A yarn hanging method with collaborative control of two composite robots

Abstract

This invention relates to the field of robot collaborative control technology and discloses a method for yarn hanging using dual composite robot collaborative control. The method includes: acquiring motion parameters of the two composite robots and preprocessing them to generate a collaborative path reference; identifying motion features and determining synchronization rules; performing collaborative path planning to form a collaborative action sequence and execution nodes; verifying the validity of execution nodes; executing corresponding operations according to the collaborative state (synchronous or asynchronous); and locating abnormal nodes, calculating phase offset, and compensating for phase offset when asynchronous. This method improves the accuracy, stability, and efficiency of dual-robot collaborative operations through steps such as motion parameter processing, motion feature identification, synchronization rule formulation, execution node verification, and phase compensation. It is suitable for industrial scenarios requiring high-precision collaborative operations, especially in yarn hanging operations, ensuring the accuracy of yarn hanging and the continuity of the operation.

Description

Technical Field

[0001] This invention relates to the field of robot collaborative control technology, specifically to a yarn hanging method for collaborative control of two composite robots. Background Technology

[0002] In modern industrial production, the application of robotics technology is becoming increasingly widespread, especially in scenarios requiring high-precision collaborative operations such as textiles and assembly. Dual-robot collaborative control technology has become crucial for improving production efficiency and quality. Traditional single-robot operation modes often suffer from low efficiency and insufficient flexibility when facing complex tasks, while dual-robot collaborative operation can significantly improve operational efficiency and precision through the rational allocation of tasks and coordinated actions. However, current dual-robot collaborative control technology still faces many challenges in practical applications.

[0003] Collaborative operation of two robots requires precise acquisition and preprocessing of motion parameters. In actual production environments, robot pose data, joint angles, and end-effector trajectories are easily affected by environmental interference and sensor errors, leading to outliers in the data. Furthermore, the units and physical meanings of different parameters may vary. Directly using unprocessed data for collaborative control will severely impact the accuracy and reliability of collaborative path planning. For example, in yarn hanging operations, deviations in the robot's pose data may result in inaccurate yarn hanging positions, affecting product quality.

[0004] The identification of motion features and the determination of synchronization rules are the core aspects of dual-robot cooperative control. Different cooperative path benchmarks have different motion features, such as continuous and discrete features. Accurately identifying these features and formulating corresponding synchronization rules is key to achieving efficient dual-robot cooperation. Traditional synchronization rule formulation methods often lack in-depth analysis of motion features, leading to problems such as motion mismatch and poor synchronization during cooperative operations, affecting the smoothness and stability of the operation.

[0005] Collaborative path planning and the validity verification of execution nodes are crucial for ensuring the smooth operation of dual-robot collaborative work. During collaborative path planning, multiple sequences of collaborative actions and execution nodes need to be formed, and the validity of these execution nodes directly affects the success of the entire operation. However, existing methods for verifying the validity of execution nodes are insufficient, failing to accurately and promptly determine their collaborative status. If asynchronous execution occurs, abnormal execution nodes cannot be quickly located and effectively compensated for, leading to operation interruptions or quality degradation.

[0006] In terms of phase offset calculation and phase compensation, existing technologies suffer from insufficient calculation accuracy and inflexible compensation strategies. When two robots execute asynchronously during operation, it is necessary to accurately calculate the phase offset and perform corresponding compensation to restore the preset phase difference. However, traditional calculation methods often overlook the importance of historical trajectory backtracking, leading to inaccurate phase offset calculations and poor compensation effects. Furthermore, the compensation strategy lacks the ability to dynamically adjust based on factors such as the number of backtracking attempts, making it unable to adapt to different degrees of anomalies. Summary of the Invention

[0007] The purpose of this invention is to provide a yarn-hanging method with collaborative control of two composite robots to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a yarn-hanging method under the cooperative control of two composite robots, the method comprising:

[0009] The motion parameters of the two composite robots are acquired, preprocessed, and a cooperative path reference is generated.

[0010] Identify the motion characteristics of each of the cooperative path references, and determine the synchronization rules of the two robots based on the motion characteristics;

[0011] Based on the synchronization rules, collaborative path planning is performed on the two robots to form multiple sets of collaborative action sequences, as well as the execution nodes of each set of collaborative action sequences;

[0012] The validity of each execution node is verified to obtain the collaborative state of each execution node, wherein the collaborative state includes synchronous execution state and asynchronous execution state;

[0013] In the synchronous execution state, the cooperative action sequence of the two robots is executed according to a preset phase difference;

[0014] In the asynchronous execution state, the abnormal execution node is located, and the phase offset of the cooperative action sequence under the abnormal execution node is calculated;

[0015] Based on the phase offset, phase compensation is performed on the abnormal execution node until the cooperative action sequence of the two robots recovers the preset phase difference and then terminates.

[0016] Preferably, the step of acquiring the motion parameters of the dual-robot composite robot, performing preprocessing, and generating a cooperative path reference includes:

[0017] The pose data, joint angles and end effector trajectories of the two robots are collected by sensors and stored as independent datasets, with a unique action identifier added to each dataset.

[0018] Outlier removal is performed on the data in each of the independent datasets, filtering out data items that exceed physical constraints;

[0019] The processed independent dataset is kinematically decomposed to extract translation and rotation components. The component parameters are then unified to the same unit before being output as a cooperative path reference.

[0020] Preferably, the identification of the action features of each of the cooperative path benchmarks includes:

[0021] Extract the key points of the trajectory under each collaborative path baseline, and generate an action node sequence according to the execution time sequence;

[0022] Calculate the spatial distance between adjacent action node sequences and define it as a synchronization constraint parameter;

[0023] Set an allowable deviation range, and normalize the synchronization constraint parameters according to the allowable deviation range to generate multiple synchronization intervals;

[0024] The distribution density of synchronization constraint parameters within each synchronization interval is statistically analyzed and labeled as classification constraint parameters. Based on the classification constraint parameters, the action characteristics of the cooperative path benchmark are determined respectively.

[0025] The action features include continuous features and discrete features, and the synchronization priority of the continuous features is higher than that of the discrete features.

[0026] Preferably, the step of determining the action features of the collaborative path benchmark based on the classification constraint parameters includes:

[0027] Obtain the classification constraint parameters corresponding to the baseline of each collaborative path;

[0028] The classification constraint parameters under the same collaborative path benchmark are sorted according to the density intervals from dense to sparse, and the coverage ratio of the maximum density interval is calculated.

[0029] Set a judgment threshold and compare the coverage ratio with the judgment threshold;

[0030] If the coverage ratio is higher than the determination threshold, then the collaborative path benchmark is determined to have continuous characteristics;

[0031] If the coverage ratio is lower than or equal to the determination threshold, then the collaborative path benchmark is determined to have discrete characteristics.

