An adaptive modular robotic grasping system and method

Abstract

The present application relates to the technical field of mechanical arm control, in particular to a self-adaptive modular robot grabbing system and method, the system comprising a vision-guided registration module, a workpiece feature perception module, a contact state perception module, a grabbing parameter learning module and a dynamic grabbing control module. The present application can convert the workpiece contour position, width and spindle orientation into posture basis that can directly constrain the grabbing action through the cooperative registration of calibration images, end reference points and pose data. The contact process can be jointly represented by the opening width ratio, displacement rollback, current change and contact deformation variable, which can more accurately distinguish the contact state and clamping trend. The lifting following result and the trial grabbing sample are regressed into the opening and closing stroke, the closing rhythm, the holding current and the release timing, so that the formal grabbing stage can continuously roll around the end deviation and the contact change to correct the position, the closing increment, the current output and the closing direction, thereby compressing the alignment error and reducing the risk of slipping and extrusion.

Description

Technical Field

[0001] This invention relates to the field of robotic arm control technology, and in particular to an adaptive modular robot grasping system and method. Background Technology

[0002] The field of robotic arm control technology involves grasping, transporting and manipulating target objects through mechanical structures and actuators, including the motion connection structure design of multi-joint robotic arms, the configuration and driving method of end effectors, the transmission and positioning coordination between joints, and the setting of motion paths based on the task.

[0003] Traditional adaptive modular robot gripping systems refer to systems designed for different shapes and sizes. These systems divide the gripping device into several detachable functional units and use standardized connection interfaces to combine and install grippers, support components, and drive components. Simultaneously, a pre-set control program sequentially controls the gripping actions. In practical applications, different workpieces are typically adapted by replacing gripper bodies of different specifications, adjusting the position of connecting flanges, setting elastic buffer structures, and limiting the gripping range using mechanical limiters. The gripping action is completed by driving the actuator through the gripping position coordinates, opening and closing stroke parameters, and action execution sequence set in the program. After component wear, individual gripping components can be replaced by disassembling fixing bolts or plug-in structures to complete structural adjustments and maintenance for diverse gripping needs.

[0004] Existing technologies rely on replacing the claw body, adjusting the flange position, setting mechanical limits, and presetting coordinate parameters to achieve adaptation. The gripping conditions are mainly maintained by the structural relationship and fixed action sequence after manual adjustment. When faced with differences in workpiece contours, deviations in placement posture, and fluctuations in contact state, the clamping range and execution trajectory lack the ability to be corrected synchronously with the process. In actual operation, it is easy to have accumulated alignment deviations, rough judgment of clamping contact, and mismatch between clamping degree and workpiece state. This may cause slippage, biased gripping, and crushing damage, and will also increase the frequency of repeated model changes, debugging, and component maintenance, thus limiting the gripping stability, versatility, and continuous operation efficiency. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an adaptive modular robot grasping system and method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an adaptive modular robot grasping system, the system comprising:

[0007] The visual-guided registration module acquires the current calibration board image, the robot end effector gripper reference point image, the robot end effector position data, and the robot end effector orientation data, and performs coordinate registration to generate pose mapping data.

[0008] The workpiece feature perception module performs feature analysis on the image of the workpiece to be grasped based on the pose mapping data, and generates workpiece contour pose data.

[0009] The contact state sensing module obtains the allowable opening value of the replaceable gripper component, the displacement data of the robot gripper encoder and the drive current data of the robot gripper, and combines the workpiece contour posture data to determine the contact response state of the robot gripper and generate gripper grasping characterization data.

[0010] The grasping parameter learning module acquires the workpiece following the robot's lifting state and the robot's gripper displacement data backtracking amount. It combines the gripper grasping characterization data and performs Gaussian process regression processing to generate a gripper grasping parameter set.

[0011] The dynamic gripping control module acquires the position of the robot's end effector, compares it with the workpiece contour posture data to determine the displacement data deviation, and simultaneously acquires the current position of the robot gripper, the robot gripper drive current, and the robot gripper contact deformation. Combined with the gripper gripping parameter set, it performs rolling prediction calculations to generate an adaptive gripping control sequence for the robot gripper.

[0012] The present invention is improved in that the pose mapping data includes the coordinates of the robot end effector reference point, the position coordinates of the robot end effector, and the orientation angle of the robot end effector in the industrial camera coordinate system; the workpiece contour posture data includes the center coordinates of the workpiece to be gripped, the width of the workpiece to be gripped, and the orientation angle of the spindle of the workpiece to be gripped; the gripper gripping characterization data includes the robot gripper opening width ratio, the robot gripper driving current change rate, the robot gripper contact deformation, and the robot gripper contact response state; the gripper gripping parameter set includes the robot gripper opening and closing stroke, the robot gripper closing speed inflection point, the robot gripper holding current threshold, and the robot gripper release delay; and the robot gripper adaptive gripping control sequence includes the robot end effector position correction amount, the robot gripper closing increment, the robot gripper holding current correction amount, and the robot gripper closing direction correction amount.

[0013] The present invention is improved in that the visual guided registration module includes:

[0014] The calibration board correction submodule acquires calibration board images, robot end gripper reference point images, robot end position data, and robot end orientation data. It extracts pixel matrix coordinate items corresponding to the corner pixel coordinates in the calibration board images, performs multi-dimensional distortion correction calculations on the pixel matrix coordinate items, and obtains camera coordinate coefficient values.

[0015] The pose calculation submodule obtains the reference point pixel coordinates in the reference point image of the robot's end gripper, performs coordinate conversion with the camera coordinate coefficient value, obtains the gripper reference point pixel coordinate item in the corresponding reference dimension, and performs multi-dimensional pose calculation by combining the robot end position data and robot end direction data to obtain the end gripper coordinate quantity.

[0016] The coordinate registration submodule extracts the position node vectors of the end gripper coordinates and the corresponding reference point pixel coordinates in the reference dimension, performs cross-coordinate system position node registration calculations on the position node vectors, obtains the relative displacement offset coefficient and direction rotation angle coefficient between each position node, and generates pose mapping data.

