Rapid calibration method for connecting rod type double-arm direct-driven manipulator

Abstract

The invention relates to the technical field of manipulator calibration, in particular to a quick calibration method for a connecting rod type double-arm direct-driven manipulator. When the joint image data are detected and the feature point coordinates of the marker are extracted, the joint image data are detected by adopting the dynamic threshold Canny edge detection algorithm, and the detection precision and robustness of the Canny edge detection algorithm on the reflective point marker are improved by combining the local image statistical characteristics and the dynamic threshold adjustment strategy, so that the detection accuracy and robustness of the reflective point marker are improved. The method is especially suitable for scenes with large illumination variation and non-uniform contrast; and further, an adjustment coefficient is determined based on the global gradient, the global variance and the global average brightness of the image, that is, the influence of the local variance and the local average brightness of each pixel point in the image on high and low thresholds is constrained through the global feature of the image, so that the sensitivity of the algorithm to different image features can be flexibly controlled, and the accuracy of the algorithm is improved. And the dynamic threshold Canny edge detection algorithm is more flexible and universal.

Description

Technical Field

[0001] The present invention relates to the technical field of manipulator calibration, and in particular to a rapid calibration method for a connecting rod type dual-arm direct-drive manipulator. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, robotic arms, as a key piece of automation equipment, have been widely used in a variety of fields, including manufacturing, logistics and warehousing, medical rehabilitation, and aerospace. Robots can replace manual labor in repetitive, high-intensity, and high-precision tasks, improving production efficiency, product quality, and safety. Among them, link-type dual-arm direct-drive robots have garnered particular attention due to their unique structural advantages, such as compactness, flexible motion, high control precision, and fast response speed. Link-type dual-arm direct-drive robots typically consist of multiple joints, each directly driven by a direct-drive motor, reducing the number of transmission links and improving transmission efficiency and control accuracy. Their dual-arm structure enables them to perform complex collaborative manipulation tasks, such as dual-arm coordinated grasping and assembly. However, because the robot's motion accuracy and repeatability directly impact its performance and application effectiveness, precise calibration is essential before commissioning.

[0003] In the existing technology, there is a solution that attaches reflective point markers to the joints of the robot and uses Canny edge detection to obtain the joint coordinates of the robot. However, for the specific application scenario of the connecting rod type dual-arm direct-drive robot with large lighting changes and uneven contrast, the traditional Canny algorithm results in insufficient recognition accuracy of reflective points and weak robustness.

[0004] At the same time, existing technologies use deep learning models to obtain the theoretical position of the manipulator. In these solutions, the Nesterov accelerated gradient method is generally used to train the deep learning model to optimize the training process. However, the optimization speed of the Nesterov accelerated gradient method does not meet the expected requirements, resulting in a long manipulator calibration time. Therefore, developing a fast, efficient, and high-precision calibration method is of great significance for improving the practicality and production efficiency of the connecting rod type dual-arm direct drive manipulator. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a rapid calibration method for a connecting rod type dual-arm direct-drive manipulator, which is used to solve the problems existing in the prior art.

[0006] The present invention provides a rapid calibration method for a connecting rod type dual-arm direct drive manipulator, comprising the following steps: S1: Initialize the structural parameters of the connecting rod type dual-arm direct drive manipulator; S2: using a visual sensor to obtain actual position data of the joints of the connecting rod type dual-arm direct drive manipulator; In S2, the joint image data is detected using a dynamic threshold Canny edge detection algorithm; S3: Establishing a kinematic model of the connecting rod type dual-arm direct drive manipulator; The kinematic model is a deep neural network model. During the training process of the deep neural network model, the improved Nesterov accelerated gradient method is used to optimize the training process of the deep neural network model, an adaptive learning rate is introduced into the momentum term, and the momentum term is adjusted; S4: calculating theoretical positions of joints of the link-type dual-arm direct-drive manipulator according to a kinematic model of the link-type dual-arm direct-drive manipulator; S5: Calibrate the link-type dual-arm direct-drive manipulator according to the actual positions and theoretical positions of the joints of the link-type dual-arm direct-drive manipulator.

