preset spatial grid, where the spatial grid refers to a grid divided by a cube with a predetermined step length in the working area of the robot;
  • the training unit 73 is configured to train to obtain an error compensation model using the moving point set, where the error compensation model is a model established based on the error compensation parameters between the theoretical position and the actual position of the flange center point of the robot;
  • the acquiring unit 75 is configured to acquire the actual coordinates of the target position that the robot intends to reach;
  • the moving unit 77 is configured to use the error compensation model to calculate compensation parameters for the robot moving to the actual coordinates, where the compensation parameters are used to compensate the positioning accuracy of the robot moving to the actual coordinates;
  • the control unit 79 is configured to control the robot to reach the target position based on the compensation parameter.
  • the positioning compensation device of the robot described above may first collect the moving point set of the robot when the robot moves on a preset spatial grid through the collecting unit 71 when compensating the positioning accuracy of the robot, and then use the moving point set to train through the training unit 73 to obtain error compensation. Afterwards, the actual coordinates of the target position that the robot intends to reach can be obtained through the acquisition unit 75, the error compensation model is used to calculate the compensation parameters for the robot to move to the actual coordinates through the moving unit 77, and the control unit 79 controls the robot to reach the target position based on the compensation parameters. .
  • moving point sets with different sampling intervals can be used as training sets for error compensation model training and compensation testing, and an error compensation model between the theoretical position and actual position of the robot flange center point can be established, and the error compensation model It can compensate the absolute positioning accuracy of any point in the working area of the robot, with fast compensation speed, high absolute positioning accuracy and good generalization performance, which can meet the positioning accuracy requirements of the robot, so as to solve the problem of robot positioning error compensation speed in related technologies. It is slow, and has low adaptability, and cannot meet the technical problems of supplementary requirements for robot positioning accuracy.
  • the acquisition unit includes: a first division module configured to divide the robot's work area into multiple spatial cubes based on the robot's basic coordinate system and a predetermined step length as a division rule; and the first acquisition module is configured to To start from the first base point of the robot's working area, collect the three-dimensional coordinates of the sampling points at the end of the robot in multiple spatial cubes according to the preset collection rules; the first determination module is set to determine the robot when it is moving based on the three-dimensional coordinates Set of moving points.
  • the positioning compensation device further includes: a second dividing module configured to divide the multiple spatial cubes into multiple sampling layers after collecting the moving point set when the robot moves on the preset spatial grid;
  • the second collection module is set to collect the change parameters and change characteristics of each coordinate system direction of the robot in each sampling layer;
  • the first establishment module is set to establish an error compensation model based on the change parameters and change characteristics.
  • the training unit includes: a second determining module, configured to use a first vector to represent the actual coordinates of each sampling point in the moving point set, and use a second vector to represent the theoretical coordinates of each sampling point in the moving point set ;
  • the third determining module is set to determine the feedback neural network for constructing the error compensation model.
  • the feedback neural network includes: an input layer, a hidden layer, and an output layer; the first input module is set to consist of a first vector and a second vector The vector set is input to the input layer of the feedback neural network; the fourth determination module is set to determine the first connection weight between the neurons of the input layer and the neurons of the hidden layer, and determine the neurons and output of the hidden layer The second connection weight between the neurons of the layer; the fifth determination module is set to determine the theoretical coordinate set and the test coordinate set of the sampling point based on the first connection weight and the second connection weight; the sixth determination module, Set to determine the error compensation model based on the theoretical coordinate set, and use the test coordinate set to test the error compensation model.
  • the fourth determining module includes: a first determining sub-module configured to determine the number of neurons in the hidden layer; and a first generating module configured to generate neurons in the input layer based on the number of neurons The first connection weight with the neurons in the hidden layer; the second determining sub-module is set to determine the activation function and the hidden layer output matrix for constructing the error compensation model; the third determining sub-module is set to be based on the activation function , The output matrix of the hidden layer, the first connection weight and the threshold of the number of neurons in the hidden layer, determine the second connection weight between the neurons in the hidden layer and the neurons in the output layer.
  • the positioning compensation device further includes: a seventh determining module, configured to determine the error corresponding to each sampling point based on the actual and theoretical coordinates of each sampling point after the error compensation model is obtained through the training of the moving point set Vector and absolute error value; the eighth determining module is set to determine the predicted deviation value corresponding to each sampling point based on the error vector and absolute error value; the calculation module is set to calculate the prediction based on the predicted deviation value of each sampling point Deviation average value; the filtering module is set to filter the target value interval based on the predicted deviation average value, where the target value interval is set to optimize the calculation accuracy of the error compensation model.
  • a seventh determining module configured to determine the error corresponding to each sampling point based on the actual and theoretical coordinates of each sampling point after the error compensation model is obtained through the training of the moving point set Vector and absolute error value
  • the eighth determining module is set to determine the predicted deviation value corresponding to each sampling point based on the error vector and absolute error value
  • the calculation module is set to calculate the prediction based on the predicted deviation value
  • the aforementioned robot positioning compensation device may also include a processor and a memory.
  • the aforementioned collection unit 71, training unit 73, acquisition unit 75, moving unit 77, control unit 79, etc. are all stored as program units in the memory, and the processor executes the storage.
  • the above-mentioned program unit in the memory realizes the corresponding function.
  • the above-mentioned processor contains a kernel, and the kernel calls the corresponding program unit from the memory.
  • the kernel can be set to one or more, and the accuracy of the robot in positioning can be compensated by adjusting the kernel parameters.
  • the aforementioned memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least A memory chip.
  • RAM random access memory
  • ROM read-only memory
  • flash RAM flash memory
  • an industrial robot including: a processor; and a memory, configured to store executable instructions of the processor; wherein the processor is configured to execute the foregoing by executing the executable instructions Any one of the robot's positioning compensation method.
  • a computer storage medium includes a stored program, wherein when the program runs, the device where the computer storage medium is located is controlled to execute any one of the above-mentioned robot positioning compensation methods .
  • This application also provides a computer program product, which when executed on a data processing device, is suitable for executing a program that initializes the following method steps: Collecting a set of moving points when the robot moves on a preset space grid, where the space network A grid refers to a grid divided by a cube with a predetermined step length in the working area of the robot; an error compensation model is obtained by training with a moving point set, where the error compensation model is based on the difference between the theoretical position and the actual position of the flange center point of the robot The model established by the error compensation parameters of the robot; obtain the actual coordinates of the target position that the robot intends to reach; use the error compensation model to calculate the compensation parameters of the robot moving to the actual coordinates, where the compensation parameters are set to compensate the positioning accuracy of the robot moving to the actual coordinates ; Based on the compensation parameters, control the robot to reach the target position.
  • the disclosed technical content can be implemented in other ways.
  • the device embodiments described above are merely illustrative.
  • the division of the units may be a logical function division.
  • multiple units or components may be combined or may be Integrate into another system, or some features can be ignored or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, units or modules, and may be in electrical or other forms.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
  • the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated unit can be implemented in the form of hardware or software functional unit.
  • the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium.
  • the technical solution of the present application essentially or the part that contributes to the existing technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , Including several instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application.
  • the aforementioned storage media include: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes. .
  • the solutions provided in the embodiments of this application can be used in the field of robot control, and are suitable for environments such as home service positioning and industrial production positioning.
  • the set of moving points when the grid moves, and then use the set of moving points to train to obtain the error compensation model. After that, the actual coordinates of the target position that the robot intends to reach can be obtained.
  • the error compensation model is used to calculate the compensation parameters for the robot to move to the actual coordinates.
  • the compensation speed is fast, and the absolute positioning accuracy is high, and the generalization performance is good, which can meet the positioning accuracy requirements of the robot. It solves the technical problem that the robot positioning error compensation speed in the related technology is slow, and the adaptability is low, which cannot meet the supplementary requirements of the robot's positioning accuracy.

