A robot pose compensation method and device

Abstract

The application provides a robot pose compensation method and device, which comprises the following steps: inputting actual poses and DH parameter prediction values and DH parameter observation values into a first model trained in advance to obtain first DH parameter error data and second DH parameter error data; performing reinforcement learning training on the first DH parameter error data and the second DH parameter error data by using a preset reward function to obtain a second model; and processing the actual poses, the DH parameter prediction values and joint angle real-time data by using the second model to determine target DH parameter error data. The deviation between the actual DH parameters of each joint of the robot and the prediction values can be clearly understood, the robot can dynamically adapt to environmental changes, the motion control strategy can be timely adjusted, the positioning accuracy can be always maintained to be high, and the stability and reliability of the robot in a complex and changeable environment can be enhanced.

Description

Technical Field

[0001] This application relates to the field of robotics, and in particular to a method and apparatus for robot pose compensation. Background Technology

[0002] In the field of robot motion control, Denavit-Hartenberg (DH) parameters are used to describe the geometric relationships between robot links. However, in practical applications, due to factors such as manufacturing tolerances, assembly deviations, and material deformation (e.g., thermal expansion caused by temperature changes), DH parameters usually have deterministic errors, which in turn lead to pose deviations of the robot's end effector, affecting the success rate of the task.

[0003] It should be noted that the above introduction to the technical background is only for the purpose of providing a clear and complete explanation of the technical solutions of this application and facilitating understanding by those skilled in the art. It should not be assumed that these technical solutions are known to those skilled in the art simply because they have been described in the background section of this application. Summary of the Invention

[0004] The purpose of this application is to at least partially solve one of the technical problems in the related art.

[0005] Therefore, the first objective of this application is to propose a robot pose compensation method.

[0006] The second objective of this application is to propose a robot pose compensation device.

[0007] The third objective of this application is to propose an electronic device.

[0008] The fourth objective of this application is to provide a non-transitory computer-readable storage medium.

[0009] The fifth objective of this application is to provide a computer program product.

[0010] To achieve the above objectives, a robot pose compensation method is proposed in the first aspect of this application, comprising:

[0011] The actual robot pose and at least one set of predicted DH parameters are input into a pre-trained first model to determine the first DH parameter error data. The actual pose and at least one set of observed DH parameters are input into the first model to determine the second DH parameter error data. The first model is trained using multiple sets of data through machine learning. The multiple sets of data include a first type of data that completes the predetermined target task and a second type of data that does not complete the predetermined target task. The first type of data and the second type of data are obtained based on the robot's historical operation data.

[0012] The actual pose, the predicted values ​​of each DH parameter, and the real-time collected data of robot joint angles are input into a pre-trained second model to determine the target DH parameter error data. The second model is obtained by reinforcement learning training on the first DH parameter error data and the second DH parameter error data through a preset reward function.

[0013] Based on the target DH parameter error data, the robot joint angles are pre-compensated.

[0014] To achieve the above objectives, a second aspect of this application provides a robot pose compensation device, comprising:

[0015] The first acquisition module is used to input the collected actual robot pose and at least one set of DH parameter prediction values ​​into a pre-trained first model to determine first DH parameter error data, and to input the actual pose and at least one set of collected DH parameter observation values ​​into the first model to determine second DH parameter error data. The first model is obtained by training multiple sets of data through machine learning. The multiple sets of data include a first type of data that completes a predetermined target task and a second type of data that does not complete the predetermined target task. The first type of data and the second type of data are obtained based on the robot's historical operation data.

[0016] The second acquisition module is used to input the actual pose, the predicted values ​​of each DH parameter, and the real-time collected data of the robot joint angle into a pre-trained second model to determine the target DH parameter error data. The second model is obtained by reinforcement learning training on the first DH parameter error data and the second DH parameter error data through a preset reward function.

[0017] The compensation module is used to pre-compensate the robot joint angles based on the target DH parameter error data.

[0018] To achieve the above objectives, a third aspect of this application provides an electronic device, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the robot pose compensation method proposed in the first aspect of this application.

[0019] To achieve the above objectives, a fourth aspect of this application provides a non-transitory computer-readable storage medium that, when the instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to perform the robot pose compensation method proposed in the first aspect of this application.

