A learning evaluation method, system and device for teaching robots to perform tasks, and a storage medium
By acquiring the operational trajectory and classroom interaction information of the teaching robot, and combining it with standard trajectory information for segmented evaluation, the problem of insufficient interpretability of evaluation results in existing evaluation methods is solved, and an accurate assessment of the mastery and cognitive dependence on the teaching robot is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- LUOYANG VOCATIONAL&TECHNICAL COLLEGE
- Filing Date
- 2026-04-23
- Publication Date
- 2026-07-14
Smart Images
Figure CN122390927A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent teaching and robot training technology, specifically to a robot practical task learning evaluation method, system, device, and storage medium for teaching, training, and skills assessment scenarios. Background Technology
[0002] Educational robot platforms are widely used in practical teaching activities such as robotic arm grasping and handling, trajectory following, assembly operations, teach programming, and process flow training. In these scenarios, the robot platform not only performs actions but also provides teaching support functions such as training guidance, process recording, prompts and feedback, and assessment. Unlike robot systems in general industrial production scenarios that aim for continuous equipment operation, educational robots target learners. Their focus extends beyond task completion to include whether the learner understands the task steps, masters key actions, relies on prompts, or repeatedly attempts or restarts the learning process.
[0003] Existing motion evaluation methods for general-purpose or industrial robots mostly focus on trajectory accuracy, path deviation, operational efficiency, posture stability, and whether the final task result meets the standards. Their evaluation objects primarily concern whether the robot's actions themselves meet control or technological requirements. This type of evaluation is suitable for production operations, path reproduction, and process manufacturing. However, in educational robot platforms, the evaluation object is not only the robot's motion result itself, but also the learning status and mastery level of the executor during task execution. In educational robot scenarios, even if the same operational trajectory ultimately completes the task, it does not necessarily mean that the executor has truly mastered the corresponding skill.
[0004] For example, in tasks involving grasping, transferring, and placing in educational robots, a subject might ultimately complete the placement action, but may pull back multiple times during the approach phase, deviate from key action nodes during the grasping phase, or only complete the task after repeatedly calling prompts before placement. For conventional robot evaluation methods, these processes often only manifest as trajectory deviation or increased time consumption; however, in educational robot scenarios, these phenomena precisely reflect differences in the subject's mastery of key actions, task comprehension, independent completion ability, and reliance on prompts. If evaluation is based solely on whether the final task is completed, or solely on overall trajectory error, it is difficult to accurately reflect the subject's true learning progress.
[0005] Furthermore, practical tasks of educational robots typically possess strong pedagogical semantics, with different training objectives and scoring focuses at different stages. For example, the assessment significance of action stages such as approaching, locating, grasping, transferring, placing, and returning differs in teaching. During training, the robot also engages in classroom interactions such as asking questions, providing prompts, confirming steps, interrupting tasks, correcting errors, and redoing. These interactions, along with the actual operation process, constitute important evidence for teaching evaluation. However, existing conventional robot evaluation methods usually only process trajectory or result data, lacking the ability to uniformly correlate classroom interaction information with action process information, making them difficult to directly apply to learning assessment tasks on educational robot platforms.
[0006] In the context of educational robots, the challenge is not simply evaluating robot motion errors, but rather how to integrate the operational trajectory information of the subject with classroom interaction information, and combine this with action stages that have pedagogical semantics to collaboratively evaluate the subject's operational mastery and cognitive dependence, thereby forming interpretable learning evaluation results oriented towards practical teaching. Summary of the Invention
[0007] The main objective of this invention is to provide a learning assessment method, system, device, and storage medium for practical tasks of educational robots, in order to solve common problems in existing practical teaching of educational robots, such as: the evaluation process relies too much on the teacher's subjective observation, making it difficult to analyze the execution subject's operation process around the action stages with pedagogical semantics; classroom interaction behavior and actual operation process are separated, making it difficult to form a unified learning assessment basis; conventional robot evaluation methods can only reflect the action results or trajectory deviations, making it difficult to accurately distinguish whether the execution subject has not yet mastered the key operations or has a strong dependence on prompts, thus resulting in insufficient interpretability and weak pertinence of learning assessment results.
[0008] To achieve the above objectives, according to a first aspect of the present invention, a learning evaluation method for a teaching robot practical task is provided, comprising: acquiring operation trajectory information and classroom interaction information of an execution subject performing a teaching robot practical task; acquiring standard trajectory information corresponding to the teaching robot practical task, wherein the standard trajectory information includes at least action stage definitions, a set of key action nodes, and task event labels; using key action nodes as stage benchmarks, and combining at least one of pause points, speed change points, direction change points, and task event points in the operation trajectory information, correcting the action stage boundaries of the execution subject's operation trajectory, and segmenting the execution subject's operation trajectory and the standard trajectory accordingly; generating trajectory evaluation results based on the matching of each action stage; extracting learning evaluation features from classroom interaction information, and generating classroom evaluation results based on the learning evaluation features; constructing an operation mastery representation based on the trajectory evaluation results, constructing a cognitive dependence representation based on the classroom evaluation results, and determining the learning evaluation results for the current practical task based on the operation mastery representation and the cognitive dependence representation; and outputting the learning evaluation results.
[0009] Furthermore, the matching status of each action phase is determined by at least two or more of the following: local shape consistency, key action node hit status, phase motion trend consistency, trajectory stability, pull-back behavior, and pause behavior.
[0010] Furthermore, the boundaries of the action stages are corrected and segmented, including: using key action nodes in the standard trajectory as the main anchor points, searching for pause points, speed change points, direction change points, and task event points in the execution subject's operation trajectory within the corresponding neighborhood range; when a candidate boundary that meets the conditions is found, the endpoints of the corresponding action stages are corrected using the candidate boundary; when no candidate boundary that meets the conditions is found, the stage boundaries obtained based on the standard trajectory mapping remain unchanged.
[0011] Furthermore, the pause point is determined by the local speed being lower than a preset pause speed threshold and the duration exceeding a preset pause duration threshold; and / or, the speed change point is determined by the speed difference between adjacent trajectory points being greater than a preset speed change threshold; and / or, the direction change point is determined by the angle between adjacent displacement vectors being greater than a preset direction change threshold; and / or, the task event point is determined by one or more of the following: gripper action event, object pick-up confirmation event, placement completion event, task pause event, prompt call event, and step confirmation event.
[0012] Furthermore, operational mastery is characterized by an operational mastery index. Cognitive dependence is represented by the cognitive dependence index. .