[0017] The present invention is improved in that the workpiece feature sensing module includes:

[0018] The pixel correction submodule acquires the image of the workpiece to be grasped, calls the relative displacement offset coefficient and direction rotation angle coefficient of each position node in the pose mapping data, performs frame-by-frame pixel correction and coordinate mapping on the pixel sequence of the image of the workpiece to be grasped, and obtains the contour envelope boundary of the image of the workpiece to be grasped.

[0019] The deviation calculation submodule extracts the lateral spacing dimension of the outline edge of the workpiece to be gripped based on the outline envelope boundary, converts the size unit to calculate the corresponding overall width value of the workpiece, and obtains the orientation angle of the corresponding workpiece main axis and the orientation angle of the preset robot gripper closing direction. It calculates the absolute value of the angle difference between the orientation angle of the workpiece main axis and the orientation angle of the preset robot gripper closing direction to generate the gripping direction deviation value.

[0020] The posture generation submodule obtains the coordinate set of each edge extreme point of the contour envelope boundary of the workpiece to be grasped, calculates the mean of the horizontal coordinate component and the vertical coordinate component in each edge extreme point coordinate set, and combines them with the clamping direction deviation value and the corresponding overall width value of the workpiece to obtain the workpiece contour posture data.

[0021] The present invention is improved in that the contact state sensing module includes:

[0022] The width matching submodule collects the maximum allowable opening value parameter of the robot gripper during the trial gripping stage after the replacement gripper component is assembled, calls the corresponding overall width value of the workpiece in the workpiece contour posture data, and calculates the opening width ratio between the maximum allowable opening value parameter of the robot gripper and the corresponding overall width value of the workpiece.

[0023] The state change submodule collects the robot gripper drive current and the robot gripper encoder displacement data before and after contact at adjacent sampling times during the trial grasping phase. It calculates the difference between the two robot gripper drive currents at adjacent sampling times based on the opening width ratio, and calculates the robot gripper drive current change rate according to the sampling time interval. At the same time, it calculates the contact deformation offset of the robot gripper encoder displacement data before and after contact.

[0024] The response determination submodule collects the gripper component numbers during the trial gripping phase, groups and normalizes the opening width ratio of each gripper component, the robot gripper drive current change rate, and the contact deformation offset according to the gripper component number, and compares the normalized robot gripper drive current change rate and contact deformation offset with a preset contact limit threshold to determine the robot gripper contact response state and generate gripper gripping characterization data.

[0025] The present invention is improved in that the parameter learning module includes:

[0026] The status acquisition submodule monitors the workpiece following the robot's lifting status during the vertical lifting phase of the robot body after each trial gripping, and collects the robot gripper displacement data and back-off amount during the clamping and holding phase after each trial gripping, and establishes a set of motion following back-off amounts.

[0027] The sample set submodule calls the contact response state in the gripper gripping characterization data, and combines it with the workpiece following robot lifting status flag in the action follow-back amount set and the robot gripper displacement data back-back amount to obtain the test gripping sample set quantity.

[0028] The regression mapping submodule collects the opening and closing stroke and closing speed inflection point, as well as the holding current threshold and release delay parameters. It then performs Gaussian process regression analysis on the sample set of the test grippers, the opening and closing stroke and closing speed inflection point, the holding current threshold, and the release delay parameters to obtain the gripper gripping parameter set.

[0029] The present invention is improved in that the dynamic grasping control module includes:

[0030] The deviation prediction submodule collects the robot end position and the current position of the robot gripper, as well as the robot gripper drive current and gripper contact deformation during the closing and clamping phases of the formal grasping process. It then calculates the end displacement deviation by referring to the three-dimensional vector of the robot end position and the coordinates of the workpiece contour center node.

[0031] The rolling simulation submodule acquires the current position of the robot gripper, the gripper drive current, and the gripper contact deformation. It combines the gripper grasping parameter set to deduce the rolling prediction closure increment parameters and the holding current adjustment parameters. At the same time, it deduces the future displacement data trajectory in the time domain length based on the end displacement deviation and calculates the expected movement orientation position adjustment parameters to obtain the end position correction amount.

[0032] The joint control submodule obtains the expected closing direction adjustment amount, integrates the end position correction amount with the estimated closing increment and holding current adjustment amount and the expected closing direction adjustment amount to construct a joint adjustment matrix, covering the original robot closing and gripping action associated control commands, and generates the robot gripper adaptive grasping control sequence.

[0033] An adaptive modular robot grasping method includes the following steps:

[0034] S1: Acquire the current calibration board image, robot end effector gripper reference point image, robot end effector position data, and robot end effector orientation data, and perform coordinate registration to generate pose mapping data;

[0035] S2: Perform feature analysis on the image of the workpiece to be grasped based on the pose mapping data to generate workpiece contour pose data;

[0036] S3: Obtain the allowable opening value of the replaceable gripper component, the displacement data of the robot gripper encoder and the drive current data of the robot gripper, and combine them with the workpiece contour posture data to determine the contact response state of the robot gripper and generate gripper grasping characterization data.

[0037] S4: Obtain the workpiece following the robot's lifting state and the robot gripper displacement data backtracking amount, combine the gripper gripping characterization data, and perform Gaussian process regression processing to generate a gripper gripping parameter set.

[0038] S5: Obtain the position of the robot end effector, compare it with the workpiece contour posture data to determine the displacement data deviation, and at the same time obtain the current position of the robot gripper, the robot gripper drive current and the robot gripper contact deformation. Combine the gripper grasping parameter group to perform rolling prediction calculation and generate the robot gripper adaptive grasping control sequence.

[0039] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0040] In this invention, by co-registering calibration images, end-point reference points, and pose data, the workpiece contour position, width, and spindle orientation can be transformed into posture criteria that can directly constrain the gripping action. Combined with the opening width ratio, displacement retraction, current change, and contact deformation to jointly characterize the contact process, the contact state and clamping trend can be more accurately distinguished. Furthermore, the lifting and following results are correlated with the trial gripping samples and regressed into opening and closing stroke, closing rhythm, holding current, and release sequence. This allows the formal gripping stage to continuously and continuously correct the position, closing increment, current output, and closing direction based on end-point deviation and contact changes, thereby reducing alignment errors, lowering the risk of slippage and squeezing, reducing the burden of changeover and adjustment, and improving the gripping stability and adaptation efficiency of various workpieces. Attached Figure Description

[0041] Figure 1 This is a system module diagram of the present invention;

[0042] Figure 2 This is a system framework diagram of the present invention;

[0043] Figure 3 This is a flowchart illustrating the visual guidance registration module of the present invention.

