Robot training action quality evaluation method and system driven by time-series sensing data

Abstract

The application discloses a time sequence sensing data driven robot training action quality evaluation method and system, relates to the technical field of time sequence data processing, and comprises the following steps: acquiring multi-source time sequence sensing data in a robot training process, performing pretreatment, and obtaining a standardized time sequence data matrix; based on the standardized time sequence data matrix, extracting an inter-frame continuity index, a motion specification index and a trajectory tracking accuracy index; performing normalization processing on the indexes respectively, performing weighted summation on the normalized indexes, taking a natural constant as a base, and obtaining a comprehensive quality score of a single training action; when the comprehensive quality score is lower than a preset threshold, calculating the angle deviation contribution degrees of all joints, determining the joints with the contribution degrees exceeding the average value as abnormal joints, and outputting training improvement suggestions. The application realizes multidimensional evaluation of the quality of robot training actions, and solves the problem that traditional methods rely on artificial subjective judgment.

Description

Technical Field

[0001] This invention relates to the field of time-series data processing technology, specifically to a method and system for evaluating the quality of robot training actions driven by time-series sensor data. Background Technology

[0002] With the rapid development of embodied intelligence and humanoid robot technology, automated evaluation of robot training motion quality has become a key bottleneck restricting skill improvement. Current technologies primarily rely on the following methods to evaluate robot training motion quality: human observation and subjective judgment, which are highly influenced by the evaluator's experience level, suffer from poor consistency, and cannot achieve large-scale parallel evaluation; simple single-dimensional threshold comparison methods, such as only detecting whether joint angles exceed safety limits, can only determine pass / fail; while a few motion capture-based evaluation methods can acquire trajectory data from the end effector, they fail to fully utilize multi-source temporal sensor data and lack a multi-dimensional temporal evaluation framework that comprehensively considers motion continuity, standardization, and trajectory accuracy. Furthermore, existing methods struggle to accurately pinpoint specific abnormal joints and time periods causing quality defects when scores are low, failing to provide targeted improvement guidance for trainers and hindering further improvements in robot training efficiency.

[0003] Therefore, there is an urgent need for a method and system that can fully utilize multi-source time-series sensor data, realize automated multi-dimensional evaluation of robot training action quality, and accurately locate the root cause of anomalies. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing a method and system for evaluating the quality of robot training actions driven by time-series sensing data.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for evaluating the quality of robot training actions driven by time-series sensor data includes the following steps: Step S1: Obtain multi-source time-series sensor data during robot training, perform timestamp alignment, filtering preprocessing, and Z-score standardization on the multi-source time-series sensor data to obtain a standardized time-series data matrix; Step S2: Based on the standardized time-series data matrix, extract the inter-frame continuity index, motion regularity index, and trajectory tracking accuracy index; Step S3: Normalize the inter-frame continuity index, the motion regularity index, and the trajectory tracking accuracy index respectively. Then, perform a weighted summation on the three normalized indices and take a negative exponent with the natural constant as the base of the weighted summation result to obtain the comprehensive quality score of a single training action. Step S4: When the overall quality score is lower than the preset threshold, calculate the contribution of the angle deviation of each joint, identify the joints whose contribution exceeds the average contribution of all joints as abnormal joints, and output training improvement suggestions.

[0006] Furthermore, in step S1, the multi-source timing sensing data includes joint angle timing data, joint angular velocity timing data, end effector pose timing data, and motor current timing data.

[0007] Furthermore, in step S2, the specific calculation method for the inter-frame continuity index includes: Based on the joint angular velocity time-series data of each joint, the square of the angular velocity difference between adjacent time frames of each joint is calculated, and the average of the square of the angular velocity differences of all joints and all adjacent frames is calculated to obtain the inter-frame continuity index.

[0008] Furthermore, in step S2, the specific calculation method for the motion standardization index includes: For each joint's joint angle in each time frame, determine whether the joint angle exceeds the angle range formed by the preset minimum and maximum allowable angles of the joint. If it exceeds, calculate the square of the ratio of the excess amount to the width of the angle range; otherwise, take zero to obtain the angle over-limit penalty value. For the joint angular velocity of each joint in each time frame, determine whether the joint angular velocity exceeds the angular velocity range formed by the preset minimum allowable angular velocity and the maximum allowable angular velocity of the joint. If it exceeds, calculate the square of the ratio of the excess amount to the width of the angular velocity range; otherwise, take zero and obtain the angular velocity over-limit penalty value. The motion standardization index is obtained by adding the angle and angular velocity over-limit penalties for the same joint in the same time frame, and then averaging the values ​​for all joints and all time frames.