[0013] Furthermore, operational mastery of the index Determine as follows:
[0014]
[0015] in, This represents the overall trajectory score. This indicates that key action nodes have been hit by the indicator. Indicators representing trajectory stability This indicates the proportion of pullback behavior. This indicator represents the proportion of abnormal pauses. to The weight coefficients are non-negative and satisfy the following conditions: .
[0016] Furthermore, cognitive dependence index Determine as follows:
[0017]
[0018] in, This indicates the rate of repeated questions. This indicates the dependency rate. This indicates the redo rate. Indicates the task interruption rate. This indicates the acceptance rate. to The weight coefficients are non-negative and satisfy the following conditions: .
[0019] Furthermore, before generating trajectory evaluation results and classroom evaluation results, the operation trajectory information and classroom interaction information are aligned to a unified time reference. The unified time reference alignment process includes: adding or correcting timestamps to data from different sources; resampling or interpolating data with different sampling frequencies; filtering and noise reduction and outlier removal for operation trajectory information; deduplicating events and mapping step numbers for classroom interaction information; and establishing the correlation between trajectory segments and interaction events based on a unified timeline.
[0020] According to a second aspect of the present invention, the present invention also provides a learning assessment system for practical tasks of teaching robots, including a data acquisition module, a standard trajectory management module, a trajectory evaluation module, a classroom evaluation module, a fusion processing module, and an output module.
[0021] According to a third aspect of the present invention, the present invention also provides an electronic device and a computer-readable storage medium for implementing the above-described method.
[0022] Compared to existing technologies, this invention does not provide a general evaluation of the robot's overall motion results. Instead, it is based on standard trajectory information and corresponding segments of the action stages, combined with key action nodes, and uses pause points, speed change points, direction change points, and task event points to correct stage boundaries. This allows for a more accurate reflection of the robot's performance in different teaching stages, facilitating the identification of problems in specific aspects such as approach, positioning, grasping, and placement. Furthermore, this invention incorporates classroom interaction information such as prompts, repeated questions, task interruptions, error correction, and redoing into a unified evaluation process. This expands the evaluation scope from simple action results to both the operational and learning processes. By constructing separate representations of operational mastery and cognitive dependence, it further distinguishes between situations where actions are not mastered and tasks are completed based on prompts, thereby improving the interpretability and teaching reference value of the evaluation results. In addition, before generating trajectory evaluation results and classroom evaluation results, this invention performs unified time benchmark alignment processing on data from different sources, enabling trajectory segments and interactive events to be correlated within the same temporal framework. This improves the accuracy and stability of the evaluation process in multi-source heterogeneous data environments. Therefore, this invention is more suitable for practical teaching evaluation tasks in teaching robot platforms. Without changing the robot's control logic, it can directly serve teaching training, practical assessment, and personalized teaching intervention, and has good engineering implementation value and platform application prospects. Attached Figure Description
[0023] The accompanying drawings, which form part of this specification, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:
[0024] Figure 1 The overall flowchart of the teaching robot practical task learning assessment method of the present invention is shown;
[0025] Figure 2 This diagram illustrates the corresponding segment matching of the operation trajectory of the present invention according to the action stage;
[0026] Figure 3 A schematic diagram illustrating the composition of the matching scores at each stage of the present invention is shown.
[0027] Figure 4 This diagram illustrates the feature extraction for learning evaluation according to the present invention.
[0028] Figure 5 This diagram illustrates how the operational mastery representation and cognitive dependence representation of the present invention jointly determine the learning evaluation results.
[0029] Figure 6 A block diagram of the learning evaluation system of the present invention is shown. Detailed Implementation
[0030] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0031] To facilitate understanding of this invention, the main terms used herein are explained first. The term "teaching robot" as used herein refers to a robot platform and its supporting teaching aid system applied in teaching, practical training, skills training, demonstration assessment, or course practice scenarios. This platform is not only used to execute preset action tasks but also to record the execution process, receive interactive information, and generate learning evaluation results. The operation trajectory information mentioned herein refers to data that characterizes the change in the movement path of the executing subject over time during the execution of the teaching robot's practical tasks. The data mentioned herein can come directly from visual tracking results, or from robot control logs, or a fusion of both. The classroom interaction information mentioned herein refers to the text, voice, prompt calls, confirmations, or error correction behavior data that occur between the executing subject and the teaching aid system before, during, or after task execution. The learning evaluation features mentioned herein are data extracted from classroom interaction information to reflect the characteristics of the executing subject's learning process. The operation mastery representation mentioned herein is an intermediate quantity that comprehensively represents the executing subject's operational performance. The cognitive dependence representation mentioned herein is an intermediate quantity that comprehensively represents the executing subject's performance in seeking help, relying on prompts, and trial-and-error progression.
[0032] The core of this invention lies not in simply collecting two types of data, but in establishing an evaluation chain with a clear structure: First, a matching evaluation based on standard trajectories and action stages is performed on the operational trajectory; second, learning evaluation features are extracted from classroom interaction information to form classroom evaluation results; third, operational mastery representations and cognitive dependence representations are formed separately; finally, the learning evaluation results are determined based on these two representations. This structure gives the evaluation process a clear technical organizational logic and forms the basis for distinguishing this invention from general teaching quality statistical systems and general educational robot interaction systems.
[0033] See also Figures 1 to 6 As shown, the present invention provides a learning assessment method, system, device, and storage medium for practical tasks of teaching robots.
[0034] like Figure 1 As shown, the teaching robot practical task learning assessment method of this embodiment includes steps S101 to S105. In a preferred embodiment, step S101A, namely unified time reference alignment processing, can also be performed before generating trajectory evaluation results and classroom evaluation results.
[0035] Step S101: Obtain the operation trajectory information and classroom interaction information of the executing entity during the execution of the teaching robot practical task.
[0036] In one embodiment, the operation trajectory information is obtained by high-frequency visual recognition and continuous coordinate tracking of the end effector, tool center point, or preset key parts by physical sensors such as depth cameras installed around the teaching robot; or, it is obtained by parsing the pose logs output in real time by the joint encoders and motion controllers at the bottom of the teaching robot. Classroom interaction information can be jointly provided by the teaching client, speech recognition module, question-and-answer module, prompt management module, and task event log module.
[0037] In another embodiment, the operation trajectory information in step S101 can also be obtained by physical operation trajectory information during the execution process through physical sensors or robot control systems deployed in the teaching robot's work space; classroom interaction information is obtained synchronously through interactive terminals, question-and-answer terminals, voice systems or prompting systems.