[0020] To achieve the above objectives, a fifth aspect of this application provides a computer program product, including a computer program that, when executed by a processor in a communication device, implements the robot pose compensation method proposed in the first aspect of this application.

[0021] In this embodiment, by inputting the actual pose, predicted DH parameter values, and observed DH parameter values ​​into a pre-trained first model, first DH parameter error data and second DH parameter error data are obtained. A second model is obtained by performing reinforcement learning training on the first and second DH parameter error data using a preset reward function. The second model is then used to process the actual pose, predicted DH parameter values, and real-time joint angle data to determine the target DH parameter error data. This allows for a clear understanding of the deviation between the actual and predicted DH parameters of each joint of the robot, enabling the robot to dynamically adapt to environmental changes, adjust motion control strategies in a timely manner, maintain high positioning accuracy, and enhance the stability and reliability of the robot in complex and changing environments.

[0022] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0023] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0024] Figure 1 A flowchart illustrating a robot pose compensation method provided in an embodiment of this application;

[0025] Figure 2 This is a flowchart illustrating another robot pose compensation method provided in an embodiment of this application.

[0026] Figure 3 This is a schematic diagram of the structure of a robot pose compensation device provided in an embodiment of this application;

[0027] Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application;

[0028] Figure 5 This is a schematic diagram of the structure of another electronic device provided according to an embodiment of this application. Detailed Implementation

[0029] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0030] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of this application. The singular forms “a” and “the” as used in the embodiments of this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0031] It should be understood that although the terms first, second, third, etc., may be used to describe various information in the embodiments of this application, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words "if" and "suppose" as used herein can be interpreted as "when," "when," or "in response to a determination."

[0032] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0033] In the field of robot motion control, accurately describing the geometric relationships between the links of a robot is fundamental to achieving precise motion control. DH parameters, as a standardized mathematical descriptive tool, are widely used to describe the geometric relationships between robot links. Through a series of specific parameters, such as link length *a*, joint angle *θ*, link offset *d*, link torsion angle *α*, and the parallelism of the plane containing the link *β*, DH parameters provide a clear framework for the kinematic modeling of the robot. With the help of DH parameters, a precise mapping relationship between the robot's joint space and Cartesian space can be established, thereby enabling precise control of the position and attitude of the robot's end effector, ensuring that the robot can complete various complex tasks according to predetermined trajectories and movements.

[0034] However, in real-world robot applications, various factors can interfere with the accuracy of the DH parameters, leading to deterministic errors. Manufacturing tolerances refer to the fact that, due to limitations in processing equipment and technology, the dimensions and shape of each link are difficult to achieve absolute precision during robot manufacturing, resulting in deviations from the theoretical design values. These deviations directly affect the accuracy of parameters such as link length and link offset in the DH parameters, causing the kinematic model built based on these parameters to deviate from reality.

[0035] Assembly deviations refer to situations where, even if each link is manufactured with high precision, inaccurate installation positions or misalignment of joint axes may occur during the assembly process of assembling individual links into a complete robot. These assembly deviations cause changes in DH parameters such as joint rotation angle θ and link torsion angle α, further compromising the accuracy of the kinematic model.

[0036] Material deformation refers to the deformation of the link material due to thermal expansion and contraction when the ambient temperature of the robot changes. This deformation alters the actual length and shape of the link, resulting in dynamic changes in the DH parameters. Furthermore, long-term use and loading can also cause fatigue deformation of the link material, which can also adversely affect the accuracy of the DH parameters.

[0037] Due to the inaccuracy of the DH parameters, a deviation will occur between the theoretical pose and the actual pose of the robot end effector calculated based on these parameters. This pose deviation may cause the robot to fail to accurately reach the target position during task execution, or to collide with the surrounding environment during operation, thus affecting the successful completion of the task. In some applications with extremely high precision requirements, such as precision machining and medical surgical robots, even a small pose deviation can lead to serious consequences, reduce the success rate of the task, or even cause irreparable losses. Therefore, how to effectively reduce the deterministic error of the DH parameters and improve the pose accuracy of the robot end effector has become an important problem that urgently needs to be solved in the field of robot motion control.

[0038] The robot pose compensation method and apparatus of this application are described below with reference to the accompanying drawings.

[0039] Figure 1 This is a flowchart illustrating a robot pose compensation method provided in an embodiment of this application.