[0009] Furthermore, in step S2, the specific calculation method for the trajectory tracking accuracy index includes: Based on the end effector pose timing data, the square of the Euclidean distance between the actual position and the expected position of the end effector in each time frame is calculated as the square of the position deviation. Calculate the angular deviation between the actual attitude rotation matrix and the desired attitude rotation matrix of the actuator at the end of each time frame; The trajectory tracking accuracy index is obtained by multiplying the squared position deviation and angle deviation of the same time frame by the position weight coefficient and attitude weight coefficient, respectively, and then summing them. The average value is then calculated over all time frames.

[0010] Furthermore, in step S4, the specific calculation method for the contribution of the angular deviation of each joint includes: For each joint, the mean of the squares of the differences between the actual angle and the expected angle of the joint across all time frames is calculated as the angle deviation term; Calculate the average value of the angle over-limit penalty for the joint across all time frames, multiply it by a preset normalized penalty weight coefficient, and use it as the over-limit penalty item; The contribution of joint angle deviation is obtained by adding the angle deviation term to the over-limit penalty term.

[0011] Furthermore, in step S4, determining joints whose contribution exceeds the average contribution of all joints as abnormal joints specifically includes: Calculate the arithmetic mean of the angular deviation contributions of all joints; The contribution of each joint's angle deviation is compared with the arithmetic mean, and joints whose contribution of angle deviation is greater than the arithmetic mean are identified as abnormal joints.

[0012] A time-series sensor data-driven robot training motion quality assessment system, used to implement any one of the time-series sensor data-driven robot training motion quality assessment methods, including: The data acquisition module is used to acquire multi-source time-series sensor data during robot training. The data preprocessing module is used to perform timestamp alignment, filtering preprocessing, and Z-score normalization on multi-source time-series sensor data to obtain a normalized time-series data matrix. The index extraction module is used to extract inter-frame continuity indexes, motion regularity indexes, and trajectory tracking accuracy indexes based on a standardized time-series data matrix. The quality scoring module is used to normalize the inter-frame continuity index, the motion regularity index, and the trajectory tracking accuracy index respectively, perform a weighted sum on the three normalized indices, and then take a negative exponent with the natural constant as the base of the weighted sum to obtain the comprehensive quality score of a single training action. The anomaly diagnosis module is used to calculate the contribution of angle deviation of each joint when the overall quality score is lower than a preset threshold. Joints whose contribution exceeds the average contribution of all joints are identified as abnormal joints, and training improvement suggestions are output.

[0013] Furthermore, the indicator extraction module specifically includes: The continuity extraction unit is used to calculate the squared mean of the angular velocity difference between adjacent frames based on the angular velocity time-series data of each joint, so as to obtain the inter-frame continuity index. The normative extraction unit is used to calculate the average of the secondary over-limit penalty based on whether the angle and angular velocity of each joint exceed the preset allowable range, and obtain the motion normative index. The accuracy extraction unit is used to calculate the weighted average value based on the position deviation and attitude deviation between the actual pose and the expected pose of the end effector, so as to obtain the trajectory tracking accuracy index.

[0014] Furthermore, the anomaly diagnosis module specifically includes: The contribution calculation unit is used to calculate the sum of the squared mean of the angle deviation and the weighted average of the over-limit penalty for each joint when the overall quality score is lower than the preset threshold, and use it as the angle deviation contribution of the joint. An anomaly determination unit is used to calculate the arithmetic mean of the contribution of all joint angle deviations, and to determine joints whose angle deviation contribution is greater than the arithmetic mean as abnormal joints. The suggested output unit is used to output the deviation curve between the actual angle and the expected angle of each abnormal joint during the abnormal time period, as well as the information on the time period exceeding the allowable range, based on the identified abnormal joints, as a training improvement suggestion.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By extracting inter-frame continuity indicators, this invention can effectively quantify the smoothness of robot actions during execution, identify subtle abnormalities such as jitter, pauses, or discontinuities in the actions, and overcome the coarse-grained defect of traditional methods that can only judge whether an action is qualified or unqualified.