Landscapes

  • Engineering & Computer Science (AREA)
  • Robotics (AREA)
  • Mechanical Engineering (AREA)
  • Manipulator (AREA)
  • Numerical Control (AREA)

Abstract

一种机器人的定位补偿方法,该方法包括:采集机器人在预设空间网格移动时的移动点集,其中,空间网格是指在机器人的工作区域按照预定步长的立方体划分的网格;利用移动点集训练得到误差补偿模型,其中,误差补偿模型是根据机器人的法兰中心点的理论位置与实际位置之间的误差补偿参数建立的模型;获取机器人拟到达的目标位置的实际坐标;利用误差补偿模型计算机器人移动至实际坐标的补偿参数,其中,补偿参数用于为对机器人移动至实际坐标的定位精度进行补偿;基于补偿参数,控制机器人到达目标位置。该方法补偿速度快,且绝对定位精度高、泛化性能好。还涉及一种机器人的定位补偿装置、一种工业机器人和一种计算机存储介质。

Description

机器人的定位补偿方法及装置
本申请要求于2020年05月26日提交中国专利局、申请号为202010456895.5、申请名称“机器人的定位补偿方法及装置”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及机器人控制技术领域,具体而言,涉及一种机器人的定位补偿方法及装置。
背景技术
相关技术中,机器人在产品生产制造、加工装配等方面的应用越来越广泛.由于采用工业机器人进行加工装配时主要依赖于绝对定位精度,若是绝对定位精度过低会严重影响产品质量,现有提高机器人定位精度的方案包括:离线误差补偿和在线误差补偿。其中,离线误差补偿即在使用机器人进行工作之前,通过一定的方法补偿机器人定位精度,如数学逼近方法和在线误差补偿等,但这些补偿机器人定位精度的方法都存在一些不足,数学逼近法求解复杂非线性模型的效果欠佳,难以满足飞机钻铆等应用的高精度要求;在线误差补偿通常需要在机器人末端增加一个实时反馈装置,使机器人在工作过程中能够不断地调整末端直至理想位置。而在线误差补偿,通常能够获得较高的定位精度,但是在机器人末端执行器上增加反馈装置使得机器人在一些复杂场景中不易操作,容易导致机器人的定位精度补偿速度慢,且适应性较低。
针对上述的问题,目前尚未提出有效的解决方案。
发明内容
本申请实施例提供了一种机器人的定位补偿方法及装置,以至少解决相关技术中机器人定位误差补偿速度较慢,且适应性低,无法满足机器人的定位精度补充需求的技术问题。
根据本申请实施例的一个方面,提供了一种机器人的定位补偿方法,包括:采集机器人在预设空间网格移动时的移动点集,其中,所述空间网格是指在机器人的工作区域按照预定步长的立方体划分的网格;利用所述移动点集训练得到误差补偿模型, 其中,所述误差补偿模型是根据所述机器人的法兰中心点的理论位置与实际位置之间的误差补偿参数建立的模型;获取机器人拟到达的目标位置的实际坐标;利用所述误差补偿模型计算所述机器人移动至所述实际坐标的补偿参数,其中,所述补偿参数用于对所述机器人移动至所述实际坐标的定位精度进行补偿;基于所述补偿参数,控制所述机器人到达所述目标位置。
可选地,采集机器人在预设空间网格移动时的移动点集的步骤,包括:以所述机器人的基坐标系为基准,以所述预定步长作为划分规则,将所述机器人的工作区域划分为多个空间立方体;从所述机器人的工作区域的第一个基点开始,按照预设采集规则在多个空间立方体中采集所述机器人末端的采样点的三维坐标;基于所述三维坐标,确定所述机器人在移动时的移动点集。
可选地,在采集机器人在预设空间网格移动时的移动点集之后,所述定位补偿方法还包括:将所述多个空间立方体划分为多个采样层;采集所述机器人在每个采样层中各坐标系方向的变化参数和变化特征;基于所述变化参数和所述变化特征,建立所述误差补偿模型。
可选地,利用所述移动点集训练得到误差补偿模型包括:以第一向量表示所述移动点集中每个采样点的对应的实际坐标,并以第二向量表示所述移动点集中每个采样点对应的理论坐标;确定构建所述误差补偿模型的反馈神经网络,所述反馈神经网络包括:输入层、隐含层、输出层;将所述第一向量和第二向量组成的向量集输入至所述反馈神经网络的输入层;确定所述输入层的神经元与所述隐含层的神经元之间的第一连接权值,并确定所述隐含层的神经元与所述输出层的神经元之间的第二连接权值;基于所述第一连接权值和所述第二连接权值,确定采样点的理论坐标集和测试坐标集;基于所述理论坐标集确定所述误差补偿模型,并利用所述测试坐标集对所述误差补偿模型进行测试。
可选地,确定所述输入层的神经元与所述隐含层的神经元之间的第一连接权值,并确定所述隐含层的神经元与所述输出层的神经元之间的第二连接权值包括:确定所述隐含层的神经元个数;基于所述神经元个数,产生所述输入层的神经元与所述隐含层的神经元之间的第一连接权值;确定构建误差补偿模型的激活函数和隐含层输出矩阵;基于所述激活函数、所述隐含层输出矩阵、所述第一连接权值和所述隐含层的神经元数量阈值,确定所述隐含层的神经元与所述输出层的神经元之间的第二连接权值。
可选地,在利用所述移动点集训练得到误差补偿模型之后,所述定位补偿方法还包括:基于每个采样点的实际坐标和理论坐标,确定与每个采样点对应的误差向量和绝对误差值;基于所述误差向量和所述绝对误差值,确定与每个采样点对应的预测偏 差值;基于每个采样点的预测偏差值,计算预测偏差平均值;基于所述预测偏差平均值,筛选目标取值区间,其中,所述目标取值区间设置为优化所述误差补偿模型的计算准确度。
可选地,利用极限学习算法ELM构建所述误差补偿模型。
根据本申请实施例的另一方面,还提供了一种机器人的定位补偿装置,包括:采集单元,设置为采集机器人在预设空间网格移动时的移动点集,其中,所述空间网格是指在机器人的工作区域按照预定步长的立方体划分的网格;训练单元,设置为利用所述移动点集训练得到误差补偿模型,其中,所述误差补偿模型是根据所述机器人的法兰中心点的理论位置与实际位置之间的误差补偿参数建立的模型;获取单元,设置为获取机器人拟到达的目标位置的实际坐标;移动单元,设置为利用所述误差补偿模型计算所述机器人移动至所述实际坐标的补偿参数,其中,所述补偿参数用于对所述机器人移动至所述实际坐标的定位精度进行补偿;控制单元,设置为基于所述补偿参数,控制所述机器人到达所述目标位置。
可选地,所述采集单元包括:第一划分模块,设置为以所述机器人的基坐标系为基准,以所述预定步长作为划分规则,将所述机器人的工作区域划分为多个空间立方体;第一采集模块,设置为从所述机器人的工作区域的第一个基点开始,按照预设采集规则在多个空间立方体中采集所述机器人末端的采样点的三维坐标;第一确定模块,设置为基于所述三维坐标,确定所述机器人在移动时的移动点集。
可选地,所述定位补偿装置还包括:第二划分模块,设置为在采集机器人在预设空间网格移动时的移动点集之后,将所述多个空间立方体划分为多个采样层;第二采集模块,设置为采集所述机器人在每个采样层中各坐标系方向的变化参数和变化特征;第一建立模块,设置为基于所述变化参数和所述变化特征,建立所述误差补偿模型。
可选地,所述训练单元包括:第二确定模块,设置为以第一向量表示所述移动点集中每个采样点的对应的实际坐标,并以第二向量表示所述移动点集中每个采样点对应的理论坐标;第三确定模块,设置为确定构建所述误差补偿模型的反馈神经网络,所述反馈神经网络包括:输入层、隐含层、输出层;第一输入模块,设置为将所述第一向量和第二向量组成的向量集输入至所述反馈神经网络的输入层;第四确定模块,设置为确定所述输入层的神经元与所述隐含层的神经元之间的第一连接权值,并确定所述隐含层的神经元与所述输出层的神经元之间的第二连接权值;第五确定模块,设置为基于所述第一连接权值和所述第二连接权值,确定采样点的理论坐标集和测试坐标集;第六确定模块,设置为基于所述理论坐标集确定所述误差补偿模型,并利用所述测试坐标集对所述误差补偿模型进行测试。
可选地,所述第四确定模块包括:第一确定子模块,设置为确定所述隐含层的神经元个数;第一产生模块,设置为基于所述神经元个数,产生所述输入层的神经元与所述隐含层的神经元之间的第一连接权值;第二确定子模块,设置为确定构建误差补偿模型的激活函数和隐含层输出矩阵;第三确定子模块,设置为基于所述激活函数、所述隐含层输出矩阵、所述第一连接权值和所述隐含层的神经元数量阈值,确定所述隐含层的神经元与所述输出层的神经元之间的第二连接权值。
可选地,所述定位补偿装置还包括:第七确定模块,设置为在利用所述移动点集训练得到误差补偿模型之后,基于每个采样点的实际坐标和理论坐标,确定与每个采样点对应的误差向量和绝对误差值;第八确定模块,设置为基于所述误差向量和所述绝对误差值,确定与每个采样点对应的预测偏差值;计算模块,设置为基于每个采样点的预测偏差值,计算预测偏差平均值;筛选模块,设置为基于所述预测偏差平均值,筛选目标取值区间,其中,所述目标取值区间设置为优化所述误差补偿模型的计算准确度。
