Robot teaching control method, device, equipment and medium based on large model

Abstract

This application discloses a robot teaching control method, device, equipment, and medium based on a large model, relating to the field of data processing technology. The method includes: converting voice teaching commands into text, parsing structured commands using a large model in conjunction with a control knowledge base, sequentially completing trajectory planning and singular pose recognition, then solving inverse motion solutions for different scenarios, driving robot movement and storing points locally, and finally implementing cyclic execution based on the point list, with full-process status monitoring and anomaly handling. This application supports natural language interaction, lowers the teaching threshold, handles singular poses hierarchically and suppresses sudden velocity changes, ensures smooth trajectory, and improves the stability and safety of robot teaching and cyclic operations.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a robot teaching control method, device, equipment and medium based on a large model. Background Technology

[0002] Currently, the mainstream teaching methods for robots rely on manual operation with dedicated teaching pendants or offline programming. Some voice control solutions only use keyword matching to recognize fixed sentence commands. The motion control process generally follows the processing logic of first judging singular poses and then planning motion trajectories. Point recording and cyclic motion are independent of each other. Point information mostly relies on external model storage. During robot operation, it simply executes the preset trajectory and lacks a hierarchical processing mechanism for singular poses and areas near singular points, as well as speed fluctuation suppression methods.

[0003] Traditional teaching methods have high operational barriers, require professional personnel, and involve cumbersome programming and position adjustment processes, resulting in low overall teaching efficiency. Keyword matching-based voice interaction cannot understand ambiguous natural language commands, leading to poor interaction flexibility. The operational logic of first identifying singular poses and then planning trajectories can cause a disconnect between trajectory planning and singularity handling. When the robot passes through singular points or their vicinity, it is prone to sudden speed changes, motion stuttering, trajectory jumps, and even safety issues such as equipment impacts and collisions. At the same time, the integration of position memory, trajectory planning, singularity handling, and loop control is low, resulting in insufficient overall operational stability and safety controllability.

[0004] Therefore, how to leverage large models to achieve natural language speech teaching, while simultaneously completing trajectory planning, singular pose hierarchical processing, and velocity fluctuation suppression according to a reasonable time sequence, thereby improving the convenience of robot teaching and operational safety, has become an urgent problem to be solved. Summary of the Invention

[0005] The main purpose of this application is to provide a robot teaching control method, device, equipment and medium based on a large model, aiming to solve the technical problem of how to rely on a large model to realize natural language voice interaction and reduce the threshold of robot teaching operation.

[0006] To achieve the above objectives, this application proposes a robot teaching control method based on a large model, comprising: Collect user natural language teaching instructions and convert the natural language teaching instructions into text instructions; The preset robot control knowledge base and the text instructions are input into the large model for semantic parsing to obtain structured control instructions; The target pose is trajectory planned by calling a preset path planning function according to the structured control command, and the planned trajectory is obtained. Singular pose identification is performed on the poses of the sampling points on the planned trajectory to obtain singular pose identification results; Based on the singular pose recognition result, the corresponding inverse solution function is called to perform inverse solution for each case, and the inverse solution result of the joint angle is obtained. The robot joints are controlled to move according to the inverse kinematics of the joint angles, and the current pose is simultaneously stored in the local memory module to form a point list. The robot is controlled to execute the point list cyclically according to the cyclic motion instructions in the structured control instructions, and the cyclic execution result is obtained. The robot's operating status is collected based on the results of the loop execution and fed back to the user terminal. When an abnormal situation is detected, the preset stop control function and the preset speed change suppression function are called to obtain the abnormal handling result.

[0007] In one embodiment, the step of inputting a preset robot control knowledge base and the text instructions into a large model for semantic parsing to obtain structured control instructions includes: Obtain the action semantic information and parameter semantic information from the text instructions; Read action definition rules, parameter mapping rules, and safety boundary rules from the robot control knowledge base; Based on the action semantic information and the parameter semantic information, the corresponding control rules are matched to obtain the semantic matching result; The semantic matching result is converted according to the preset structured control instruction parsing format to obtain the initial structured control instruction; The initial structured control command is subjected to safety boundary verification to check whether the position parameters, attitude parameters, velocity parameters and acceleration parameters in the initial structured control command exceed the preset safety boundary parameters, thereby obtaining the verified structured control command.

[0008] In one embodiment, the step of calling a preset path planning function according to the structured control command to plan the trajectory of the target pose and obtain the planned trajectory includes: The target pose parameters and motion parameters are extracted from the structured control commands to obtain trajectory planning input data; Based on the trajectory planning input data, a preset path planning function is called to perform interpolation calculation on the motion path between the current pose and the target pose to obtain the initial trajectory curve. The initial trajectory curve is smoothed to obtain a smoothed trajectory curve; Based on the smooth trajectory curve and the preset sampling step size, the pose of the sampling points is extracted on the motion path to obtain the planned trajectory.