[0016] 2. This invention constructs a motion standardization index by introducing a secondary limit violation penalty mechanism, and applies nonlinear penalties to the limit violation behavior of each joint angle and angular velocity. The more severe the limit violation, the more squarely the penalty increases. Compared with the traditional hard threshold method, it can more accurately characterize the standardization of the movement.

[0017] 3. When the score is lower than the preset threshold, the present invention calculates the contribution of the angle deviation of each joint and judges the joints with a contribution higher than the average value as abnormal joints. It adopts an adaptive statistical standard higher than the average value instead of a fixed threshold, which can automatically adapt to different action types and individual differences of different robots, avoiding misjudgment or omission caused by subjectively setting thresholds. Attached Figure Description

[0018] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating an embodiment of the present invention; Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention; Figure 3 This is a flowchart of the comprehensive scoring and anomaly diagnosis process according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] like Figure 1 As shown, the method for evaluating the quality of robot training actions driven by time-series sensor data includes the following steps: Step S1: Obtain multi-source time-series sensor data during robot training, perform timestamp alignment, filtering preprocessing, and Z-score standardization on the multi-source time-series sensor data to obtain a standardized time-series data matrix; Step S2: Based on the standardized time-series data matrix, extract the inter-frame continuity index, motion regularity index, and trajectory tracking accuracy index; Step S3: Normalize the inter-frame continuity index, the motion regularity index, and the trajectory tracking accuracy index respectively. Then, perform a weighted summation on the three normalized indices and take a negative exponent with the natural constant as the base of the weighted summation result to obtain the comprehensive quality score of a single training action. Step S4: When the overall quality score is lower than the preset threshold, calculate the contribution of the angle deviation of each joint, identify the joints whose contribution exceeds the average contribution of all joints as abnormal joints, and output training improvement suggestions.

[0021] The multi-source timing sensing data specifically includes: the angle values ​​of each joint of the robot in each time frame, i.e., joint angle timing data; the angular velocity values ​​of each joint of the robot in each time frame, i.e., joint angular velocity timing data; the three-dimensional position coordinates and attitude rotation matrix of the end effector in each time frame, i.e., end effector pose timing data; and the current values ​​of each drive motor in each time frame, i.e., motor current timing data. The multi-source sensing data reflects the kinematic and dynamic state of the robot when performing actions from different perspectives. Since the sampling frequency and start time of different sensors may differ, the timestamp alignment adopts the linear interpolation method to unify all sensor data to the same time baseline, ensuring that the data of each channel corresponds one-to-one in time. The filtering adopts the sliding window mean filtering method to eliminate high-frequency noise interference and retain the main motion characteristics during the execution of the action. After preprocessing, a standardized time series data matrix is ​​obtained. To eliminate the impact of dimensional differences between sensor channels on subsequent evaluation, the time-series data of each sensor channel were Z-score standardized, eliminating the influence of the original dimensions and making the data from different sensor channels comparable.

[0022] In step S2, the specific calculation method for the inter-frame continuity index includes: Based on the joint angular velocity time-series data of each joint, the square of the angular velocity difference between adjacent time frames of each joint is calculated, and the average of the square of the angular velocity differences of all joints and all adjacent frames is calculated to obtain the inter-frame continuity index.

[0023] The inter-frame continuity index is used to quantify the smoothness of motion transitions between adjacent time frames during robot execution. In ideal smooth motion, joint angular velocities should not undergo drastic changes. The smaller the change in angular velocity between adjacent frames, the better the continuity of the motion. This index evaluates the quality of motion execution from the perspective of motion smoothness and can effectively identify abnormal phenomena such as jitter, pauses, or discontinuities in the motion. The specific calculation formula is as follows: in, This represents the inter-frame continuity index; the smaller the value, the smoother the temporal transition of the action, i.e., the better the continuity. This represents the total number of joints in the robot. Indicates a joint index. Indicates the total number of time frames. This represents the time frame index. Since the calculation is based on the difference between adjacent frames, the traversal starts from the second frame. This represents the angular velocity of the j-th joint in frame t. Let represent the angular velocity of the j-th joint in frame t-1. During the calculation, all physical quantities are taken as their corresponding numerical values, i.e., they are dimensionless.