可选地,利用极限学习算法ELM构建所述误差补偿模型。
根据本申请实施例的另一方面,还提供了一种工业机器人,包括:处理器;以及存储器,设置为存储所述处理器的可执行指令;其中,所述处理器配置为经由执行所述可执行指令来执行上述任意一项所述的机器人的定位补偿方法。
根据本申请实施例的另一方面,还提供了一种计算机存储介质,所述计算机存储介质包括存储的程序,其中,在所述程序运行时控制所述计算机存储介质所在设备执行上述任意一项所述的机器人的定位补偿方法。
本申请实施例中,在对机器人定位精度进行补偿时,先采集机器人在预设空间网格移动时的移动点集,然后利用移动点集训练得到误差补偿模型,之后可获取机器人拟到达的目标位置的实际坐标,利用误差补偿模型计算机器人移动至实际坐标的补偿参数,基于补偿参数,控制机器人到达目标位置。在该实施例中,可以采用不同采样间隔的移动点集作为训练集进行误差补偿模型训练和补偿测试,并建立机器人法兰中心点理论位置与实际位置之间的误差补偿模型,通过误差补偿模型可对机器人的工作区域内任意一点的绝对定位精度进行补偿,补偿速度快,且绝对定位精度高、泛化性能好,能够满足机器人的定位精度需求,从而解决相关技术中机器人定位误差补偿速度较慢,且适应性低,无法满足机器人的定位精度补充需求的技术问题。
附图说明
此处所说明的附图用来提供对本申请的进一步理解,构成本申请的一部分,本申 请的示意性实施例及其说明设置为解释本申请,并不构成对本申请的不当限定。在附图中:
图1是根据本申请实施例的一种可选的机器人的定位补偿方法的流程图;
图2是根据本申请实施例的一种可选的空间网格采样点规划的示意图;
图3是根据本申请实施例的一种可选的反馈神经网络的结构示意图;
图4是根据本申请实施例的一种可选的构建误差补偿模型的示意图;
图5是根据本申请实施例的一种可选的工业机器人各坐标系转换关系的示意图;
图6是根据本申请实施例的一种可选的机器人各坐标系的转换方法的流程图;
图7是根据本申请实施例的一种可选的机器人的定位补偿装置的示意图。
具体实施方式
为了使本技术领域的人员更好地理解本申请方案,下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本申请一部分的实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都应当属于本申请保护的范围。
需要说明的是,本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是设置为区别类似的对象,而不必设置为描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换,以便这里描述的本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外,术语“包括”和“具有”以及他们的任何变形,意图在于覆盖不排他的包含,例如,包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元,而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
为便于本领域技术人员理解本申请,下面对本申请实施例中涉及的部分术语或名词做出解释:
ELM,Extreme Learing Machine,简称极限学习机,是一种基于前馈神经网络构建的机器学习系统或方法。本申请中将ELM应用于机器人的定位精度补偿,建立基于ELM算法的误差补偿模型,可对工作区域内任意一点的绝对定位精度进行补偿。
本申请实施例涉及的误差补偿模型,可优化隐含层神经元取值,提高定位精度。
根据本申请实施例,提供了一种机器人的定位补偿方法实施例,需要说明的是,在附图的流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行,并且,虽然在流程图中示出了逻辑顺序,但是在某些情况下,可以以不同于此处的顺序执行所示出或描述的步骤。
图1是根据本申请实施例的一种可选的机器人的定位补偿方法的流程图,如图1所示,该方法包括如下步骤:
步骤S102,采集机器人在预设空间网格移动时的移动点集,其中,空间网格是指在机器人的工作区域按照预定步长的立方体划分的网格;
步骤S104,利用移动点集训练得到误差补偿模型,其中,误差补偿模型是根据机器人的法兰中心点的理论位置与实际位置之间的误差补偿参数建立的模型;
步骤S106,获取机器人拟到达的目标位置的实际坐标;
步骤S108,利用误差补偿模型计算机器人移动至实际坐标的补偿参数,其中,补偿参数用于对机器人移动至实际坐标的定位精度进行补偿;
步骤S110,基于补偿参数,控制机器人到达目标位置。
通过上述步骤,可以在对机器人定位精度进行补偿时,先采集机器人在预设空间网格移动时的移动点集,然后利用移动点集训练得到误差补偿模型,之后可获取机器人拟到达的目标位置的实际坐标,利用误差补偿模型计算机器人移动至实际坐标的补偿参数,基于补偿参数,控制机器人到达目标位置。在该实施例中,可以采用不同采样间隔的移动点集作为训练集进行误差补偿模型训练和补偿测试,并建立机器人法兰中心点理论位置与实际位置之间的误差补偿模型,通过误差补偿模型可对机器人的工作区域内任意一点的绝对定位精度进行补偿,补偿速度快,且绝对定位精度高、泛化性能好,能够满足机器人的定位精度需求,从而解决相关技术中机器人定位误差补偿速度较慢,且适应性低,无法满足机器人的定位精度补充需求的技术问题。
本申请实施例涉及的机器人包括但不限于:工业机器人(如六轴机器人)、教育机器人,利用机器人可以实现焊接、钻铆、打磨、喷漆等场景的应用,提高机器人在零部件自动化作业时的绝对定位精度。
下面结合上述各实施步骤来说明本申请实施例。
步骤S102,采集机器人在预设空间网格移动时的移动点集,其中,空间网格是指在机器人的工作区域按照预定步长的立方体划分的网格。
通过对机器人的空间网格进行采样,可以得到移动点集。
可选的,采集机器人在预设空间网格移动时的移动点集的步骤,包括:以机器人的基坐标系为基准,以预定步长作为划分规则,将机器人的工作区域划分为多个空间立方体;从机器人的工作区域的第一个基点开始,按照预设采集规则在多个空间立方体中采集机器人末端的采样点的三维坐标;基于三维坐标,确定机器人在移动时的移动点集。
图2是根据本申请实施例的一种可选的空间网格采样点规划的示意图,如图2所示,以机器人基坐标系为基准,在机器人工作空间中首先规划出实际工作区域,即立方体P1P2P3P4P5P6P7P8,在工作区域中以距离d为预定步长(固定采样步长),采样路径可以为:在面P1P2P3P4上进行Z字形采样,即从起始点P1开始至P2再回到P1P3上距P1预定步长的位置采集第2行,直至采样至P4,再以相同的方式采集第2层,直至采集完采样终点P8。每次采样均以机器人零点位置为起始位置,以相同的姿态到达采样点,然后使用测量设备对机器人末端的实际位置进行三维坐标测量,从而得到移动点集。
可选的,在采集机器人在预设空间网格移动时的移动点集之后,定位补偿方法还包括:将多个空间立方体划分为多个采样层;采集机器人在每个采样层中各坐标系方向的变化参数和变化特征;基于变化参数和变化特征,建立误差补偿模型。
上述各采样层的坐标系方向包括:X方向、Y方向、Z方向。
另一种可选的,在利用移动点集训练得到误差补偿模型之后,定位补偿方法还包括:基于每个采样点的实际坐标和理论坐标,确定与每个采样点对应的误差向量和绝对误差值;基于误差向量和绝对误差值,确定与每个采样点对应的预测偏差值;基于每个采样点的预测偏差值,计算预测偏差平均值;基于预测偏差平均值,筛选目标取值区间,其中,目标取值区间用于优化误差补偿模型的计算准确度。
基于图2所示的采样空间,记第i个采样点(即图中黑色圆点)理论坐标为T i(x i,y i,z i)(i=1:n),实际坐标为Pi(x′ i,y′ i,z′ i)(i=1:n),通过第一公式表示第i个采样点的误差向量,其中第一公式为:
e i=P i-T i=(x′ i-x i,y′ i-y i,z′ i-z i),式(1);
令Δx i=x′ i-x i,Δy i=y′ i-y i,Δz i=z′ i-z i,则第一公式可表示如下第二公式:
e i=(Δx i,Δy i,Δz i),式(2);
然后,通过第三公式表示绝对误差值,该绝对误差值为欧氏距离,第三公式为:
Figure PCTCN2020139939-appb-000001
结合各采样点的采集路径,比较各采样层(与面P1P2P3P4平行)之间的绝对误差值可反映定位误差大小沿机器人基坐标系Z方向的变化情况,比较单独采样层上各采样行(与线段P1P2平行)之间的绝对误差值可反映定位误差大小沿机器人基坐标系X方向的变化情况,比较单独采样行上各采样点之间误差绝对值可反映定位误差大小沿机器人基坐标系Y方向的变化情况。