[0009] In one embodiment, the step of performing singular pose identification on the sampling point poses on the planned trajectory to obtain singular pose identification results includes: The position coordinates and attitude angles of each sampling point are extracted from the planned trajectory to obtain the pose data to be identified. The determinant of the robot Jacobian matrix is ​​calculated based on the pose data to be identified, and the pose singularity metric is obtained. The pose singularity metric is compared with a preset singular pose threshold to obtain the comparison result; Based on the comparison results and the preset threshold near the singular point, a hierarchical judgment is made to obtain the singular pose recognition result.

[0010] In one embodiment, the step of calling the corresponding inverse kinematics function to perform case-by-case inverse kinematics solution based on the singular pose recognition result to obtain the joint angle inverse kinematics result includes: The singular pose recognition results are analyzed to obtain pose classification information; When the pose classification information is a normal pose, the preset conventional inverse kinematics function is called to perform conventional inverse kinematics solution to obtain the first inverse kinematics result; When the pose classification information is a singular pose, the preset singular point inverse kinematics function is called to perform the singular pose inverse kinematics solution and obtain the second inverse solution result. When the pose classification information is a pose near a singular point, the preset velocity mutation suppression function is called to replan the planned trajectory to obtain the replanned trajectory, and the preset conventional inverse solution function is called to solve the inverse solution of the pose of the sampling point on the replanned trajectory to obtain the third inverse solution result. The joint angle inverse solution result is obtained by taking the first inverse solution result, the second inverse solution result, or the third inverse solution result.

[0011] In one embodiment, the step of controlling the robot to cyclically execute according to the point list based on the cyclic motion instructions in the structured control instructions, and obtaining the cyclic execution result, includes: The cyclic motion instruction is parsed from the structured control instruction to obtain the cyclic control parameters, wherein the cyclic control parameters include the number of cycles parameter and the cyclic mode parameter; The point list is retrieved from the local memory module according to the loop control parameters, and the trajectory planning parameters associated with each point in the point list are extracted to obtain a set of reused trajectory parameters. Based on the set of reused trajectory parameters, a preset path planning function is called to perform trajectory reuse planning, resulting in a loop trajectory. Singular pose identification is performed on the poses of the sampling points on the cyclic trajectory to obtain the cyclic singular pose identification result. Based on the cyclic singular pose recognition result, the corresponding inverse solution function is called to perform inverse solution for each case, and the cyclic joint angle inverse solution result is obtained. Based on the inverse kinematics solution of the cyclic joint angles, the robot joints are controlled to move to the current cyclic target point to obtain the single-point execution result. When the single-point execution result satisfies the termination condition corresponding to the loop count parameter, the loop execution result is output.

[0012] In one embodiment, the steps of collecting the robot's operating status based on the loop execution result and feeding it back to the user terminal, and calling a preset stop control function and a preset speed mutation suppression function when an abnormal situation is detected to obtain the abnormal handling result include: Based on the results of the loop execution, real-time operation status data is obtained, wherein the operation status data includes the robot's current pose parameters, joint angle parameters, motion speed parameters, and trajectory execution deviation parameters; The operational status data is compared with preset safety boundary parameters to obtain the status comparison result; When any of the following conditions are met: the state comparison result exceeds the safety boundary, the trajectory execution deviation is greater than the preset deviation threshold, the point list is empty, or a singular pose causes motion abnormality, an abnormal judgment result is obtained. Based on the anomaly type of the anomaly judgment result, a preset stop control function is invoked to terminate the robot's current motion, and a preset speed mutation suppression function is invoked to smooth the decay of the robot's current motion speed parameters, thereby obtaining the motion suppression result; An anomaly alert is generated based on the motion suppression result, and the anomaly alert and the running status data are fed back to the user terminal to obtain the anomaly handling result.

[0013] Furthermore, to achieve the above objectives, this application also proposes a robot teaching control device based on a large model, the robot teaching control device based on a large model comprising: The voice acquisition module is used to acquire the user's natural language teaching instructions and convert the natural language teaching instructions into text instructions; The semantic parsing module is used to input the preset robot control knowledge base and the text instructions into the large model for semantic parsing to obtain structured control instructions; The trajectory planning module is used to call a preset path planning function according to the structured control instructions to plan the trajectory of the target pose and obtain the planned trajectory. The singularity identification module is used to identify singular poses of the sampling points on the planned trajectory and obtain singular pose identification results. The inverse solution module is used to call the corresponding inverse solution function to perform case-by-case inverse solution based on the singular pose recognition result, and obtain the joint angle inverse solution result. The motion control module is used to control the movement of the robot joints based on the inverse kinematics results of the joint angles, and simultaneously store the current pose to the local memory module to form a point list; The loop execution module is used to control the robot to execute the point list cyclically according to the cyclic motion instructions in the structured control instructions, and obtain the cyclic execution result; The status feedback module is used to collect the robot's running status based on the loop execution results and feed it back to the user terminal. When an abnormal situation is detected, the preset stop control function and the preset speed change suppression function are called to obtain the abnormal handling result.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the robot teaching control method based on a large model as described above.