[0032] Preferably, determining the synchronization rules for the two robots based on the motion features includes:

[0033] Obtain the action characteristics of each collaborative path baseline;

[0034] The minimum value of the synchronization interval corresponding to the continuous feature is used as the synchronization trigger threshold;

[0035] Extract the synchronization trigger thresholds under all continuous features and arrange them in ascending order to generate a synchronization trigger sequence;

[0036] Action matching is performed on the two robots according to the synchronous trigger sequence, and after the collaborative path under the continuous feature is executed, the collaborative path under the discrete feature is executed.

[0037] Preferably, the validity verification of each execution node includes:

[0038] Collect the actual pose data of the two robots at each execution node, perform coordinate system alignment transformation, and generate multiple verification parameters;

[0039] Call the validation model, input the validation parameters into the model, and mark the matching degree output by the model as the validation result;

[0040] Set a verification threshold and compare the verification result with the verification threshold;

[0041] If the verification result is higher than the verification threshold, the execution node is determined to be in a synchronous execution state.

[0042] If the verification result is lower than or equal to the verification threshold, the execution node is determined to be in an asynchronous execution state.

[0043] Preferably, calculating the phase offset of the cooperative action sequence under the abnormal execution node includes:

[0044] Obtain the target trajectory points where the two robots did not execute synchronously under the abnormal execution node;

[0045] Calculate the spatial offset between the target trajectory point and the current trajectory point, and mark it as the real-time phase offset;

[0046] The historical trajectory of the abnormal execution node is traced back, and the historical phase offset of its predecessor node is extracted.

[0047] Set a maximum allowable offset, and stop backtracking when the historical phase offset is less than or equal to the maximum allowable offset;

[0048] The compensation algorithm is invoked, and the real-time phase offset and historical phase offset are input into the algorithm. The algorithm output is then marked as the total phase offset.

[0049] Preferably, the step of performing phase compensation on the abnormal execution node based on the phase offset includes:

[0050] Obtain the backtracking count of the abnormal execution node and mark it as a compensation priority parameter;

[0051] Set a compensation threshold and compare the compensation priority parameter with the compensation threshold;

[0052] When the compensation priority parameter exceeds the compensation threshold, the trajectory parameters of the abnormal execution node are adjusted according to the total phase offset to generate an updated execution node.

[0053] When the compensation priority parameter does not exceed the compensation threshold, the phase offset is continuously monitored until the priority parameter reaches the compensation threshold and compensation is triggered.

[0054] Preferably, the step of generating the synchronization trigger sequence further includes:

[0055] Extract the robot end effector velocity curves corresponding to each synchronization trigger threshold;

[0056] The synchronization trigger thresholds are sorted a second time based on the smoothness of the speed curve, and the synchronization trigger thresholds with a speed change rate lower than the set value are selected first.

[0057] Preferably, the coordinate system alignment transformation step includes:

[0058] Map the pose data of the dual robot end effectors to the same global coordinate system;

[0059] The relative pose deviation between the two robot base coordinate systems is eliminated by homogeneous matrix transformation and unified into the same reference datum.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] In terms of motion parameter processing, sensors collect motion parameters such as pose data, joint angles, and end-effector trajectories of the two robots, and store them as independent datasets, providing a rich and accurate data foundation for subsequent processing and analysis. Outlier removal is performed on the data within each independent dataset, filtering out data items that exceed physical constraints, effectively ensuring the reliability and validity of the data and avoiding collaborative control errors caused by data anomalies. The processed independent datasets are then kinematically decomposed, extracting translational and rotational components, and outputting them as a collaborative path reference after unifying them to the same units. This allows different types of motion parameters to be analyzed and planned under the same reference, improving the accuracy and consistency of collaborative path planning.

[0062] In terms of motion feature recognition and synchronization rule determination, key trajectory points under each collaborative path baseline are extracted and motion node sequences are generated. Multiple synchronization intervals are generated by calculating and normalizing the spatial distance between adjacent motion node sequences, and then the distribution density of synchronization constraint parameters within the synchronization intervals is statistically analyzed to determine motion features. This method can accurately identify continuous and discrete features of the collaborative path baseline, providing a basis for formulating reasonable synchronization rules. Based on motion features, the synchronization rules for the two robots are determined. The minimum value of the synchronization interval corresponding to continuous features is used as the synchronization trigger threshold, and synchronization trigger sequences are generated in ascending order. Simultaneously, after the collaborative path under continuous features is executed, the collaborative path under discrete features is executed. This synchronization rule fully considers the priority of different motion features, ensuring that the two robots can execute actions in a reasonable order and rhythm during collaborative operation, improving the smoothness and stability of collaborative operation.

[0063] In terms of collaborative path planning and execution node verification, collaborative path planning is performed on the two robots according to synchronization rules, forming multiple sets of collaborative action sequences and execution nodes, providing a clear execution flow for the collaborative operation of the two robots. The effectiveness of each execution node is verified by collecting actual pose data and performing coordinate system alignment transformation to generate verification parameters, and then calling the verification model to judge the synchronization state of the execution nodes. This verification method can promptly and accurately detect abnormal situations of execution nodes. When an execution node is in an asynchronous execution state, it can quickly locate the abnormal execution node and calculate the phase offset, providing strong support for subsequent phase compensation.

[0064] In terms of phase offset calculation and phase compensation, the spatial offset is calculated and historical trajectory backtracking is performed by acquiring the target trajectory points where the two robots are not executed synchronously under abnormal execution nodes. Historical phase offsets are extracted, and a compensation algorithm is called to generate the total phase offset, ensuring the accuracy and comprehensiveness of the phase offset calculation. Phase compensation is performed on abnormal execution nodes based on the phase offset. A compensation priority parameter is set according to the number of backtracking attempts. When the compensation priority parameter exceeds the compensation threshold, the trajectory parameters are adjusted to generate an updated execution node; otherwise, monitoring continues until compensation is triggered. This compensation strategy is flexible and dynamically adaptable, enabling timely adjustments to compensation measures based on the severity of the abnormal situation. This ensures that the cooperative action sequence of the two robots can quickly recover the preset phase difference, guaranteeing the continuity and stability of the operation.

[0065] Furthermore, during the generation of the synchronization trigger sequence, the robot end effector velocity curves corresponding to each synchronization trigger threshold are extracted. The synchronization trigger thresholds are then reordered based on the smoothness of the velocity curves, prioritizing those with a velocity change rate lower than a set value. This method further optimizes the motion trajectory of the two robots, reduces the impact of sudden velocity changes on collaborative operation, and improves the stability and accuracy of the operation. During the coordinate system alignment transformation, the pose data of the two robot end effectors are mapped to the same global coordinate system. Homogeneous matrix transformation eliminates the relative pose deviation between the base coordinate systems, unifying them into a single reference datum. This ensures accurate matching of the pose data of the two robots in different coordinate systems, providing a reliable coordinate foundation for collaborative control. Attached Figure Description

[0066] Figure 1 This is a schematic diagram illustrating the working principle of the yarn-hanging method with dual composite robot collaborative control as described in this invention.