本申请实施例,通过分别取与P1P5或P1P3或P1P2平行的线段上的点进行定位误差向量方向的比较,得到定位误差方向沿机器人基坐标系各方向的变化特征,机器人定位误差的大小和方向沿其基坐标系不同坐标轴方向都存在着确定的连续变化特征,该变化特征可以包括:
(1)、绝对误差值随着采样点Z坐标的增加基本不变;
(2)、绝对误差值随着采样点X坐标的减小而增大;
(3)、绝对误差值随着采样点Y坐标的增大而减小,但在Y=0处会突变增大。
在得到变化特征后,可以基于变化参数和变化特征,进一步建立误差模型进而进行机器人定位精度补偿。
步骤S104,利用移动点集训练得到误差补偿模型,其中,误差补偿模型是根据机器人的法兰中心点的理论位置与实际位置之间的误差补偿参数建立的模型。
在本申请实施例,可利用极限学习算法ELM构建误差补偿模型,该ELM的结构可以理解为单隐含层的反馈神经网络。
可选的,利用移动点集训练得到误差补偿模型包括:以第一向量表示移动点集中每个采样点的对应的实际坐标,并以第二向量表示移动点集中每个采样点对应的理论坐标;确定构建误差补偿模型的反馈神经网络,反馈神经网络包括:输入层、隐含层、输出层;将第一向量和第二向量组成的向量集输入至反馈神经网络的输入层;确定输入层的神经元与隐含层的神经元之间的第一连接权值,并确定隐含层的神经元与输出层的神经元之间的第二连接权值;基于第一连接权值和第二连接权值,确定采样点的理论坐标集和测试坐标集;基于理论坐标集确定误差补偿模型,并利用测试坐标集对误差补偿模型进行测试。
图3是根据本申请实施例的一种可选的反馈神经网络的结构示意图,如图3所示,其包括输入层、隐含层、输出层。
输入层与隐含层、隐含层与输出层神经元间全连接。通过I 1,I 2,I 3……I N表示输入层神经元,通过O 1,O 2,O 3……O M表示输出层神经元,通过q 1,q 2,q 3……q L表示隐含层神经元,w ij表示输入层第i(i=1:N)个神经元与隐含层第j(j=1:L)个神经元之间的第一连接权值,β jk表示隐含层第j(j=1:L)个神经元与输出层第k(k=1:M)个神经元之间的第二连接权值.令向量P i、t i(i=1:n)分别表示采样点的实际坐标和理论坐标。
下面分别通过第四公式确定实际坐标,第五公式确定理论坐标,
第四公式:P i=[x′ i y′ i z′ i] T,i=1:n,式(4);
第五公式:t i=[x i y i z i] T,i=1:n,式(5);
以p i、t i为ELM算法的一组对应输入和输出,本申请可先定义M=N=3,x′ i y′ i z′ i和x i y i z i分别对应着I 1,I 2,I 3和O 1,O 2,O 3,则ELM算法的全部输入P(通过第六公式表示)和输出T(通过第七公式表示)为:
第六公式:
Figure PCTCN2020139939-appb-000002
第七公式:
Figure PCTCN2020139939-appb-000003
另一种可选的,确定输入层的神经元与隐含层的神经元之间的第一连接权值,并确定隐含层的神经元与输出层的神经元之间的第二连接权值包括:确定隐含层的神经元个数;基于神经元个数,产生输入层的神经元与隐含层的神经元之间的第一连接权值;确定构建误差补偿模型的激活函数和隐含层输出矩阵;基于激活函数、隐含层输出矩阵、第一连接权值和隐含层的神经元数量阈值,确定隐含层的神经元与输出层的神经元之间的第二连接权值。
设定隐含层神经元个数L,并随机产生输入层与隐含层之间的第一连接权值w、隐含层神经元阈值b。通过第八公式表示第一连接权值w,通过第九公式表示神经元阈值,其中,第八公式为:
Figure PCTCN2020139939-appb-000004
第九公式为:
Figure PCTCN2020139939-appb-000005
设定ELM算法的激活函数为f(x)(f(x)在任意区间无限可微),则此时仅有隐含层与输出层之间的第二连接权值β没有确定,且β为L×3维矩阵。根据图3的线性网络,可通过第十公式重新表示第五公式,第十公式为:
Figure PCTCN2020139939-appb-000006
其中:w j=[w j1 w j2 w 3],j=1:L。
式(10)可进一步表示为式(11):Hβ=T。
其中,H和T均为已知量,H为反馈神经网络中的隐含层输出矩阵,由输入P、激活函数f(x)、第一连接权值w和阈值b四个已知量表示。
为防止ELM算法出现过拟合的问题,隐层神经元个数L的取值小于训练样本数量n,因此隐含层和输出层之间的第二连接权值β的解为式(12):
Figure PCTCN2020139939-appb-000007
其中,H +为输出矩阵的广义逆矩阵。
由于作为ELM算法参数的隐含层神经元个数L的取值对误差补偿模型泛化性能影响较大,使用过少的隐含层神经元会导致拟合模型误差较大,过多的隐含层神经元会使误差模型出现过拟合,且降低模型训练效率,因此本申请实施例中,进一步调整L数值对补偿模型进行优化。
图4是根据本申请实施例的一种可选的构建误差补偿模型的示意图,如图4所示,在使用ELM算法进行训练时,通过采样点理论坐标和采样点实际坐标,结合选取的激活函数和设定的隐含层神经元个数进行训练,得到误差补偿模型,而在误差补偿模型的优化阶段,结合测试点的实际坐标,得到预测后的测试点预测坐标,并基于测试点的理论坐标,判断定位精度是否符合要求,若是符合,则可以进一步优化该误差补偿 模型,若是不符合,则需要重新调整隐含层的神经元个数,从而进一步对补偿模型进行优化。
为优化神经元个数L取值,取采样点包络范围内的另一组采集点作为误差补偿算法中优化目标的测试点,分别记其理论坐标集和实际测量坐标集为X t和Y p。通过式(13)表示理论坐标集,通过式(14)表示实际测量坐标集。
Figure PCTCN2020139939-appb-000008
Figure PCTCN2020139939-appb-000009
其中,m表示测试点集样本数。
以实际坐标集Y p为ELM算法的输入,则由训练后的式(10)可计算得到误差模型预测的理论坐标,记为X f。通过式(15)表示理论坐标,
Figure PCTCN2020139939-appb-000010
误差补偿模型的预测偏差E为测试点集的理论坐标X t和误差模型预测的理论坐标X f的差通过式(16)表示:
Figure PCTCN2020139939-appb-000011
由式(1)和(3),可得第j(j=1:m)个测试点的预测偏差通过式(17)表示:
Figure PCTCN2020139939-appb-000012
其中,
Figure PCTCN2020139939-appb-000013
预测偏差平均值E A通过式(18)表示:
Figure PCTCN2020139939-appb-000014
预测偏差E A是神经元个数L的函数,该值越小,表明在相应的L取值下建立的误差模型泛化性能越好.编写程序自动调节隐含层神经元个数L。具体是从训练集样本数n的1%开始取值,至与训练集样本数n相同为止,根据E A值的大小筛选出泛化性能好的L取值区间。由于L取值决定ELM算法线性方程组中方程的个数,因此为提高训练求解及预测效率,在泛化性能好的L取值区间中取L值较小时的权值w、β、阈值b代入式(10),得到优化后的误差补偿模型。
步骤S106,获取机器人拟到达的目标位置的实际坐标。
在确定机器人拟到达的目标位置的实际坐标后,通过误差补偿模型,计算机器人移动时的理论坐标,然后通过补偿参数,控制机器人精确到达目标位置。
步骤S108,利用误差补偿模型计算机器人移动至实际坐标的补偿参数,其中,补偿参数设置为对机器人移动至实际坐标的定位精度进行补偿。
步骤S110,基于补偿参数,控制机器人到达目标位置。
下面通过另一种可选的实施例来说明本申请。
本申请实施例,以GR20A型6自由度工业机器人为例进行说明,以分析机器人定位误差规律并完成基于ELM算法的误差补偿实验,测量设备使用的是3维三光学测量仪。
在采集得到多个采样点数据(例如,采集13点×13行×13层=2197组采样点数据)后,在所规划的采样立方体中,随机采集第一组数据点作为误差模型优化测试的测试点,再随机采集第二组数据点作为误差补偿模型实际补偿效果的验证点,图5是根据本申请实施例的一种可选的工业机器人各坐标系转换关系的示意图,如图5所示,包括建立以机器人底座中心点的基坐标系,法兰盘中心坐标系,世界坐标系,和测量系统坐标系。
图6是根据本申请实施例的一种可选的机器人各坐标系的转换方法的流程图,如图6所示,可以先由测量系统建立法兰坐标系、世界坐标系及测量系统坐标系三者之间的关系,进而分别转动机器人A1、A6轴,同时测量各转动位置下标记点在世界坐标系下的坐标,测量世界坐标系下标记点的初始位置坐标,同时测量各理论位置下标记点在世界坐标系下坐标。在测量各转动位置下标记点在世界坐标系下的坐标后,可以分别对A1、A6轴转动时测量得到的标记点作圆拟合,进而得到世界坐标系下的机器人x轴和z轴,结合初始位置下的法兰盘中心在世界坐标系下的坐标,可以确定世界坐 标系与机器人基坐标系的转换关系,确定世界坐标系与机器人基坐标系之间的平移关系。而在测量各理论位置下标记点在世界坐标系下坐标之后,结合标记点与法兰盘中心偏差参数,确定各理论位置下法兰盘中心在世界坐标系下的坐标,进而确定各理论位置下机器人法兰盘中心在其基坐标系下的实际坐标。