[0024] In step S2, the specific calculation method for the motion conformity index includes: For each joint's joint angle in each time frame, determine whether the joint angle exceeds the angle range formed by the preset minimum and maximum allowable angles of the joint. If it exceeds, calculate the square of the ratio of the excess amount to the width of the angle range; otherwise, take zero to obtain the angle over-limit penalty value. For the joint angular velocity of each joint in each time frame, determine whether the joint angular velocity exceeds the angular velocity range formed by the preset minimum allowable angular velocity and the maximum allowable angular velocity of the joint. If it exceeds, calculate the square of the ratio of the excess amount to the width of the angular velocity range; otherwise, take zero and obtain the angular velocity over-limit penalty value. The motion standardization index is obtained by adding the angle and angular velocity over-limit penalties for the same joint in the same time frame, and then averaging the values ​​for all joints and all time frames.

[0025] Motion standardization indicators are used to quantify whether the angles and angular velocities of each joint during motion execution are within a preset safe and reasonable range. Each joint of the robot has physical limits and speed restrictions; exceeding these limits not only affects motion quality but may also cause equipment damage or safety accidents. By introducing a limit-crossing penalty mechanism, the degree of standardization in motion execution is comprehensively evaluated. The specific formula is as follows: in, This indicates the standardization of movement; the smaller the value, the more standardized the movement, and the more within the permissible range of joint movement. This represents the angle value of the j-th joint in frame t. and Let represent the preset minimum and maximum allowable angles of the j-th joint, respectively. This represents the angular velocity value of the j-th joint in frame t. and These represent the preset minimum and maximum permissible angular velocities of the j-th joint, respectively. Let represent the quadratic over-limit penalty function, which quantifies the degree of over-limit violation as a non-negative penalty value. The specific formula is as follows: in, This refers to the physical quantity to be checked, namely angle or angular velocity. and These represent the preset lower and upper limits of the physical quantity, respectively. When the quantity is within the closed interval of the lower and upper limits, the function value is 0, indicating no penalty. When the quantity exceeds the interval, the function value increases with the square of the ratio of the excess quantity to the interval width. The squaring operation amplifies the impact of the degree of exceeding the limit, so that the penalty increases non-linearly when the excess is more severe. Dimensionless processing is also performed in the calculation.

[0026] In step S2, the specific calculation method for the trajectory tracking accuracy index includes: Based on the end effector pose timing data, the square of the Euclidean distance between the actual position and the expected position of the end effector in each time frame is calculated as the square of the position deviation. Calculate the angular deviation between the actual attitude rotation matrix and the desired attitude rotation matrix of the actuator at the end of each time frame; The trajectory tracking accuracy index is obtained by multiplying the squared position deviation and angle deviation of the same time frame by the position weight coefficient and attitude weight coefficient, respectively, and then summing them. The average value is then calculated over all time frames.

[0027] The trajectory tracking accuracy index is used to quantify the deviation between the actual trajectory of the end effector and the desired trajectory. Robot training actions are typically evaluated based on whether the end effector can accurately reproduce the desired pose. Position deviation and attitude deviation together reflect the accuracy level of trajectory tracking. The specific calculation formula is as follows: in, This represents the trajectory tracking accuracy index. The smaller the value, the closer the actual trajectory of the end effector is to the expected trajectory, and the higher the tracking accuracy. This represents the three-dimensional position vector of the end effector during actual motion in frame t. This represents the expected position vector of the actuator at the end of frame t. This represents the square of the Euclidean distance. The weighting coefficient represents the positional deviation. Let represent the rotation matrix indicating the actual attitude of the end effector in frame t. Each column of the matrix represents the direction vector of each axis of the end effector's local coordinate system in the world coordinate system. Let the rotation matrix represent the desired orientation of the end effector in frame t. This represents the angular deviation between two rotation matrices. The weighting coefficients representing attitude deviations. and All are dimensionless positive numbers, and their relative magnitudes reflect the emphasis placed on positional and attitude accuracy in the evaluation. They can be appropriately increased during spraying operations. The size can be appropriately increased during handling operations. The calculation also involves dimensionless processing.