[0067] Figure 2 Design drawings generated for the collaborative path baseline of the dual composite robots;

[0068] Figure 3 Design diagram for collaborative path baseline action feature recognition;

[0069] Figure 4 Design drawings for determining the synchronization rules of the two robots. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0071] Please see Figures 1-4 This invention relates to a yarn-hanging method for collaborative control of two composite robots. By acquiring the motion parameters of the two composite robots and preprocessing them to generate a collaborative path reference, identifying motion characteristics to determine synchronization rules, performing collaborative path planning, and verifying the effectiveness of execution nodes, precise collaborative control of the two robots in synchronous or asynchronous states is achieved. The following detailed description, in conjunction with specific steps and embodiments, illustrates this method:

[0072] The pose, joint angles, and end-effector trajectories of the two robots are collected in real time by sensors. These three types of data are stored as independent datasets, and each dataset is assigned a unique motion identifier for subsequent tracking. Outlier removal is performed on each dataset, using statistical methods (such as the Z-score algorithm) to filter data items that exceed the robot's physical constraints (such as joint range of motion and end-effector velocity limits). Subsequently, kinematic decomposition is performed on the processed datasets to separate translational components (such as XYZ coordinates in Cartesian coordinates) and rotational components (such as Euler angles or quaternions). All component parameters are then standardized to the same unit (such as millimeters or radians), and the final output is the cooperative path reference.

[0073] Key trajectory points (such as trajectory turning points and velocity change points) under each collaborative path baseline are extracted and action node sequences are generated according to the execution time sequence. The Euclidean distance between adjacent action node sequences is calculated as a synchronization constraint parameter, and the parameter is normalized by setting an allowable deviation range (such as ±5mm) to generate multiple synchronization intervals (such as 0-5mm, 5-10mm, etc.). The distribution density of parameters within each interval is statistically analyzed and marked as classification constraint parameters. Based on these parameters, action features are determined: if the coverage ratio of the maximum density interval under a certain baseline is higher than the determination threshold (such as 70%), it is determined as a continuous feature; otherwise, it is a discrete feature. The synchronization priority of continuous features is higher than that of discrete features. Based on this, the synchronization rule is determined: the minimum value of the synchronization interval corresponding to the continuous feature is used as the synchronization trigger threshold, and synchronization trigger sequences are generated in ascending order of threshold. Continuous feature paths are executed first, and discrete feature paths are executed after completion.

[0074] Based on synchronization rules, cooperative path planning is performed on the two robots, combining continuous and discrete feature paths into multiple sets of cooperative action sequences, and setting execution nodes for each set of sequences. The execution nodes contain the pose, joint angles, and end-effector trajectory parameters of the two robots at a specific moment, serving as a benchmark for subsequent verification.

[0075] The actual pose data of the two robots at each execution node are collected, and the data is mapped to the same global coordinate system through homogeneous matrix transformation to eliminate base coordinate system deviation. The processed verification parameters are input into the verification model (such as a matching degree calculation model based on a neural network), and the matching degree is output as the verification result. A verification threshold is set (e.g., 85%). If the matching degree is higher than the threshold, it is determined to be a synchronous execution state, and the two robots execute according to the preset phase difference; if it is lower than or equal to the threshold, it is determined to be an asynchronous execution state, and the exception handling process is triggered.

[0076] In asynchronous mode, the abnormal execution node is located, and the target trajectory point that the two robots did not execute synchronously is obtained. The spatial offset between the target point and the current point is calculated as the real-time phase offset. The preceding node is traced back, and historical phase offsets are extracted until the historical offset exceeds the maximum allowable offset (e.g., 10mm), at which point the tracing stops. A compensation algorithm (e.g., PID control algorithm) is invoked, and the total phase offset is calculated by combining the real-time and historical offsets. Based on the comparison between the number of tracing iterations (compensation priority parameter) and the compensation threshold (e.g., 3 times), when the priority parameter exceeds the threshold, the trajectory parameters of the abnormal node are adjusted (e.g., extending the execution time, adjusting the movement speed), and an updated execution node is generated; if the threshold is not exceeded, monitoring continues until the threshold is reached, triggering compensation until the two robots restore the preset phase difference.

[0077] The present invention will be further described below with reference to Examples 1 to 6:

[0078] Example 1:

[0079] In the yarn-hanging method of cooperative control of two composite robots, acquiring the motion parameters of the two robots and preprocessing them to generate a cooperative path reference is a fundamental step. This embodiment describes in detail the specific implementation of this step, including steps such as data acquisition, storage, outlier handling, and kinematic decomposition.

[0080] First, data acquisition is accomplished through a sensor system. The sensors employ a combination of high-precision inertial measurement units (IMUs) and vision cameras. The IMUs are mounted near each joint of the robot to collect data such as joint angular velocity and acceleration in real time, with a sampling frequency set to 100Hz to ensure the capture of subtle changes in joint movement. The vision cameras use a binocular vision module, positioned above the robot's end effector. They calculate the three-dimensional coordinates of the end effector trajectory using the principle of binocular parallax. The camera resolution is 1920×1080 pixels, and the frame rate is 60 frames per second, meeting real-time requirements. Each robot in the dual-robot system is independently configured with a sensor system to achieve synchronous acquisition of its own motion parameters.

[0081] The collected data needs to be categorized and stored. Specifically, the pose data, joint angle data, and end-effector trajectory data of the two robots are stored as separate datasets. Pose data includes the robot's position (x, y, z) in the global coordinate system and its attitude (roll angle, pitch angle, yaw angle); joint angle data consists of the real-time angle or displacement values ​​of each rotary and translator joint; end-effector trajectory data is a continuous sequence of coordinate points of the end effector during motion. Each dataset is stored in CSV format, with the file naming convention being "robot identifier_data type_acquisition time_serial number," for example, "RobotA_Pose_202505281430_001.csv" and "RobotB_JointAngles_202505281435_002.csv." The robot identifier distinguishes between the two robots (e.g., RobotA and RobotB), the data type clearly specifies the stored content (pose, joint angle, or end-effector trajectory), the acquisition time is accurate to the minute, and the serial number distinguishes different datasets of the same type, ensuring that each dataset has a unique motion identifier for easy data tracing and matching later.

[0082] Next, outlier removal is performed. For joint angle data, the physical constraint range of each joint is first set according to the robot's mechanical design parameters. For example, the mechanical limit of a certain rotary joint is -180° to 180°. If the collected angle value exceeds this range, it is judged as an outlier and directly removed. For pose data and end-effector trajectory data, the Z-score algorithm from statistical methods is used for processing. Taking the X coordinate of the end-effector trajectory as an example, the mean μ and standard deviation σ of the dataset are first calculated. If the X value of a data point satisfies |X-μ|>3σ, it is judged as an outlier and removed. When removing outliers, the timestamp information in the original data must be retained to ensure the continuity of the data in the time series for subsequent processing.