对实验采集的第一组验证点的3维坐标进行坐标转换之后,由式(1)-(3)进行绝对误差计算,得出补偿前的误差范围和误差值。根据提出的误差补偿模型,求得补偿后的绝对误差情况,将其与误差补偿前的绝对误差值进行对比,结果显示利用ELM算法采用不同采样间隔的点集作为训练集进行误差补偿模型训练和补偿测试均能有效提高机器人绝对定位精度。
下面通过另一种可选的实施例来说明本申请。
图7是根据本申请实施例的一种可选的机器人的定位补偿装置的示意图,如图7所示,该定位补偿装置可以包括:采集单元71、训练单元73、获取单元75、移动单元77、控制单元79,其中,
采集单元71,设置为采集机器人在预设空间网格移动时的移动点集,其中,空间网格是指在机器人的工作区域按照预定步长的立方体划分的网格;
训练单元73,设置为利用移动点集训练得到误差补偿模型,其中,误差补偿模型是根据机器人的法兰中心点的理论位置与实际位置之间的误差补偿参数建立的模型;
获取单元75,设置为获取机器人拟到达的目标位置的实际坐标;
移动单元77,设置为利用误差补偿模型计算机器人移动至实际坐标的补偿参数,其中,补偿参数用于对机器人移动至实际坐标的定位精度进行补偿;
控制单元79,设置为基于补偿参数,控制机器人到达目标位置。
上述机器人的定位补偿装置,可以在对机器人定位精度进行补偿时,先通过采集单元71采集机器人在预设空间网格移动时的移动点集,然后通过训练单元73利用移动点集训练得到误差补偿模型,之后可通过获取单元75获取机器人拟到达的目标位置的实际坐标,通过移动单元77利用误差补偿模型计算机器人移动至实际坐标的补偿参数,通过控制单元79基于补偿参数,控制机器人到达目标位置。在该实施例中,可以采用不同采样间隔的移动点集作为训练集进行误差补偿模型训练和补偿测试,并建立机器人法兰中心点理论位置与实际位置之间的误差补偿模型,通过误差补偿模型可对机器人的工作区域内任意一点的绝对定位精度进行补偿,补偿速度快,且绝对定位精 度高、泛化性能好,能够满足机器人的定位精度需求,从而解决相关技术中机器人定位误差补偿速度较慢,且适应性低,无法满足机器人的定位精度补充需求的技术问题。
可选的,采集单元包括:第一划分模块,设置为以机器人的基坐标系为基准,以预定步长作为划分规则,将机器人的工作区域划分为多个空间立方体;第一采集模块,设置为从机器人的工作区域的第一个基点开始,按照预设采集规则在多个空间立方体中采集机器人末端的采样点的三维坐标;第一确定模块,设置为基于三维坐标,确定机器人在移动时的移动点集。
在本申请实施例中,定位补偿装置还包括:第二划分模块,设置为在采集机器人在预设空间网格移动时的移动点集之后,将多个空间立方体划分为多个采样层;第二采集模块,设置为采集机器人在每个采样层中各坐标系方向的变化参数和变化特征;第一建立模块,设置为基于变化参数和变化特征,建立误差补偿模型。
可选的,训练单元包括:第二确定模块,设置为以第一向量表示移动点集中每个采样点的对应的实际坐标,并以第二向量表示移动点集中每个采样点对应的理论坐标;第三确定模块,设置为确定构建误差补偿模型的反馈神经网络,反馈神经网络包括:输入层、隐含层、输出层;第一输入模块,设置为将第一向量和第二向量组成的向量集输入至反馈神经网络的输入层;第四确定模块,设置为确定输入层的神经元与隐含层的神经元之间的第一连接权值,并确定隐含层的神经元与输出层的神经元之间的第二连接权值;第五确定模块,设置为基于第一连接权值和第二连接权值,确定采样点的理论坐标集和测试坐标集;第六确定模块,设置为基于理论坐标集确定误差补偿模型,并利用测试坐标集对误差补偿模型进行测试。
另一种可选的,第四确定模块包括:第一确定子模块,设置为确定隐含层的神经元个数;第一产生模块,设置为基于神经元个数,产生输入层的神经元与隐含层的神经元之间的第一连接权值;第二确定子模块,设置为确定构建误差补偿模型的激活函数和隐含层输出矩阵;第三确定子模块,设置为基于激活函数、隐含层输出矩阵、第一连接权值和隐含层的神经元数量阈值,确定隐含层的神经元与输出层的神经元之间的第二连接权值。
可选的,定位补偿装置还包括:第七确定模块,设置为在利用移动点集训练得到误差补偿模型之后,基于每个采样点的实际坐标和理论坐标,确定与每个采样点对应的误差向量和绝对误差值;第八确定模块,设置为基于误差向量和绝对误差值,确定与每个采样点对应的预测偏差值;计算模块,设置为基于每个采样点的预测偏差值,计算预测偏差平均值;筛选模块,设置为基于预测偏差平均值,筛选目标取值区间,其中,目标取值区间设置为优化误差补偿模型的计算准确度。
可选的,利用极限学习算法ELM构建误差补偿模型。
上述的机器人的定位补偿装置还可以包括处理器和存储器,上述采集单元71、训练单元73、获取单元75、移动单元77、控制单元79等均作为程序单元存储在存储器中,由处理器执行存储在存储器中的上述程序单元来实现相应的功能。
上述处理器中包含内核,由内核去存储器中调取相应的程序单元。内核可以设置一个或以上,通过调整内核参数来补偿机器人在定位时的精度。
上述存储器可能包括计算机可读介质中的非永久性存储器,随机存取存储器(RAM)和/或非易失性内存等形式,如只读存储器(ROM)或闪存(flash RAM),存储器包括至少一个存储芯片。
根据本申请实施例的另一方面,还提供了一种工业机器人,包括:处理器;以及存储器,设置为存储处理器的可执行指令;其中,处理器配置为经由执行可执行指令来执行上述任意一项的机器人的定位补偿方法。
根据本申请实施例的另一方面,还提供了一种计算机存储介质,计算机存储介质包括存储的程序,其中,在程序运行时控制计算机存储介质所在设备执行上述任意一项的机器人的定位补偿方法。
本申请还提供了一种计算机程序产品,当在数据处理设备上执行时,适于执行初始化有如下方法步骤的程序:采集机器人在预设空间网格移动时的移动点集,其中,空间网格是指在机器人的工作区域按照预定步长的立方体划分的网格;利用移动点集训练得到误差补偿模型,其中,误差补偿模型是根据机器人的法兰中心点的理论位置与实际位置之间的误差补偿参数建立的模型;获取机器人拟到达的目标位置的实际坐标;利用误差补偿模型计算机器人移动至实际坐标的补偿参数,其中,补偿参数设置为对机器人移动至实际坐标的定位精度进行补偿;基于补偿参数,控制机器人到达目标位置。
上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。
在本申请的上述实施例中,对各个实施例的描述都各有侧重,某个实施例中没有详述的部分,可以参见其他实施例的相关描述。
在本申请所提供的几个实施例中,应该理解到,所揭露的技术内容,可通过其它的方式实现。其中,以上所描述的装置实施例仅仅是示意性的,例如所述单元的划分,可以为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。另一点,所 显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,单元或模块的间接耦合或通信连接,可以是电性或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本申请各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可为个人计算机、服务器或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括:U盘、只读存储器(ROM,Read-Only Memory)、随机存取存储器(RAM,Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述仅是本申请的优选实施方式,应当指出,对于本技术领域的普通技术人员来说,在不脱离本申请原理的前提下,还可以做出若干改进和润饰,这些改进和润饰也应视为本申请的保护范围。
工业实用性
本申请实施例提供的方案可以用于机器人的控制领域,适用于家庭服务定位、工业生产定位等环境,在本申请实施例中,对机器人定位精度进行补偿时,先采集机器人在预设空间网格移动时的移动点集,然后利用移动点集训练得到误差补偿模型,之后可获取机器人拟到达的目标位置的实际坐标,利用误差补偿模型计算机器人移动至实际坐标的补偿参数,基于补偿参数,控制机器人到达目标位置,从而通过误差补偿模型对机器人的工作区域内任意一点的绝对定位精度进行补偿,补偿速度快,且绝对定位精度高、泛化性能好,能够满足机器人的定位精度需求,从而解决相关技术中机器人定位误差补偿速度较慢,且适应性低,无法满足机器人的定位精度补充需求的技术问题。

