Collaborative robot self-calibration method based on improved random sampling consensus algorithm

Abstract

The invention discloses a collaborative robot self-calibration method based on an improved random sampling consistency algorithm. The collaborative robot self-calibration method comprises the steps of collecting m groups of configuration data of a six-degree-of-freedom collaborative robot; constructing a sampling pool in the configuration data corresponding to each single-point constraint; determining a selection strategy of configuration data in the sampling pool; randomly selecting configuration data of a preset temporary configuration data sample number from each sampling pool to form a temporary configuration data set; according to the temporary configuration data set, kinematics parameters based on temporary configuration data set identification are obtained; calculating the spherical profile tolerance error of the m groups of configuration data according to the kinematics parameters; constructing a score function according to the spherical profile tolerance error; calculating a configuration score under the kinematic parameters according to the score function; obtaining a model score according to m groups of configuration scores under kinematic parameters; and judging the model score until the number of sampled times reaches the maximum number of sampled times. According to the method, the evaluation accuracy of a sample estimation model and the reasonability of inner and outer point thresholds are enhanced, the accuracy of abnormal configuration identification is effectively improved by combining a semi-random sampling strategy, and support is provided for improving the kinematics parameter identification accuracy and the collaborative robot calibration precision.

Description

Technical Field

[0001] This invention relates to a self-calibration method for collaborative robots based on an improved random sampling consensus algorithm, belonging to the field of robot calibration technology. Background Technology

[0002] Collaborative robots emphasize human-robot collaboration and offer advantages such as high safety and flexibility, leading to their widespread application in complex and dynamic industrial production and medical services. In the face of changing application demands and long-term use, the positioning accuracy of collaborative robots, as a crucial indicator affecting their operational performance, is primarily maintained through kinematic calibration.

[0003] Leveraging the draggable, lightweight, and flexible characteristics of collaborative robots, the self-calibration method based on physical single-point constraints significantly reduces usage costs, improves calibration efficiency, and is more suitable for the frequent calibration needs of collaborative robots under different tasks. While manually dragging the collaborative robot to collect configuration data shortens operation time, it also more easily introduces complex human-induced data acquisition errors. Abnormal configuration data caused by acquisition errors may reduce the accuracy of collaborative robot kinematic calibration.

[0004] In view of this, the present invention is hereby proposed. Summary of the Invention

[0005] This invention provides a self-calibration method for collaborative robots based on an improved random sampling consensus algorithm. For collaborative robots under drag-and-drop physical single-point constraints, it can effectively identify abnormal configuration data, enhance the evaluation accuracy of the sample estimation model and the rationality of the inside and outside point thresholds, and effectively improve the accuracy of abnormal configuration identification by combining a semi-random sampling strategy.

[0006] The technical solution of this invention is:

[0007] According to a first aspect of the present invention, a self-calibration method for collaborative robots based on an improved random sampling consensus algorithm is provided, comprising:

[0008] Step 1, in Under single-point constraints, a total of six degrees of freedom collaborative robots were collected. Group configuration data; among which, ≥2.

[0009] Step 2: Set initial thresholds, reset loop count, number of samples, and thresholds for inside and outside points.

[0010] Step 3: Increment the sampling count by 1. Based on the current sampling count, construct a sampling pool in the configuration data corresponding to each single-point constraint; where the number of configuration data in the sampling pool is... As the number of samples changes.

[0011] Step 4: Reset the number of loops based on the threshold to determine the selection strategy for configuration data in the sampling pool; randomly select a preset number of configuration data samples from each sampling pool to form a temporary configuration dataset.

[0012] Step 5: Based on the temporary configuration dataset, obtain the kinematic parameters identified from the temporary configuration dataset; calculate the kinematic parameters based on the kinematic parameters. Spherical profile error in group configuration data.

[0013] Step 6: Construct a scoring function based on the spherical profile error; calculate the configuration score under kinematic parameters based on the scoring function; calculate the configuration score under kinematic parameters... The model score is obtained by analyzing the configuration scores.