[0083] After outlier removal, kinematic decomposition is performed on the processed independent datasets. The purpose of kinematic decomposition is to break down complex motion parameters into translational and rotational components to facilitate subsequent cooperative path planning. For pose data, the position coordinates are directly extracted as the translational component (unit: millimeters), and the attitude angles are converted to radians as the rotational component. For joint angle data, the position and attitude of the end effector are calculated using the robot's forward kinematic model, thus obtaining the translational and rotational components. Specifically, assuming the robot has n joints, the joint angle vector is θ=[θ1,θ2,...,θ...]. n ]ᵀ, through the forward kinematics formula:

[0084]

[0085] Where R is the rotation matrix and p is the translation vector, the pose of the end effector in the global coordinate system is calculated using this formula, and then the translation component p(x,y,z) and the Euler angles or quaternions (converted to radians) corresponding to the rotation matrix R are separated. For end effector trajectory data, the position of each coordinate point is directly extracted as the translation component. If the trajectory data contains end effector attitude information, the rotation component is also separated; if it only contains position information, the rotation component is set to zero by default or calculated based on the kinematic model.

[0086] After extracting the translational and rotational components, all component parameters need to be standardized to the same unit. Translational components are standardized to millimeters (mm), and rotational components to radians (rad). For rotational parameters expressed in degrees (°) in the original data, the conversion is performed using the formula: radians = angle × π / 180. For example, a pitch angle of 30° for a joint is converted to π / 6 rad. After unit standardization, each dataset will output structured data containing timestamps, translational components (x, y, z), and rotational components (α, β, γ), with timestamps accurate to milliseconds to ensure time synchronization of the motion parameters between the two robots.

[0087] The generated collaborative path reference data needs to meet the following requirements: First, the time series of the data must be continuous without interruption or jump; second, the accuracy of the translation and rotation components must meet the requirements of robot collaborative control, for example, the error of the translation component should not exceed ±0.1mm and the error of the rotation component should not exceed ±0.001rad; third, each dataset must be accurately matched through a unique action identifier to ensure that the motion parameters of the two robots correspond one-to-one in time and action.

[0088] During data processing, the following details need attention: The installation location and calibration accuracy of sensors directly affect the accuracy of the data. Therefore, strict calibration is required after sensor installation. For example, the intrinsic and extrinsic parameters of the vision camera should be calibrated using a calibration board, and the IMU should be calibrated for zero bias and scale factor using standard parts. The parameter settings of the outlier removal algorithm need to be adjusted according to the actual physical performance and motion characteristics of the robot to avoid erroneously removing normal data or retaining outlier data. The kinematic decomposition model needs to accurately reflect the mechanical structure of the robot. For different types of robots (such as serial robots and parallel robots), the corresponding kinematic formulas need to be used for calculation.

[0089] Example 2:

[0090] In the yarn-hanging method of cooperative control of two composite robots, identifying the motion features of the cooperative path reference is a key step in determining the synchronization rules of the two robots. This embodiment describes in detail the specific implementation of motion feature identification, including the processes of trajectory key point extraction, synchronization constraint parameter calculation, synchronization interval generation, classification constraint parameter statistics, and motion feature determination.

[0091] The extraction of key points on the trajectory employs a curvature detection algorithm. Curvature is a crucial parameter describing the degree of trajectory bending; a larger curvature indicates a more severe trajectory bending, typically corresponding to the turning points of action nodes. The terminal trajectory curve of each cooperative path baseline is represented in discrete point form. , where each point Given three-dimensional spatial coordinates, calculate the curvature at each point. When this is the case, the difference method of adjacent three points is used to approximate the solution:

[0092]

[0093] Where "×" represents the cross product of vectors, the numerator is the magnitude of the cross product, and the denominator is the cube of the distance between two adjacent points. When the curvature of a point... If the curvature of a point exceeds a set threshold (e.g., 0.1 / mm), the point is determined to be a key point in the trajectory. For example, if the curvature of a point in a certain trajectory is calculated to be 0.15 / mm, exceeding the threshold of 0.1 / mm, then the point is marked as a key point and used as part of the action node sequence.

[0094] After extracting key points, an action node sequence is generated according to the execution time sequence. Each node This includes the timestamp, translation component, and rotation component of the key point. For continuous trajectories, the action node sequence is dense and continuous; for discrete trajectories, the node sequence is sparse and irregularly spaced.

[0095] Calculate the spatial distance between adjacent action node sequences as a synchronization constraint parameter. For the action node sequences of two robots... and Assuming the current process is the first... Node pairs Its spatial distance Calculated using the three-dimensional Euclidean distance formula:

[0096]

[0097] This distance value reflects the synchronization error of the two robots at the same action node's end-effector position. For example, if The coordinates are (100, 200, 300) mm. Given coordinates (102, 201, 303) mm, the spatial distance is... mm.

[0098] After calculating the spatial distances between all adjacent node pairs, the synchronization constraint parameters need to be normalized. First, a tolerance range is set, for example, ±5mm, meaning the maximum allowable synchronization error is 5mm. During normalization, each distance value... Mapping to the interval [0,1], the mapping formula is:

[0099]

[0100] in, It is the minimum value among all distance values. It is the maximum value. If mm, mm, then in the above example After normalization, the parameters are divided into multiple synchronization intervals based on their numerical range, for example, into three intervals: [0, 0.3), [0.3, 0.6), and [0.6, 1.0]. These intervals correspond to different synchronization constraint strengths, with smaller interval values ​​indicating higher synchronization requirements.

[0101] The distribution density of synchronization constraint parameters within each synchronization interval is statistically analyzed and labeled as classification constraint parameters. The distribution density is calculated using the kernel density estimation (KDE) method. By setting the kernel function (such as a Gaussian kernel) and bandwidth, the probability density value of the parameter within each interval is calculated. For example, for the set of synchronization constraint parameters of a certain cooperative path benchmark, after kernel density estimation, the density value of the interval [0,0.3) is 0.8, the density value of the interval [0.3,0.6) is 0.2, and the density value of the interval [0.6,1.0] is 0.05. Then, the classification constraint parameters of this benchmark are the density value array of each interval [0.8,0.2,0.05].

[0102] The action characteristics of the collaborative path baseline are determined based on classification constraint parameters. The specific steps are as follows: First, obtain the classification constraint parameters corresponding to the baseline and sort the intervals from densest to sparsest according to their density values. For example, in the example above, the intervals are sorted as [0, 0.3), [0.3, 0.6), and [0.6, 1.0]. Then, calculate the coverage ratio of the maximum density interval, which is the proportion of the maximum density value to the sum of the density values ​​of all intervals. In the example, the sum of density values ​​is 0.8 + 0.2 + 0.05 = 1.05, and the coverage ratio of the maximum density interval [0, 0.3) is 0.8 / 1.05 ≈ 76.19%. A judgment threshold of 70% is set. If the coverage ratio is higher than the threshold, the collaborative path baseline is judged to have continuous characteristics, indicating that the actions of the two robots on this path need to be closely synchronized; if the coverage ratio is lower than or equal to the threshold, it is judged to have discrete characteristics, allowing for a certain degree of asynchrony in action execution.