[0007] Preferably, in S2, the position data of the joints of the connecting rod type dual-arm direct drive manipulator are obtained by using a visual sensor: S2.1: Install multiple vision sensors in the workspace of the linkage-type dual-arm direct-drive manipulator and calibrate them; S2.2: Install markers on each joint of the linkage-type dual-arm direct-drive manipulator; S2.3: Using a visual sensor to collect joint image data of the linkage-type dual-arm direct-drive manipulator; S2.4: Detect the joint image data, extract the feature point coordinates of the marker, convert the feature point coordinates of the marker into world coordinates, and calculate the actual position data of the joint.

[0008] Preferably, the dynamic threshold Canny edge detection algorithm detects the joint image data specifically as follows: For each pixel in the joint image data, calculate the local variance σ and the local average brightness L in its neighborhood; Dynamically adjust the high and low thresholds of the Canny algorithm based on the local variance and average brightness of each pixel in its neighborhood; The calculated high threshold and low threshold are used to perform edge detection on the joint image data.

[0009] Preferably, the high threshold and low threshold of the Canny algorithm are dynamically adjusted according to the local variance and average brightness of each pixel in its neighborhood as follows: Set a base threshold T base , and then according to the local variance of each pixels and average brightness L The relative size of is used to adjust the high and low thresholds of the Canny algorithm; Specifically, the high threshold T of the Canny algorithm is calculated using the following formula: high and low threshold T low :

[0010] ; Where, T base is the basic threshold, k 1 and k 2 is the adjustment coefficient used to adjust the impact of local variance and average brightness on the threshold, d for An adjustment factor that controls the difference between the high and low thresholds.

[0011] Preferably, the adjustment coefficient is calculated k 1 and k The specific steps of 2 are: Calculating the global gradient G of the joint image; Calculate the global average brightness of the joint image data and global variance ; According to the global gradient G and global average brightness of the joint image and global variance Calculate the adjustment coefficient k 1 and k The value of 2.

[0012] Preferably, the calculation formula is:

[0013] ; Where, c 1 is a constant used to adjust the global gradient k The impact of 1, c 2 and c 3 is a constant used to adjust the global variance and global average brightness k 2. The degree of impact.

[0014] Preferably, the S3 is specifically: S3.1: Build a deep neural network model; S3.2 obtaining a data set for training the deep neural network model; S3.3: Use the dataset to train the deep neural network model.

[0015] Preferably, the formula of the improved Nesterov accelerated gradient method is:

[0016] ; Where t is the number of iterations, v t is the momentum term of the t-th iteration, v t+1 is the momentum term calculated before the t+1th iteration, β is the momentum coefficient, or t is the learning rate, is the gradient of the loss function E, i t Indicates in t The model parameters at the iteration, i t+1 It is in t +1 updated model parameters, G t It is in t The cumulative amount of gradient at iterations, is a constant.

[0017] Preferably, the learning rate or t Calculated as follows: .

[0018] The embodiments of the present invention have the following technical effects: The present invention detects joint image data, extracts the feature point coordinates of the marker, converts the feature point coordinates of the marker into world coordinates, and calculates the actual position data of the joint; and adopts a dynamic threshold Canny edge detection algorithm to detect the joint image data. By combining local image statistical characteristics and a dynamic threshold adjustment strategy, the detection accuracy and robustness of the Canny edge detection algorithm for reflective point markers are improved, and the method is particularly suitable for scenes with large lighting changes and uneven contrast; further, an adjustment coefficient is determined based on the global gradient, global variance and global average brightness of the image, that is, the influence of the local variance and local average brightness of each pixel point in the image on high and low thresholds is constrained by the global features of the image, so that the sensitivity of the algorithm to different image characteristics can be flexibly controlled, making the dynamic threshold Canny edge detection algorithm more flexible and universal.

[0019] In the process of motion model training, the improved Nesterov accelerated gradient method is used to introduce adaptive learning rate in the momentum term, and the momentum term is adjusted to achieve iterative optimization, that is, through the accumulation of gradient G t To adjust the learning rate or, which reduces the learning rate when the gradient changes significantly and increases it when the gradient changes slightly. This approach facilitates rapid convergence in the early stages of training and allows for more refined adjustments in the later stages. The goal is to better control changes in the learning rate during training, thereby improving optimization efficiency and stability, and ultimately increasing the calibration speed of the linkage-type dual-arm direct-drive manipulator. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 This is a flow chart of a rapid calibration method for a connecting rod type dual-arm direct drive manipulator provided by an embodiment of the present invention; Figure 2 This is a flow chart of using a visual sensor to obtain position data of the joints of the connecting rod type dual-arm direct drive manipulator provided by an embodiment of the present invention; Figure 3 This is a flow chart of establishing the kinematic model of the connecting rod type dual-arm direct-drive manipulator provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention are described clearly and completely below. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are also within the scope of protection of the present invention.