[0044] Figure 4 This is a flowchart illustrating the workpiece feature sensing module of the present invention.

[0045] Figure 5 This is a flowchart illustrating the contact state sensing module of the present invention.

[0046] Figure 6 This is a flowchart illustrating the parameter learning module of the present invention.

[0047] Figure 7 This is a flowchart illustrating the dynamic capture control module of the present invention.

[0048] Figure 8 This is a flowchart of the method of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0050] Please see Figure 1 This invention provides a technical solution: an adaptive modular robot grasping system, the system comprising:

[0051] The visual-guided registration module acquires the current calibration board image, the robot end effector gripper reference point image, the robot end effector position data, and the robot end effector orientation data, and performs coordinate registration to generate pose mapping data.

[0052] The workpiece feature perception module performs feature analysis on the image of the workpiece to be grasped based on the pose mapping data, and generates workpiece contour pose data.

[0053] The contact state sensing module obtains the allowable opening value of the replaceable gripper component, the displacement data of the robot gripper encoder and the drive current data of the robot gripper, and combines the workpiece contour posture data to determine the contact response state of the robot gripper and generate gripper grasping characterization data.

[0054] The grasping parameter learning module acquires the workpiece following the robot's lifting state and the robot's gripper displacement data backtracking amount. It combines the gripper grasping characterization data and performs Gaussian process regression processing to generate a set of gripper grasping parameters.

[0055] The dynamic gripping control module acquires the position of the robot's end effector, compares it with the workpiece contour posture data to determine the displacement data deviation, and simultaneously acquires the current position of the robot gripper, the robot gripper drive current, and the robot gripper contact deformation. Combined with the gripper gripping parameter set, it performs rolling prediction calculations to generate an adaptive gripping control sequence for the robot gripper.

[0056] The pose mapping data includes the coordinates of the robot end effector reference point, the robot end effector position coordinates, and the robot end effector orientation angle in the industrial camera coordinate system. The workpiece contour posture data includes the center coordinates of the workpiece to be gripped, the width of the workpiece to be gripped, and the orientation angle of the spindle of the workpiece to be gripped. The gripper gripping characterization data includes the robot gripper opening width ratio, the robot gripper drive current change rate, the robot gripper contact deformation, and the robot gripper contact response state. The gripper gripping parameter set includes the robot gripper opening and closing stroke, the robot gripper closing speed inflection point, the robot gripper holding current threshold, and the robot gripper release delay. The robot gripper adaptive gripping control sequence includes the robot end effector position correction, the robot gripper closing increment, the robot gripper holding current correction, and the robot gripper closing direction correction.

[0057] Please see Figure 2 and Figure 3 The visual guidance registration module includes:

[0058] The calibration board correction submodule acquires calibration board images, robot end gripper reference point images, robot end position data, and robot end orientation data. It extracts pixel matrix coordinate items corresponding to the corner pixel coordinates in the calibration board images, performs multi-dimensional distortion correction calculations on the pixel matrix coordinate items, and obtains camera coordinate coefficient values.

[0059] The calibration plate image is a surface pattern image of a standard visual calibration reference set independently of the workpiece and the robot end gripper.

[0060] The surface pattern image of a standard industrial vision calibration reference object, independent of the workpiece and the robot end effector gripper, is acquired as the calibration board image. Simultaneously, an industrial camera acquires reference point images of the robot end effector gripper. The 3D spatial coordinates of the robot end effector are read via a communication node as its position data, the Euler angles as its orientation data, and the 2D coordinates of locations where grayscale gradients abruptly change within the calibration board image are read as corner pixel coordinates. All corner pixel coordinates are arranged in row and column order to construct a pixel matrix coordinate system. The radial and tangential distortion coefficients of the camera lens are read, and the lateral coordinate difference is calculated as: lateral corner pixel coordinate value - lateral corner pixel coordinate value. The difference between the horizontal and vertical coordinates of the image optical center pixel is calculated as follows: The difference between the horizontal and vertical coordinates is calculated as follows: The difference between the vertical coordinates of the corner pixels and the vertical coordinates of the image optical center pixel. To conform to the correct expression of Euclidean distance in physical space, the original logic of direct summation is corrected, and a geometric distance formula is adopted: Corner radial distance = (square of the horizontal coordinate difference + square of the vertical coordinate difference) , Radial offset = radial distortion coefficient multiplied by the square of the corner radial distance, Tangential offset = tangential distortion coefficient multiplied by (horizontal coordinate difference multiplied by vertical coordinate difference), Distortion correction pixel coordinates = corner pixel coordinates + radial offset + tangential offset, and Camera coordinate coefficient value = distortion correction pixel coordinates multiplied by the inverse of the camera intrinsic parameter matrix.

[0061] The pose calculation submodule obtains the reference point pixel coordinates in the reference point image of the robot's end gripper, performs coordinate conversion with the camera coordinate coefficient value, obtains the gripper reference point pixel coordinate item in the corresponding reference dimension, and performs multi-dimensional pose calculation by combining the robot end position data and robot end orientation data to obtain the end gripper coordinate quantity.

[0062] The horizontal and vertical coordinates of a specific highly reflective marker in the reference point image of the robot's end effector gripper are obtained as the reference point pixel coordinates. The normalized planar coordinates are calculated as the reference point pixel coordinates multiplied by the inverse matrix of the camera coordinate coefficient values. The gripper reference point pixel coordinates in the corresponding reference dimension are calculated as the normalized planar coordinates multiplied by the camera focal length physical parameter value. The horizontal, vertical, and longitudinal components of the robot's end effector position data are read. The roll, pitch, and yaw angles of the robot's end effector orientation data are read. The sine and cosine values ​​corresponding to the roll, pitch, and yaw angles are calculated. The three-dimensional rotation matrix is ​​calculated as the rotation matrix corresponding to the roll angle multiplied by the rotation matrix corresponding to the pitch angle multiplied by the rotation matrix corresponding to the yaw angle. The three components of the robot's end effector position data are directly used as the translation vector. The end effector gripper coordinates are calculated as the gripper reference point pixel coordinates in the corresponding reference dimension multiplied by the three-dimensional rotation matrix plus the translation vector.