[0103] In practical applications, the following details should be noted: The curvature threshold for extracting key trajectory points needs to be adjusted according to the accuracy requirements of the yarn hanging process. For example, for high-precision processes, the threshold can be set to 0.05 / mm, while for low-precision processes, it can be relaxed to 0.2 / mm. The allowable deviation range of the synchronization constraint parameters needs to be determined in conjunction with the robot's motion accuracy. For example, for a robot with a repeatability accuracy of ±0.1mm, the allowable deviation can be set to ±0.5mm to reserve a safety margin. The choice of bandwidth for kernel density estimation will affect the density calculation results. The optimal bandwidth is usually determined by cross-validation to avoid overfitting or underfitting.

[0104] The determination of motion features directly affects the formulation of synchronization rules. Continuous features have a higher synchronization priority than discrete features. Therefore, in collaborative path planning, the sequence of motion nodes corresponding to continuous features should be processed first to ensure that the two robots strictly execute according to the preset phase difference in high synchronization requirements. For example, in the yarn hanging process, the initial clamping and termination release actions of the yarn usually have continuous features and require precise synchronization; while the intermediate transportation actions may have discrete features and allow for a certain degree of asynchronous adjustment.

[0105] Example 3:

[0106] In the yarn-hanging method of collaborative control of two composite robots, determining the synchronization rules of the two robots based on motion features is the core step in achieving precise collaboration. This embodiment describes in detail the process of determining the synchronization rules, including steps such as motion feature acquisition, synchronization trigger threshold setting, synchronization trigger sequence generation, and motion matching strategy, focusing on the priority differences between continuous and discrete features.

[0107] The motion characteristics of each collaborative path baseline are obtained. These motion characteristics are determined through the curvature detection, spatial distance calculation, and classification constraint parameter statistics process described in Example 2. Each collaborative path baseline is labeled as either a continuous or discrete feature. Continuous features indicate that the actions of the two robots on this path must be closely synchronized, such as critical actions like yarn clamping and trajectory tracking. Discrete features indicate that actions are allowed to be executed asynchronously to a certain extent, such as adjustments or waiting actions on non-critical paths.

[0108] For the collaborative path benchmark corresponding to a continuous feature, the minimum value of its synchronization interval is extracted as the synchronization trigger threshold. The synchronization interval is the parameter interval after normalization processing as described in Example 2. For example, if the synchronization constraint parameters of a certain continuous feature are distributed in the interval [0, 0.4) after normalization, then the minimum value of this interval is 0 (corresponding to the minimum value of the actual spatial distance). It should be noted that the synchronization trigger threshold needs to be mapped back to the actual physical unit. For example, if the normalized interval [0, 0.4) corresponds to an actual spatial distance of 0-4mm, then the threshold is 0mm (indicating that the positions of the two robot end effects must be completely synchronized).

[0109] After extracting the synchronization trigger thresholds for all continuous features, the thresholds need to be sorted to generate a synchronization trigger sequence. The first stage of sorting arranges the thresholds in ascending order. For example, if the thresholds for the three continuous features are 1.5mm, 2mm, and 3mm, the initial sequence is [1.5mm, 2mm, 3mm]. The second stage involves a secondary sorting based on the smoothness of the robot's end-effector velocity curve to avoid mechanical shocks or synchronization errors caused by sudden velocity changes. Specifically, the robot's end-effector velocity curve corresponding to each synchronization trigger threshold is extracted. The velocity curve is calculated using the translation components after kinematic decomposition, using the following formula:

[0110]

[0111] in, This indicates the terminal velocity (unit: mm / s). This indicates the change in translational component within adjacent time intervals (unit: mm). Indicates the time interval (unit: seconds). The smoothness of the velocity curve is measured by the rate of change of velocity. (Unit: mm / s²) Measured by the rate of change, a smaller rate of change indicates a more stable velocity. The threshold for the rate of change of velocity is set as follows: (e.g., 50 mm / s²), prioritize synchronous triggering thresholds with a velocity change rate lower than this threshold. For example, if the velocity change rate corresponding to a threshold of 1.5 mm is 40 mm / s², and the velocity change rate corresponding to a threshold of 2 mm is 60 mm / s², then the former will be prioritized at the beginning of the sequence because it meets the smoothness requirement. The adjusted synchronous triggering sequence is [1.5 mm, 2 mm, 3 mm] (assuming the velocity change rate of the threshold of 3 mm is 45 mm / s², which is still lower than the threshold).

[0112] After the synchronous trigger sequence is generated, motion matching is performed on the two robots based on this sequence. The core of motion matching is to ensure that continuous feature paths are executed sequentially according to the synchronous trigger sequence, with each trigger threshold corresponding to an action node that the two robots must reach synchronously. For example, when the synchronous trigger threshold is 1.5mm, the end effectors of the two robots must simultaneously reach the spatial position corresponding to that threshold, and the error must be controlled within the allowable deviation range (e.g., ±0.5mm). When executing continuous feature paths, the pose data of the two robots is monitored in real time (as described in Example 1, sensor-collected data) to determine whether the synchronous trigger conditions are met. If the threshold is reached and the pose error of the two robots is within the allowable range, the next node action is triggered; if the error exceeds the range, an asynchronous processing flow (as described in Example 1, phase compensation mechanism) is triggered.

[0113] After executing all continuous feature paths, collaborative paths for discrete features are executed. The execution order of discrete feature paths is determined by the distribution density of their classification constraint parameters; paths with higher density (indicating stronger synchronization constraints) are executed first. For example, if the density of a discrete feature's synchronization constraint parameter is 0.6 in the interval [0.3, 0.6) and 0.4 in the interval [0.6, 1.0], then this path has a higher priority than discrete paths with a density distribution of [0.5, 0.5]. During discrete path execution, a certain phase difference between the two robots is allowed, for example, a preset phase difference of ±50ms. This allows for a delay or advance of no more than 50ms in the action execution time between the two robots to accommodate the adjustment needs of non-critical actions.

[0114] In the practical application of synchronization rules, the following details should be noted: the physical meaning of the synchronization trigger threshold must be strictly matched with the precision requirements of the yarn hanging process. For example, for yarn splicing actions with micron-level precision, the threshold may be set to 0.01mm, while for roving transport actions, the threshold can be relaxed to 10mm; the smoothness analysis of the speed curve should take into account the dynamic characteristics of the robot to avoid motor overload or trajectory deviation due to sudden speed changes; the phase difference tolerance of discrete feature paths should be verified by robot kinematic simulation to ensure that asynchronous execution will not cause mechanical interference or process failure.

[0115] The generation of synchronization trigger sequences needs to dynamically adapt to real-time operating conditions. For example, when two robots detect external disturbances (such as sudden changes in yarn tension) while executing a continuous path, causing the current synchronization trigger threshold to be unmet, the system needs to automatically pause sequence execution, trigger error compensation, and then re-evaluate the validity of the threshold. This dynamic adjustment mechanism ensures the robustness of synchronization rules in complex environments.