[0023] Example 1, attached Figure 1 A flow chart of a fast calibration method for a connecting rod type dual-arm direct drive manipulator is shown in the attached figure. Figure 1 As shown, a rapid calibration method for a connecting rod type dual-arm direct drive manipulator includes the following steps: S1: Initialize the structural parameters of the connecting rod type dual-arm direct drive manipulator; Before calibrating the linkage type dual-arm direct-drive manipulator, it is necessary to initialize the structural parameters of the linkage type dual-arm direct-drive manipulator; specifically: place the linkage type dual-arm direct-drive manipulator in a known and repeatable initial position and posture to provide a reference benchmark for subsequent measurements, and at the same time obtain the initial position and status information of each joint of the linkage type dual-arm direct-drive manipulator. These data will be used as benchmark data in the calibration process for subsequent calculations and error compensation. Through initialization, the movement speed and joint angle of the linkage type dual-arm direct-drive manipulator are set within a safe range to avoid collisions or other accidents during the calibration process.

[0024] Among them, the structural parameters of the link-type dual-arm direct-drive manipulator include: geometric parameters and kinematic parameters; the geometric parameters include link lengths (LinkLengths), joint offsets (JointOffsets) and link twist angles (LinkTwistAngles), etc.; the kinematic parameters include joint angles (JointAngles), joint velocities (JointVelocities) and joint accelerations (JointAccelerations), etc.

[0025] Furthermore, the link length parameters of the link-type dual-arm direct-drive manipulator are initialized as follows: a laser rangefinder is used to measure the actual length of each link, and the measured link length is recorded in the control system of the link-type dual-arm direct-drive manipulator as an initial parameter.

[0026] Initializing the joint offset parameters of the linkage-type dual-arm direct-drive manipulator specifically includes: using a laser tracker to measure the offset of each joint, and recording the measured joint offset in the control system of the linkage-type dual-arm direct-drive manipulator.

[0027] Initializing the link torsion angle parameters of the link-type dual-arm direct-drive manipulator is specifically as follows: using a laser tracker to measure the torsion angle of each link, and recording the measured link torsion angle in the control system of the link-type dual-arm direct-drive manipulator.

[0028] The joint angle parameters of the linkage-type dual-arm direct-drive manipulator are initialized as follows: the angle of each joint is set to a known initial value (such as 0 degrees) through the control system of the linkage-type dual-arm direct-drive manipulator, and then the set joint angle is recorded in the control system of the linkage-type dual-arm direct-drive manipulator.

[0029] The joint speed and joint acceleration parameters of the linkage-type dual-arm direct-drive manipulator are initialized specifically as follows: the speed and acceleration of each joint are set to a low-speed mode (such as 10% to 20% of the maximum speed) through the control system of the linkage-type dual-arm direct-drive manipulator, and then the set joint speed and acceleration are recorded in the control system of the linkage-type dual-arm direct-drive manipulator.

[0030] The above method can accurately initialize the structural parameters of the connecting rod type dual-arm direct drive manipulator. The initialization of these parameters ensures the accuracy and efficiency of subsequent calibration.

[0031] S2: using a visual sensor to obtain actual position data of the joints of the connecting rod type dual-arm direct drive manipulator; Vision-based joint position measurement uses a visual sensor (such as a camera) to capture image information about the manipulator's joint positions and then uses image processing algorithms to calculate the precise joint positions. This method offers the advantages of being non-contact, highly accurate, and highly real-time.

[0032] Specifically, as attached Figure 2 As shown, the position data of the joints of the connecting rod type dual-arm direct drive manipulator are obtained by using a visual sensor as follows: S2.1: Install multiple vision sensors in the workspace of the linkage-type dual-arm direct-drive manipulator and calibrate them; Among them, multiple vision sensors are installed in the workspace of the manipulator to ensure that the field of view of the vision sensors can cover all joints of the manipulator; the vision sensors are calibrated through a calibration plate to determine the intrinsic parameters (such as focal length, principal point) and extrinsic parameters (such as the position and posture of the sensor in the world coordinate system) of the vision sensors.