[0015] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the robot teaching control method based on a large model as described above.

[0016] This application converts voice teaching commands into text, combines a control knowledge base with a large model to parse structured instructions, sequentially completes trajectory planning and singular pose recognition, then solves inverse kinematics for different scenarios, drives robot movement and stores points locally, and finally achieves cyclic execution based on the point list, with full-process status monitoring and anomaly handling. This application supports natural language interaction, lowers the teaching threshold, handles singular poses in a hierarchical manner and suppresses sudden velocity changes, ensures smooth trajectory, and improves the stability and safety of robot teaching. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the first embodiment of the robot teaching and control method based on a large model according to this application. Figure 2 This is a flowchart illustrating the second embodiment of the robot teaching and control method based on a large model according to this application. Figure 3 This is a schematic diagram of the modular structure of the robot teaching and control device based on a large model in this application; Figure 4 This is a schematic diagram of the device structure of the hardware operating environment involved in the robot teaching control method based on a large model in the embodiments of this application.

[0019] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0021] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0022] Currently, mainstream robot teaching methods rely on manual operation with dedicated teach pendants or offline programming. Some voice control solutions only use keyword matching to recognize fixed sentence commands. Motion control generally follows a processing logic of first determining singular poses and then planning the trajectory. Point recording and cyclic movement are independent, and point information mostly relies on external model storage. During robot operation, it simply executes preset trajectories, lacking hierarchical processing mechanisms and speed fluctuation suppression measures for singular poses and areas near singular points. Traditional teaching methods have high operational barriers, require professional personnel, and involve cumbersome programming and point adjustment processes, resulting in low overall teaching efficiency. Keyword matching voice interaction cannot understand ambiguous natural language commands, leading to poor interaction flexibility. The logic of first determining singular poses and then planning trajectories causes a disconnect between trajectory planning and singularity handling. When the robot passes through singular points or their vicinity, it is prone to sudden speed changes, motion stuttering, trajectory jumps, and even safety issues such as equipment impacts and collisions. Furthermore, the low integration of point memory, trajectory planning, singularity handling, and cyclic control results in insufficient overall operational stability and safety controllability.

[0023] Therefore, how to leverage large models to achieve natural language speech teaching, while simultaneously completing trajectory planning, singular pose hierarchical processing, and velocity fluctuation suppression according to a reasonable time sequence, thereby improving the convenience of robot teaching and operational safety, has become an urgent problem to be solved.

[0024] Based on the above, this application also provides a robot teaching control method based on a large model, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the robot teaching and control method based on a large model according to this application.

[0025] In this embodiment, the robot teaching control method based on a large model includes steps S10 to S80: Step S10: Collect the user's natural language teaching instructions and convert the natural language teaching instructions into text instructions.

[0026] Specifically, first, the system collects natural language teaching commands from the user via a microphone array, such as "record current location" or "move to point A," obtaining the raw speech signal. Then, the controller performs noise reduction and enhancement processing on the raw speech signal to eliminate environmental noise interference, resulting in a denoised speech signal. Finally, the denoised speech signal is sent to the speech-to-text module, which performs speech recognition to obtain text commands for subsequent parsing. This is done because natural language is the most familiar interaction method for humans, allowing robot operation without specialized programming knowledge, significantly lowering the teaching threshold. Simultaneously, noise reduction processing ensures the accuracy of speech recognition in noisy industrial environments.

[0027] Step S20: Input the preset robot control knowledge base and text instructions into the large model for semantic parsing to obtain structured control instructions.

[0028] It should be noted that step S20 includes: obtaining action semantic information and parameter semantic information from the text instruction; reading action definition rules, parameter mapping rules, and safety boundary rules from the robot control knowledge base; matching the corresponding control rules according to the action semantic information and parameter semantic information to obtain semantic matching results; converting the semantic matching results into a format according to a preset structured control instruction parsing format to obtain initial structured control instructions; performing safety boundary verification on the initial structured control instructions to verify whether the position parameters, attitude parameters, velocity parameters, and acceleration parameters in the initial structured control instructions exceed the preset safety boundary parameters, and obtaining the verified structured control instructions.