[0028] After obtaining the inter-frame continuity index, motion regularity index, and trajectory tracking accuracy index respectively, it is necessary to integrate these three indices with different dimensions and magnitudes into a unified comprehensive quality score to intuitively reflect the overall quality level of a single training action. Since all three indices are of the "smaller is better" type, when constructing the comprehensive score, this "smaller is better" relationship needs to be mapped to a positive evaluation relationship where a higher score represents better quality. Divide each of the three indicators by its corresponding historical standard deviation to achieve dimensionless normalization, making indicators of different scales comparable. Then, perform a weighted sum on the three normalized indicators. Finally, take a negative exponent with the natural constant as the base of the weighted sum to monotonically map the degree of comprehensive deviation to the (0, 1] interval, obtaining a positive comprehensive quality score. The specific formula is as follows: in, This represents the overall quality score of a single training exercise. A higher value indicates a higher overall quality of the exercise, and a value closer to 1 indicates a more ideal performance. denoted by , respectively, represent the sample standard deviations of the inter-frame continuity metric, motion regularity metric, and trajectory tracking accuracy metric on the historical training dataset. This maps three metrics with different dimensions to a comparable dimensionless scale. A smaller denominator indicates less fluctuation in the metric within historical data, and the normalized value of the current bias will be amplified, meaning it will be more sensitive. These represent the weight coefficients corresponding to the inter-frame continuity index, motion regularity index, and trajectory tracking accuracy index, respectively. They are all positive numbers and their sum is 1. They can be set according to actual application requirements. For example, the trajectory accuracy weight can be appropriately increased in precision assembly tasks, and the continuity weight can be appropriately increased in high-speed motion tasks.

[0029] In step S4, the specific calculation method for the contribution of the angle deviation of each joint includes: For each joint, the mean of the squares of the differences between the actual angle and the expected angle of the joint across all time frames is calculated as the angle deviation term; Calculate the average value of the angle over-limit penalty for the joint across all time frames, multiply it by a preset normalized penalty weight coefficient, and use it as the over-limit penalty item; The contribution of joint angle deviation is obtained by adding the angle deviation term to the over-limit penalty term.

[0030] In step S4, determining joints whose contribution exceeds the average contribution of all joints as abnormal joints specifically includes: Calculate the arithmetic mean of the angular deviation contributions of all joints; The contribution of each joint's angle deviation is compared with the arithmetic mean, and joints whose contribution of angle deviation is greater than the arithmetic mean are identified as abnormal joints.

[0031] When the overall quality score is lower than a preset threshold such as 0.7, the specific value of the preset threshold can be adjusted according to the actual application scenario. It is determined that the quality of the current single training action is substandard. By calculating the contribution of each joint to the overall angle deviation, abnormal joints with significantly higher contribution are identified. The contribution is calculated by considering two factors: 1. The deviation between the actual and expected angles of the joint; a larger deviation indicates a greater tracking error for the joint. 2. The degree to which the joint angle exceeds the allowable range; a more severe deviation indicates poorer motion regularity of the joint. The specific calculation formula is as follows: in, This represents the contribution of the angle deviation of the j-th joint. The larger the value, the greater the contribution of that joint to the motion quality defect. This represents the expected angle value of the j-th joint in frame t, obtained from the standard motion template. This represents the quadratic over-limit penalty function. This represents the weighting coefficient of normative penalties. The larger the value, the higher the weight of the unauthorized behavior in the contribution determination. In practical applications, it can be set according to the importance attached to security, and is a dimensionless positive number. After obtaining the deviation contribution of all joints, the arithmetic mean is calculated. The contribution of each joint is compared with the arithmetic mean. If the contribution of a certain joint is greater than the arithmetic mean, the joint is determined to be an abnormal joint. The specific formula is as follows: in, This represents a set of abnormal joint indices. This represents the arithmetic mean of the deviation contribution of all joints. Joints with a contribution higher than the average level are considered as the main abnormal joints causing motion quality defects. Using a value higher than the mean as the judgment standard instead of a fixed threshold, it can automatically adapt to different motion types and individual differences of different robots, avoiding misjudgment or omission caused by subjectively setting a fixed threshold.