[0033] S2.2: Install markers on each joint of the linkage-type dual-arm direct-drive manipulator; In this step, the marker can be a reflective point, and the marker is installed on each joint of the connecting rod type dual-arm direct-drive manipulator to ensure that the position of the marker is fixed and easy to identify.

[0034] S2.3: Using a visual sensor to collect joint image data of the linkage-type dual-arm direct-drive manipulator; Wherein, in this step, the visual sensor is a binocular camera.

[0035] S2.4: Detecting the joint image data, extracting the coordinates of the feature points of the markers, converting the coordinates of the feature points of the markers into world coordinates, and calculating the actual position data of the joint; Among them, when detecting the joint graphic data, extracting the markers in the joint graphic is the key to extracting the coordinates of the markers; in the prior art, for reflective point markers, edge detection is a key step in extracting their precise position, and Canny edge detection is a classic edge detection algorithm, which is widely used for its good detection effect and computational efficiency; however, for the specific application scenario of the connecting rod type dual-arm direct-drive manipulator with large lighting changes and uneven contrast, the traditional Canny algorithm leads to insufficient recognition accuracy of reflective points and weak robustness; therefore, this embodiment proposes a dynamic threshold Canny edge detection algorithm to detect the joint image data to improve the detection accuracy and robustness of reflective point markers on the joints of the connecting rod type dual-arm direct-drive manipulator.

[0036] Specifically, the dynamic threshold Canny edge detection algorithm detects the joint image data as follows: For each pixel in the joint image data, calculate the local variance σ and the local average brightness L in its neighborhood; The size of the neighborhood is 3×3; Dynamically adjust the high and low thresholds of the Canny algorithm based on the local variance and average brightness of each pixel in its neighborhood; Among them, set a basic threshold T base , and then according to the local variance of each pixel s and average brightness L The relative size of is used to adjust the high and low thresholds of the Canny algorithm; Specifically, the high threshold T of the Canny algorithm is calculated using the following formula: high and low threshold T low :

[0037] ; Where, T base is the basic threshold, k 1 and k 2 is the adjustment coefficient used to adjust the impact of local variance and average brightness on the threshold, d for An adjustment factor that controls the difference between the high and low thresholds.

[0038] The calculated high threshold and low threshold are used to perform edge detection on the joint image data; this step is a prior art and will not be discussed in detail here.

[0039] Furthermore, this embodiment also provides a method for calculating the adjustment coefficient. k 1 and k The specific solution of 2 is: Calculating the global gradient G of the joint image; Use the Sobel operator to calculate the horizontal gradient of the joint image G x and vertical gradient G y , and then calculate the global gradient of the joint image; the specific formula is: ; Calculate the global average brightness of the joint image data and global variance ; According to the global gradient G and global average brightness of the joint image and global variance Calculate the adjustment coefficient k 1 and k The value of 2.

[0040] The calculation formula is:

[0041] ; Where, c 1 is a constant used to adjust the global gradient k The impact of 1, c 2 and c 3 is a constant used to adjust the global variance and global average brightness k 2. The degree of impact.

[0042] By combining local image statistical characteristics with a dynamic threshold adjustment strategy, this solution can improve the detection accuracy and robustness of the Canny edge detection algorithm for reflective point markers, making it particularly suitable for scenes with large illumination changes and uneven contrast. Furthermore, the adjustment coefficient is determined based on the global gradient, global variance, and global average brightness of the image. That is, by constraining the influence of the local variance and local average brightness of each pixel in the image on the high and low thresholds through the global features of the image, the algorithm's sensitivity to different image characteristics can be flexibly controlled, making the algorithm more flexible and versatile.

[0043] Coordinate transformation is to convert the detected image coordinates into world coordinates and calculate the three-dimensional position of the joint. This process is a key step in mapping two-dimensional image information into three-dimensional space. In this step, the intrinsic and extrinsic parameters of the two cameras of the binocular camera are determined by the marker, and then the parallax of the marker is calculated by matching the images of the two cameras. The feature point coordinates of the marker are calculated using the parallax and the baseline distance of the camera, and then the feature point coordinates of the marker are converted into world coordinates to obtain the position data of the joint.