[0014] Step 7: Judge the model score: If the model score is not the current maximum value, then judge the number of samples already taken: If the number of samples already taken has reached the maximum number of samples, select the kinematic parameters corresponding to the maximum model score under the current threshold reset loop number, and calculate the corresponding spherical profile error. Use the latest interior and exterior point thresholds to judge the obtained spherical profile error to determine the interior point configuration data and exterior point configuration data; use the interior point configuration data to re-identify the kinematic parameters and output the final identification result; otherwise, execute Step 3; otherwise, increment the threshold reset loop number by 1, update the maximum score based on the current model score, update the interior and exterior point thresholds according to the interior and exterior point threshold update formula, and set the number of samples already taken to zero before executing Step 3.

[0015] Furthermore, the number of configuration data in the sampling pool As the number of samples changes, the expression becomes:

[0016] ;

[0017] in, This represents the total number of configuration data collected under a single point constraint. For interior point ratio; and These represent the number of samples already taken and the maximum number of samples taken, respectively.

[0018] The maximum number of samples The calculation method is as follows:

[0019] ;

[0020] in, For confidence rate, To preset the number of temporary configuration data samples, This is the interior point ratio.

[0021] Furthermore, Step 4 specifically involves: if the threshold reset loop count is taken as the initial value, then the configuration data in each sampling pool is adjusted according to the number of data points. A temporary configuration dataset is formed by randomly selecting configuration data from the configuration data collected under the corresponding single-point constraints, and then randomly selecting a preset number of temporary configuration data samples from each sampling pool; otherwise, the configuration data in each sampling pool is selected based on the number of data points. A semi-random selection strategy is adopted to randomly select a preset number of temporary configuration data samples from each sampling pool to form a temporary configuration dataset.

[0022] Furthermore, the scoring function expression is:

[0023] ;

[0024] in, Represents the kinematic parameters of the first Individual configuration scores; For the first Spherical profile error of individual configuration data; The threshold values ​​are for both internal and external points.

[0025] Furthermore, the model score is the negative mean of the scores for all configurations, expressed as:

[0026] ;

[0027] in, Score the model; Represents the kinematic parameters of the first Individual configuration scores.

[0028] Furthermore, the formula for updating the thresholds of the inner and outer points is:

[0029] ;

[0030] in, It is a function of standard deviation; Specifically: sequentially The spherical profile error of the group configuration data is modulated and sorted according to the interior point ratio. The statistical model shows a smaller spherical profile error.

[0031] Furthermore, the configuration data in each sampling pool is determined according to the number of data points. A semi-random strategy is adopted for selection. Specifically, for configuration data under any single-point constraint, the spherical profile error calculated using the kinematic parameters with the maximum score of the current model is moduloed and sorted, and then ranked according to the inlier ratio. The spherical profile error with a smaller statistical modulus is the first spherical profile error. The remaining spherical profile error is considered as the second spherical profile error. Among them, the number of configuration data belonging to the first spherical profile error is: , To represent the total number of configuration data collected under a single point constraint; to include all configuration data corresponding to the first spherical profile error and those belonging to the second spherical profile error. A sampling pool is constructed from the configuration data; among which... .

[0032] According to a second aspect of the present invention, a collaborative robot self-calibration system based on an improved random sampling consensus algorithm is provided, the system comprising modules of any of the methods described above.

[0033] The beneficial effects of this invention are: 1. The proposed algorithm based on improved random sampling consensus (RANSAC) enhances the evaluation accuracy of the sample estimation model and the rationality of the thresholds for inside and outside points. Combined with a semi-random sampling strategy, it effectively improves the accuracy of abnormal configuration identification.

[0034] 2. This invention eliminates abnormal configurations based on the identification results of the improved RANSAC algorithm, avoiding the impact of abnormal configurations on kinematic calibration during the drag-and-collection process of the collaborative robot. The use of interior point configuration data for calibration effectively improves the accuracy of the collaborative robot.