[0038] Step S101A: Unified time base alignment processing.
[0039] Before generating trajectory evaluation results and classroom evaluation results, the system prioritizes aligning the operation trajectory information and classroom interaction information with a unified time reference. This unified time reference alignment includes: adding or correcting timestamps to data from different sources; resampling or interpolating data with different sampling frequencies; filtering and noise reduction and outlier removal for operation trajectory information; deduplicating events and mapping step numbers to classroom interaction information; and establishing the correlation between trajectory segments and interaction events based on a unified timeline.
[0040] In one embodiment, data from the depth camera, the robot controller's pose log, prompt call log, question-and-answer log, and task event log have different sampling frequencies and recording granularities, and the system preferably aligns them with a unified timeline. Specifically, the controller log can be used as a high-frequency master time reference, and the visual tracking results can be mapped to the same timeline through time interpolation. At the same time, events such as prompt calls, task confirmations, pause and resume, and step completions can be mapped to the corresponding trajectory time intervals, thereby providing a unified temporal basis for subsequent action phase corrections and classroom behavior analysis.
[0041] As can be seen from the above description, this invention enables multi-source heterogeneous data to be jointly processed within the same temporal framework by adding a unified time benchmark alignment process before trajectory evaluation and classroom evaluation. This avoids misalignment between trajectory segments and interactive events, and improves the accuracy and engineering feasibility of subsequent evaluation results.
[0042] Step S102: Obtain the standard trajectory information corresponding to the teaching robot's practical task, and use key action nodes as stage benchmarks. Combine at least one of the pause points, speed change points, direction change points, and task event points in the execution subject's operation trajectory to correct the action stage boundary of the execution subject's operation trajectory. Then, divide the execution subject's operation trajectory into corresponding segments with the standard trajectory, and generate trajectory evaluation results based on the matching of each action stage.
[0043] To support the segmented matching and evaluation of action stages, this invention preferably pre-establishes standard trajectory information corresponding to the practical tasks of the teaching robot. The standard trajectory information includes not only a sequence of trajectory points, but also the definition of action stages, a set of key action nodes, stage teaching weights, allowable deviation ranges for each stage, and task event labels.
[0044] In one embodiment, standard trajectory information is obtained through teacher demonstration. That is, the instructor performs a standardized operation on the teaching robot platform and records the trajectory as a standard trajectory sample; then the system performs stage annotation and key node annotation on the sample to form standard trajectory information.
[0045] In another embodiment, standard trajectory information is obtained through the execution of a standard program. That is, the robot control program runs the task with preset standard parameters, records its control trajectory, and then converts it into standard trajectory information.
[0046] In another embodiment, the standard trajectory information can also be formed by aggregating multiple high-quality samples. Specifically, multiple excellent trajectories can first be time-normalized, action phase aligned, and outlier removed, and then the phase template trajectory and node template position can be calculated to reduce the impact of random deviations of single samples on the standard trajectory information.
[0047] In another embodiment, the standard trajectory information further includes stage teaching weights and stage allowable deviation ranges.
[0048] In terms of task modeling, the system preferably abstracts each teaching robot practice task into several action stages with clear pedagogical meaning. For example, for grasping, transporting, and placing tasks, an approach stage, a localization stage, a grasping stage, a lifting stage, a transfer stage, a placement stage, and a return stage can be defined; for trajectory following tasks, a start stage, a straight-line stage, a turning stage, a correction stage, and an end stage can be defined. The introduction of stage modeling allows subsequent segmented matching to no longer rely on pure time slices, but rather on a structured task representation with pedagogical semantics.
[0049] In one embodiment, the action stage boundary correction follows this logic: using key action nodes in the standard trajectory as the main anchor points, the system searches for pause points, speed change points, direction change points, and task event points in the execution subject's operation trajectory within the corresponding neighborhood; when a candidate boundary that meets the criteria is found, the endpoint of the corresponding action stage is corrected using the candidate boundary; when no candidate boundary that meets the criteria is found, the stage boundary obtained based on the standard trajectory mapping remains unchanged. The basic idea is to first provide a stage benchmark using predefined key action nodes in the standard trajectory, and then correct the stage boundary based on the behavioral changes during the actual execution process of the execution subject, thereby achieving a more individual-friendly segmentation.
[0050] Let the execution entity's operation trajectory be: ,in Indicates the first A trajectory point, Let the corresponding timestamp be used. Then the local velocity between adjacent points can be expressed as:
[0051]
[0052] If satisfied And continuously satisfy within the time window If so, the corresponding interval is determined to be a pause interval, where Indicates the pause speed threshold. This represents the threshold for the duration of the pause. The pause interval often corresponds to the hesitation, checking, or waiting behavior of the executing entity before and after a critical action, and therefore can serve as a candidate boundary for action phase switching.
[0053] The point of velocity change can be determined by the velocity difference. When the following conditions are met... At that time, the corresponding point can be marked as the velocity change point, where This refers to the speed change threshold. Speed change points typically correspond to situations where the executing entity transitions from an approaching action to an executing action, or from an executing action to a withdrawing action.
[0054] Points of directional change can be determined by the angle between adjacent displacement vectors. Let the local displacement vector be... Then when When this happens, the corresponding point can be considered as a candidate point for direction change, where This is the threshold for direction change. Points of direction change often correspond to behaviors such as path turning, attitude correction, and target alignment.
[0055] Task event points can be provided by system logs, prompt management module, or controller event logs. For example, events such as gripper closure, object pickup confirmation, placement completion, task pause, prompt invocation, and step confirmation can all serve as auxiliary information for the boundaries of the action phase.
[0056] As can be seen from the above description, by incorporating both the standard action semantics and the actual operational behavior of the executing subject into the stage division process, this invention can effectively avoid stage misalignment caused by simple time equalization, and establish stage matching on a more accurate structural basis.
[0057] In one embodiment, the matching status of each action stage is determined by at least two or more of the following: local shape consistency, key action node hit rate, stage motion trend consistency, trajectory stability, pull-back behavior, and pause behavior. The indicator system is not a simple application of general path similarity indicators, but rather a teaching semantic evaluation structure established in conjunction with the standardized operational requirements in the practical tasks of teaching robots.
[0058] Local shape consistency is mainly used to determine how similar the shape of the executing subject's trajectory is to the standard trajectory during a certain action phase. For example, during the approach phase, if the shape of the executing subject's trajectory obviously deviates or frequently turns back, the local shape consistency will decrease.