[0029] It's important to understand that action semantic information is the operational intent extracted from text commands, clearly defining the type of robot action the user wants the robot to perform. Parameter semantic information is the action-related constraints extracted from text commands, including descriptions of position, posture, motion, and speed. The robot control knowledge base is a pre-built information set that includes robot action specification parameters and various judgment rules. Action definition rules are unified definitions of robot-executable actions, clarifying the function and execution logic of each type of action. Safety boundary rules are constraints that define the robot's normal operating range, limiting the legal intervals of various operating parameters. The structured control instruction parsing format is a pre-defined unified instruction layout format used to standardize the organization and content arrangement of instructions. Position parameters describe the robot's spatial coordinates, used to determine the robot's target spatial position. Posture parameters describe the robot's spatial rotation angle, used to determine the robot's spatial orientation when performing actions. Velocity parameters describe the robot's speed, used to limit the robot's travel rate during movement. Acceleration parameters describe the rate of change of the robot's velocity, used to limit the magnitude of acceleration and deceleration. The preset safety boundary parameters are the pre-defined limits of various robot operating indicators, used to define the numerical range for normal robot operation.

[0030] Specifically, first, the action semantic information and parameter semantic information are parsed from the acquired text instructions. For example, from "move to point A, speed 50", the action semantic "move" and the parameter semantic "target point A, speed 50 mm / s" are extracted. Then, action definition rules, parameter mapping rules, and safety boundary rules are read from a preset robot control knowledge base. For example, the function name, coordinate parameter format, speed range (1 to 100 mm / s), and safety boundary parameters corresponding to the "move" action are read. Second, the corresponding control rules are matched according to the action semantic information and parameter semantic information. For example, "move" is matched to a movement control function, "speed 50" is mapped to the value 50, and "point A" is mapped to a preset coordinate value to obtain the semantic matching result. Then, the semantic matching result is format-converted according to a preset structured control instruction parsing format. For example, it is converted into a standardized data structure containing function name, position parameter, attitude parameter, speed parameter, and acceleration parameter to obtain the initial structured control instruction. Finally, a safety boundary check is performed on the initial structured control commands. This check verifies whether the position, attitude, velocity, and acceleration parameters in the initial structured control commands exceed the safety boundary parameters. For example, it determines whether the velocity of 50 is within the allowable range and whether the target coordinates are within the workspace. The result is a verified structured control command. This is done because the large model is only responsible for semantic parsing and does not directly participate in control decisions. The rigid constraints of the knowledge base ensure that the parsing results comply with the robot's physical limitations and safety requirements, preventing the generation of dangerous commands that exceed the robotic arm's range of motion or speed limits, thus ensuring the inherent safety of the teaching process.

[0031] Step S30: According to the structured control instructions, the preset path planning function is called to plan the trajectory of the target pose and obtain the planned trajectory.

[0032] It should be noted that in step S30: the target pose parameters and motion parameters are extracted from the structured control instructions to obtain trajectory planning input data; the preset path planning function is called according to the trajectory planning input data to perform interpolation calculation on the motion path between the current pose and the target pose to obtain the initial trajectory curve; the initial trajectory curve is smoothed to obtain the smooth trajectory curve; the pose of the sampling points is extracted on the motion path according to the smooth trajectory curve and the preset sampling step size to obtain the planned trajectory.

[0033] It's important to understand that the target pose parameters are data used to determine the robot's final spatial position and orientation, including coordinate and rotation angle data. Motion parameters are data that defines the robot's motion state, including velocity and acceleration data. The preset path planning function is pre-configured computational logic that uses mature industry algorithms to calculate the robot's motion path. The target pose is the standard stationary pose of the robot recorded and saved during the teaching process, belonging to the preset work points.

[0034] Specifically, firstly, target pose parameters and motion parameters are extracted from the structured control instructions. The target pose parameters include the target position coordinates and the target attitude angle, while the motion parameters include the motion velocity and motion acceleration, thus obtaining trajectory planning input data. Then, based on the trajectory planning input data, a preset path planning function is called to interpolate the motion path between the current pose and the target pose. For example, in 3D space, a continuous curve connecting the start and end points is generated through polynomial interpolation or spline interpolation, resulting in an initial trajectory curve. Secondly, the initial trajectory curve is smoothed, for example, by using curvature constraint optimization or acceleration limiting filtering to eliminate abrupt inflection points in the trajectory, ensuring continuous velocity transition and obtaining a smooth trajectory curve. Finally, based on the smooth trajectory curve and a preset sampling step size (10 mm), the pose of sampling points is extracted segment by segment along the motion path. The pose of each sampling point includes position coordinates and attitude angles, thus obtaining the planned trajectory. This is done because planning the trajectory first and then processing subsequent issues aligns with the logical framework of robot motion control. Smoothing ensures seamless connection between the positions and postures of adjacent teaching points, while fixed-step sampling provides uniform and dense pose data for subsequent singular pose recognition, avoiding speed fluctuations and equipment wear caused by sudden trajectory changes.

[0035] Step S40: Perform singular pose identification on the poses of the sampling points on the planned trajectory to obtain the singular pose identification result.

[0036] It should be noted that step S40 includes: extracting the position coordinates and attitude angles of each sampling point from the planned trajectory to obtain the pose data to be identified; calculating the determinant value of the robot Jacobian matrix based on the pose data to be identified to obtain the pose singularity metric; comparing the pose singularity metric with a preset singular pose threshold to obtain the comparison result; and classifying and judging the singular pose based on the comparison result and the preset threshold near the singular point to obtain the singular pose recognition result.