[0040] like Figure 1 As shown, the robot pose compensation method includes, but is not limited to, the following steps:

[0041] S101, the actual pose of the robot and at least one set of predicted DH parameters are input into a pre-trained first model to determine the first DH parameter error data, and the actual pose and at least one set of observed DH parameters are input into the first model to determine the second DH parameter error data. The first model is obtained by machine learning training using multiple sets of data. The multiple sets of data include a first type of data that completes the predetermined target task and a second type of data that does not complete the predetermined target task. The first type of data and the second type of data are obtained based on the robot's historical operation data.

[0042] In one feasible implementation, a laser tracker, a vision sensor, or an inertial measurement unit (IMU) is used to acquire the robot's actual pose during movement. The laser tracker determines the position and orientation of the robot's end effector in space by emitting a laser beam and measuring the time or angle of the reflected light. The vision sensor uses image processing technology to analyze images of the robot and its surrounding environment to obtain the robot's pose data. The inertial measurement unit measures the acceleration and angular velocity of the robot's end effector, and the actual pose is obtained by integrating the acceleration and angular velocity.

[0043] In one feasible implementation, the predicted DH parameters are typically derived from the robot's theoretical model and initial design parameters. During the robot design phase, a set of theoretical DH parameters is determined based on the robot's structural characteristics and motion requirements. These parameters include link length *a*, joint angle *θ*, link offset *d*, link torsion angle *α*, and the parallelism *β* of the plane containing the link. In practical applications, considering factors such as the robot's operating environment and load conditions, these theoretical parameters are modified and predicted. For example, if the robot needs to operate in a high-temperature environment, the link length may change due to the thermal expansion effect of materials. In this case, the theoretical link length can be modified based on the material's coefficient of thermal expansion to obtain the predicted DH parameters. These predicted values ​​reflect the expected state of the robot's DH parameters under ideal conditions.

[0044] In one feasible implementation, the observed DH parameters are obtained through actual measurement and estimation. In some embodiments, by measuring the angles of the robot joints and the actual pose of the end effector, and combining this with the forward kinematics equations, a system of equations concerning the DH parameters can be established, and then the observed DH parameters can be obtained by solving them through an optimization algorithm.

[0045] In some embodiments, a calibration device, such as a high-precision coordinate measuring machine, is used to measure the links of the robot to obtain the actual size and position information of the links, thereby obtaining the observed values ​​of the DH parameters.

[0046] It should be noted that when collecting DH parameter observations, in order to avoid the influence of external interference on the measurement results, multiple measurements can be performed and the average value calculated to reduce measurement error.

[0047] In one feasible implementation, the acquired actual robot pose and at least one set of predicted DH parameters are input into a pre-trained first model. The first model processes and analyzes the actual pose and predicted DH parameters based on the internally learned mapping relationship.

[0048] In some embodiments, the first model employs a neural network algorithm. Each neuron in the neural network performs a weighted summation and activation function operation on the actual pose and at least one set of predicted DH parameters. The result is then passed to the next layer, ultimately outputting the first DH parameter error data. This first DH parameter error data reflects the potential errors in the DH parameters under the current input conditions during learning based on historical data showing both completion and non-completion of the predetermined target task.

[0049] In one feasible implementation, since the first model has learned the mapping relationship between input features and DH parameter errors during training, based on the learning mechanism within the first model, the collected actual robot pose and at least one set of DH parameter observations are input into the pre-trained first model to obtain second DH parameter error data. This second DH parameter error data is then used as expert data. This expert data reflects the actual error status of the robot's DH parameters based on real-time measurement data and the patterns learned by the model.

[0050] S102, the actual pose, predicted values ​​of each DH parameter and real-time collected data of robot joint angles are input into the pre-trained second model to determine the target DH parameter error data. The second model is obtained by reinforcement learning training on the first DH parameter error data and the second DH parameter error data through a preset reward function.

[0051] In one feasible implementation, joint angle data is acquired in real time by encoders installed at the robot joints. The encoders convert the joint rotation angles into electrical signals, which are then processed to obtain precise joint angle values. This data reflects the actual motion state of each joint of the robot and is an important basis for determining the robot's actual pose and DH parameter errors.