[0035] 3. Compared with traditional kinematic calibration methods, the accuracy of this invention after calibration is significantly improved.

[0036] Specifically, this invention introduces a scoring function that considers the "quality" of interior points, reducing the interference of extreme outliers and improving the evaluation accuracy of the sample estimation model. Furthermore, based on the model score, an adaptive threshold adjustment strategy is designed to avoid the influence of experience-dependent interior / exterior point threshold settings on the configuration recognition results. On the other hand, combined with the proposed semi-random sampling strategy, the algorithm quickly converges to a relatively optimal kinematic parameter and interior / exterior point threshold in the early stages of iteration, and the sampling pool gradually approaches the full configuration pool in the later stages of iteration, avoiding convergence to local optima. Finally, compared with traditional kinematic calibration methods under physical single-point constraints, this invention enhances the evaluation accuracy of the sample estimation model and the rationality of the interior / exterior point thresholds. Combined with the semi-random sampling strategy, it effectively improves the accuracy of abnormal configuration recognition, providing support for improving the accuracy of kinematic parameter identification and the calibration accuracy of collaborative robots. Attached Figure Description

[0037] Figure 1 This is a flowchart of the present invention.

[0038] Figure 2The L2 norm curves of the collaborative robot's positional error before and after calibration using different methods are shown. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other.

[0040] Example 1: As Figures 1-2 As shown, according to a first aspect of the present invention, a collaborative robot self-calibration method based on an improved random sampling consensus algorithm is provided, comprising:

[0041] Step 1, in Under single-point constraints, a total of six degrees of freedom collaborative robots were collected. Group configuration data; among which, ≥2.

[0042] Step 2: Set initial thresholds, reset loop count, number of samples, and thresholds for inside and outside points.

[0043] Step 3: Increment the sampling count by 1. Based on the current sampling count, construct a sampling pool in the configuration data corresponding to each single-point constraint; where the number of configuration data in the sampling pool is... As the number of samples changes.

[0044] Step 4: Reset the number of loops based on the threshold to determine the selection strategy for configuration data in the sampling pool; randomly select a preset number of configuration data samples from each sampling pool to form a temporary configuration dataset.

[0045] Step 5: Based on the temporary configuration dataset, obtain the kinematic parameters identified from the temporary configuration dataset; calculate the kinematic parameters based on the kinematic parameters. Spherical profile error in group configuration data.

[0046] Step 6: Construct a scoring function based on the spherical profile error; calculate the configuration score under kinematic parameters based on the scoring function; calculate the configuration score under kinematic parameters... The model score is obtained by analyzing the configuration scores.

[0047] Step 7: Judge the model score: If the model score is not the current maximum value, then judge the number of samples already taken: If the number of samples already taken has reached the maximum number of samples, select the kinematic parameters corresponding to the maximum model score under the current threshold reset loop number, and calculate the corresponding spherical profile error. Use the latest interior and exterior point thresholds to judge the obtained spherical profile error to determine the interior point configuration data and exterior point configuration data; use the interior point configuration data to re-identify the kinematic parameters and output the final identification result; otherwise, execute Step 3; otherwise, increment the threshold reset loop number by 1, update the maximum score based on the current model score, update the interior and exterior point thresholds according to the interior and exterior point threshold update formula, and set the number of samples already taken to zero before executing Step 3.

[0048] For example, the A value greater than or equal to 2 indicates that, under any single-point constraint, the configuration data collected by the drag-and-drop collaborative robot is greater than 50. For example... With a value of 3, the drag-and-drop collaborative robot collects 100 sets of configuration data under the first single-point constraint, 100 sets of configuration data under the second single-point constraint, and 60 sets of configuration data under the third single-point constraint. That is, a total of 260 sets of configuration data are collected under the three single-point constraints.