[0032] Based on the identified abnormal joints, targeted training improvement suggestions are output. For each abnormal joint, the deviation curve between the actual angle and the expected angle of the joint in the continuous time frame interval with significantly higher deviation during the abnormal period is output, as well as the information on the time period exceeding the allowable range. This provides trainers with precise feedback guidance to help them focus on the performance of these joints in specific time periods during subsequent training. Based on the obtained comprehensive quality score and the identified abnormal joint set, a quality assessment report for a single training exercise is generated. This report includes the comprehensive quality score of the training exercise, detailed scores for each dimension indicator, the conclusion of whether the standard is met, a list of abnormal joints and their deviation contribution, as well as a visualized deviation curve and over-limit time annotation for each abnormal joint. This provides trainees with systematic quality feedback and supports targeted improvements in subsequent training.

[0033] like Figure 2 As shown, a time-series sensor data-driven robot training motion quality assessment system is used to implement any one of the time-series sensor data-driven robot training motion quality assessment methods, including: The data acquisition module is used to acquire multi-source time-series sensor data during robot training. The data preprocessing module is used to perform timestamp alignment, filtering preprocessing, and Z-score normalization on multi-source time-series sensor data to obtain a normalized time-series data matrix. The index extraction module is used to extract inter-frame continuity indexes, motion regularity indexes, and trajectory tracking accuracy indexes based on a standardized time-series data matrix. The quality scoring module is used to normalize the inter-frame continuity index, the motion regularity index, and the trajectory tracking accuracy index respectively, perform a weighted sum on the three normalized indices, and then take a negative exponent with the natural constant as the base of the weighted sum to obtain the comprehensive quality score of a single training action. The anomaly diagnosis module is used to calculate the contribution of angle deviation of each joint when the overall quality score is lower than a preset threshold. Joints whose contribution exceeds the average contribution of all joints are identified as abnormal joints, and training improvement suggestions are output.

[0034] like Figure 3 The diagram shows a detailed flowchart of the comprehensive quality scoring and anomaly diagnosis of this invention. First, the inter-frame continuity index, motion regularity index, and trajectory tracking accuracy index are input into the quality scoring module. After normalization, they are weighted and summed. Then, a negative exponent with the natural constant as the base is taken to obtain the comprehensive quality score. ;judge Is it below the preset threshold? If the action quality is ≥ a preset threshold, the overall quality score is directly output. <A preset threshold is set to trigger the anomaly diagnosis module, which calculates the contribution of angle deviation of each joint, identifies joints whose contribution exceeds the average contribution of all joints as abnormal joints, and finally outputs training improvement suggestions.

[0035] The indicator extraction module specifically includes: The continuity extraction unit is used to calculate the squared mean of the angular velocity difference between adjacent frames based on the angular velocity time-series data of each joint, so as to obtain the inter-frame continuity index. The normative extraction unit is used to calculate the average of the secondary over-limit penalty based on whether the angle and angular velocity of each joint exceed the preset allowable range, and obtain the motion normative index. The accuracy extraction unit is used to calculate the weighted average value based on the position deviation and attitude deviation between the actual pose and the expected pose of the end effector, so as to obtain the trajectory tracking accuracy index.

[0036] The anomaly diagnosis module specifically includes: The contribution calculation unit is used to calculate the sum of the squared mean of the angle deviation and the weighted average of the over-limit penalty for each joint when the overall quality score is lower than the preset threshold, and use it as the angle deviation contribution of the joint. An anomaly determination unit is used to calculate the arithmetic mean of the contribution of all joint angle deviations, and to determine joints whose angle deviation contribution is greater than the arithmetic mean as abnormal joints. The suggested output unit is used to output the deviation curve between the actual angle and the expected angle of each abnormal joint during the abnormal time period, as well as the information on the time period exceeding the allowable range, based on the identified abnormal joints, as a training improvement suggestion.

[0037] Based on the deviation curve and information on periods exceeding limits, suggestions for parameter adjustment directions can also be provided, including: If the joint continues to deviate positively during a specific period, the desired angle command of the joint during that period can be appropriately reduced or the feedforward gain can be adjusted. If the deviation curve shows a sudden spike at the beginning of the period, the angular acceleration command of the joint during that period can be reduced and the acceleration transition time extended. If the deviation curve fluctuates periodically, the joint transmission system can be checked for hysteresis or friction abnormalities, or the PID control parameters can be adjusted. If the end effector position deviation continues to increase, the tool center point can be recalibrated or the linkage parameters can be checked.