[0063] The coordinate registration submodule extracts the position node vectors of the end gripper coordinates and the corresponding reference point pixel coordinates of the gripper in the reference dimension, performs cross-coordinate system position node registration calculations on the position node vectors, obtains the relative displacement offset coefficient and direction rotation angle coefficient between each position node, and generates pose mapping data.

[0064] The three-axis coordinates of the end gripper coordinates in 3D space are extracted as the first position node vector. The three-axis coordinates of the gripper reference point pixel coordinates in the corresponding reference dimension in the camera coordinate system are extracted as the second position node vector. The first centroid coordinate = the sum of all coordinate values ​​in the first position node vector / the total number of coordinates in the first position node vector. The second centroid coordinate = the sum of all coordinate values ​​in the second position node vector / the total number of coordinates in the second position node vector. The first decentroid vector = the first position node vector - the first centroid coordinate. The second decentroid vector = the second position node vector - the first centroid coordinate. Two centroid coordinates, covariance matrix = first decentroid vector multiplied by the transpose of second decentroid vector, extract the eigenvectors of the product of the transpose of the covariance matrix and the original covariance matrix to construct the right singular matrix, extract the eigenvectors of the product of the original covariance matrix and its transpose matrix to construct the left singular matrix, orientation rotation angle coefficient = left singular matrix multiplied by the transpose of the right singular matrix, relative displacement offset coefficient = first centroid coordinate - second centroid coordinate multiplied by orientation rotation angle coefficient, arrange the matrix elements of orientation rotation angle coefficient and the vector elements of relative displacement offset coefficient in order to generate pose mapping data.

[0065] Please see Figure 2 and Figure 4 The workpiece feature sensing module includes:

[0066] The pixel correction submodule acquires the image of the workpiece to be grasped, calls the relative displacement offset coefficient and direction rotation angle coefficient of each position node in the pose mapping data, performs frame-by-frame pixel correction and coordinate mapping on the pixel sequence of the image of the workpiece to be grasped, and obtains the contour envelope boundary of the image of the workpiece to be grasped.

[0067] A 2D color image of the workpiece to be gripped on the worktable is acquired as the workpiece image. The relative displacement offset coefficient and direction rotation angle coefficient of each position node in the pose mapping data are retrieved. The 2D coordinates of all pixels in the workpiece image are read as the pixel sequence of the workpiece image. The rotated pixel coordinates are calculated as the pixel sequence of the workpiece image multiplied by the direction rotation angle coefficient matrix. The corrected mapped coordinates are calculated as the rotated pixel coordinates plus the relative displacement offset coefficient. The grayscale difference between adjacent corrected mapped coordinates is calculated as the grayscale value of the current corrected mapped coordinate minus the grayscale value of the adjacent corrected mapped coordinate. The coordinate grayscale value is retrieved, and the preset grayscale threshold of 50 is selected. This threshold is set based on the acquisition of 100 images of the workbench without a workpiece background, the statistical maximum grayscale gradient value of the background noise is 30, and the anti-interference margin of 20 is added to determine the boundary value of 50. When the grayscale difference of adjacent corrected mapped coordinates is greater than 50, the corresponding coordinate points are extracted to form an edge pixel point set. When the grayscale difference of adjacent corrected mapped coordinates is less than or equal to 50, the corresponding coordinate points are discarded. The outermost closed continuous pixel point set of the edge pixel point set is extracted to obtain the contour envelope boundary of the workpiece image to be captured.

[0068] The deviation calculation submodule extracts the lateral spacing dimension of the outline edge of the workpiece to be gripped based on the contour envelope boundary, converts the size unit to calculate the corresponding overall width value of the workpiece, and obtains the orientation angle of the corresponding workpiece main axis and the preset robot gripper closing orientation angle. It calculates the absolute value of the angle difference between the orientation angle of the workpiece main axis and the preset robot gripper closing orientation angle to generate the gripping direction deviation value.

[0069] Based on the contour envelope boundary, extract the x-coordinates of the leftmost and rightmost pixels of the envelope polygon. The horizontal spacing dimension of the workpiece contour edge to be grasped = the x-coordinate of the rightmost pixel - the x-coordinate of the leftmost pixel. Call the physical millimeter value corresponding to a single pixel. The overall width value of the corresponding workpiece = the horizontal spacing dimension of the workpiece contour edge to be grasped multiplied by the physical millimeter value corresponding to a single pixel. Extract the angle value between the long side of the smallest bounding rectangle of the contour envelope boundary and the horizontal X-axis as the orientation angle of the corresponding workpiece main axis. Read the angle value between the line connecting the two fingers of the end effector closing action and the horizontal X-axis in the configuration file through the control node as the preset robot gripper closing orientation angle. The gripping direction deviation value = the absolute value of the difference between the workpiece main axis orientation angle and the preset robot gripper closing orientation angle.

[0070] The posture generation submodule obtains the coordinate set of each edge extreme point of the contour envelope boundary of the workpiece to be grasped, calculates the mean of the horizontal coordinate component and the vertical coordinate component in each edge extreme point coordinate set, and combines them with the clamping direction deviation value and the corresponding overall width value of the workpiece to obtain the workpiece contour posture data.

[0071] The two-dimensional plane coordinates of the upper extreme point of the maximum Y coordinate, the lower extreme point of the minimum Y coordinate, the left extreme point of the minimum X coordinate, and the right extreme point of the maximum X coordinate of the contour envelope boundary of the workpiece to be grasped are obtained to form the edge extreme point coordinate set. The mean of the horizontal coordinate component is equal to the sum of the horizontal coordinate values ​​of the edge extreme point coordinate set / 4, and the mean of the vertical coordinate component is equal to the sum of the vertical coordinate values ​​of the edge extreme point coordinate set / 4. The floating-point number corresponding to the clamping direction deviation value and the floating-point number corresponding to the overall width value of the workpiece in millimeters are extracted. The mean of the horizontal coordinate component, the mean of the vertical coordinate component, the clamping direction deviation value, and the corresponding overall width value of the workpiece are written into the same data array in a fixed order to obtain the workpiece contour posture data.