[0052] In one feasible implementation, reinforcement learning is a machine learning method that learns optimal policies through interaction between an agent and its environment. In reinforcement learning, the process of determining the target DH parameter error data is regarded as the agent's decision-making process. The environment is the robot's motion control system, including the actual pose, predicted DH parameter values, and real-time joint angle data. The agent is a second model whose state consists of the input actual pose, predicted DH parameter values, and real-time joint angle data, and whose action is the output of the target DH parameter error data.

[0053] In one feasible implementation, the reward function is used to evaluate the quality of the agent's actions and guide the agent to learn the optimal policy. If a key indicator (such as mean squared error) in the error data of the first or second DH parameters decreases, a positive reward is assigned to the second model; conversely, if a key indicator (such as mean squared error) increases, a negative reward is assigned to the second model.

[0054] In some embodiments, the parameters of the second model are randomly initialized, and the initial state is set to the actual pose at a certain moment, a set of predicted DH parameters at that moment, and real-time collected data of robot joint angles. Based on the current state at that moment, the second model correlates the probability γ with the reduction in the error data of the first DH parameter and correlates the probability (1-γ) with the reduction in the error data of the second DH parameter.

[0055] In one feasible implementation, the actual robot pose, predicted values ​​of each DH parameter, and real-time acquired data of robot joint angles are input into the second model. The second model analyzes and processes the input data based on the excitation function to obtain the DH parameter error determination strategy that optimizes robot performance and outputs the target DH parameter error data.

[0056] S103, based on the target DH parameter error data, pre-compensates the robot joint angles.

[0057] In one feasible implementation, the Jacobian matrix at the current joint angle is determined based on the relationship between the robot's end effector and the joint angular velocity. The pose change that the end effector needs to compensate for is determined based on the target DH parameter error data. If the Jacobian matrix is ​​directional and invertible, the joint angle change is determined based on the inverse of the Jacobian matrix and the pose change; if the Jacobian matrix is ​​not directional or is not invertible, the joint angle change is determined based on the pseudo-inverse of the Jacobian matrix and the pose change. The joint angle change is added to the current joint angle to obtain the compensated joint angle, thus completing the pre-compensation of the robot's joint angle.

[0058] In summary, the robot pose compensation method provided in this application obtains first DH parameter error data and second DH parameter error data by inputting the actual pose, predicted DH parameter values, and observed DH parameter values ​​into a pre-trained first model. A second model is obtained by training the first and second DH parameter error data using reinforcement learning through a preset reward function. The second model is then used to process the actual pose, predicted DH parameter values, and real-time joint angle data to determine the target DH parameter error data. This allows for a clear understanding of the deviation between the actual and predicted DH parameters of each joint of the robot, enabling the robot to dynamically adapt to environmental changes, adjust motion control strategies in a timely manner, maintain high positioning accuracy, and enhance the stability and reliability of the robot in complex and changing environments.

[0059] Figure 2 This is a flowchart illustrating another robot pose compensation method provided in an embodiment of this application.

[0060] like Figure 2 As shown, the robot pose compensation method includes, but is not limited to, the following steps:

[0061] S201, acquire the actual pose of the robot during its movement and at least one set of predicted DH parameters, wherein the predicted DH parameters are determined based on the nominal values ​​in the robot design drawings.

[0062] In one feasible implementation, the spatial coordinates and attitude angles of the robot's end effector or key positions are directly acquired through external measuring devices (such as laser trackers, vision sensors, inertial measurement units, etc.). The spatial coordinates and attitude angles are used as the actual pose, which is represented as p = (x, y, z, qw, qx, qy, qz), where x, y, and z are spatial coordinates; qw, qx, qy, and qz are quaternions. The angular velocity is determined by the attitude angles of the end effector or key positions, and the quaternion is obtained by integrating the angular velocity.

[0063] In one feasible implementation, the lengths of the common perpendiculars of adjacent joint axes, the axial distances along the joint axes, the spatial angles between two adjacent joint axes, the marked joint zero-position angles, and the parallelism deviations between the reference planes of adjacent links are obtained from the robot design drawings. The nominal value of the link length *a* is determined based on the length of the common perpendicular of the adjacent joint axes; the nominal value of the link offset *d* is determined based on the axial distance along the joint axes; the nominal value of the link torsion angle *α* is determined based on the spatial angles between two adjacent joint axes; the nominal value of the joint rotation angle *θ* is determined based on the marked joint zero-position angles; and the nominal value of the parallelism *β* of the plane containing the link is determined based on the parallelism deviations between the reference planes of adjacent links.