Claims (10)

  1. 一种机器人的定位补偿方法,包括:
    采集机器人在预设空间网格移动时的移动点集,其中,所述空间网格是指在机器人的工作区域按照预定步长的立方体划分的网格;
    利用所述移动点集训练得到误差补偿模型,其中,所述误差补偿模型是根据所述机器人的法兰中心点的理论位置与实际位置之间的误差补偿参数建立的模型;
    获取机器人拟到达的目标位置的实际坐标;
    利用所述误差补偿模型计算所述机器人移动至所述实际坐标的补偿参数,其中,所述补偿参数用于对所述机器人移动至所述实际坐标的定位精度进行补偿;
    基于所述补偿参数,控制所述机器人到达所述目标位置。
  2. 根据权利要求1所述的定位补偿方法,其中,采集机器人在预设空间网格移动时的移动点集的步骤,包括:
    以所述机器人的基坐标系为基准,以所述预定步长作为划分规则,将所述机器人的工作区域划分为多个空间立方体;
    从所述机器人的工作区域的第一个基点开始,按照预设采集规则在多个空间立方体中采集所述机器人末端的采样点的三维坐标;
    基于所述三维坐标,确定所述机器人在移动时的移动点集。
  3. 根据权利要求2所述的定位补偿方法,其中,在采集机器人在预设空间网格移动时的移动点集之后,所述定位补偿方法还包括:
    将所述多个空间立方体划分为多个采样层;
    采集所述机器人在每个采样层中各坐标系方向的变化参数和变化特征;
    基于所述变化参数和所述变化特征,建立所述误差补偿模型。
  4. 根据权利要求2所述的定位补偿方法,其中,利用所述移动点集训练得到误差补偿模型包括:
    以第一向量表示所述移动点集中每个采样点的对应的实际坐标,并以第二向量表示所述移动点集中每个采样点对应的理论坐标;
    确定构建所述误差补偿模型的反馈神经网络,所述反馈神经网络包括:输入层、隐含层、输出层;
    将所述第一向量和第二向量组成的向量集输入至所述反馈神经网络的输入层;
    确定所述输入层的神经元与所述隐含层的神经元之间的第一连接权值,并确定所述隐含层的神经元与所述输出层的神经元之间的第二连接权值;
    基于所述第一连接权值和所述第二连接权值,确定采样点的理论坐标集和测试坐标集;
    基于所述理论坐标集确定所述误差补偿模型,并利用所述测试坐标集对所述误差补偿模型进行测试。
  5. 根据权利要求4所述的定位补偿方法,其中,确定所述输入层的神经元与所述隐含层的神经元之间的第一连接权值,并确定所述隐含层的神经元与所述输出层的神经元之间的第二连接权值包括:
    确定所述隐含层的神经元个数;
    基于所述神经元个数,产生所述输入层的神经元与所述隐含层的神经元之间的第一连接权值;
    确定构建误差补偿模型的激活函数和隐含层输出矩阵;
    基于所述激活函数、所述隐含层输出矩阵、所述第一连接权值和所述隐含层的神经元数量阈值,确定所述隐含层的神经元与所述输出层的神经元之间的第二连接权值。
  6. 根据权利要求4所述的定位补偿方法,其中,在利用所述移动点集训练得到误差补偿模型之后,所述定位补偿方法还包括:
    基于每个采样点的实际坐标和理论坐标,确定与每个采样点对应的误差向量和绝对误差值;
    基于所述误差向量和所述绝对误差值,确定与每个采样点对应的预测偏差值;
    基于每个采样点的预测偏差值,计算预测偏差平均值;
    基于所述预测偏差平均值,筛选目标取值区间,其中,所述目标取值区间设置为优化所述误差补偿模型的计算准确度。
  7. 根据权利要求1所述的定位补偿方法,其中,利用极限学习算法ELM构建所述误 差补偿模型。
  8. 一种机器人的定位补偿装置,包括:
    采集单元,设置为采集机器人在预设空间网格移动时的移动点集,其中,所述空间网格是指在机器人的工作区域按照预定步长的立方体划分的网格;
    训练单元,设置为利用所述移动点集训练得到误差补偿模型,其中,所述误差补偿模型是根据所述机器人的法兰中心点的理论位置与实际位置之间的误差补偿参数建立的模型;
    获取单元,设置为获取机器人拟到达的目标位置的实际坐标;
    移动单元,设置为利用所述误差补偿模型计算所述机器人移动至所述实际坐标的补偿参数,其中,所述补偿参数设置为对所述机器人移动至所述实际坐标的定位精度进行补偿;
    控制单元,设置为基于所述补偿参数,控制所述机器人到达所述目标位置。
  9. 一种工业机器人,包括:
    处理器;以及
    存储器,设置为存储所述处理器的可执行指令;
    其中,所述处理器配置为经由执行所述可执行指令来执行权利要求1至7中任意一项所述的机器人的定位补偿方法。
  10. 一种计算机存储介质,所述计算机存储介质包括存储的程序,其中,在所述程序运行时控制所述计算机存储介质所在设备执行权利要求1至7中任意一项所述的机器人的定位补偿方法。
PCT/CN2020/139939 2020-05-26 2020-12-28 机器人的定位补偿方法及装置 WO2021238191A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010456895.5A CN111590581B (zh) 2020-05-26 2020-05-26 机器人的定位补偿方法及装置
CN202010456895.5 2020-05-26