[0072] Please see Figure 2 and Figure 5 The contact state sensing module includes:

[0073] The width matching submodule collects the maximum allowable opening value parameter of the robot gripper during the trial gripping stage after the replacement gripper component is assembled, calls the corresponding overall width value of the workpiece in the workpiece contour posture data, and calculates the opening width ratio between the maximum allowable opening value parameter of the robot gripper and the corresponding overall width value of the workpiece.

[0074] The control node reads the distance between the left and right gripper fingers in millimeters when the servo motor drives the slider to the position of the physical stop in the forward stroke. This distance is used as the maximum allowable opening value parameter of the robot gripper during the trial gripping stage after the replacement gripper component is assembled. The corresponding overall width value in millimeters is extracted from the workpiece contour posture data. The opening width ratio is calculated as: overall width value of the corresponding workpiece / maximum allowable opening value parameter of the robot gripper.

[0075] The state change submodule collects the robot gripper drive current and the robot gripper encoder displacement data before and after contact at adjacent sampling times during the trial grasping phase. It calculates the difference between the two robot gripper drive currents at adjacent sampling times based on the opening width ratio, and calculates the robot gripper drive current change rate based on the sampling time interval. At the same time, it calculates the contact deformation offset of the robot gripper encoder displacement data before and after contact.

[0076] The stator current ampere value of the current sampling period and the stator current ampere value of the previous sampling period are read by the current transmitter during the trial gripping phase as the robot gripper drive current at adjacent sampling moments. The number of motor rotor pulses at the moment the gripper contacts the workpiece is read by the encoder as the robot gripper encoder displacement data before contact. The number of motor rotor pulses when the gripper clamps the workpiece and triggers the stop command is read as the robot gripper encoder displacement data after contact. Combined with the opening width ratio, the difference between the two robot gripper drive currents at adjacent sampling moments = the stator current ampere value of the current sampling period - the stator current ampere value of the previous sampling period. The sampling period in milliseconds configured in the control node is read as the sampling time interval. The robot gripper drive current change rate = the difference between the two robot gripper drive currents at adjacent sampling moments / the sampling time interval. The contact deformation offset = the absolute value of the difference between the robot gripper encoder displacement data before contact and the robot gripper encoder displacement data after contact.

[0077] The response determination submodule collects the gripper component numbers during the trial gripping phase. It groups and normalizes the opening width ratio, robot gripper drive current change rate, and contact deformation offset of each gripper component according to the gripper component number. It compares the normalized robot gripper drive current change rate and contact deformation offset with the preset contact limit threshold to determine the robot gripper contact response state and generate gripper gripping characterization data.

[0078] The hexadecimal device identifier code sent by the gripper communication serial port is used as the gripper component number during the trial gripping phase. According to the gripper component number, the corresponding opening width ratio, robot gripper drive current change rate, and contact deformation offset are stored in the corresponding data group with the same number. The historical maximum and minimum values ​​of the robot gripper drive current change rate within the same data group are extracted. After normalization, the robot gripper drive current change rate = (current robot gripper drive current change rate - historical minimum value) / (historical maximum value - historical minimum value). The historical maximum and minimum values ​​of the contact deformation offset within the same data group are extracted. After normalization, the contact deformation offset = (current contact deformation offset - historical minimum value) / (historical maximum value - historical minimum value). The first preset touch limit threshold and the second preset touch limit threshold defined in the control node are extracted. The threshold setting is based on collecting current and deformation data records from 500 normal gripping and 500 slip-off states, plotting probability density function curves, and defining the coordinates of the intersection points of the two normal distribution curves as the first and second preset contact limit thresholds, respectively. If the normalized rate of change of the robot gripper driving current is greater than the first preset contact limit threshold and the normalized contact deformation offset is greater than the second preset contact limit threshold, then the robot gripper contact response state is determined to be a valid gripping state and the valid gripping state is assigned the number 1. If the normalized rate of change of the robot gripper driving current is less than or equal to the first preset contact limit threshold or the normalized contact deformation offset is less than or equal to the second preset contact limit threshold, then the robot gripper contact response state is determined to be an invalid slip-off state and the invalid slip-off state is assigned the number 0, and the gripper gripping characterization data is saved.

[0079] Please see Figure 2 and Figure 6 The parameter learning module includes:

[0080] The status acquisition submodule monitors the workpiece following the robot's lifting status during the vertical lifting phase of the robot body after each trial gripping, and collects the robot gripper displacement data and back-off amount during the clamping and holding phase after each trial gripping, and establishes a set of motion following back-off amounts.

[0081] Extract the Z-axis movement distance of the workpiece's center of mass during the vertical lifting phase of the robot's Z-axis direction after each gripping attempt. Read the distance following tolerance value. This tolerance value is set based on the physical parameter of 5 mm, which is the maximum compressible deformation of the flexible pad on the inner side of the gripper. When the Z-axis movement distance of the workpiece's center of mass is greater than the distance following tolerance value, write the workpiece following robot lifting status indicator to the character "following successfully". When the Z-axis movement distance of the workpiece's center of mass is less than or equal to the distance following tolerance value, write the workpiece following robot lifting status indicator to the character "falling failed". Extract the highest and lowest points of the servo motor encoder pulse values ​​during the holding static maintenance phase after each gripping attempt through the control node. The robot gripper displacement data backoff amount = (highest point of servo motor encoder pulse value - lowest point of servo motor encoder pulse value) multiplied by the millimeter equivalent of a single pulse. Create a new blank database table and insert the character writing result and the corresponding robot gripper displacement data backoff amount row by row into the data table to establish the motion following backoff amount set.

[0082] The sample set submodule calls the contact response state in the gripper gripping characterization data, and combines it with the workpiece following the robot lifting status flag and the robot gripper displacement data back-off amount in the motion follow-back amount set to obtain the test gripping sample set amount.

[0083] The system retrieves the contact response status (1 or 0) stored in the gripper gripping data. It then extracts the workpiece following the robot's lifting status identifier character and the robot gripper displacement data retraction floating-point value corresponding to the gripping round from the action follow-back set in the database table. A one-dimensional data array is created in memory, with the contact response status number placed at the first index, the boolean value converted from the workpiece following the robot's lifting status identifier character placed at the second index, and the robot gripper displacement data retraction floating-point value placed at the third index. The system performs sequential concatenation of memory addresses, aggregating all the concatenated one-dimensional data arrays into a two-dimensional matrix to obtain the set of test gripping samples.