[0064] In one feasible implementation, a coordinate system is established based on the robot design drawings. The direction along the joint axis is defined as the Z-axis, the X-axis is determined based on the direction of the common perpendicular of the Z-axis, and the Y-axis is determined based on the Z-axis and X-axis. Based on the coordinate system, the nominal values ​​of the link length *a*, link offset *d*, link torsion angle *α*, joint rotation angle *θ*, and the parallelism *β* of the plane containing the link, the predicted values ​​of the DH parameters are determined.

[0065] S202, the actual pose and the predicted values ​​of each DH parameter are input into the pre-trained first model to determine the first DH parameter error data. The first model is obtained by machine learning training using multiple sets of data. The multiple sets of data include a first type of data that completes the predetermined target task and a second type of data that does not complete the predetermined target task. The first type of data and the second type of data are obtained based on the robot's historical operation data.

[0066] For details on the process of obtaining the first DH parameter error data, please refer to the relevant descriptions in the above embodiments, which will not be repeated here.

[0067] In one feasible implementation, the actual pose p is generated by the forward kinematics function f (DH parameters). If there is an error in the DH parameters, the DH parameters can be represented as "DH parameter prediction value + first DH parameter error data", then p = f(DH parameter prediction value + first DH parameter error data). Each set of DH parameter prediction values ​​corresponds to a first DH parameter error data.

[0068] Exemplary illustration, Where n can be 30, This represents the iterative learning operation of a neural network.

[0069] Exemplary illustration,

[0070] S203, acquire at least one set of DH parameter observations based on a preset robot simulation platform.

[0071] In one feasible implementation, at least one set of DH parameter observations is obtained based on the robot model library built into the preset robot simulation platform or a custom imported model file.

[0072] S204, input the actual pose and the observed values ​​of each DH parameter into the first model to determine the error data of the second DH parameter.

[0073] For details on the process of obtaining the second DH parameter error data, please refer to the relevant descriptions in the above embodiments, which will not be repeated here.

[0074] S205, the actual pose, predicted values ​​of each DH parameter, and real-time collected data of robot joint angles are input into the pre-trained second model to determine the target DH parameter error data. The second model is obtained by reinforcement learning training on the first DH parameter error data and the second DH parameter error data through a preset reward function.

[0075] In one feasible implementation, the process of obtaining the reward function includes: determining the loss function during the robot's motion process based on the robot's actual pose and the predicted values ​​of each DH parameter.

[0076] In some embodiments, the predicted pose of the robot is determined based on the predicted values ​​of each DH parameter, and the loss function is determined based on a first deviation between the actual pose and the predicted pose.

[0077] For example, the loss function L can be characterized as:

[0078]

[0079] Where T is the transition matrix, which is obtained by the translation between the coordinates of the end effector and the coordinates of the robot base. The specific acquisition process will not be described here.

[0080] In one feasible implementation, the degree of difference between the first DH parameter error data and the second DH parameter error data is determined.

[0081] In some embodiments, a second deviation between the first DH parameter error data and the second DH parameter error data is obtained. The square root of the second deviation is then calculated to obtain the degree of difference.

[0082] For example, the degree of difference can be expressed as:

[0083] sqrt(first DH parameter error data - second DH parameter error data)

[0084] In one feasible implementation, a Bayesian method is used to perform confidence analysis on the loss function and the degree of difference to obtain the first dependency ratio γ related to the loss function and the second dependency ratio (1-γ) related to the degree of difference.

[0085] In one feasible implementation, the (1-γ) reward function Rew is determined based on the loss function L, the first dependency ratio γ, the difference degree, and the second dependency ratio.

[0086] In some embodiments, the reward function can be expressed as:

[0087]

[0088] For details regarding the training process of the second model and the acquisition process of the target DH parameter error data, please refer to the relevant descriptions in the above embodiments; they will not be repeated here.

[0089] S206, based on the target DH parameter error data, pre-compensate the robot joint angles.

[0090] In one feasible implementation, the compensation amount of the robot joint angle is determined based on the target DH parameter error data, and the compensation amount of the robot joint angle is used to pre-compensate the robot joint angle.