[0059] The critical action node hit rate is mainly used to determine whether the executor is performing critical actions in the correct position. For example, in a grasping task, grasping nodes, placement nodes, and lifting starting nodes are all critical action nodes; if the executor deviates significantly near these nodes, it will significantly affect the critical action node hit rate score.
[0060] The consistency of movement trends during a phase is mainly used to reflect whether the overall movement direction of the executing entity in that phase conforms to the standard. For example, in the transfer phase, if the overall movement direction of the executing entity deviates significantly from the standard, it indicates that there is a problem with its understanding of the task path.
[0061] Trajectory stability is used to reflect whether the operation is smooth. When the executor is not proficient enough, frequent fine adjustments, severe local jitter, and sudden changes in speed often occur, all of which can be reflected by stability indicators.
[0062] Pullback behavior is used to characterize repetitive and ineffective movements that are opposite to the main direction of the task. For example, excessive withdrawal before grasping or repeated probing back and forth during a transfer can be classified as pullback behavior.
[0063] Pausing behavior is used to characterize abnormal stagnation within a phase. For example, prolonged hesitation before a critical action or pauses caused by insufficient confirmation can be identified through pausing behavior.
[0064] As can be seen from the above description, this invention establishes an index set that combines geometric features and pedagogical semantics, enabling trajectory evaluation to move beyond a single distance error level and thus more comprehensively characterize the operational quality of the executing subject at different stages of the action.
[0065] In one embodiment, such as Figure 2 As shown, a specific work scenario demonstrates the calculation logic of the stage matching score. Here, 22 represents the standard trajectory, 32 represents the execution subject's operation trajectory, 24, 34, 36, 38, and 40 represent key action nodes, 42, 44, 46, 48, and 50 represent pause points, speed change points, or direction change points, and 26, 28, and 30 represent action stage boundaries. The system initially divides the stage based on key action nodes in the standard trajectory 22, such as 34, 24, and 36. Then, by identifying feature points in the execution subject's operation trajectory 32, such as pause points or speed change points 44, it establishes a temporal or semantic correspondence between these feature points and the standard key action nodes 24, thereby determining the stage boundary 26. Similarly, the association between feature point 48 and key action nodes 38 can be used to determine the stage boundary 30. This method ensures that even if the execution subject's operation is delayed or advanced in time, the segmented evaluation can still be based on the same action semantics.
[0066] exist Figure 2 The first one shown Within each action phase, for example, within the region between phase boundaries 26 and 28, the system calculates the spatial distance between the executing subject's operation trajectory 32 and the standard critical action node 36. If the executing subject's operation trajectory 32 deviates significantly near the critical action node 36, the critical action node hit score for that phase will decrease. Observing the executing subject's operation trajectory 32, it is evident that it exhibits significant local fluctuations relative to the standard trajectory 22. These fluctuations reflect the instability of the executing subject's operation; the system can quantify the lower trajectory stability score by calculating the difference in the rate of change of the direction angle between adjacent trajectory points.
[0067] Figure 2 Feature point 46 can also identify abnormal pauses in the execution of tasks by the executing entity; if the pause occurs in a non-standard defined dwell area, the system calculates the pause behavior penalty item based on the pause duration and deducts it from the total score for that stage.
[0068] exist Figure 2 The first one shown Within each action phase, the system can calculate the spatial distance between the trajectory point of the executing entity and the standard key action node, and obtain the hit score of the key action node. The trajectory stability score is calculated by the difference in the rate of change of the direction angle between adjacent trajectory points. If a pause is detected within a non-standard defined stopping area, a pause behavior penalty item can be generated. .
[0069] In one embodiment, such as Figure 3 As shown, it illustrates the first embodiment of this example. A schematic diagram illustrating the logical composition of matching scores for each action stage. Figure 3In the text, 300 represents the number 100. The action phase matching score is as follows: 301 represents the local shape consistency score, 302 represents the key action node hit score, 303 represents the phase motion trend consistency score, 304 represents the trajectory stability score, 305 represents the pullback behavior penalty item, and 306 represents the pause behavior penalty item.
[0070] In this embodiment, the first The matching situation within each action phase includes the first... Matching score for each action phase The matching score of 300 is a weighted aggregation of multiple evaluation dimensions. Matching score for each action phase Determine as follows:
[0071]
[0072] in, Indicates the action phase index. The local shape consistency score is 301. This indicates a critical action node hit score of 302. The consistency score for the phased movement trend is 303. This indicates a trajectory stability score of 304. This indicates that the penalty for pulling back is item 305. Item 306 indicates a penalty for pausing behavior. , , , , , For the first The weighting coefficients for the indicators corresponding to each action stage can be set or obtained through training based on the course type, task type, stage importance, and teaching objectives. To make the formula easier to implement in engineering, each indicator is preferably normalized to the same numerical range.
[0073] The local shape consistency score 301 is used to measure the similarity between the trajectory of the executing subject and the standard trajectory in terms of micro-geometry; the key action node hit score 302 is used to evaluate whether the executing subject accurately triggered the preset action node in this stage; the stage motion trend consistency score 303 is used to measure the fit between the direction of the overall motion vector and the standard path; the trajectory stability score 304 is used to evaluate whether the operation process is smooth and whether there is meaningless jitter; the pullback behavior penalty item 305 and the pause behavior penalty item 306 are used to deduct the current stage score when pullback behavior or abnormal pause behavior is detected.
[0074] The optimal calculation methods and derivation logic for each indicator are given below.
[0075] (1) Local shape consistency score. To eliminate the influence of execution speed differences on trajectory shape comparison, the trajectory of the main execution stage is first evaluated. and standard phase trajectory Normalized resampling based on arc length yields L sampling points:
[0076]
[0077]
[0078] Calculate the orientation angles of adjacent points after resampling to obtain:
[0079]
[0080]
[0081] After calculating the orientation angles of adjacent points after resampling, the local shape consistency score can be expressed as:
[0082]
[0083] In this formula, the more consistent the directions, the closer the fraction is to 1.
[0084] (2) Key action node hit score. Let the first... Phase includes The key action nodes correspond to the hit points of the executing entity as follows: The key points of the standard are Then we can define:
[0085]
[0086] in, This is the stage tolerance parameter. The closer the hit point is to the standard point, the higher the score.
[0087] (3) Consistency score of phase movement trend. Let the overall displacement vector of the executing subject in the phase be... The overall displacement vector in the standard stage is :
[0088]
[0089]
[0090] The trend consistency score can be expressed as:
[0091]
[0092] This formula maps the direction cosine from [-1,1] to [0,1].