[0044] S3: Establishing a kinematic model of the connecting rod type dual-arm direct drive manipulator; Traditional robot kinematic models rely on precise geometric parameters and assumptions to establish the robot kinematic model, but actual robots may have manufacturing errors and assembly errors. Therefore, in order to improve the adaptability and accuracy of the kinematic model, this embodiment adopts a data-driven method to establish the kinematic model, that is, collecting a large amount of end position and posture data of link-type dual-arm direct-drive robots at different joint angles, and using this data to train a data-driven kinematic model. The kinematic model uses a neural network model to establish a mapping relationship between input (joint position data) and output (end position and posture).

[0045] In this step, the kinematic model of the linkage-type dual-arm direct-drive manipulator is a deep neural network (DNN). This DNN uses a large amount of sample data to train a neural network model, enabling it to quickly and accurately estimate the manipulator's kinematic parameters. This approach is particularly suitable for complex nonlinear systems, automatically learning the mapping relationship between input (joint angles) and output (end position and posture), reducing the complex geometric modeling and parameter estimation steps required in traditional kinematic models.

[0046] Among them, as attached Figure 3 As shown, the S3 is specifically: S3.1: Build a deep neural network model; The deep learning network model includes an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is equal to the number of joint angles n. For example, for a 6-joint linkage dual-arm direct-drive manipulator, the input layer has 6 neurons, corresponding to 6 joint angles.

[0047] The hidden layer is a key part of the deep learning network model, generally a multi-layer structure, each layer contains multiple neurons, and the activation function of the hidden layer is generally a ReLU function; Furthermore, the hidden layer of this embodiment is a 3-layer architecture, with 128, 64 and 32 neurons in each layer respectively.

[0048] The number of neurons in the output layer is the same as the dimensions of the end position and posture of the linkage-type dual-arm direct-drive manipulator. For example, for a 6-dimensional output of position and posture [x, y, z, roll, pitch, yaw], the output layer has 6 neurons. Because the end position and posture of the linkage-type dual-arm direct-drive manipulator are continuous values, the activation function of the output layer is a linear activation function (LinearActivation).

[0049] S3.2: Obtain a data set for training the deep neural network model; Among them, the joint angles of the connecting rod type dual-arm direct drive manipulator in different working states are recorded i =[ i 1, i 2,…, θn ], and the position of the end effector of the connecting rod type dual-arm direct drive manipulator is recorded by a high-precision measuring device (such as a visual sensor) [ x , y , z ] and postures [roll, pitch, yaw] as data samples; collect enough data samples to cover the working space and motion range of the manipulator, and use the data samples as a data set.

[0050] S3.3: Using the dataset to train the deep neural network model; The weights and biases of the deep neural network model are randomly set, a data set for training is input, a predicted value is calculated by the deep neural network model, the error between the predicted value and the true value is calculated using a loss function, the gradient is calculated using a backpropagation algorithm, the network parameters are updated, and the above process is repeated until the loss function converges or a preset number of training rounds is reached.

[0051] During model training, an optimizer is generally used to accelerate the convergence of the model. In the prior art, the Nesterov accelerated gradient method is generally used for training. However, the above method cannot achieve the expected optimization speed during the optimization process. Therefore, this embodiment proposes an improved optimization algorithm to reduce the number of iterations. Specifically, in order to further improve the performance of the Nesterov accelerated gradient method, this embodiment adopts an improved Nesterov accelerated gradient method, introduces an adaptive learning rate in the momentum term, and adjusts the momentum term; The formula of the improved Nesterov accelerated gradient method is expressed as follows:

[0052] ; Where t is the number of iterations, v t is the momentum term of the t-th iteration, indicating the current “speed”, v t+1 is the momentum term calculated before the t+1th iteration, which is calculated based on the parameters and gradient of the tth iteration. β is the momentum coefficient, usually ranging from 0.8 to 0.99, or t is the learning rate, is the gradient of the loss function E, i tIndicates in t The model parameters at the iteration, i t+1 It is in t +1 updated model parameters, G t It is in t The cumulative amount of gradient at the iteration is used to calculate the learning rate, is a small constant used to prevent division by zero errors.