Publications (1)

Publication Number Publication Date
WO2021238191A1 true WO2021238191A1 (zh) 2021-12-02

Family

ID=72181550

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/139939 WO2021238191A1 (zh) 2020-05-26 2020-12-28 机器人的定位补偿方法及装置

Country Status (2)

Country Link
CN (1) CN111590581B (zh)
WO (1) WO2021238191A1 (zh)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114211173A (zh) * 2022-01-27 2022-03-22 上海电气集团股份有限公司 一种确定焊接位置的方法、装置及系统
CN114310873A (zh) * 2021-12-17 2022-04-12 上海术航机器人有限公司 位姿转换模型的生成方法、控制方法、系统、设备和介质
CN114347018A (zh) * 2021-12-20 2022-04-15 上海大学 一种基于小波神经网络的机械臂扰动补偿方法
CN114378816A (zh) * 2021-12-27 2022-04-22 歌尔股份有限公司 装备运动轨迹的纠偏方法、控制装置及存储介质
CN114474004A (zh) * 2022-03-18 2022-05-13 河北工业大学 一种多因素耦合车载建筑机器人误差补偿规划控制策略
CN114523475A (zh) * 2022-03-01 2022-05-24 南京理工大学 一种机器人装配系统误差自动标定与补偿装置及方法
CN114683279A (zh) * 2022-02-17 2022-07-01 广东博智林机器人有限公司 导航精度的确定方法、装置和电子设备
CN114734440A (zh) * 2022-04-15 2022-07-12 同济大学 一种基于upf-rbf组合模型的混联双臂搬运机器人运动学参数精准标定方法
CN114993166A (zh) * 2022-04-11 2022-09-02 中国船舶重工集团公司第七一九研究所 位敏探测器的标定方法、装置及系统
CN115179289A (zh) * 2022-07-18 2022-10-14 天津大学 一种面向数字孪生的机器人标定方法
CN115183716A (zh) * 2022-09-10 2022-10-14 武汉光昱明晟智能科技有限公司 一种基于智能导航机器人的土方测量方法及系统
CN115543096A (zh) * 2022-12-02 2022-12-30 世优(北京)科技有限公司 基于神经网络的动捕数据处理方法、装置及系统
CN116061176A (zh) * 2022-12-27 2023-05-05 哈尔滨思哲睿智能医疗设备股份有限公司 运动补偿方法、装置、电子设备及存储介质
CN116113229A (zh) * 2023-04-14 2023-05-12 合肥安迅精密技术有限公司 保证多把飞达供料点一致性的方法及装置、存储介质
CN117518984A (zh) * 2023-11-29 2024-02-06 山东大学 机器人加工轮廓误差补偿方法、装置及介质

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111590581B (zh) * 2020-05-26 2021-10-22 珠海格力智能装备有限公司 机器人的定位补偿方法及装置
CN112643669B (zh) * 2020-12-04 2022-01-18 广州机械科学研究院有限公司 一种机器人位置偏差补偿方法、系统、装置及存储介质
CN112551044B (zh) * 2020-12-10 2022-09-27 惠州市恒泰科技股份有限公司 待化成电芯上料方法及装置
CN113514050B (zh) * 2021-04-01 2024-05-28 佛山中车四方轨道车辆有限公司 定位方法、系统、介质、设备、移动平台以及检修生产线
CN113635312B (zh) * 2021-10-18 2022-02-15 中国电子产品可靠性与环境试验研究所((工业和信息化部电子第五研究所)(中国赛宝实验室)) 工业机器人运动精度补偿方法、系统和计算机设备
CN114001655B (zh) * 2021-11-02 2022-11-15 深圳市腾盛精密装备股份有限公司 确定机台设备的精度的方法及装置
CN115416018B (zh) * 2022-08-17 2024-03-15 雅客智慧(北京)科技有限公司 末端执行器形变补偿方法、装置、电子设备和存储介质
CN115648228B (zh) * 2022-12-28 2023-05-26 广东隆崎机器人有限公司 工业机器人多源误差补偿方法、装置、设备及存储介质
CN116079696B (zh) * 2023-04-07 2023-06-13 广东毕要科技有限公司 基于振动信号的scara机器人定位更新方法及装置
CN117482417B (zh) * 2023-12-14 2024-07-30 合肥中科离子医学技术装备有限公司 治疗床摆位系统的空间移动精度补偿方法及精度补偿装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH07261821A (ja) * 1994-03-16 1995-10-13 Fanuc Ltd 負荷による撓みを考慮したロボット軌道計画方法
CN108818540A (zh) * 2018-08-28 2018-11-16 珠海格力智能装备有限公司 补偿参数的方法及装置、处理器和存储介质
CN110385720A (zh) * 2019-07-26 2019-10-29 南京航空航天大学 一种基于深度神经网络的机器人定位误差补偿方法
CN111055293A (zh) * 2019-11-21 2020-04-24 东南大学 一种基于曲面自适应的工业机器人高精度恒力打磨方法
CN111590581A (zh) * 2020-05-26 2020-08-28 珠海格力智能装备有限公司 机器人的定位补偿方法及装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0589394A1 (en) * 1992-09-21 1994-03-30 Toyoda Koki Kabushiki Kaisha Method and apparatus for controlling a robot using a neural network
CN105353301B (zh) * 2015-10-23 2018-03-02 华北电力大学(保定) 一种断路器状态评估参数获取方法
CN106737855B (zh) * 2016-08-22 2019-07-02 南京理工大学 一种综合位姿误差模型与刚度补偿的机器人精度补偿方法
CN106407659A (zh) * 2016-08-31 2017-02-15 中国地质大学(武汉) 一种空气质量指数预测方法和装置
CN110039537B (zh) * 2019-03-15 2021-07-13 北京精密机电控制设备研究所 一种基于神经网络的在线自学习多关节运动规划方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH07261821A (ja) * 1994-03-16 1995-10-13 Fanuc Ltd 負荷による撓みを考慮したロボット軌道計画方法
CN108818540A (zh) * 2018-08-28 2018-11-16 珠海格力智能装备有限公司 补偿参数的方法及装置、处理器和存储介质
CN110385720A (zh) * 2019-07-26 2019-10-29 南京航空航天大学 一种基于深度神经网络的机器人定位误差补偿方法
CN111055293A (zh) * 2019-11-21 2020-04-24 东南大学 一种基于曲面自适应的工业机器人高精度恒力打磨方法
CN111590581A (zh) * 2020-05-26 2020-08-28 珠海格力智能装备有限公司 机器人的定位补偿方法及装置