[0116] Example 4:

[0117] In the yarn-hanging method using collaborative control of two composite robots, validating the execution nodes is a crucial step in determining the collaborative state of the two robots. This embodiment uses the actual process of collaborative yarn hanging by two robots, Robot A and Robot B, as an example to describe in detail the specific implementation methods for validating the execution nodes and determining the asynchronous state.

[0118] Suppose that in a yarn loading task, two robots need to work together to transfer the yarn from the pay-off stand to the loom. The coordinated action sequence is divided into three stages: Robot A picks up the yarn from the pay-off stand (node ​​1), Robot B moves to a designated position to receive the yarn (node ​​2), and both move synchronously to the loom's yarn feed port to release the yarn (node ​​3). Each stage corresponds to an execution node, and the system needs to verify the validity of these three nodes sequentially.

[0119] First, data acquisition and coordinate system alignment transformation are performed. After node 1 is executed, the sensors acquire the end effector pose data of RobotA and RobotB in real time. The end effector pose data of RobotA is based on its base coordinate system (denoted as coordinate system A), with coordinate values ​​of (100mm, 200mm, 300mm, 0°, 0°, 0°) (representing X, Y, Z coordinates and roll, pitch, and yaw angles, respectively); the end effector pose data of RobotB is based on its base coordinate system (denoted as coordinate system B), with coordinate values ​​of (150mm, 250mm, 350mm, 0°, 0°, 0°). Since the bases of the two robots are located in different positions in the global coordinate system, their pose data need to be mapped to the same global coordinate system.

[0120] Coordinate system alignment transformation is achieved through homogeneous matrix transformation. First, the pose parameters of coordinate system B relative to the global coordinate system are measured: Assume the origin of coordinate system B is located at (50mm, 50mm, 0mm) in the global coordinate system, and its pose is consistent with the global coordinate system (i.e., the rotation matrix is ​​the identity matrix). Then, the pose parameters of coordinate system B are transformed using a homogeneous matrix. Transform the end effector pose of RobotB from coordinate system B to global coordinate system G. The transformed coordinates are:

[0121]

[0122] The orientation remains unchanged at (0°, 0°, 0°). At this time, the end poses of RobotA and RobotB in the global coordinate system are (100, 200, 300, 0°, 0°) and (200, 300, 350, 0°, 0°, 0°) respectively, completing coordinate system one.

[0123] Next, verification parameters are generated. These parameters include the dual-robot end-effector position deviation, attitude deviation, and time synchronization error. The position deviation is calculated as the absolute value of the difference between the X, Y, and Z coordinates in the global coordinate system, i.e.:

[0124]

[0125] Since both are 0°, the attitude deviation is 0. The time synchronization error is the time difference between the arrival of the two robots at node 1. Assuming Robot A arrives at t=1.0s and Robot B arrives at t=1.2s, the error is 0.2s.

[0126] Input the above verification parameters into the verification model. The verification model is a machine learning model (such as a support vector machine) trained based on historical synchronous data. The model outputs a matching degree value (range 0-100%), with a higher value indicating better collaboration. Assuming that the verification result of node 1 is a matching degree of 90%, and the verification threshold is set to 85%, since 90% is higher than the threshold, node 1 is determined to be in a synchronous execution state, and the two robots continue to execute the next node according to the preset phase difference.

[0127] During execution at node 2, assume RobotA needs to move to (150mm, 220mm, 320mm) and RobotB needs to move to (180mm, 280mm, 330mm). After collecting the actual pose data and performing coordinate system transformation, the actual coordinates of RobotA are (155mm, 225mm, 325mm) and the actual coordinates of RobotB are (190mm, 290mm, 340mm). Calculate the position deviation:

[0128]

[0129] The time synchronization error is 0.1s (Robot A arrives at t=2.5s, and Robot B arrives at t=2.6s). After inputting the validation model, the output matching degree is 82%, which is lower than the validation threshold of 85%. Therefore, it is determined that node 2 is in an asynchronous execution state, triggering the exception handling process.

[0130] After locating the abnormal execution node, the system first obtains the target trajectory points of the two robots that were not executed synchronously under that node. The target trajectory points of node 2 are RobotA (150mm, 220mm, 320mm) and RobotB (180mm, 280mm, 330mm), while the actual trajectory points are (155mm, 225mm, 325mm) and (190mm, 290mm, 340mm) collected above. When calculating the spatial offset, using the target point as the reference, the offset of RobotA is √[(155-150)²+(225-220)²+(325-320)²]=√(25+25+25)=√75≈8.66mm, and the offset of RobotB is √[(190-180)²+(290-280)²+(340-330)²]=√(100+100+100)=√300≈17.32mm. The larger value of 17.32mm is taken as the real-time phase offset.

[0131] Then, historical trajectory backtracking is performed to extract the historical phase offset of the predecessor node (node ​​1) of node 2. The offset of node 1 is already controlled within the allowable range under synchronization (e.g., the preset maximum allowable offset is 20mm), so the backtracking is performed once, and the maximum allowable offset is not exceeded, so the backtracking stops. At this time, the system records the real-time phase offset and the historical offset, providing a data basis for subsequent phase compensation.

[0132] Throughout the verification process, the following details should be noted: The accuracy of coordinate system alignment transformation depends on the initial calibration of the robot base. If there is an error in the calibration, it will lead to deviation in the verification parameters. Therefore, the base pose needs to be calibrated regularly using equipment such as laser trackers. The training data of the verification model needs to cover normal synchronous samples under different process scenarios to ensure that the model can accurately identify abnormal states. The calculation of real-time phase offset needs to be based on pose data with the same timestamp to avoid error amplification due to asynchronous sampling.

[0133] Example 5:

[0134] In the yarn-hanging method of dual-robot collaborative control, phase compensation for abnormal execution nodes in asynchronous execution states is a key step to ensure the restoration of collaboration between the two robots. This embodiment takes the asynchronous state of dual robots RobotA and RobotB in a yarn-hanging task as an example to describe in detail the specific implementation process of phase offset calculation and compensation.

[0135] Assume that during node 4 of a yarn-hanging task, Robot A and Robot B need to synchronously move the yarn to the designated release position. The preset execution time for node 4 is t=4.0s, and the target trajectory points for the two robots are Robot A (300mm, 400mm, 500mm) and Robot B (350mm, 450mm, 550mm), with a preset phase difference of 0ms (i.e., they need to arrive simultaneously). When the system determines that node 4 is in an asynchronous execution state (e.g., the verification matching degree is lower than the threshold), the phase offset calculation and compensation process is initiated.

[0136] First, locate the abnormal execution node and obtain the unsynchronized target trajectory points. Real-time acquisition of the actual pose data of RobotA and RobotB at t=4.0s is performed. After coordinate system alignment transformation, the actual coordinates of RobotA are (310mm, 415mm, 505mm), and the actual coordinates of RobotB are (365mm, 470mm, 560mm). The unsynchronized target trajectory points are the preset (300mm, 400mm, 500mm) and (350mm, 450mm, 550mm). The spatial offset between the actual points and the target points is calculated.