[0053] In each iteration, the improved Nesterov accelerated gradient method will be based on the current gradient and the previous momentum v t To calculate the new momentum v t+1 , and then use this momentum to update the model parameters i t+1 This process is repeated until the stopping condition is met, such as reaching the maximum number of iterations or the value of the loss function no longer decreases significantly.

[0054] Among them, the learning rate or t Calculated as follows:

[0055] Through the above learning rate formula, the learning rate is dynamically adjusted according to the iteration t of each step, that is, through the accumulation of gradient G t To adjust the learning rate or , which reduces the learning rate when the gradient changes significantly, and increases it when the gradient changes slightly. This approach helps achieve rapid convergence in the early stages of training and allows for more refined adjustments in the later stages of training. The goal is to better control changes in the learning rate during training, thereby improving optimization efficiency and stability.

[0056] Therefore, the process of using the data set to train the deep neural network model is as follows: by inputting the data set into the deep neural network model, the deep neural network model can calculate the current parameters i t The predicted value under ; Then calculate the gradient of the loss function E , then update the gradient accumulation; calculate the learning rate for each iteration t and update the momentum term v t+1 and parameters i t+1 ; Thus, a training process is completed.

[0057] S4: calculating theoretical positions of joints of the link-type dual-arm direct-drive manipulator according to a kinematic model of the link-type dual-arm direct-drive manipulator; Among them, the S4 is: obtaining the angle data of the joints of the connecting rod type dual-arm direct-drive manipulator as the input data of the kinematic model of the connecting rod type dual-arm direct-drive manipulator, and the output of the kinematic model of the connecting rod type dual-arm direct-drive manipulator is the theoretical position of the joints of the connecting rod type dual-arm direct-drive manipulator; through the above process, the theoretical position of the joints of the manipulator can be quickly acquired.

[0058] S5: Calibrate the link-type dual-arm direct-drive manipulator according to the actual positions and theoretical positions of the joints of the link-type dual-arm direct-drive manipulator.

[0059] Among them, S5 is: calculating the interpolation of the actual position and theoretical position of the joint of the link type dual-arm direct-drive manipulator as the error of the control system of the link type dual-arm direct-drive manipulator, and adjusting the joint control parameters of the control system according to the error so that it can compensate for the error in subsequent movement to achieve the calibration of the link type dual-arm direct-drive manipulator.

[0060] Example 2: The present invention also provides an electronic device, including one or more processors and a memory.

[0061] The processor may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0062] The memory may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may, for example, include read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may run the program instructions to implement a fast calibration method for a connecting rod type dual-arm direct-drive manipulator of any embodiment of the present application described above and / or other desired functions. Various contents such as initial external parameters, thresholds, etc. may also be stored in the computer-readable storage medium.

[0063] In one example, the electronic device may further include an input device and an output device, these components interconnected via a bus system and / or other connection mechanisms (not shown). The input device may include, for example, a keyboard, a mouse, etc. The output device may output various information to the outside, including warning information, braking force, etc. The output device may include, for example, a display, a speaker, a printer, a communication network, and remote output devices connected thereto.

[0064] Of course, for the sake of simplicity, components such as buses, input / output interfaces, etc. are omitted. In addition, the electronic device may further include any other appropriate components according to specific application scenarios.

[0065] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to implement the function of a rapid calibration method for a link-type dual-arm direct-drive manipulator provided in any embodiment of the present application.

[0066] The computer program product may be written in any combination of one or more programming languages ​​to implement the program code for performing the operations of the embodiments of the present application, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0067] In addition, an embodiment of the present application may also be a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor implements a rapid calibration method for a link-type dual-arm direct-drive manipulator provided in any embodiment of the present application.