[0093] (4) Trajectory stability score. Define the change in direction:
[0094]
[0095]
[0096] After defining the local directional changes of the executing entity and the standard trajectory, the stability score can be expressed as:
[0097]
[0098] This score is used to describe the degree of consistency between the local variation rhythm of the trajectory and the standard variation rhythm.
[0099] (5) Pullback behavior penalty term. The system can perform smoothing filtering on the underlying trajectory coordinates collected by the sensor before calculation. Let the unit vector of the main direction in the standard stage be... The smoothed local displacement vector of the execution subject is ,like ,and If the noise dead zone exceeds the preset physical displacement threshold, it is determined to be a pullback segment. The number of pullback segments within this stage is counted. With the total number of segments ,but:
[0100]
[0101] (6) Penalty for Stall Behavior. The system calculates the instantaneous velocity of the physical trajectory point through differential calculation. When the velocity is lower than the preset static judgment threshold and the duration exceeds the judgment period, a physical pause is detected. Furthermore, if the coordinates of the physical pause exceed the spatial tolerance neighborhood of the preset key action node, it is defined as an abnormal pause. Let the first... The total duration of the phase is The sum of the durations of all pauses defined as abnormal pauses in this phase is ,but:
[0102]
[0103] As can be seen from the above description, this invention provides a clear mathematical definition and calculation process for each evaluation component, and deeply integrates physical data filtering and spatial teaching semantic filtering mechanisms. This makes the stage matching score not only feasible, but also more accurately eliminates hardware interference and reasonable pauses, thereby reflecting the actual geometric deviations and behavioral abnormalities of the executing subject in the action stage.
[0104] Obtain matching scores for each action phase. Then, the trajectory is further weighted according to the importance of each phase to generate a comprehensive trajectory score. The preferred method is as follows:
[0105]
[0106] in, Indicates the total number of action phases. Indicates the action phase index. Indicates the first The teaching weight of each action stage should be optimized to meet the requirements. and , Indicates the first Matching score for each action phase.
[0107] The weighting of instructional objectives should reflect the differences in contribution of each stage to the learning objectives in different tasks. For example, in a grasping-carrying-placing task, the grasping and placement stages are usually more critical than the return stage, and therefore can be given higher weights; in a trajectory following task, the turning and correction stages often better reflect the level of control of the executing entity than the straight-line stage, and therefore can be given higher weights.
[0108] Teaching weights can be determined in the following ways: First, by teachers setting them directly based on their teaching experience; second, by calculating the contribution of each stage to the final success rate or teacher rating through statistical analysis of historical high-performing samples, and then normalizing the results; third, by fitting the teacher rating results through machine learning to find the optimal stage weights.
[0109] As can be seen from the above description, the above-described embodiments of the present invention, by introducing a teaching weight aggregation mechanism based on stage matching, can make the overall trajectory score more accurately reflect the teaching focus and avoid the dilution of key stage errors by non-key stage performance.
[0110] Step S103: Extract learning evaluation features from classroom interaction information and generate classroom evaluation results based on the learning evaluation features.
[0111] In one embodiment, such as Figure 4 As shown, the system first acquires classroom interaction information 401. It should be noted that this classroom interaction information does not include operation trajectory information, but rather includes at least teacher-student dialogue text, speech recognition results, system prompt trigger records, prompt acceptance records, task pause and resume records, step confirmation records, error feedback records, and redo records. Subsequently, the system uses a preset feature extraction operator to extract four core features in parallel from the classroom interaction information 401: knowledge understanding features 402, prompt dependence features 403, task error correction features 404, and independent completion features 405. After the extraction of these four features, the system performs feature fusion using a preset evaluation model, ultimately generating a classroom evaluation result 406.
[0112] In another embodiment, classroom interaction information is not directly used as the final evaluation result. Instead, it is first converted into learning evaluation features, and then classroom evaluation results are generated based on these features. This design transforms originally discrete and heterogeneous interaction events into quantifiable and fusionable structured representations.
[0113] Learning evaluation features should include at least one or more of the following: knowledge comprehension features, prompt dependence features, task error correction features, and independent completion features. For ease of engineering implementation, it is preferable to further extract three or more of the following from classroom interaction information: percentage of conceptual questions, percentage of operational questions, repeat questioning rate, prompt dependence rate, prompt adoption rate, task interruption rate, error correction rate, independent completion rate, and redo rate.
[0114] The preferred feature definition method is given below. Let... This indicates the total number of questions asked. Indicates the number of conceptual questions. Indicates the number of action-oriented questions. Indicates the number of times the question was asked repeatedly. This indicates the number of times the function can be called. Indicates the total number of task steps. This indicates the number of times the prompt was accepted and successfully executed. This indicates the number of steps completed independently without relying on the prompts. Indicates the number of times the task was interrupted. Indicates the number of error corrections. If the number of repetitions is indicated, then:
[0115]
[0116]
[0117]
[0118]
[0119]
[0120]
[0121]
[0122]
[0123]
[0124] Conceptual questions can relate to principles, coordinate systems, execution logic, and the meaning of rules; operational questions can relate to button locations, next steps, and trajectory adjustment methods. Duplicate questions can be detected by setting a semantic similarity threshold and a time window; for example, within a preset time window, if the semantic similarity between two questions exceeds the threshold, it is considered a duplicate question.
[0125] The prompt dependency rate reflects the intensity of the execution subject's call to system prompts; the prompt adoption rate reflects the actual absorption and use of prompt information by the execution subject; the independent completion rate reflects the execution subject's ability to complete task steps without external prompts; the task interruption rate, error correction rate, and redo rate can be obtained directly from the task log and normalized. Among them, the task error correction feature can be further refined into post-prompt correction and autonomous correction.
[0126] In one embodiment, classroom interaction information can first be categorized into conceptual, operational, confirmatory, and other types using a combination of rules and semantic classification models, and then corresponding features can be generated. The advantage of this approach is that it retains the controllability of rule-based methods while incorporating the semantic model's ability to recognize complex expressions.
[0127] It should be noted that classroom assessment results can serve as a reference for personalized calibration of parameters in subsequent similar tasks, or as a basis for teachers to adjust their teaching strategies, but should not be used to negatively impact the current round. Figure 3 The instant generation of matching scores in the middle stage avoids circular dependencies in the current round of evaluation.
[0128] As can be seen from the above description, this invention transforms classroom interaction information into explicit learning evaluation features, making classroom behavior no longer just text or log records, but structured inputs that can further participate in learning evaluation, thereby providing a reliable foundation for the subsequent construction of cognitive dependency representations.