[0137] The offset of RobotA is the square root of the sum of the squares of the differences in the coordinates of each axis, that is, √[(310-300)²+(415-400)²+(505-500)²]=√(100+225+25)=√350≈18.71mm;

[0138] The offset of RobotB is √[(365-350)²+(470-450)²+(560-550)²]=√(225+400+100)=√725≈26.93mm. Taking the larger of the two values, 26.93mm, as the real-time phase offset, this value reflects the spatial deviation between the current execution state and the target state of the two robots.

[0139] Next, historical trajectory backtracking is performed to extract the historical phase offset of the preceding nodes. Assuming node 4's preceding node is node 3, and node 3 experienced an asynchronous state due to slight yarn entanglement during execution, the recorded historical phase offset was 15mm (not exceeding the maximum allowable offset of 20mm). Backtracking continues to node 2, with a historical offset of 8mm (all within the allowable range), until backtracking reaches node 1 (the initial node, with an offset of 0mm), at which point backtracking stops. At this point, the cumulative backtracking count is 3 (nodes 3, 2, and 1), which serves as the compensation priority parameter.

[0140] The compensation threshold is set to 3 times. When the compensation priority parameter (3 times) equals the compensation threshold, the phase compensation operation is triggered. The compensation strategy adjusts the trajectory parameters of the abnormal execution node based on the total phase offset. The total phase offset considers both the real-time offset and the historical offset. Here, the sum of the real-time offset of 26.93 mm and the historical maximum offset of 15 mm is taken (assuming the system uses an accumulation algorithm), resulting in a total offset of 41.93 mm. The system replans the execution path of node 4 based on this offset. The specific adjustment method is as follows:

[0141] Extend the execution time of RobotB to reduce its movement speed and decrease the displacement deviation per unit time;

[0142] The trajectory of RobotA is fine-tuned by adding intermediate transition points, so that the offset between the two gradually decreases in subsequent movements.

[0143] The adjusted execution node parameters include new pose coordinates and timestamps. For example, RobotB's target trajectory is adjusted to arrive in stages: first, it moves to (357.5mm, 457.5mm, 555mm) at t=4.2s, and then reaches the final target point (350mm, 450mm, 550mm) at t=4.4s, compensating for the phase difference through segmented deceleration. RobotA maintains its original speed, but pauses for 0.2s at t=4.2s, waiting for RobotB to adjust, ensuring that both arrive at the adjusted target point synchronously at t=4.4s.

[0144] During the compensation process, the system continuously monitors the phase offset. If the offset exceeds the allowable range (e.g., 20mm) again after compensation, the backtracking count increases to 4 times. If it exceeds the compensation threshold by 3 times, a deeper compensation is triggered: the inverse kinematics solution of the two robots is recalculated, joint angle parameters are adjusted, and new execution nodes are generated. For example, a joint angle of RobotB is adjusted from 60° to 65° to change the curvature of the end effector trajectory, shorten the actual path length, and speed up the arrival at the target point.

[0145] Homogeneous matrix transformations are used throughout the coordinate system alignment process. For example, after each pose data acquisition, the data of RobotB is transformed into the base coordinate system of RobotA using a pre-calibrated robot base relative pose matrix, ensuring that the offset calculation is based on the same reference datum. If errors are found in the base calibration parameters during the compensation process (such as base displacement due to mechanical vibration), the system will automatically trigger a recalibration process, measuring the real-time distance between the bases using a laser rangefinder and updating the transformation matrix parameters to avoid compensation failure due to reference deviation.

[0146] In practical applications, the following details should be noted: The setting of the compensation threshold needs to balance the system response speed and stability. If the threshold is too low, it may lead to frequent compensation and affect efficiency, while if it is too high, it may lead to error accumulation. The trajectory parameter adjustment should avoid exceeding the robot's physical limits (such as maximum speed and acceleration) to prevent mechanical overload. The time interval of the segmented compensation strategy should be determined according to the robot's motion control cycle, and is usually set to an integer multiple of the control cycle (such as 100ms) to ensure the synchronization of command execution.

[0147] This embodiment demonstrates a complete phase compensation process, from real-time offset calculation and historical trajectory backtracking to trajectory parameter adjustment, through a specific asynchronous node compensation case. By dynamically matching priority parameters with compensation thresholds, the system can flexibly select compensation strategies based on the degree of error. Combined with precise calibration of the coordinate system reference, this ensures that the two robots can quickly restore their collaborative state under complex working conditions, guaranteeing the continuity and accuracy requirements of the yarn-hanging task.

[0148] Example 6:

[0149] In a specific implementation scenario, two composite robots need to work together to complete the task of hanging yarn. Both composite robots consist of an AGV mobile platform and a collaborative robotic arm, and are equipped with a laser SLAM navigation system, vision sensors, and force sensors.

[0150] First, motion parameters are acquired and preprocessed. A laser SLAM navigation system is used to collect the AGV's position and attitude information in real time. Simultaneously, an encoder is used to acquire the angle data of each joint of the collaborative robot arm, while the trajectory of the end effector is collected by both vision and force sensors. This data is stored in separate datasets, with each dataset assigned a unique action identifier, such as "AGV moves to point A" or "robot arm grasps yarn." Next, the data is preprocessed to remove outliers; for example, joint angles exceeding physical limits are identified as outliers and removed. Finally, the processed data undergoes kinematic decomposition, unifying the data into a collaborative path reference in meters and radians.

[0151] Then, the action features of the collaborative path baseline are identified, and synchronization rules are determined. Key points on the trajectory are extracted, such as the start, end, and turning points of the path, and an action node sequence is generated according to the execution order. The spatial distance between adjacent action nodes is calculated and used as a synchronization constraint parameter. An allowable deviation range is set, and the synchronization constraint parameter is normalized to form multiple synchronization intervals. The distribution density of the synchronization constraint parameter within each synchronization interval is statistically analyzed. If the distribution density of an interval is high, it is marked as a continuous feature; otherwise, it is marked as a discrete feature. The synchronization priority of continuous features is higher than that of discrete features. Based on this, the minimum value of the synchronization interval corresponding to the continuous feature is used as the synchronization trigger threshold. These thresholds are arranged in ascending order to generate a synchronization trigger sequence. When executing the collaborative path, the path corresponding to the continuous feature is executed first, followed by the path corresponding to the discrete feature.

[0152] Next, collaborative path planning and execution node verification are performed. Based on the synchronous trigger sequence, multiple sets of collaborative action sequences and execution nodes for each sequence are planned. During execution, the actual pose data of the two robots are collected in real time. Homogeneous matrix transformation is used to map this data to the same global coordinate system, eliminating the relative pose deviation between the two robots' base coordinate systems. The processed data is input into the verification model, and the matching degree is used as the verification result. If the verification result is higher than the set verification threshold, the execution node is in synchronous execution mode; if it is lower than or equal to the verification threshold, it is in asynchronous execution mode.