[0037] Specifically, firstly, the position coordinates and attitude angles of each sampling point are extracted from the planned trajectory. The position coordinates contain coordinate values ​​in three-dimensional space, and the attitude angles contain rotation angles around each axis, thus obtaining the pose data to be identified. Then, the determinant of the robot's Jacobian matrix is ​​calculated based on the pose data. The Jacobian matrix describes the mapping relationship between the robot's joint velocities and the end effector velocities, and its determinant value reflects the degree to which the pose approaches a singular state, thus obtaining the pose singularity metric. Secondly, the pose singularity metric is compared with a preset singular pose threshold (e.g., 0.01). If the absolute value of the determinant is less than this threshold, it is determined to be a singular pose, and the comparison result is obtained. Finally, based on the comparison results and a preset singularity threshold (e.g., 0.05), a graded judgment is made. If the absolute value of the determinant is less than the singular pose threshold, it is determined to be a singular pose; if it is between the singular pose threshold and the singularity threshold, it is determined to be a singularity pose; and if it is greater than the singularity threshold, it is determined to be a normal pose, thus obtaining the singular pose recognition result. This is done because when the robot encounters a singular pose during movement, the joint velocity increases sharply, which may lead to mechanical impact or even equipment damage. Early identification of singular poses is a key step in ensuring teaching safety. The graded judgment provides a basis for subsequent differentiated inverse kinematics solutions, ensuring that poses with different levels of danger are handled in a targeted manner.

[0038] Step S50: Based on the singular pose recognition result, call the corresponding inverse solution function to perform case-by-case inverse solution solving to obtain the joint angle inverse solution result.

[0039] It should be noted that step S50 includes: parsing the singular pose recognition result to obtain pose classification information; when the pose classification information is a normal pose, calling a preset conventional inverse kinematics function to perform conventional inverse kinematics solution to obtain a first inverse kinematics result; when the pose classification information is a singular pose, calling a preset singular point inverse kinematics function to perform singular pose inverse kinematics solution to obtain a second inverse kinematics result; when the pose classification information is a pose near a singular point, calling a preset velocity mutation suppression function to replan the planned trajectory to obtain a replanned trajectory, and calling a preset conventional inverse kinematics function to perform inverse kinematics solution on the pose of the sampling points on the replanned trajectory to obtain a third inverse kinematics result; and using the first inverse kinematics result, the second inverse kinematics result, or the third inverse kinematics result as the joint angle inverse kinematics result.

[0040] It's important to understand that the preset conventional inverse kinematics function is a pre-configured computational logic for normal poses, used to complete kinematic solutions under standard conditions. This function is the core computational carrier for acquiring joint data in normal scenarios. The preset singularity inverse kinematics function is a dedicated computational logic pre-configured for singular poses, adapting to the solution requirements under special poses. This function specifically addresses the issue of general solution methods failing under singular poses. Pose near singular points are transitional poses close to the singular pose interval, prone to abnormal phenomena such as sudden velocity changes. For these poses, the trajectory needs to be adjusted before performing the solution computation. The preset velocity change suppression function is a pre-configured computational logic that can adjust the motion path to mitigate the problem of drastic velocity changes. This function is used to optimize the trajectory and avoid operational failures around singular points.

[0041] Specifically, firstly, the singular pose recognition results are analyzed to extract the pose classification information contained therein, i.e., determining whether each sampling point belongs to a normal pose, a singular pose, or a pose near a singular point. Then, when the pose classification information is a normal pose, a preset conventional inverse kinematics function is called to perform conventional inverse kinematics solving. The joint angles are calculated directly from the end-effector pose parameters using a numerical iteration method, yielding the first inverse kinematics result. Secondly, when the pose classification information is a singular pose, a preset singular point inverse kinematics function is called to perform singular pose inverse kinematics solving. Damped least squares is used to regularize the Jacobian matrix, suppressing joint velocity divergence while maintaining end-effector pose accuracy, yielding the second inverse kinematics result. Then, when the pose classification information is a pose near a singular point, a preset velocity mutation suppression function is called to replan the planned trajectory. By introducing preset transition curve parameters to adjust the interpolation function or adding intermediate transition points to change the local shape of the trajectory, the trajectory still passes through the teaching point but avoids the influence area of ​​the singular point, resulting in a replanned trajectory. A preset conventional inverse kinematics function is then called to solve the inverse kinematics of the sampled point poses on the replanned trajectory, obtaining the third inverse kinematics result. Finally, the first, second, or third inverse kinematics result is used as the joint angle inverse kinematics result for subsequent joint control. This is done because different pose types require differentiated inverse kinematics strategies. Direct solution for normal poses is the most efficient. For singular point poses, damped least squares method is used to avoid numerical instability when inverting the Jacobian matrix. For singular point poses, the geometric characteristics of the motion path are changed by replanning the trajectory, turning the pose that was originally close to a singular into a normal pose, thus reusing the conventional inverse kinematics function. This ensures smooth motion while reducing computational complexity. Handling these three cases separately ensures that the robot can move safely and smoothly in any pose.