[0049] Furthermore, the number of configuration data in the sampling pool As the number of samples changes, the expression becomes:

[0050] ;

[0051] in, This represents the total number of configuration data collected under a single point constraint. For interior point ratio; and These represent the number of samples already taken and the maximum number of samples taken, respectively.

[0052] The maximum number of samples The calculation method is as follows:

[0053] ;

[0054] in, For confidence rate, To preset the number of temporary configuration data samples, This is the interior point ratio.

[0055] Furthermore, Step 4 specifically involves: if the threshold reset loop count is taken as the initial value, then the configuration data in each sampling pool is adjusted according to the number of data points. A temporary configuration dataset is formed by randomly selecting configuration data from the configuration data collected under the corresponding single-point constraints, and then randomly selecting a preset number of temporary configuration data samples from each sampling pool; otherwise, the configuration data in each sampling pool is selected based on the number of data points. A semi-random selection strategy is adopted to randomly select a preset number of temporary configuration data samples from each sampling pool to form a temporary configuration dataset.

[0056] Furthermore, the scoring function expression is:

[0057] ;

[0058] in, Represents the kinematic parameters of the first Individual configuration scores; For the first Spherical profile error of individual configuration data; The threshold values ​​are for both internal and external points.

[0059] Furthermore, the model score is the negative mean of the scores for all configurations, expressed as:

[0060] ;

[0061] in, Score the model; Represents the kinematic parameters of the first Individual configuration scores.

[0062] Furthermore, the formula for updating the thresholds of the inner and outer points is:

[0063] ;

[0064] in, It is a function of standard deviation; Specifically: sequentially The spherical profile error of the group configuration data is modulated and sorted according to the interior point ratio. The statistical model shows a smaller spherical profile error.

[0065] Furthermore, the configuration data in each sampling pool is determined according to the number of data points. A semi-random strategy is adopted for selection. Specifically, for configuration data under any single-point constraint, the spherical profile error calculated using the kinematic parameters with the maximum score of the current model is moduloed and sorted, and then ranked according to the inlier ratio. The spherical profile error with a smaller statistical modulus is the first spherical profile error. The remaining spherical profile error is considered as the second spherical profile error. Among them, the number of configuration data belonging to the first spherical profile error is: , To represent the total number of configuration data collected under a single point constraint; to include all configuration data corresponding to the first spherical profile error and those belonging to the second spherical profile error. A sampling pool is constructed from the configuration data; among which... .

[0066] According to a second aspect of the present invention, a collaborative robot self-calibration system based on an improved random sampling consensus algorithm is provided, the system comprising modules of any one of the methods described above, specifically including:

[0067] The first module is used to execute Step 1: Under single-point constraints, a total of m sets of configuration data for a six-DOF collaborative robot were collected; among them... ≥2.

[0068] The second module is used to execute Step 2: set the initial threshold, reset the number of loops, the number of samples, and the thresholds for inside and outside points.

[0069] The third module is used to execute Step 3: Increment the number of samples by 1, and construct a sampling pool in the configuration data corresponding to each single-point constraint based on the current number of samples; wherein, the number of configuration data in the sampling pool is... As the number of samples changes.

[0070] The fourth module is used to execute Step 4: reset the number of loops based on the threshold, determine the selection strategy for configuration data in the sampling pool, and randomly select a preset number of configuration data samples from each sampling pool to form a temporary configuration dataset.

[0071] The fifth module is used to perform Step 5: based on the temporary configuration dataset, obtain the kinematic parameters identified from the temporary configuration dataset; and calculate based on the kinematic parameters. Spherical profile error in group configuration data.

[0072] The sixth module is used to execute Step 6: constructing a scoring function based on the spherical profile error; calculating the configuration score under kinematic parameters based on the scoring function; and calculating the configuration score under kinematic parameters. The model score is obtained by analyzing the configuration scores.