[0091] In summary, the robot pose compensation method provided in this application obtains first DH parameter error data and second DH parameter error data by inputting the actual pose, predicted DH parameter values, and observed DH parameter values ​​into a pre-trained first model. A second model is obtained by training the first and second DH parameter error data using reinforcement learning through a preset reward function. The second model is then used to process the actual pose, predicted DH parameter values, and real-time joint angle data to determine the target DH parameter error data. This allows for a clear understanding of the deviation between the actual and predicted DH parameters of each joint of the robot, enabling the robot to dynamically adapt to environmental changes, adjust motion control strategies in a timely manner, maintain high positioning accuracy, and enhance the stability and reliability of the robot in complex and changing environments.

[0092] Figure 3 This is a schematic diagram of the structure of a robot pose compensation device provided in an embodiment of this application. Figure 3 As shown, the robot pose compensation device 300 includes:

[0093] The first acquisition module 301 is used to input the actual pose of the robot and at least one set of predicted DH parameters into a pre-trained first model to determine the first DH parameter error data, and to input the actual pose and at least one set of observed DH parameters into the first model to determine the second DH parameter error data. The first model is obtained by training multiple sets of data through machine learning. The multiple sets of data include a first type of data that completes the predetermined target task and a second type of data that does not complete the predetermined target task. The first type of data and the second type of data are obtained based on the robot's historical operation data.

[0094] The second acquisition module 302 is used to input the actual pose, the predicted values ​​of each DH parameter and the real-time collected data of the robot joint angle into the pre-trained second model to determine the target DH parameter error data. The second model is obtained by reinforcement learning training on the first DH parameter error data and the second DH parameter error data through a preset reward function.

[0095] The compensation module 303 is used to pre-compensate the robot joint angles based on the target DH parameter error data.

[0096] Optionally, the first acquisition module 301 is further configured to:

[0097] The robot's actual pose and at least one set of predicted DH parameters are obtained during the robot's motion process, wherein the predicted DH parameters are determined based on the nominal values ​​in the robot's design drawings;

[0098] The actual pose and the predicted values ​​of each DH parameter are input into the pre-trained first model to determine the error data of the first DH parameter.

[0099] Optionally, the first acquisition module 301 is further configured to:

[0100] At least one set of DH parameter observations are obtained based on a pre-set robot simulation platform;

[0101] The actual pose and the observed values ​​of each DH parameter are input into the first model to determine the error data of the second DH parameter.

[0102] Figure 4 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Figure 4 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0103] like Figure 4 As shown, the electronic device 400 includes a processor 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from memory 406 into a random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the electronic device 400. The processor 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0104] The following components are connected to I / O interface 405: memory 406 including hard disks, etc.; and communication section 407 including network interface cards such as LAN (Local Area Network) cards, modems, etc., which performs communication processing via a network such as the Internet; and driver 408 is also connected to I / O interface 405 as needed.

[0105] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 407. When the computer program is executed by processor 401, it performs the functions defined in the methods of this application.

[0106] In an exemplary embodiment, a storage medium including instructions is also provided, such as a memory including instructions, which can be executed by a processor 401 of an electronic device 400 to perform the above-described method. Optionally, the storage medium may be a non-transitory computer-readable storage medium, such as a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device.

[0107] In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can transmit, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wireline, optical fiber, RF, etc., or any suitable combination thereof.

[0108] Figure 5 This is a schematic diagram of the structure of another electronic device provided according to an embodiment of this application. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this application. Figure 5 As shown, the electronic device 500 includes a processor 501 and a memory 502. The memory 502 is used to store program code, and the processor 501 is connected to the memory 502 to read program code from the memory 502 to implement the robot pose compensation method in the above embodiment.

[0109] Optionally, the number of processors 501 can be one or more.

[0110] Optionally, the electronic device may also include an interface 503, and there may be multiple interfaces 503. The interface 503 can be connected to an application and can receive data from external devices such as sensors.