[0042] In step S60, the robot joints are controlled to move according to the inverse kinematics solution of the joint angles, and the current pose is simultaneously stored in the local memory module to form a point list.

[0043] Specifically, firstly, joint motion control commands are generated based on the inverse kinematics (IK) results of the joint angles. These commands include target angles and velocity parameters for each joint. The commands are then sent to the robot driver, which drives the joint motors to rotate according to the commands, controlling the robot's end effector to move to the target pose. Next, during robot movement, the robot controller collects current pose parameters in real time. These parameters include current position coordinates, current attitude angles, current velocity, and a singular pose identifier. The current pose parameters are then associated with and stored in the local memory module along with the corresponding motion parameters, including linear velocity and linear acceleration, resulting in point-based data. Finally, the point-based data is added to the point list in the local memory module in the order of recording. Each point-based data entry in the list includes position coordinates, attitude angles, velocity, acceleration, and a singular pose identifier, forming a traceable set of teaching points. This is done because the inverse kinematics solution of joint angles serves as a bridge connecting trajectory planning and physical execution. The actuator directly controls the motor to achieve precise movement based on the joint angle commands, while simultaneously storing pose data in real time to build a local point list. This ensures that the teaching process does not rely on the memory of a large model, guaranteeing the controllability and reliability of the point data. It also supports subsequent cyclic execution and point editing, achieving a closed loop between teaching and execution.

[0044] Step S70: Control the robot to execute the point list cyclically according to the cyclic motion instructions in the structured control instructions, and obtain the cyclic execution result.

[0045] It should be noted that the cyclic motion instruction is parsed from the structured control instruction to obtain the cyclic control parameters, which include the cycle count parameter and the cycle mode parameter. Based on the cyclic control parameters, the point list is retrieved from the local memory module, and the trajectory planning parameters associated with each point in the point list are extracted to obtain a set of reused trajectory parameters. Based on the set of reused trajectory parameters, a preset path planning function is called to perform trajectory reuse planning, resulting in a cyclic trajectory. Singular pose identification is performed on the poses of the sampled points on the cyclic trajectory to obtain the cyclic singular pose identification result. Based on the cyclic singular pose identification result, the corresponding inverse kinematics function is called to perform case-by-case inverse kinematics solution to obtain the cyclic joint angle inverse kinematics result. Based on the cyclic joint angle inverse kinematics result, the robot joints are controlled to move to the current cyclic target point to obtain the single-point execution result. When the single-point execution result meets the termination condition corresponding to the cycle count parameter, the cyclic execution result is output.

[0046] It's important to understand that the local memory module is an independently operating local storage area specifically for storing various point and trajectory-related data generated during the teaching process. This area does not rely on external models, ensuring stable storage and retrieval of point data. The reused trajectory parameter set is a unified data set formed by integrating all trajectory planning parameters from the point list, and is used uniformly for path calculation in the iterative phase. This set provides complete historical parameter support for iterative trajectory planning.

[0047] Specifically, firstly, the cyclic motion instruction is parsed from the structured control instructions to extract the cycle count parameter and cycle mode parameter. The cycle mode parameter includes two modes: fixed-cycle and infinite cycle, thus obtaining the cyclic control parameters. Then, based on the cyclic control parameters, the point list is retrieved from the local memory module, and the trajectory planning parameters associated with each point in the point list are extracted. The trajectory planning parameters include the interpolation function type, smoothing coefficient, and planning step size between adjacent points, resulting in a set of reused trajectory parameters. Secondly, based on the set of reused trajectory parameters, a preset path planning function is called to perform trajectory reuse planning. The motion trajectory between adjacent points is regenerated according to the interpolation function type and parameters determined in the teaching phase, resulting in the cyclic trajectory. Then, singular pose identification is performed on the poses of the sampling points on the cyclic trajectory. The same singular pose determination logic as in the teaching phase is used to calculate the pose singularity metric value of each sampling point and perform hierarchical determination, resulting in the cyclic singular pose identification result. Then, based on the cyclic singular pose recognition results, the corresponding inverse kinematics function is called to perform case-specific inverse kinematics solutions. When the recognition result is a normal pose, a preset conventional inverse kinematics function is called; when the recognition result is a singular pose, a preset singular point inverse kinematics function is called; when the recognition result is a pose near a singular point, a preset velocity mutation suppression function is called to replan the trajectory, and then the preset conventional inverse kinematics function is called to obtain the cyclic joint angle inverse kinematics results. Next, joint motion control commands are generated based on the cyclic joint angle inverse kinematics results and sent to the robot driver to control the robot joints to move to the current cyclic target point, obtaining the single-point execution result. Finally, it is determined whether the single-point execution result meets the termination condition corresponding to the loop count parameter. If the current number of times the point list has been traversed has reached the loop count parameter, the cyclic execution result is output and the cyclic motion terminates; if the current number of traversals has not reached the loop count parameter, the step of retrieving the point list is returned, and the cyclic execution continues to traverse the next target point until the termination condition is met. This is done because reusing the trajectory parameters generated during the teaching phase can avoid repeated planning and calculation, improve the efficiency of loop execution, and at the same time, singular pose identification and inverse solution for each loop are performed again to ensure that pose safety is continuously verified during loop execution. The judgment of meeting the termination condition ensures that the loop motion is executed accurately according to the preset number of times, and if it is not met, the loop will automatically continue to achieve automated repetitive operation.