[0073] The seventh module is used to execute Step 1: Judging the model score. If the model score is not the current maximum value, the number of samples is then judged. If the number of samples has reached the maximum number of samples, the kinematic parameters corresponding to the maximum model score under the current threshold reset loop number are selected, and the corresponding spherical profile error is calculated. The obtained spherical profile error is judged using the latest interior and exterior point thresholds to determine the interior point configuration data and exterior point configuration data. The kinematic parameters are re-identified using the interior point configuration data, and the final identification result is output. Otherwise, Step 3 is executed. Otherwise, the threshold reset loop number is incremented by 1, the maximum score is updated according to the current model score, the interior and exterior point thresholds are updated according to the interior and exterior point threshold update formula, and the number of samples is set to zero before executing Step 3.

[0074] Example 2: Figure 1-2 As shown, a self-calibration method for collaborative robots based on an improved random sampling consensus algorithm includes:

[0075] S1, Number of single-point constraints =2, 100 sets of configuration data were collected under each single-point constraint, for a total of 2. =200 sets of configuration data, execute S2. Group configuration data is the first Joint angle data of robot configuration θ j = [ θ 1 θ 2 ⋯ θ 6 ] T :in, , ,……, These are the joint angle data for this configuration.

[0076] S2, Set threshold to reset loop count =1, number of samples already taken =0, set the initial threshold for inside and outside points, and execute S3.

[0077] S3. Increment the number of samples by 1, construct a sampling pool in the configuration data corresponding to each single-point constraint, and execute S4; where the number of configuration data in the sampling pool is... As the number of samples changes, the expression becomes: ;in, This represents the total number of configuration data collected under a single point constraint. This is the interior point ratio. When... When the calculation result is not an integer, it is rounded up. In this example, =100, =0.8. and These represent the number of samples already taken and the maximum number of samples taken, respectively.

[0078] Furthermore, the maximum number of sampling times The calculation method is as follows: ;in, For confidence rate, This sets the preset number of temporary configuration data samples. In this example, =0.99, =12, corresponding to the calculated value. The value is 64.6841, and this parameter is rounded up to 65 when it is used.

[0079] S4. If the threshold reset loop count is 1, then the configuration data in each sampling pool is determined according to the number of data points. Randomly select configuration data from the configuration data collected under the corresponding single-point constraints, and randomly select a preset number of temporary configuration data samples from each sampling pool to form a temporary configuration dataset, then execute S5; otherwise, the configuration data in each sampling pool are selected according to the number of data points. A semi-random strategy is adopted to randomly select a preset number of configuration data samples from each sampling pool to form a temporary configuration dataset, and then execute S5;

[0080] In this embodiment, when the threshold reset loop count is 1, the two sampling pools randomly select 81 data points from 100 sets of configuration data collected under the corresponding single-point constraint; 12 first configuration data points are randomly selected from the 81 configuration data points in the first sampling pool, and 12 second configuration data points are randomly selected from the 81 configuration data points in the second sampling pool. The first and second configuration data points are then merged to form a temporary configuration dataset.

[0081] S5. Based on the temporary configuration dataset and a system of linear equations constructed using the virtual spherical profile error model, the system of linear equations is solved using the least squares method to obtain the kinematic parameter error vector; after compensating the kinematic parameters using the kinematic parameter error vector, the kinematic parameters identified based on the temporary configuration dataset are obtained; and calculations are performed based on the kinematic parameters. To check the spherical profile error of the group configuration data, execute S6.

[0082] The system of linear equations is expressed as follows: ;in, and These are the Jacobian matrix and the spherical profile error vector corresponding to the temporary configuration dataset, respectively. This is the kinematic parameter error vector.

[0083] Solve the system of linear equations using the least squares method. The least squares solution is: .