[0153] When asynchronous execution occurs, the phase offset is calculated and compensated. The non-synchronously executed target trajectory point under the abnormal execution node is obtained, and the spatial offset between this point and the current trajectory point is calculated to obtain the real-time phase offset. Historical trajectory backtracking is performed on the predecessor node of the abnormal execution node to extract the historical phase offset. Backtracking stops when the historical phase offset is less than or equal to the maximum allowable offset. The real-time phase offset and historical phase offset are input into the compensation algorithm to obtain the total phase offset. The compensation priority parameter is determined based on the number of backtracking attempts of the abnormal execution node. When this parameter exceeds the compensation threshold, the trajectory parameters of the abnormal execution node are adjusted according to the total phase offset to generate an updated execution node; when the compensation threshold is not exceeded, the phase offset is continuously monitored until it is reached before compensation is triggered. Through these steps, the dual composite robots can efficiently and stably collaborate to complete the yarn-hanging task.

[0154] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0155] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for hanging a yarn by dual-composite robot cooperative control, characterized by: The method comprises the following steps: acquiring motion parameters of a dual-combined robot and preprocessing the motion parameters to generate a cooperative path reference; identifying motion features of each cooperative path reference and determining synchronization rules of the dual robot based on the motion features; planning a cooperative path for the dual robot according to the synchronization rules to form a plurality of cooperative action sequences and execution nodes of each cooperative action sequence; validating each execution node to obtain a cooperative state of each execution node, wherein the cooperative state comprises a synchronous execution state and an asynchronous execution state; in the synchronous execution state, the cooperative action sequences of the dual robot are executed with a preset phase difference; in the asynchronous execution state, an abnormal execution node is located, and a phase offset of the cooperative action sequence under the abnormal execution node is calculated; based on the phase offset, phase compensation is performed on the abnormal execution node until the cooperative action sequences of the dual robot restore the preset phase difference and the phase compensation is terminated; the step of acquiring motion parameters of a dual-combined robot and preprocessing the motion parameters to generate a cooperative path reference comprises the following steps: collecting pose data, joint angles and end trajectory of the dual robot through a sensor, and storing the data as independent data sets respectively, and adding a unique action identifier to each data set; performing outlier rejection on the data in each independent data set to filter out data items exceeding physical constraints; kinematically decomposing the processed independent data sets to extract translation components and rotation components, and outputting the component parameters in the same unit as the cooperative path reference; the step of identifying motion features of each cooperative path reference comprises the following steps: extracting trajectory key points under each cooperative path reference, and generating an action node sequence according to the execution time sequence; calculating the spatial distance between adjacent action node sequences and labeling it as a synchronization constraint parameter; setting a tolerance range, normalizing the synchronization constraint parameter according to the tolerance range to generate a plurality of synchronization intervals; statistically analyzing the distribution density of the synchronization constraint parameter in each synchronization interval and labeling it as a classification constraint parameter, and determining the motion features of the cooperative path reference based on the classification constraint parameter; wherein the motion features include continuous features and discrete features, and the synchronization priority of the continuous features is higher than that of the discrete features.

2. The method of claim 1, wherein the double compound robot cooperative control is characterized in that: the step of determining the motion features of the cooperative path reference based on the classification constraint parameter comprises the following steps: obtaining the classification constraint parameter corresponding to each cooperative path reference; sorting the classification constraint parameters under the same cooperative path reference according to the density interval from high to low, and calculating the coverage ratio of the maximum density interval; setting a determination threshold and comparing the coverage ratio with the determination threshold; if the coverage ratio is higher than the determination threshold, it is determined that the cooperative path reference has continuous features; if the coverage ratio is lower than or equal to the determination threshold, it is determined that the cooperative path reference has discrete features.

3. The method of claim 1, wherein the method further comprises: determining a position of the first robot and the second robot; and determining a position of the first yarn and the second yarn. the step of determining the synchronization rules of the dual robot based on the motion features comprises the following steps: obtaining the motion features of each cooperative path reference; taking the minimum value of the synchronization interval corresponding to the continuous features as a synchronization trigger threshold; Extract the synchronization trigger threshold under all continuous features, and arrange them in ascending order to generate a synchronization trigger sequence; According to the synchronization trigger sequence, the dual-robot is matched in action, and after the cooperative path under the continuous feature is executed, the cooperative path under the discrete feature is supplemented.

4. The method of claim 1, wherein the dual-robot cooperative control of the yarn hanging is characterized by: The validity verification of each execution node includes: Collect the actual pose data of the dual-robot under each execution node, and perform coordinate system alignment conversion to generate multiple verification parameters; Call the verification model, input the verification parameters into the model, and mark the matching degree output by the model as the verification result; Set a verification threshold, and compare the verification result with the verification threshold; If the verification result is higher than the verification threshold, it is determined that the execution node is in a synchronous execution state; If the verification result is lower than or equal to the verification threshold, it is determined that the execution node is in an asynchronous execution state.

5. The method of claim 1, wherein the dual-robot cooperative control of the yarn hanging is characterized by: The calculation of the phase offset of the cooperative action sequence under the abnormal execution node includes: Obtain the target trajectory point of the dual-robot that is not synchronized under the abnormal execution node; Calculate the spatial offset of the target trajectory point and the current trajectory point, and mark it as the real-time phase offset; Backtrack the history trajectory of the abnormal execution node to extract the historical phase offset of its predecessor node; Set a maximum allowable offset, and stop backtracking when the historical phase offset is less than or equal to the maximum allowable offset; Call the compensation algorithm, input the real-time phase offset and historical phase offset into the algorithm, and mark the algorithm output result as the total phase offset.

6. The method of claim 5, wherein the method further comprises: The phase compensation of the abnormal execution node based on the phase offset includes: Obtain the number of backtracking of the abnormal execution node, and mark it as a compensation priority parameter; Set a compensation threshold, and compare the compensation priority parameter with the compensation threshold; When the compensation priority parameter exceeds the compensation threshold, adjust the trajectory parameters of the abnormal execution node according to the total phase offset to generate an updated execution node; When the compensation priority parameter does not exceed the compensation threshold, continuously monitor the phase offset until the priority parameter reaches the compensation threshold to trigger compensation.

7. The method of claim 3, wherein the method further comprises: determining a position of the first robot and the second robot; and determining a position of the first yarn and the second yarn. The generation step of the synchronization trigger sequence further includes: Extract the robot end speed curve corresponding to each synchronization trigger threshold; According to the smoothness of the speed curve, the synchronization trigger threshold is sorted again, and the synchronization trigger threshold with a speed change rate lower than a set value is preferentially selected.

8. The method of claim 4, wherein the method further comprises: determining a position of the first robot and the second robot; and determining a position of the first yarn and the second yarn. The steps of coordinate system alignment conversion include: Map the pose data of the dual-robot end effector to the same global coordinate system; Eliminate the relative pose deviation of the dual-robot base coordinate system through homogeneous matrix transformation, and unify it to the same reference datum.

Images