[0084] The regression mapping submodule collects the opening and closing stroke and closing speed inflection point, as well as the holding current threshold and release delay parameters. It performs Gaussian process regression analysis on the sample set size of the test grasping, the opening and closing stroke and closing speed inflection point, the holding current threshold and release delay parameters, and obtains the gripper grasping parameter set.

[0085] The process of Gaussian process regression includes:

[0086] Map the size of the sample set to an input feature vector;

[0087] The opening and closing stroke, the inflection point of the closing speed, the holding current threshold, and the release delay parameter are combined and mapped into a target observation vector;

[0088] Calculate the covariance mapping matrix between the input feature vector and the target observation vector;

[0089] The regression prediction values ​​under the condition of multidimensional feature space are solved based on the covariance mapping matrix;

[0090] Merge each regression prediction value to generate a gripper grasping parameter set;

[0091] The control node reads the maximum finger displacement (in millimeters) of the end effector driver as the opening / closing stroke, the pulse frequency value where the first derivative of the velocity-position curve abruptly changes as the closing velocity inflection point, the rated current (amperes) set during the steady-state clamping phase as the holding current threshold, and the millisecond interval from receiving the release signal to the motor spindle reversal as the release delay parameter. A Gaussian process regression model is constructed, including collecting historical test grasp samples as training input data, and using the opening / closing stroke and closing velocity inflection point as training target data. A covariance matrix is ​​constructed using a zero-mean prior assumption. Hyperparameter optimization training is performed by maximizing the marginal likelihood function. In application, the test samples are input into the model for prediction via covariance mapping. The two-dimensional matrix of the test grasp sample set is converted into a floating-point data matrix to construct the input feature vector. The specific values ​​of the opening / closing stroke, closing velocity inflection point, holding current threshold, and release delay parameter are arranged into a single-column matrix to construct the target observation vector. The covariance mapping matrix = (input feature vector matrix / target observation vector matrix) / total number of values. , in the formula, The kernel function evaluation value has the physical dimension of square millimeters and is obtained by reading the calculation result from memory. The input feature vector matrix contains historical trial sample vectors, which are dimensionless. The input vector for the new sample to be tested is dimensionless; For signal amplitude coefficient, As the amplitude base, To correspond to the signal attribute identifier, the overall unit is millimeters. The setting is based on the maximum fluctuation range of the target regression parameters. The more violent the fluctuation of the target amplitude, the greater its weight value. The value range is set between 0.5 and 5.0. is the cyclic index of the dimension of the vector, which is dimensionless; is the total number of feature dimensions of the vector, which is dimensionless; For vectors The first in The characteristic element values ​​are dimensionless. For vectors The first in The characteristic element values ​​are dimensionless. For length scale parameters, As the scale cardinality, This is a correlation distance identifier, dimensionless, set based on the correlation decay rate of sample features in multidimensional space. The smoother the correlation between features, the larger the value. The value range is from 0.1 to 10.0. If the hardware input signal is disconnected, causing the parameter to be 0, it will be forcibly assigned a value of 0.1 to avoid mathematical denominator errors. This is the noise tolerance limit. As the tolerance base, This is an indicator of ambient background noise, with the overall unit being millimeters. The setting is based on the basic measurement deviation limit given by the sensor hardware itself. The higher the sensor's noise floor, the larger the value, with a range of 0.01 to 0.5. This is an equivalence judgment function, dimensionless, and the judgment condition is: if vectors... with vector The value is 1 if the corresponding elements are completely equal, and 0 if there are differences; get parameter settings. , , , , , The logical judgment determines that the two vectors are inconsistent. Substitute the above data into the formula to obtain the result. The product of the row vector formed by the kernel function evaluation value and the inverse matrix of the covariance mapping matrix is ​​extracted. The product result is multiplied with the target observation vector to solve the regression prediction value under the multidimensional feature space condition. The regression prediction values ​​corresponding to the parameters of each dimension are merged by column to generate the gripper grasping parameter group.

[0092] Please see Figure 2 and Figure 7 The dynamic capture control module includes:

[0093] The deviation prediction submodule collects the robot end position and the current position of the robot gripper, as well as the robot gripper drive current and gripper contact deformation during the closing and clamping phases of the formal grasping process. It then calculates the end displacement deviation by referring to the three-dimensional vector of the robot end position and the coordinates of the workpiece contour center node.

[0094] The robot end-effector center coordinates in the world coordinate system are read in real time during the closing and gripping phases of the formal grasping process by the control node. This is used as the robot end-effector position. The mechanical distance is converted by the encoder to obtain the current position of the robot gripper. The real-time ampere reading of the drive coil is read by the current transmitter to obtain the robot gripper drive current. The change in resistance of the strain gauge of the flange torque sensor is read to calculate the corresponding deformation in micrometers as the gripper contact deformation. The three-dimensional vector X, Y, and Z components of the robot end-effector position are extracted. The lateral and longitudinal projection coordinates of the workpiece contour center position node coordinates are extracted in the same three-dimensional reference space. The end-effector displacement deviation is calculated as the square root of ((robot end-effector X component - workpiece contour center position node coordinate lateral projection value) + (robot end-effector Y component - workpiece contour center position node coordinate longitudinal projection value) + robot end-effector Z component square).

[0095] The rolling simulation submodule obtains the current position of the robot gripper, the gripper drive current and the gripper contact deformation. It combines the gripper grasping parameter set to deduce the rolling prediction closure increment parameters and the holding current adjustment parameters. At the same time, it deduces the future displacement data trajectory in the time domain length based on the end displacement deviation and calculates the expected movement orientation position adjustment parameters to obtain the end position correction amount.