[0084] Using the kinematic parameter error vector After compensating for the kinematic parameters, the kinematic parameters identified from the temporary configuration dataset are obtained. According to kinematic parameters calculate The spherical profile error of the group configuration data is specifically: the first In the configuration, using kinematic parameters and joint angle data Calculate the origin position of the robot flange coordinate system and the center point position of virtual tools For any single-point constraint, use the corresponding... Position of the origin of the robot flange coordinate system calculated from the configuration data and the center point position of virtual tools Fitted sphere center And calculate the radius of the sphere corresponding to the origin of the flange coordinate system. The radius of the sphere corresponding to the center point of the virtual tool Calculate the first... Spherical profile error corresponding to the origin of the flange coordinate system in the configuration data Spherical profile error corresponding to the center point of the virtual tool ,get Set of spherical profile error in configuration data The calculation of spherical profile error is referenced in " and ".

[0085] S6. Construct a scoring function based on the spherical profile error; calculate the configuration score under kinematic parameters based on the scoring function; calculate the configuration score under kinematic parameters... Group configuration score, obtain model score, execute S7.

[0086] The scoring function expression is: ;in, Represents the kinematic parameters of the first Individual configuration scores; For the first The spherical profile error of each configuration data. In the above formula, if If the absolute value of is greater than or equal to the threshold of the in- and out-of-point points, then the kinematic parameters are... Each configuration score takes an in-situ threshold; otherwise, the configuration score is... The absolute value of.

[0087] The model score is the negative mean of the scores for all configurations: .

[0088] S7. Determine the model score: If the model score is not the current maximum value, proceed directly to S8; otherwise, increment the threshold reset loop count by 1 and update the maximum score based on the current model score. Update the inner and outer point thresholds according to the inner and outer point threshold update formula, and set the number of samples already taken. After setting the value to zero, execute S3. The initial value is negative infinity.

[0089] The formula for updating the thresholds of the inner and outer points is: ;middle, It is a function of standard deviation; Specifically: sequentially The spherical profile error of the group configuration data is modulated and sorted according to the interior point ratio. The statistical modulus of the smaller spherical profile error (if sorted in ascending order, the one ranked first) is the one with the largest spherical profile error. The spherical profile error is used as If sorted in descending order, then the last few digits will be sorted. The spherical profile error is used as In this example, the interior point ratio =0.8, =200, Spherical profile error There are 160 groups in total.

[0090] The configuration data in each sampling pool is based on the number of data points. A semi-random strategy is adopted for selection. Specifically, for configuration data under any single-point constraint, the spherical profile error calculated using the kinematic parameters with the maximum score of the current model is moduloed and sorted, and then ranked according to the inlier ratio. The spherical profile error with a smaller statistical modulus is the first spherical profile error. The remaining spherical profile error is considered as the second spherical profile error. Among them, the number of configuration data belonging to the first spherical profile error is: , To represent the total number of configuration data collected under a single point constraint; to include all configuration data corresponding to the first spherical profile error and those belonging to the second spherical profile error. A sampling pool is constructed from the configuration data; among which... .

[0091] For example, =65, if the threshold reset loop count is greater than 1, then... As the number of samples increases from 1 to 65, the sampling pool... The number of configuration data points increased from 81 to 100, meaning the number of configuration data points increased from... The corresponding 80 groups and The corresponding set of configuration data has grown to The corresponding 80 groups and The corresponding 20 sets of configuration data.

[0092] S8. Determine the number of samples already taken: If the number of samples already taken has reached the maximum number of samples, select the kinematic parameters corresponding to the maximum model score under the current threshold reset loop count, and calculate the corresponding spherical profile error. Use the latest inner and outer point thresholds to determine the obtained spherical profile error to identify the inner point configuration data and outer point configuration data. Use the inner point configuration data to re-identify the kinematic parameters and output the final identification result; otherwise, execute S3.

[0093] The process of using the latest inside / outside point thresholds to determine the obtained spherical profile error is as follows: if the absolute value of the spherical profile error is less than the inside / outside point threshold, then the spherical profile error is an inside-point spherical profile error, and the corresponding configuration data is inside-point configuration data; otherwise, it is an outside-point spherical profile error and outside-point configuration data. That is, outside-point configuration data is recorded as abnormal configuration data. After removing abnormal configuration data from the group configuration data, the remaining configuration data is used as interior point configuration data.