[0038] If the joint angle exceeds the upper limit, the desired angle setting value for that joint during that period can be reduced; if the joint angle exceeds the lower limit, the desired angle setting value for that joint during that period can be increased; if the joint angular velocity exceeds the upper limit, the motion speed command for that joint during that period can be reduced; if both the angle and angular velocity exceed the limits, the angular velocity limit can be reduced first.

[0039] The examples described herein are merely preferred embodiments of the invention and are not intended to limit the concept and scope of the invention. Any modifications and improvements made by those skilled in the art to the technical solutions of the invention without departing from the design concept of the invention should fall within the protection scope of the invention.

Claims

1. A method for evaluating the quality of robot training actions driven by time-series sensor data, characterized in that, Includes the following steps: Step S1: Obtain multi-source time-series sensor data during robot training, perform timestamp alignment, filtering preprocessing, and Z-score standardization on the multi-source time-series sensor data to obtain a standardized time-series data matrix; Step S2: Based on the standardized time-series data matrix, extract the inter-frame continuity index, motion regularity index, and trajectory tracking accuracy index; The specific calculation method for the inter-frame continuity index includes: Based on the time-series data of joint angular velocity for each joint, the square of the difference in angular velocity between adjacent time frames for each joint is calculated. The average of the squares of the angular velocity differences for all joints and all adjacent frames is then calculated to obtain the inter-frame continuity index. The specific calculation formula is as follows: ； in, Indicates the inter-frame continuity index. This represents the total number of joints in the robot. Indicates a joint index. Indicates the total number of time frames. Indicates the time frame index. This represents the angular velocity of the j-th joint in frame t. This represents the angular velocity of the j-th joint in frame t-1; The specific calculation method for the aforementioned sports performance indicators includes: For each joint's joint angle in each time frame, determine whether the joint angle exceeds the angle range formed by the preset minimum and maximum allowable angles of the joint. If it exceeds, calculate the square of the ratio of the excess amount to the width of the angle range; otherwise, take zero to obtain the angle over-limit penalty value. For the joint angular velocity of each joint in each time frame, determine whether the joint angular velocity exceeds the angular velocity range formed by the preset minimum allowable angular velocity and the maximum allowable angular velocity of the joint. If it exceeds, calculate the square of the ratio of the excess amount to the width of the angular velocity range; otherwise, take zero and obtain the angular velocity over-limit penalty value. The motion compliance index is obtained by adding the angular and angular velocity violation penalties for the same joint in the same time frame, and then averaging the values ​​for all joints and all time frames. The specific formula is as follows: ； in, Indicators representing the standardization of movement, This represents the angle value of the j-th joint in frame t. and Let represent the preset minimum and maximum allowable angles of the j-th joint, respectively. This represents the angular velocity value of the j-th joint in frame t. and These represent the preset minimum and maximum permissible angular velocities of the j-th joint, respectively. The quadratic over-limit penalty function is represented by the following formula: ； in, This refers to the physical quantity to be checked, namely angle or angular velocity. and These represent the preset lower and upper limits of the physical quantity, respectively. The specific calculation method for the trajectory tracking accuracy index includes: Based on the end effector pose timing data, the square of the Euclidean distance between the actual position and the expected position of the end effector in each time frame is calculated as the square of the position deviation. Calculate the angular deviation between the actual attitude rotation matrix and the desired attitude rotation matrix of the actuator at the end of each time frame; The trajectory tracking accuracy index is obtained by multiplying the squared position deviation and angle deviation of the same time frame by the position weight coefficient and attitude weight coefficient, respectively, summing them, and then averaging them over all time frames. The specific calculation formula is as follows: ； in, Indicates the trajectory tracking accuracy index. This represents the three-dimensional position vector of the end effector during actual motion in frame t. This represents the expected position vector of the actuator at the end of frame t. This represents the square of the Euclidean distance. The weighting coefficient represents the positional deviation. Let represent the rotation matrix indicating the actual attitude of the end effector in frame t. Each column of the matrix represents the direction vector of each axis of the end effector's local coordinate system in the world coordinate system. Let the rotation matrix represent the desired orientation of the end effector in frame t. This represents the angular deviation between two rotation matrices. Weighting coefficients representing attitude deviation; Step S3: Normalize the inter-frame continuity index, the motion regularity index, and the trajectory tracking accuracy index respectively. Then, perform a weighted summation on the three normalized indices and take a negative exponent with the natural constant as the base of the weighted summation result to obtain the comprehensive quality score of a single training action. Step S4: When the overall quality score is lower than the preset threshold, calculate the contribution of the angle deviation of each joint, identify the joints whose contribution exceeds the average contribution of all joints as abnormal joints, and output training improvement suggestions.