[0129] Step S104: Construct an operational mastery representation based on the trajectory evaluation results, construct a cognitive dependence representation based on the classroom evaluation results, and determine the learning evaluation results for the current practical task based on the operational mastery representation and the cognitive dependence representation.
[0130] In one embodiment, the operational mastery representation is preferably the Operational Mastery Index (OMI), and the cognitive dependence representation is preferably the Cognitive Dependence Index (CDI). For example... Figure 5 As shown, the joint determination logic 503 has two core inputs: operational mastery representation 501 and cognitive dependence representation 502. The input to operational mastery representation 501 comes from... Figure 3The stage matching score and comprehensive trajectory score represent the standardization and accuracy of the executing entity at the physical operation level, reflecting its actual mastery of robot motion control skills. The input of the cognitive-dependent representation 502 comes from... Figure 4 The classroom assessment results, which integrate characteristics of knowledge comprehension, prompting dependence, task error correction, and independent completion, reflect the depth of thinking and the degree of dependence on external assistance by the subject in completing the task.
[0131] The joint determination logic 503 can perform correlation analysis on the operational mastery representation 501 and the cognitive dependence representation 502 through a preset fusion algorithm, such as weighted summation, fuzzy comprehensive evaluation, or a multimodal neural network model. For example, when the operational mastery level is high but the cognitive dependence level is also high, it can be determined that the executing subject is in the mechanical imitation stage rather than the stage of complete autonomous mastery.
[0132] Furthermore, the Operational Mastery Index (OMI) and the Cognitive Dependence Index (CDI) are preferably presented in index form for ease of engineering calculation and interpretation.
[0133] The Operational Mastery Index (OMI) comprehensively reflects the operational performance quality of an executor in a current practical task. It is calculated by combining the overall trajectory score, the hit rate of key action nodes, trajectory stability, pullback behavior, and abnormal pauses. The preferred formula is:
[0134]
[0135] in, The key action node hit index is preferably obtained by normalizing the key action node hit scores at each stage. The trajectory stability index is preferably obtained by normalizing the stability scores at each stage. The indicator representing the proportion of pullback behavior is preferably obtained by combining the pullback penalty items at each stage. This indicator represents the proportion of abnormal pauses, ideally derived from a combination of pause penalty factors across all stages. Weighting to Preferred satisfaction:
[0136]
[0137] The Cognitive Dependency Index (CDI) comprehensively reflects an actor's reliance on prompts, repeated questioning, and trial-and-error progress in the current task. The preferred formula is:
[0138]
[0139] in, To calculate the rate of repeated questions, To indicate dependency rate, For redo rate, For task interruption rate, To indicate the adoption rate. Weight to Preferred satisfaction:
[0140]
[0141] A higher OMI indicates a greater degree of mastery over the operational process by the executing entity; a higher CDI indicates a stronger reliance on external prompts, trial-and-error progress, and interruption recovery by the executing entity. By constructing OMI and CDI separately, we can characterize the quality of performance and the degree of reliance separately, avoiding the reduction of interpretability caused by mixing multiple dimensions in a single score.
[0142] As can be seen from the above description, by projecting the operation trajectory evaluation results and classroom evaluation results onto two representation spaces with clear meanings, the present invention can make the formation process of the final learning evaluation results transparent and structured, making it easier for teachers to understand and use.
[0143] In one embodiment, the learning evaluation results include at least one of the following: learning score, learning status classification, error diagnosis results, and personalized learning suggestions. Preferably, the system outputs all of the above results so that teachers and implementers can understand the evaluation results from three levels: numerical, categorical, and textual suggestions.
[0144] In another embodiment, a comprehensive learning score can be further constructed based on OMI, CDI, task completion rate, independent completion rate, and knowledge comprehension. Specifically, it can be expressed as:
[0145]
[0146] in, Indicates the task completion rate. Indicates the percentage of work completed independently. The degree of knowledge comprehension can be determined by the accuracy rate of answering conceptual questions, the results of quizzes, or the results of teacher verification. to These are non-negative weighting coefficients.
[0147] Learning states can be categorized into at least three types: mastery, clear understanding but unstable operation, reliance on prompts, and trial-and-error learning. The following rules can be used to select the best approach:
[0148] like and and Then it is judged as mastery type.
[0149] like and If so, it is judged as a type with clear cognition but unstable operation.
[0150] like 2 and If so, it is judged as a dependency hint type.
[0151] like or If so, it is judged as a trial-and-error type.
[0152] in, to A preset threshold is set. This threshold can be determined by teacher experience, historical sample statistics, or machine learning methods. Based on the learning state classification, the system can further output error diagnosis results and personalized suggestions. For example, for learners with clear cognition but unstable operation, the system can point out insufficient hits in key action nodes or weak trajectory stability, and suggest strengthening repeated training of local paths; for learners who rely on prompts, the system can point out frequent prompt calls and a high rate of repeated questioning, and suggest reducing the prompt frequency and increasing independent planning steps.
[0153] As can be seen from the above description, by refining the learning evaluation results into scores, status, diagnoses, and suggestions, this invention enables the evaluation results to directly serve subsequent teaching interventions, rather than simply remaining at the level of score output.
[0154] Step S105: Output learning evaluation results. Output formats may include numerical scores, text-based diagnostic results, learning status types, and personalized suggestions. The output can provide real-time feedback to the implementing entity or generate visual analysis results and teaching intervention suggestions for the teacher.
[0155] In one embodiment, the operation trajectory information can be either two-dimensional operation trajectory information or three-dimensional operation trajectory information further obtained based on two-dimensional operation trajectory information. The purpose of setting this feature is to ensure that the main patent is compatible with evaluation results obtained from different trajectory morphology inputs, without requiring three-dimensional mapping as an essential feature of an independent claim.
[0156] In scenarios using only two-dimensional operation trajectory information, the system can directly complete action stage segmentation and stage matching evaluation on a two-dimensional plane. In scenarios with further spatial mapping capabilities, the system can also use three-dimensional operation trajectory information as input, thereby improving the ability to identify spatial deviations. Regardless of whether the input is two-dimensional or three-dimensional, the core of this invention lies in: action stage corresponding segmentation and matching evaluation based on standard trajectories and learning evaluation feature extraction based on classroom interaction information, and on this basis, constructing operation mastery representation and cognitive dependence representation.