[0094] In summary, this invention first rapidly collects configuration data by dragging a collaborative robot under physical single-point constraints, and then uses an improved Random Sample Consensus Algorithm (RANSAC) to identify abnormal configuration data. Based on the identification results, abnormal configurations are removed from the initial configuration data, and the optimal solution for kinematic parameters is calculated using interior point configuration data and the least squares method. Applying the above technical solution, it can be seen that this invention improves the accuracy of abnormal configuration identification and enhances the accuracy of kinematic calibration of collaborative robots when performing self-calibration based on physical single-point constraints.

[0095] To verify the effectiveness of the proposed method, this experiment was conducted on a general-purpose six-DOF collaborative robot. A total of 100 sets of joint angle data for different robot configurations under two constraint points were collected, and 50 different positioning points were collected in the robot's main workspace using a laser tracker as validation set data. The improved RANSAC algorithm of this invention includes an ingress / exgress point threshold. The initial value is set to 0.5, and the interior point ratio is... The value is 0.8, and the number of configuration samples is [missing information]. The value is 12. After calibrating the kinematic parameters of the collaborative robot using different methods, its validation set position error is as follows: Figure 2 As shown in Table 1, the data is organized as follows. It can be seen from the figure that the kinematic calibration method proposed in this invention can effectively avoid the influence of abnormal configurations on parameter identification, further improving the absolute position accuracy of the collaborative robot.

[0096] Table 1 Comparison of the L2 error of the collaborative robot validation set before and after calibration.

[0097]

[0098] As shown in the table, under physical single-point constraints, dragging collaborative robots to collect configuration data easily introduces abnormal configuration data. When using all configuration parameters for identification, abnormal configurations affect the accuracy of kinematic parameter identification, reducing the calibration accuracy of the collaborative robot. The recognition performance of the traditional RANSAC algorithm is affected by the initial values ​​of parameters such as inside and outside point thresholds. After calibration using this method, compared with all configurations, the maximum and average L2 norm of the validation set error are reduced by 14.6% and 5.3%, respectively. This invention proposes a collaborative robot self-calibration method based on an improved random sampling consensus algorithm, mitigating the impact of initial algorithm parameter values ​​on the accuracy of abnormal configuration identification, improving the accuracy of abnormal configuration identification, and thus ensuring the accuracy and reliability of parameter identification. Compared with calibration using all configurations, the proposed method reduces the maximum error by 43.4%, the average error by 33.6%, and the standard deviation by 30.9%. Compared with the traditional RANSAC algorithm, the proposed method reduces these three indicators by 33.7%, 29.9%, and 31.2%, respectively. Therefore, the self-calibration method for collaborative robots based on an improved random sampling consensus algorithm proposed in this invention can more effectively improve the position accuracy of the robot's end effector.

[0099] Experiments show that this invention is applicable to the kinematic calibration method of collaborative robots based on physical single-point constraints, which can significantly improve the position accuracy of the robot end effector.

[0100] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A self-calibration method for collaborative robots based on an improved random sampling consensus algorithm, characterized in that, include: Step 1, in Under single-point constraints, a total of six degrees of freedom collaborative robots were collected. Group configuration data; among which, ≥2; Step 2: Set initial thresholds, reset loop count, number of samples, and inner / outer point thresholds; Step 3: Increment the sampling count by 1. Based on the current sampling count, construct a sampling pool in the configuration data corresponding to each single-point constraint; where the number of configuration data in the sampling pool is... As the number of samples changes; Step 4: Reset the number of loops based on the threshold to determine the selection strategy for configuration data in the sampling pool; randomly select a preset number of configuration data samples from each sampling pool to form a temporary configuration dataset; Step 5: Based on the temporary configuration dataset, obtain the kinematic parameters identified from the temporary configuration dataset; calculate the kinematic parameters based on the kinematic parameters. Spherical profile error in group configuration data; Step 6: Construct a scoring function based on the spherical profile error; calculate the configuration score under kinematic parameters based on the scoring function; calculate the configuration score under kinematic parameters... The model score is obtained by analyzing the configuration scores. Step 7: Judge the model score: If the model score is not the current maximum value, then the number of samples already taken is judged: if the number of samples already taken has reached the maximum number of samples, the kinematic parameters corresponding to the maximum model score under the current threshold are selected, and the corresponding spherical profile error is calculated. The obtained spherical profile error is judged using the latest interior and exterior point thresholds to determine the interior point configuration data and the exterior point configuration data; the kinematic parameters are re-identified using the interior point configuration data, and the final identification result is output; otherwise, Step 3 is executed. Otherwise, increment the threshold reset loop count by 1, update the maximum score based on the current model score, update the inner and outer point thresholds according to the inner and outer point threshold update formula, and set the number of samples to zero before proceeding to Step 3.