[0048] Step S80: Collect the robot's running status based on the loop execution results and feed it back to the user terminal. When an abnormal situation is detected, call the preset stop control function and the preset speed change suppression function to obtain the abnormal handling result.

[0049] Specifically, firstly, the robot's operational status data is collected in real time based on the loop execution results. This data includes the current position coordinates, current attitude angles, joint angles, actual movement speed, trajectory execution deviation, and singular pose identifier status. Then, the operational status data is compared item by item with preset safety boundary parameters, including the position boundary range, attitude angle limit range, speed upper limit threshold, and trajectory deviation allowable threshold, yielding a status comparison result. Secondly, based on the status comparison result, anomalies are determined. Anomalies include at least one of the following: actual position exceeding the position boundary range, actual attitude angle exceeding the attitude angle limit range, actual movement speed exceeding the speed upper limit threshold, trajectory execution deviation exceeding the trajectory deviation allowable threshold, an empty point list, or an abnormal singular pose identifier status, resulting in an anomaly judgment result. Then, when the anomaly judgment result indicates the presence of an anomaly, a preset stop control function is called to send an emergency stop command to the robot driver. Simultaneously, a preset speed mutation suppression function is called to smoothly decay the current movement speed, yielding a motion suppression result. Finally, based on the motion suppression results, an anomaly alert is generated, containing an anomaly type identifier and current operating status data. This alert and data are then fed back to the user terminal to obtain the anomaly handling result. This is done because real-time status feedback allows the user to monitor the robot's operation, immediate anomaly detection and graded braking mechanisms effectively prevent equipment damage and safety accidents, and smooth speed decay avoids mechanical shocks from sudden stops, ensuring the stability and controllability of the entire teaching process.

[0050] This embodiment converts voice teaching commands into text, combines a control knowledge base with a large model to parse structured instructions, sequentially completes trajectory planning and singular pose recognition, then solves inverse kinematics for different scenarios, drives robot movement and stores points locally, and finally achieves cyclic execution based on the point list, with full-process status monitoring and anomaly handling. This application supports natural language interaction, lowers the teaching threshold, handles singular poses in a hierarchical manner and suppresses sudden velocity changes, ensures smooth trajectory, and improves the stability and safety of robot teaching.

[0051] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 The robot teaching control method based on a large model, step S80, further includes steps S201 to S205: Step S201: Obtain running status data in real time based on the results of the loop execution.

[0052] It should be noted that the operational status data includes the robot's current pose parameters, joint angle parameters, motion speed parameters, and trajectory execution deviation parameters. The current pose parameters include the current position coordinates and the current attitude angle; the joint angle parameters include the actual rotation angle of each joint; the motion speed parameters include the actual rotation speed of each joint and the actual linear velocity of the end effector; and the trajectory execution deviation parameters include the position deviation and attitude deviation values ​​between the actual pose and the planned pose.

[0053] Specifically, firstly, operational status data is acquired in real time based on the results of the loop execution. Then, the operational status data is compared item by item with preset safety boundary parameters, including the position boundary range, attitude angle limit range, speed upper limit threshold, and trajectory deviation allowable threshold, to obtain the status comparison result. Secondly, based on the status comparison result, it is determined whether there are any abnormal situations. Abnormal situations include at least one of the following: the current position coordinates exceed the position boundary range, the current attitude angle exceeds the attitude angle limit range, the motion speed parameter exceeds the speed upper limit threshold, the trajectory execution deviation parameter exceeds the trajectory deviation allowable threshold, the point list is empty, or the singular pose identifier is abnormal, to obtain the anomaly judgment result. Then, when the anomaly judgment result indicates the existence of an abnormal situation, a preset stop control function is called to send an emergency stop command to the robot driver, and a preset speed mutation suppression function is called to smoothly decay the current motion speed parameter, to obtain the motion suppression result. Finally, based on the motion suppression result, an anomaly prompt message is generated, which includes an anomaly type identifier and operational status data. The anomaly prompt message and operational status data are then fed back to the user terminal to obtain the anomaly handling result. This is done because real-time acquisition of operational status data ensures that the controller understands the actual operation of the robot, and the immediate detection of abnormal situations and graded braking mechanisms effectively prevent equipment damage and safety accidents. Smooth speed decay avoids mechanical shock caused by sudden stops, ensuring the stability and controllability of the entire teaching process.