[0068] The computer-readable storage medium may be any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may include, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A fast calibration method for a connecting rod type dual-arm direct drive manipulator, characterized in that: The following steps are involved: S1: Initialize the structural parameters of the connecting rod type dual-arm direct drive manipulator; S2: using a visual sensor to obtain actual position data of the joints of the connecting rod type dual-arm direct drive manipulator; In S2, the joint image data is detected using a dynamic threshold Canny edge detection algorithm; S3: Establishing a kinematic model of the connecting rod type dual-arm direct drive manipulator; The kinematic model is a deep neural network model. During the training process of the deep neural network model, the improved Nesterov accelerated gradient method is used to optimize the training process of the deep neural network model, an adaptive learning rate is introduced into the momentum term, and the momentum term is adjusted; S4: calculating theoretical positions of joints of the link-type dual-arm direct-drive manipulator according to a kinematic model of the link-type dual-arm direct-drive manipulator; S5: Calibrate the link-type dual-arm direct-drive manipulator according to the actual positions and theoretical positions of the joints of the link-type dual-arm direct-drive manipulator.

2. The rapid calibration method of a connecting rod type dual-arm direct drive manipulator according to claim 1 is characterized in that: In S2, the position data of the joints of the connecting rod type dual-arm direct drive manipulator are obtained by using a visual sensor: S2.1: Install multiple vision sensors in the workspace of the linkage-type dual-arm direct-drive manipulator and calibrate them; S2.2: Install markers on each joint of the linkage-type dual-arm direct-drive manipulator; S2.3: Using a visual sensor to collect joint image data of the connecting rod type dual-arm direct drive manipulator; S2.4: Detect the joint image data, extract the feature point coordinates of the marker, convert the feature point coordinates of the marker into world coordinates, and calculate the actual position data of the joint.

3. The rapid calibration method of a connecting rod type dual-arm direct drive manipulator according to claim 2 is characterized in that: The dynamic threshold Canny edge detection algorithm detects the joint image data specifically as follows: For each pixel in the joint image data, calculate the local variance σ and the local average brightness L in its neighborhood; Dynamically adjust the high and low thresholds of the Canny algorithm based on the local variance and average brightness of each pixel in its neighborhood; The calculated high threshold and low threshold are used to perform edge detection on the joint image data.

4. The rapid calibration method for a connecting rod type dual-arm direct drive manipulator according to claim 3 is characterized in that: According to the local variance and average brightness of each pixel in its neighborhood, the high and low thresholds of the Canny algorithm are dynamically adjusted as follows: Set a base threshold T base , and then according to the local variance of each pixel σ and average brightness L The relative size of is used to adjust the high and low thresholds of the Canny algorithm; Specifically, the high threshold T of the Canny algorithm is calculated using the following formula: high and low threshold T low : ； ； Where, T base is the basic threshold, k 1 and k 2 is the adjustment coefficient used to adjust the impact of local variance and average brightness on the threshold, d An adjustment coefficient for controlling the gap between high and low thresholds.

5. The rapid calibration method of a connecting rod type dual-arm direct drive manipulator according to claim 4 is characterized in that: Calculate the adjustment coefficient k 1 and k The specific steps of 2 are: Calculating the global gradient G of the joint image; Calculate the global average brightness of the joint image data and global variance ; According to the global gradient G and global average brightness of the joint image and global variance Calculate the adjustment coefficient k 1 and k The value of 2.

6. The rapid calibration method of a connecting rod type dual-arm direct drive manipulator according to claim 5, characterized in that: The calculation formula is: ； ； Where, c 1 is a constant used to adjust the global gradient k The impact of 1, c 2 and c 3 is a constant used to adjust the global variance and global average brightness k 2. The degree of impact.

7. The rapid calibration method for a connecting rod type dual-arm direct drive manipulator according to claim 1, characterized in that: The S3 is specifically: S3.1: Build a deep neural network model; S3.2 obtaining a data set for training the deep neural network model; S3.3: Use the dataset to train the deep neural network model.

8. The rapid calibration method for a connecting rod type dual-arm direct drive manipulator according to claim 1, characterized in that: The formula of the improved Nesterov accelerated gradient method is: ； ； Where t is the number of iterations, v t is the momentum term of the t-th iteration, v t+1 is the momentum term calculated before the t+1th iteration, β is the momentum coefficient, η t is the learning rate, is the gradient of the loss function E, θ t Indicates in t The model parameters at the iteration, θ t+1 It is in t +1 updated model parameters, G t It is in t The cumulative amount of gradient at iterations, is a constant.

9. The rapid calibration method of a connecting rod type dual-arm direct drive manipulator according to claim 8, characterized in that: Learning rate η t Calculated as follows: 。

Images