2. The self-calibration method for collaborative robots based on an improved random sampling consensus algorithm according to claim 1, characterized in that, The number of configuration data in the sampling pool As the number of samples changes, the expression becomes: ； in, This represents the total number of configuration data collected under a single point constraint. For interior point ratio; and These represent the number of samples already taken and the maximum number of samples taken, respectively.

3. The self-calibration method for collaborative robots based on an improved random sampling consensus algorithm according to claim 1, characterized in that, The maximum number of samples The calculation method is as follows: ； in, For confidence rate, To preset the number of temporary configuration data samples, This is the interior point ratio.

4. The self-calibration method for collaborative robots based on an improved random sampling consensus algorithm according to claim 1, characterized in that, Step 4 specifically includes: If the threshold reset loop count is taken as the initial value, the configuration data in each sampling pool will be based on the number of data points. A temporary configuration dataset is formed by randomly selecting configuration data from the configuration data collected under the corresponding single-point constraints, and then randomly selecting a preset number of temporary configuration data samples from each sampling pool; otherwise, the configuration data in each sampling pool is selected based on the number of data points. A semi-random selection strategy is adopted to randomly select a preset number of temporary configuration data samples from each sampling pool to form a temporary configuration dataset.

5. The self-calibration method for collaborative robots based on an improved random sampling consensus algorithm according to claim 1, characterized in that, The scoring function expression is: ； in, Represents the kinematic parameters of the first Individual configuration scores; For the first Spherical profile error of individual configuration data; The threshold values ​​are for both internal and external points.

6. The self-calibration method for collaborative robots based on an improved random sampling consensus algorithm according to claim 1, characterized in that, The model score is the negative mean of the scores for all configurations, expressed as: ； in, Score the model; Represents the kinematic parameters of the first Individual configuration scores.

7. The self-calibration method for collaborative robots based on an improved random sampling consensus algorithm according to claim 1, characterized in that, The formula for updating the thresholds of the inner and outer points is: ； in, It is a function of standard deviation; Specifically: sequentially The spherical profile error of the group configuration data is modulated and sorted according to the interior point ratio. The statistical model shows a smaller spherical profile error.

8. The self-calibration method for collaborative robots based on an improved random sampling consensus algorithm according to claim 1, characterized in that, The configuration data in each sampling pool is based on the number of data points. A semi-random selection strategy is adopted, specifically: For configuration data under any single-point constraint, the spherical profile error calculated using the kinematic parameters with the maximum score of the current model is moduloed and sorted, according to the interior point ratio. The spherical profile error with a smaller statistical modulus is the first spherical profile error. The remaining spherical profile error is considered as the second spherical profile error. Among them, the number of configuration data belonging to the first spherical profile error is: , This represents the total number of configuration data collected under a single point constraint. All configuration data corresponding to the first spherical profile error and those belonging to the second spherical profile error. A sampling pool is constructed from the configuration data; among which... .

9. A self-calibration system for collaborative robots based on an improved random sampling consensus algorithm, characterized in that, The module includes the method described in any one of claims 1-8.

Images