Cited By (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114310873A (zh) * 2021-12-17 2022-04-12 上海术航机器人有限公司 位姿转换模型的生成方法、控制方法、系统、设备和介质
CN114310873B (zh) * 2021-12-17 2024-05-24 上海术航机器人有限公司 位姿转换模型的生成方法、控制方法、系统、设备和介质
CN114347018A (zh) * 2021-12-20 2022-04-15 上海大学 一种基于小波神经网络的机械臂扰动补偿方法
CN114347018B (zh) * 2021-12-20 2024-04-16 上海大学 一种基于小波神经网络的机械臂扰动补偿方法
CN114378816A (zh) * 2021-12-27 2022-04-22 歌尔股份有限公司 装备运动轨迹的纠偏方法、控制装置及存储介质
CN114211173B (zh) * 2022-01-27 2024-05-31 上海电气集团股份有限公司 一种确定焊接位置的方法、装置及系统
CN114211173A (zh) * 2022-01-27 2022-03-22 上海电气集团股份有限公司 一种确定焊接位置的方法、装置及系统
CN114683279B (zh) * 2022-02-17 2023-05-23 广东博智林机器人有限公司 导航精度的确定方法、装置和电子设备
CN114683279A (zh) * 2022-02-17 2022-07-01 广东博智林机器人有限公司 导航精度的确定方法、装置和电子设备
CN114523475A (zh) * 2022-03-01 2022-05-24 南京理工大学 一种机器人装配系统误差自动标定与补偿装置及方法
CN114474004B (zh) * 2022-03-18 2023-10-03 河北工业大学 一种多因素耦合车载建筑机器人误差补偿规划控制策略
CN114474004A (zh) * 2022-03-18 2022-05-13 河北工业大学 一种多因素耦合车载建筑机器人误差补偿规划控制策略
CN114993166A (zh) * 2022-04-11 2022-09-02 中国船舶重工集团公司第七一九研究所 位敏探测器的标定方法、装置及系统
CN114734440A (zh) * 2022-04-15 2022-07-12 同济大学 一种基于upf-rbf组合模型的混联双臂搬运机器人运动学参数精准标定方法
CN114734440B (zh) * 2022-04-15 2023-09-05 同济大学 一种混联双臂搬运机器人运动学参数精准标定方法
CN115179289A (zh) * 2022-07-18 2022-10-14 天津大学 一种面向数字孪生的机器人标定方法
CN115183716B (zh) * 2022-09-10 2023-01-13 武汉光昱明晟智能科技有限公司 一种基于智能导航机器人的土方测量方法及系统
CN115183716A (zh) * 2022-09-10 2022-10-14 武汉光昱明晟智能科技有限公司 一种基于智能导航机器人的土方测量方法及系统
CN115543096A (zh) * 2022-12-02 2022-12-30 世优(北京)科技有限公司 基于神经网络的动捕数据处理方法、装置及系统
CN116061176B (zh) * 2022-12-27 2024-02-23 哈尔滨思哲睿智能医疗设备股份有限公司 运动补偿方法、装置、电子设备及存储介质
CN116061176A (zh) * 2022-12-27 2023-05-05 哈尔滨思哲睿智能医疗设备股份有限公司 运动补偿方法、装置、电子设备及存储介质
CN116113229B (zh) * 2023-04-14 2023-06-09 合肥安迅精密技术有限公司 保证多把飞达供料点一致性的方法及装置、存储介质
CN116113229A (zh) * 2023-04-14 2023-05-12 合肥安迅精密技术有限公司 保证多把飞达供料点一致性的方法及装置、存储介质
CN117518984A (zh) * 2023-11-29 2024-02-06 山东大学 机器人加工轮廓误差补偿方法、装置及介质

Also Published As

Publication number Publication date
CN111590581A (zh) 2020-08-28
CN111590581B (zh) 2021-10-22

Similar Documents

Publication Publication Date Title
WO2021238191A1 (zh) 机器人的定位补偿方法及装置
CN110900598B (zh) 机器人三维运动空间动作模仿学习方法和系统
Wang et al. A screw axis identification method for serial robot calibration based on the POE model
CN111260649B (zh) 一种近距离机械臂感知与校准方法
Liu et al. Human-centric collaborative assembly system for large-scale space deployable mechanism driven by Digital Twins and wearable AR devices
CN106826812A (zh) 机械学习器及机械学习方法、层叠芯制造装置及系统
CN112926152B (zh) 一种数字孪生驱动的薄壁件装夹力精准控制与优化方法
CN108717262B (zh) 一种基于矩特征学习神经网络的异形曲面跟踪方法及系统
CN113910218B (zh) 一种基于运动学与深度神经网络融合的机器人标定方法和装置
CN111241747A (zh) 机械臂逆运动学求解方法、系统、计算机设备及存储介质
Liu et al. A mixed perception-based human-robot collaborative maintenance approach driven by augmented reality and online deep reinforcement learning
Zhang et al. Deep learning-based robot vision: High-end tools for smart manufacturing
Wang et al. Simulation and deep learning on point clouds for robot grasping
CN113609999B (zh) 基于姿态识别的人体模型建立方法
Hsieh et al. Deep convolutional generative adversarial network for inverse kinematics of self-assembly robotic arm based on the depth sensor
Duan et al. A technique for inversely identifying joint stiffnesses of robot arms via two-way TubeNets
Yanzhe et al. Robot trajectory planning for autonomous 3D reconstruction of cockpit in aircraft final assembly testing
Tarokh et al. Classification and characterization of inverse kinematics solutions for anthropomorphic manipulators
CN104715133B (zh) 一种待辨识对象的运动学参数在轨辨识方法和装置
CN114030008B (zh) 一种基于数据驱动的工业机器人实训能耗测量方法
Bataineh et al. Artificial neural network-based prediction of human posture
CN114536351A (zh) 冗余双臂机器人示教方法、装置、电子设备及系统
Hlavac Kinematics control of a redundant planar manipulator with a MLP neural network
Pedro et al. Learning how to grasp based on neural network retraining
Wang Comparison of Positioning Error Prediction Results of Industrial Robots Based on Three Different Types of Neural Networks

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: 20937477

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 17/03/2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20937477

Country of ref document: EP

Kind code of ref document: A1