[0157] As can be seen from the above description, by preserving the compatibility of two-dimensional or three-dimensional trajectory input, this invention can be adapted to different hardware conditions and different system architectures without changing its core learning and evaluation mechanism.
[0158] In one embodiment, such as Figure 6 As shown, the present invention also provides a learning assessment system for practical tasks of teaching robots. Figure 6 In this diagram, 100 represents the learning assessment system for practical tasks of the teaching robot, 110 represents the data acquisition module, 115 represents the standard trajectory management module, 116 represents the time reference alignment processing unit, 120 represents the trajectory evaluation module, 130 represents the classroom evaluation module, 140 represents the fusion processing module, and 150 represents the output module; among these, 111 represents the camera, 112 represents the robot control system, 113 represents the speech recognition component, and 114 represents the log component; 1151 represents the action stage definition, 1152 represents the set of key action nodes, 1153 represents the task event label, and 1154 represents... The display shows the teaching weight and allowable deviation range for each stage; 121 represents the stage determination unit, 122 represents the indicator calculation unit, 123 represents the single-stage scoring unit, and 124 represents the comprehensive score unit; 131 represents the interactive parsing unit, 132 represents the feature extraction unit, and 133 represents the classroom evaluation generation unit; 141 represents the OMI calculation unit, 142 represents the CDI calculation unit, 143 represents the learning outcome determination unit, and 144 represents the suggestion generation unit; 151 represents the report generation unit; 161 represents the display on the execution subject's end, 162 represents the teacher's visualization panel, and 163 represents the report file.
[0159] The data acquisition module 110 is used to acquire the operation trajectory information and classroom interaction information of the executing entity during the teaching robot practical task. Preferably, the data acquisition module 110 includes a camera 111, a robot control system 112, a speech recognition component 113, and a log component 114. The camera 111 and the robot control system 112 can respectively provide visual trajectory data and pose log data, the speech recognition component 113 can provide speech transcription results, and the log component 114 can provide event logs such as prompt calls, step confirmations, error feedback, pause / resume, and retry.
[0160] In a preferred embodiment, the learning assessment system 100 for practical tasks of teaching robots may further include a time alignment processing unit 116, used to perform unified time reference alignment processing on data from different sources before the trajectory evaluation module 120 and the classroom evaluation module 130. This time alignment processing unit 116 can perform timestamp appending or correction, data resampling or interpolation, trajectory data filtering and noise reduction and outlier removal, deduplication of classroom interaction data events and step number mapping, and unified timeline association between trajectory segments and interaction events.
[0161] The standard trajectory management module 115 is used to call or create standard trajectory information corresponding to the teaching robot's practical tasks. The standard trajectory information includes at least the definition of action stages, the set of key action nodes, and task event labels. In a preferred embodiment, the standard trajectory information also includes stage teaching weights and stage allowable deviation ranges.
[0162] The trajectory evaluation module 120 uses key action nodes as stage benchmarks and incorporates at least one of the following from the operation trajectory information: pause points, speed change points, direction change points, and task event points. It corrects the action stage boundaries of the executing subject's operation trajectory, segments the trajectory to correspond with the standard trajectory, and generates trajectory evaluation results based on the matching of each action stage. The trajectory evaluation module 120 further includes a stage determination unit 121, an index calculation unit 122, a single-stage scoring unit 123, and a comprehensive scoring unit 124. The stage determination unit 121 dynamically determines stage boundaries by combining standard key action nodes with feature points in the actual operation of the executing subject; the index calculation unit 122 calculates features such as local shape consistency, key action node hit rate, stage motion trend consistency, trajectory stability, pull-back behavior, and pause behavior; the single-stage scoring unit 123 generates matching scores for each action stage; and the comprehensive scoring unit 124 combines teaching weights to generate a comprehensive trajectory score.
[0163] The classroom evaluation module 130 is used to extract learning evaluation features from classroom interaction information and generate classroom evaluation results based on these features. The classroom evaluation module 130 further includes an interaction parsing unit 131, a feature extraction unit 132, and a classroom evaluation generation unit 133. The interaction parsing unit 131 is used to perform semantic classification and intent recognition on text and event logs; the feature extraction unit 132 is used to quantify features such as the proportion of conceptual questions, the proportion of operational questions, the rate of repeated questions, the rate of reliance on prompts, the rate of acceptance of prompts, the rate of task interruption, the rate of error correction, the proportion of independent completion, and the rate of rework; the classroom evaluation generation unit 133 is used to output the classroom evaluation results.
[0164] The fusion processing module 140 is used to construct an operational mastery representation based on the trajectory evaluation results, construct a cognitive dependence representation based on the classroom evaluation results, and determine the learning evaluation result based on both. The fusion processing module 140 further includes an OMI calculation unit 141, a CDI calculation unit 142, a learning result determination unit 143, and a suggestion generation unit 144. Specifically, the OMI calculation unit 141 generates the operational mastery index, the CDI calculation unit 142 generates the cognitive dependence index, the learning result determination unit 143 outputs the learning score and learning status classification, and the suggestion generation unit 144 generates error diagnosis results and personalized learning suggestions.
[0165] The output module 150 is used to output learning evaluation results. Preferably, the output module 150 includes a report generation unit 151. The output results can be sent to the execution subject's end display 161, or to the teacher's end visualization panel 162, or a report file 163 can be generated for archiving or teaching management.
[0166] As can be seen from the above description, the present invention uses a modular system structure to describe the method steps in a concrete way, which facilitates system deployment, interface design and functional expansion, and makes each technical feature have a clear correspondence at the engineering implementation level.
[0167] In one embodiment of the present invention, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable by the processor. When the processor executes the computer program, it implements the teaching robot practical task learning evaluation method of any of the above embodiments.
[0168] Electronic devices can be teaching robot control terminals, edge computing devices, industrial control computers, servers, laptops, or other devices with data processing capabilities. Memory can include read-only memory, random access memory, flash memory, solid-state drives, or other non-volatile storage media.
[0169] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the above-mentioned learning evaluation method for practical tasks of teaching robots is implemented.
[0170] In another embodiment, the threshold parameter for action phase segmentation is as follows: , , and The settings can be adjusted based on the specific course content, the robotic arm's movement speed range, and the camera's sampling frequency. For tasks with small movement amplitudes and high precision requirements, the settings can be appropriately reduced. and To improve the sensitivity of stage boundaries; for faster-paced tasks, adjustments can be made appropriately. This is to avoid misjudging brief, natural pauses as abnormal stops.