[0096] The sensor hardware serial port reads the distance value corresponding to the current position of the robot gripper, the ampere value corresponding to the gripper drive current, and the micrometer value corresponding to the gripper contact deformation. Combined with the regression prediction baseline values ​​of the gripping parameter group within the control environment, the rolling estimated closure increment parameter is calculated as: closure stroke baseline value - robot gripper current position distance value. The stiffness compensation coefficient corresponding to the gripper contact deformation is extracted. This coefficient, with dimensions in amperes / micrometers, indicates the current compensation required per unit micrometer of deformation. The holding current adjustment parameter is calculated as: (holding current baseline value - gripper drive current amperes) + ... The gripper contact deformation is multiplied by the stiffness compensation coefficient. The predicted time step value of the control node clock is read. The predicted time domain length of the future displacement data trajectory is calculated as: the current end-effector's speed components along the three axes multiplied by the predicted time step value + the corresponding three components of the current end-effector's three-dimensional coordinates. The expected movement orientation position adjustment parameter is calculated as: each component of the three-dimensional coordinates of the future displacement data trajectory endpoint minus each corresponding component of the end-effector displacement deviation. The values ​​of the rolling estimated closure increment parameter, the holding current adjustment parameter, and the expected movement orientation position adjustment parameter are filled into the same update array to obtain the end-effector position correction.

[0097] The joint control submodule obtains the expected closing direction adjustment amount, integrates the end position correction amount with the estimated closing increment and holding current adjustment amount and the expected closing direction adjustment amount to construct a joint adjustment matrix, covering the original robot closing and gripping action associated control commands, and generates the robot gripper adaptive grasping control sequence.

[0098] Extract the grasping yaw angle correction value issued by the upper-level planning node as the expected closing direction adjustment value, extract the end position correction value update array, establish a new data overlay matrix in the control node memory, fill each element of the end position correction value update array into the first row of the matrix, fill the rolling estimated closing increment parameter value into the second row of the matrix, fill the holding current adjustment value into the third row of the matrix, and fill the expected closing direction adjustment value into the fourth row of the matrix to construct a joint adjustment matrix. Locate the register memory address range in the underlying drive board that originally stored the original robot closing and gripping action associated control instructions, write the data in the joint adjustment matrix into the above register memory address range in row order to perform data overlay writing, extract the updated dataset in the register and arrange it in the order of execution timestamps to generate the robot gripper adaptive grasping control sequence.

[0099] Please see Figure 8 An adaptive modular robot grasping method includes the following steps:

[0100] S1: Acquire the current calibration board image, robot end effector gripper reference point image, robot end effector position data, and robot end effector orientation data, and perform coordinate registration to generate pose mapping data;

[0101] S2: Perform feature analysis on the image of the workpiece to be grasped based on the pose mapping data to generate workpiece contour pose data;

[0102] S3: Obtain the allowable opening value of the replaceable gripper component, the displacement data of the robot gripper encoder and the drive current data of the robot gripper, combine them with the workpiece contour posture data, determine the contact response state of the robot gripper, and generate gripper grasping characterization data.

[0103] S4: Obtain the workpiece following the robot's lifting state and the robot's gripper displacement data and backtracking amount, combine the gripper gripping characterization data, and perform Gaussian process regression processing to generate a set of gripper gripping parameters.

[0104] S5: Obtain the position of the robot end effector, compare it with the workpiece contour posture data to determine the displacement data deviation, and at the same time obtain the current position of the robot gripper, the robot gripper drive current and the robot gripper contact deformation. Combine the gripper grasping parameter set to perform rolling prediction calculation and generate the robot gripper adaptive grasping control sequence.

[0105] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An adaptive modular robot grasping system, characterized in that, The system includes: The visual-guided registration module acquires the current calibration board image, the robot end effector gripper reference point image, the robot end effector position data, and the robot end effector orientation data, and performs coordinate registration to generate pose mapping data. The workpiece feature perception module performs feature analysis on the image of the workpiece to be grasped based on the pose mapping data, and generates workpiece contour pose data. The contact state sensing module obtains the allowable opening value of the replaceable gripper component, the displacement data of the robot gripper encoder and the drive current data of the robot gripper, and combines the workpiece contour posture data to determine the contact response state of the robot gripper and generate gripper grasping characterization data. The grasping parameter learning module acquires the workpiece following the robot's lifting state and the robot's gripper displacement data backtracking amount. It combines the gripper grasping characterization data and performs Gaussian process regression processing to generate a gripper grasping parameter set. The dynamic gripping control module acquires the position of the robot's end effector, compares it with the workpiece contour posture data to determine the displacement data deviation, and simultaneously acquires the current position of the robot gripper, the robot gripper drive current, and the robot gripper contact deformation. Combined with the gripper gripping parameter set, it performs rolling prediction calculations to generate an adaptive gripping control sequence for the robot gripper.

2. The adaptive modular robot grasping system according to claim 1, characterized in that: The pose mapping data includes the coordinates of the robot end effector reference point, the robot end effector position coordinates, and the robot end effector orientation angle in the industrial camera coordinate system. The workpiece contour posture data includes the center coordinates of the workpiece to be gripped, the width of the workpiece to be gripped, and the orientation angle of the spindle of the workpiece to be gripped. The gripper gripping characterization data includes the robot gripper opening width ratio, the robot gripper drive current change rate, the robot gripper contact deformation, and the robot gripper contact response state. The gripper gripping parameter set includes the robot gripper opening and closing stroke, the robot gripper closing speed inflection point, the robot gripper holding current threshold, and the robot gripper release delay. The robot gripper adaptive gripping control sequence includes the robot end effector position correction, the robot gripper closing increment, the robot gripper holding current correction, and the robot gripper closing direction correction.

3. The adaptive modular robot grasping system according to claim 1, characterized in that: The visual guided registration module includes: The calibration board correction submodule acquires calibration board images, robot end gripper reference point images, robot end position data, and robot end orientation data. It extracts pixel matrix coordinate items corresponding to the corner pixel coordinates in the calibration board images, performs multi-dimensional distortion correction calculations on the pixel matrix coordinate items, and obtains camera coordinate coefficient values. The pose calculation submodule obtains the reference point pixel coordinates in the reference point image of the robot's end gripper, performs coordinate conversion with the camera coordinate coefficient value, obtains the gripper reference point pixel coordinate item in the corresponding reference dimension, and performs multi-dimensional pose calculation by combining the robot end position data and robot end direction data to obtain the end gripper coordinate quantity. The coordinate registration submodule extracts the position node vectors of the end gripper coordinates and the corresponding reference point pixel coordinates in the reference dimension, performs cross-coordinate system position node registration calculations on the position node vectors, obtains the relative displacement offset coefficient and direction rotation angle coefficient between each position node, and generates pose mapping data.