2. The method according to claim 1, characterized in that, In step S1, the multi-source timing sensing data includes joint angle timing data, joint angular velocity timing data, end effector pose timing data, and motor current timing data.

3. The method according to claim 2, characterized in that, In step S4, the specific calculation method for the contribution of the angle deviation of each joint includes: For each joint, the mean of the squares of the differences between the actual angle and the expected angle of the joint across all time frames is calculated as the angle deviation term; Calculate the average value of the angle over-limit penalty for the joint across all time frames, multiply it by a preset normalized penalty weight coefficient, and use it as the over-limit penalty item; The contribution of joint angle deviation is obtained by adding the angle deviation term to the over-limit penalty term.

4. The method according to claim 3, characterized in that, In step S4, determining joints whose contribution exceeds the average contribution of all joints as abnormal joints specifically includes: Calculate the arithmetic mean of the angular deviation contributions of all joints; The contribution of each joint's angle deviation is compared with the arithmetic mean, and joints whose contribution of angle deviation is greater than the arithmetic mean are identified as abnormal joints.

5. A robot training motion quality evaluation system driven by time-series sensor data, used to implement the robot training motion quality evaluation method driven by time-series sensor data as described in any one of claims 1-4, characterized in that, include: The data acquisition module is used to acquire multi-source time-series sensor data during robot training. The data preprocessing module is used to perform timestamp alignment, filtering preprocessing, and Z-score normalization on multi-source time-series sensor data to obtain a normalized time-series data matrix. The index extraction module is used to extract inter-frame continuity indexes, motion regularity indexes, and trajectory tracking accuracy indexes based on a standardized time-series data matrix. The quality scoring module is used to normalize the inter-frame continuity index, the motion regularity index, and the trajectory tracking accuracy index respectively, perform a weighted sum on the three normalized indices, and then take a negative exponent with the natural constant as the base of the weighted sum to obtain the comprehensive quality score of a single training action. The anomaly diagnosis module is used to calculate the contribution of angle deviation of each joint when the overall quality score is lower than a preset threshold. Joints whose contribution exceeds the average contribution of all joints are identified as abnormal joints, and training improvement suggestions are output.

6. The system according to claim 5, characterized in that, The indicator extraction module specifically includes: The continuity extraction unit is used to calculate the squared mean of the angular velocity difference between adjacent frames based on the angular velocity time-series data of each joint, so as to obtain the inter-frame continuity index. The normative extraction unit is used to calculate the average of the secondary over-limit penalty based on whether the angle and angular velocity of each joint exceed the preset allowable range, and obtain the motion normative index. The accuracy extraction unit is used to calculate the weighted average value based on the position deviation and attitude deviation between the actual pose and the expected pose of the end effector, so as to obtain the trajectory tracking accuracy index.

7. The system according to claim 6, characterized in that, The anomaly diagnosis module specifically includes: The contribution calculation unit is used to calculate the sum of the squared mean of the angle deviation and the weighted average of the over-limit penalty for each joint when the overall quality score is lower than the preset threshold, and use it as the angle deviation contribution of the joint. An anomaly determination unit is used to calculate the arithmetic mean of the contribution of all joint angle deviations, and to determine joints whose angle deviation contribution is greater than the arithmetic mean as abnormal joints. The suggested output unit is used to output the deviation curve between the actual angle and the expected angle of each abnormal joint during the abnormal time period, as well as the information on the time period exceeding the allowable range, based on the identified abnormal joints, as a training improvement suggestion.

Images