[0171] In another embodiment, the weighting coefficients in the stage-matched score formula can be preset by the teacher or obtained through sample fitting. If sample fitting is used, historical teacher ratings can be used as a monitoring signal, and optimization is achieved by minimizing the difference between the system rating and the teacher rating. , , , , , .
[0172] In another embodiment, the learning evaluation feature extraction can employ a combination of rules, statistical models, or semantic models. For example, the identification of conceptual and operational questions can be achieved using a combination of keyword dictionaries and intent classification models; the identification of repeated questions can be achieved using a combination of semantic similarity calculation and time window constraints.
[0173] In another embodiment, the learning state classification can be performed not only using fixed threshold rules, but also using decision trees, logistic regression, or other lightweight models, as long as they still use operational mastery representations and cognitive dependence representations as core inputs.
[0174] The above embodiments do not change the core technical idea of the present invention, that is, in the practical task of teaching robots, the trajectory evaluation result is generated by segment matching corresponding to the action stage based on standard trajectory information, the learning evaluation result is generated by extracting learning evaluation features through classroom interaction information, and the learning evaluation result is determined by operation mastery representation and cognitive dependence representation.
[0175] As can be seen from the above description, the present invention achieves the following technical effects: it can perform fine-grained analysis of the operation process of the executing subject, and can perform structured representation of the learning process of the executing subject, and finally output interpretable learning evaluation results. It has clear technical logic, good engineering feasibility and high teaching application value.
[0176] Obviously, the embodiments described above are merely some, not all, embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.
[0177] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0178] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A learning assessment method for practical tasks of teaching robots, characterized in that, include: Acquire operational trajectory information and classroom interaction information of the implementing entity during the execution of practical tasks using the teaching robot; Obtain the standard trajectory information corresponding to the practical task of the teaching robot. The standard trajectory information includes at least the definition of the action stage, the set of key action nodes, and the task event label. Using the key action nodes as the stage benchmark, and combining at least one of the pause points, speed change points, direction change points, and task event points in the operation trajectory information, correct the action stage boundary of the execution subject's operation trajectory, divide the execution subject's operation trajectory into corresponding segments with the standard trajectory, and generate trajectory evaluation results based on the matching of each action stage. Learning evaluation features are extracted from the classroom interaction information, and classroom evaluation results are generated based on the learning evaluation features; Based on the trajectory evaluation results, an operational mastery representation is constructed; based on the classroom evaluation results, a cognitive dependence representation is constructed; and based on the operational mastery representation and the cognitive dependence representation, a learning evaluation result for the current practical task is determined. Output the learning evaluation results.
2. The learning assessment method for practical tasks of teaching robots according to claim 1, characterized in that, The matching condition is determined by at least two or more of the following: local shape consistency, key action node hit status, phase motion trend consistency, trajectory stability, pullback behavior, and pause behavior.
3. The learning assessment method for practical tasks of teaching robots according to claim 1, characterized in that, The process of correcting and segmenting the boundaries of the action phases includes: Using key action nodes in the standard trajectory as the main anchor points, search for pause points, speed change points, direction change points, and task event points in the operation trajectory of the executing entity within the corresponding neighborhood range; When a candidate boundary that meets the criteria is found, the endpoint of the corresponding action stage is corrected using the candidate boundary. When no candidate boundary that meets the criteria is found, the stage boundary obtained based on the standard trajectory mapping remains unchanged.
4. The learning assessment method for practical tasks of teaching robots according to claim 3, characterized in that, The pause point is determined by a local velocity being lower than a preset pause velocity threshold and a duration exceeding a preset pause duration threshold; and / or, The velocity change point is determined when the velocity difference between adjacent trajectory points is greater than a preset velocity change threshold; and / or, The point of directional change is determined by the fact that the angle between adjacent displacement vectors is greater than a preset directional change threshold; and / or, The task event points are determined by one or more of the following: gripper action event, object retrieval confirmation event, placement completion event, task pause event, prompt call event, and step confirmation event.
5. The learning assessment method for practical tasks of teaching robots according to claim 1, characterized in that, The operational mastery is characterized by the operational mastery index. ; The operational mastery index is determined as follows: in, This represents the overall trajectory score. This indicates that key action nodes have been hit by the indicator. Indicators representing trajectory stability This indicates the proportion of pullback behavior. This indicator represents the proportion of abnormal pauses. to The weight coefficients are non-negative and satisfy the following conditions: .
6. The learning assessment method for practical tasks of teaching robots according to claim 1, characterized in that, The cognitive dependence is characterized by the cognitive dependence index. ; The cognitive dependence index is determined as follows: in, This indicates the rate of repeated questions. This indicates the dependency rate. This indicates the redo rate. Indicates the task interruption rate. This indicates the acceptance rate. to The weight coefficients are non-negative and satisfy the following conditions: .
7. The learning assessment method for practical tasks of teaching robots according to claim 1, characterized in that, Before generating trajectory evaluation results and classroom evaluation results, the operation trajectory information and classroom interaction information are subjected to unified time reference alignment processing, which includes: Add or correct timestamps to data from different sources; resample or interpolate data with different sampling frequencies; filter and reduce noise and remove outliers from operation trajectory information; Classroom interaction information is deduplicated and mapped with step numbers; the relationship between trajectory segments and interaction events is established based on a unified timeline.
8. A learning assessment system for practical tasks of teaching robots, characterized in that, include: The data acquisition module is used to acquire the operation trajectory information and classroom interaction information of the executor during the teaching robot practical task; The standard trajectory management module is used to call or create standard trajectory information corresponding to the practical task of the teaching robot. The standard trajectory information includes at least the action stage definition, the set of key action nodes, and the task event label. The trajectory evaluation module is used to correct the action stage boundary of the execution subject's operation trajectory by using the key action nodes as the stage benchmark and combining at least one of the pause points, speed change points, direction change points and task event points in the operation trajectory information, and to divide the execution subject's operation trajectory into corresponding segments with the standard trajectory, and to generate trajectory evaluation results based on the matching of each action stage. The classroom evaluation module is used to extract learning evaluation features from the classroom interaction information and generate classroom evaluation results based on the learning evaluation features. The fusion processing module is used to construct an operational mastery representation based on the trajectory evaluation results, construct a cognitive dependence representation based on the classroom evaluation results, and determine the learning evaluation results based on the operational mastery representation and the cognitive dependence representation. The output module is used to output the learning evaluation results.
9. An electronic device, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein when the processor executes the computer program, it implements the learning assessment method for the practical task of the teaching robot as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the learning assessment method for the teaching robot practical task as described in any one of claims 1 to 7.