[0111] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0112] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A robot pose compensation method, characterized in that, include: The actual robot pose and at least one set of predicted DH parameters are input into a pre-trained first model to determine the first DH parameter error data. The actual pose and at least one set of observed DH parameters are input into the first model to determine the second DH parameter error data. The first model is trained using multiple sets of data through machine learning. The multiple sets of data include a first type of data that completes the predetermined target task and a second type of data that does not complete the predetermined target task. The first type of data and the second type of data are obtained based on the robot's historical operation data. The actual pose, the predicted values ​​of each DH parameter, and the real-time collected data of robot joint angles are input into a pre-trained second model to determine the target DH parameter error data. The second model is obtained by reinforcement learning training on the first DH parameter error data and the second DH parameter error data through a preset reward function. Based on the target DH parameter error data, the robot joint angles are pre-compensated.

2. The method according to claim 1, characterized in that, The step of inputting the collected actual robot pose and at least one set of predicted DH parameters into a pre-trained first model to determine the first DH parameter error data includes: The robot's actual pose and at least one set of predicted DH parameters are obtained during its movement, wherein the predicted DH parameters are determined based on the nominal values ​​in the robot's design drawings. The actual pose and the predicted values ​​of each DH parameter are input into a pre-trained first model to determine the first DH parameter error data.

3. The method according to claim 1, characterized in that, The step of inputting the actual pose and at least one set of acquired DH parameter observations into the second model to determine the second DH parameter error data includes: At least one set of DH parameter observations are obtained based on a pre-set robot simulation platform; The actual pose and the observed values ​​of each DH parameter are input into the first model to determine the second DH parameter error data.

4. The method according to claim 1, characterized in that, The process of obtaining the reward function includes: The loss function during the robot's motion is determined based on the robot's actual pose and the predicted values ​​of each DH parameter. Determine the degree of difference between the first DH parameter error data and the second DH parameter error data; A confidence analysis is performed on the loss function and the difference using the Bayesian method to obtain a first dependency rate related to the loss function and a second dependency rate related to the difference, wherein the sum of the first dependency rate and the second dependency rate is 1. The reward function is determined based on the loss function, the first dependency rate, the difference, and the second dependency rate.

5. The method according to claim 4, characterized in that, The step of determining the loss function during robot motion based on the robot's actual pose and the predicted values ​​of each DH parameter includes: The predicted pose of the robot is determined based on the predicted values ​​of each of the DH parameters; The loss function is determined based on a first deviation between the actual pose and the predicted pose.

6. The method according to claim 4, characterized in that, Determining the degree of difference between the first DH parameter error data and the second DH parameter error data includes: Obtain the second deviation between the first DH parameter error data and the second DH parameter error data; The degree of difference is obtained by taking the square root of the second deviation.

7. The method according to claim 1, characterized in that, The step of pre-compensating the robot joint angles based on the target DH parameter error data includes: The compensation amount for the robot joint angle is determined based on the target DH parameter error data; The robot joint angle is pre-compensated based on the compensation amount of the robot joint angle.

8. A robot pose compensation device, characterized in that, include: The first acquisition module is used to input the collected actual robot pose and at least one set of DH parameter prediction values ​​into a pre-trained first model to determine first DH parameter error data, and to input the actual pose and at least one set of collected DH parameter observation values ​​into the first model to determine second DH parameter error data. The first model is obtained by training multiple sets of data through machine learning. The multiple sets of data include a first type of data that completes a predetermined target task and a second type of data that does not complete the predetermined target task. The first type of data and the second type of data are obtained based on the robot's historical operation data. The second acquisition module is used to input the actual pose, the predicted values ​​of each DH parameter, and the real-time collected data of the robot joint angle into a pre-trained second model to determine the target DH parameter error data. The second model is obtained by reinforcement learning training on the first DH parameter error data and the second DH parameter error data through a preset reward function. The compensation module is used to pre-compensate the robot joint angles based on the target DH parameter error data.

9. The apparatus according to claim 8, characterized in that, The first acquisition module is further configured to: The robot's actual pose and at least one set of predicted DH parameters are obtained during its movement, wherein the predicted DH parameters are determined based on the nominal values ​​in the robot's design drawings. The actual pose and the predicted values ​​of each DH parameter are input into a pre-trained first model to determine the first DH parameter error data.

10. The apparatus according to claim 8, characterized in that, The first acquisition module is further configured to: At least one set of DH parameter observations are obtained based on a pre-set robot simulation platform; The actual pose and the observed values ​​of each DH parameter are input into the first model to determine the second DH parameter error data.

Images