4. The adaptive modular robot grasping system according to claim 1, characterized in that: The workpiece feature sensing module includes: The pixel correction submodule acquires the image of the workpiece to be grasped, calls the relative displacement offset coefficient and direction rotation angle coefficient of each position node in the pose mapping data, performs frame-by-frame pixel correction and coordinate mapping on the pixel sequence of the image of the workpiece to be grasped, and obtains the contour envelope boundary of the image of the workpiece to be grasped. The deviation calculation submodule extracts the lateral spacing dimension of the outline edge of the workpiece to be gripped based on the outline envelope boundary, converts the size unit to calculate the corresponding overall width value of the workpiece, and obtains the orientation angle of the corresponding workpiece main axis and the orientation angle of the preset robot gripper closing direction. It calculates the absolute value of the angle difference between the orientation angle of the workpiece main axis and the orientation angle of the preset robot gripper closing direction to generate the gripping direction deviation value. The posture generation submodule obtains the coordinate set of each edge extreme point of the contour envelope boundary of the workpiece to be grasped, calculates the mean of the horizontal coordinate component and the vertical coordinate component in each edge extreme point coordinate set, and combines them with the clamping direction deviation value and the corresponding overall width value of the workpiece to obtain the workpiece contour posture data.

5. The adaptive modular robot grasping system according to claim 1, characterized in that: The contact state sensing module includes: The width matching submodule collects the maximum allowable opening value parameter of the robot gripper during the trial gripping stage after the replacement gripper component is assembled, calls the corresponding overall width value of the workpiece in the workpiece contour posture data, and calculates the opening width ratio between the maximum allowable opening value parameter of the robot gripper and the corresponding overall width value of the workpiece. The state change submodule collects the robot gripper drive current and the robot gripper encoder displacement data before and after contact at adjacent sampling times during the trial grasping phase. It calculates the difference between the two robot gripper drive currents at adjacent sampling times based on the opening width ratio, and calculates the robot gripper drive current change rate according to the sampling time interval. At the same time, it calculates the contact deformation offset of the robot gripper encoder displacement data before and after contact. The response determination submodule collects the gripper component numbers during the trial gripping phase, groups and normalizes the opening width ratio of each gripper component, the robot gripper drive current change rate, and the contact deformation offset according to the gripper component number, and compares the normalized robot gripper drive current change rate and contact deformation offset with a preset contact limit threshold to determine the robot gripper contact response state and generate gripper gripping characterization data.

6. The adaptive modular robot grasping system according to claim 1, characterized in that: The capture parameter learning module includes: The status acquisition submodule monitors the workpiece following the robot's lifting status during the vertical lifting phase of the robot body after each trial gripping, and collects the robot gripper displacement data and back-off amount during the clamping and holding phase after each trial gripping, and establishes a set of motion following back-off amounts. The sample set submodule calls the contact response state in the gripper gripping characterization data, and combines it with the workpiece following robot lifting status flag in the action follow-back amount set and the robot gripper displacement data back-back amount to obtain the test gripping sample set quantity. The regression mapping submodule collects the opening and closing stroke and closing speed inflection point, as well as the holding current threshold and release delay parameters. It then performs Gaussian process regression analysis on the sample set of the test grippers, the opening and closing stroke and closing speed inflection point, the holding current threshold, and the release delay parameters to obtain the gripper gripping parameter set.

7. The adaptive modular robot grasping system according to claim 1, characterized in that: The dynamic capture control module includes: The deviation prediction submodule collects the robot end position and the current position of the robot gripper, as well as the robot gripper drive current and gripper contact deformation during the closing and clamping phases of the formal grasping process. It then calculates the end displacement deviation by referring to the three-dimensional vector of the robot end position and the coordinates of the workpiece contour center node. The rolling simulation submodule acquires the current position of the robot gripper, the gripper drive current, and the gripper contact deformation. It combines the gripper grasping parameter set to deduce the rolling prediction closure increment parameters and the holding current adjustment parameters. At the same time, it deduces the future displacement data trajectory in the time domain length based on the end displacement deviation and calculates the expected movement orientation position adjustment parameters to obtain the end position correction amount. The joint control submodule obtains the expected closing direction adjustment amount, integrates the end position correction amount with the estimated closing increment and holding current adjustment amount and the expected closing direction adjustment amount to construct a joint adjustment matrix, covering the original robot closing and gripping action associated control commands, and generates the robot gripper adaptive grasping control sequence.

8. The adaptive modular robot grasping system according to claim 3, characterized in that: The calibration plate image is a surface pattern image of a standard visual calibration reference set independently of the workpiece and the robot end gripper.

9. The adaptive modular robot grasping system according to claim 6, characterized in that: The Gaussian process regression process includes: Map the size of the sample set to an input feature vector; The opening and closing stroke, the inflection point of the closing speed, the holding current threshold, and the release delay parameter are combined and mapped into a target observation vector; Calculate the covariance mapping matrix between the input feature vector and the target observation vector; The regression prediction values ​​under the condition of multidimensional feature space are solved based on the covariance mapping matrix; Merge each regression prediction value to generate a gripper grabbing parameter set.

10. An adaptive modular robot grasping method, characterized in that, The adaptive modular robot grasping system according to any one of claims 1-9 is executed. Includes the following steps: S1: Acquire the current calibration board image, robot end effector gripper reference point image, robot end effector position data, and robot end effector orientation data, and perform coordinate registration to generate pose mapping data; S2: Perform feature analysis on the image of the workpiece to be grasped based on the pose mapping data to generate workpiece contour pose data; S3: Obtain the allowable opening value of the replaceable gripper component, the displacement data of the robot gripper encoder and the drive current data of the robot gripper, and combine them with the workpiece contour posture data to determine the contact response state of the robot gripper and generate gripper grasping characterization data. S4: Obtain the workpiece following the robot's lifting state and the robot gripper displacement data backtracking amount, combine the gripper gripping characterization data, and perform Gaussian process regression processing to generate a gripper gripping parameter set. S5: Obtain the position of the robot end effector, compare it with the workpiece contour posture data to determine the displacement data deviation, and at the same time obtain the current position of the robot gripper, the robot gripper drive current and the robot gripper contact deformation. Combine the gripper grasping parameter group to perform rolling prediction calculation and generate the robot gripper adaptive grasping control sequence.

Images