[0054] Step S202: Compare the running status data with the preset safety boundary parameters to obtain the status comparison result.

[0055] Specifically, the operational status data is first compared item by item with preset safety boundary parameters, which include the position boundary range, attitude angle limit range, speed upper limit threshold, and trajectory deviation allowable threshold. The specific comparison process is as follows: the current position coordinates are compared with the position boundary range to determine if they exceed the preset workspace boundary; the current attitude angle is compared with the attitude angle limit range to determine if it exceeds the allowable range for each joint's attitude angle; the actual rotational speed of each joint and the actual linear speed of the end effector are compared with the speed upper limit threshold to determine if the motion speed parameter exceeds the safe speed limit; and the position deviation value and attitude deviation value are compared with the trajectory deviation allowable threshold to determine if the trajectory execution deviation parameter is within the allowable error range. Through the above item-by-item comparison, a status comparison result is obtained, which includes a comparison status identifier for each parameter item, including normal status identifiers and abnormal status identifiers. This is done because there are many potential risk factors during robot operation, including mechanical overtravel, speed runaway, and trajectory deviation. Each item is compared to ensure that each safety dimension is monitored independently. Any parameter exceeding the safety boundary can be identified in a timely manner, providing an accurate data basis for subsequent anomaly judgment and avoiding the risk of missed detection caused by monitoring a single threshold.

[0056] Step S203: When the state comparison result exceeds the safety boundary, the trajectory execution deviation is greater than the preset deviation threshold, the point list is empty, or the singular pose causes motion abnormality, an abnormal judgment result is obtained.

[0057] Specifically, the judgment logic is as follows: when the current position coordinates exceed the position boundary range, or the current attitude angle exceeds the attitude angle limit range, or the motion speed parameter exceeds the speed upper limit threshold, it is determined that the safety boundary has been exceeded; when the position deviation value is greater than the trajectory deviation allowable threshold, or the attitude deviation value is greater than the attitude deviation allowable threshold, it is determined that the trajectory execution deviation is greater than the preset deviation threshold; when no point record data is stored in the local memory module, it is determined that the point list is empty; when the singular pose identifier is abnormal and accompanied by a sharp fluctuation in joint speed parameters, it is determined that the singular pose has caused a motion abnormality. If any of the above four judgments is true, an abnormal judgment result is obtained, which includes an abnormality type identifier and abnormality triggering parameter information. This is done because multiple abnormal factors may occur independently or in combination during robot operation, including mechanical overtravel, trajectory deviation, data loss, and singular pose loss of control. Judging each factor individually and comprehensively ensures that any abnormal situation can be captured in a timely manner, providing accurate abnormality type information for subsequent emergency braking and smooth speed decay, ensuring the targetedness and effectiveness of abnormality handling.

[0058] Step S204: Based on the anomaly type of the anomaly judgment result, call the preset stop control function to terminate the robot's current motion, and call the preset speed mutation suppression function to perform smooth decay processing on the robot's current motion speed parameters to obtain motion suppression results.

[0059] Specifically, the anomaly detection results are first analyzed to extract anomaly type identifiers, including those for exceeding safety boundaries, excessive trajectory execution deviation, missing point list, and singular pose motion anomalies. Then, based on the anomaly type identifier, a preset stop control function is invoked to terminate the robot's current motion. Specifically, an emergency stop command is sent to the robot driver, containing braking signals and power-off protection signals for each joint motor, causing the robot driver to immediately cut off the power output of each joint motor and activate the mechanical braking device. Simultaneously, a preset speed mutation suppression function is invoked to smoothly decay the robot's current motion speed parameters. Specifically, a decay gradient is calculated based on the current motion speed parameters, and the actual rotational speed of each joint and the actual linear velocity of the end effector are gradually reduced according to a preset decay time (e.g., 100 milliseconds) to avoid mechanical shock and vibration caused by a sudden drop in speed to zero. Through the coordinated processing of the above stop control and smooth decay, a motion suppression result is obtained, which includes a stop completion status identifier and the final speed parameters. This is done because different types of anomalies require a unified emergency response mechanism to immediately stop movement and prevent the anomaly from escalating and causing equipment damage or personal injury. Smooth speed decay rather than instantaneous emergency stop avoids secondary damage to the robotic arm joints and end effector caused by inertial impact. The coordinated invocation of the two functions takes into account both safety and equipment protection.

[0060] Step S205: Generate an abnormal prompt message based on the motion inhibition result, and feed back the abnormal prompt message and running status data to the user terminal to obtain the abnormal handling result.

[0061] Specifically, the motion suppression results are first analyzed to extract the stop completion status identifier and final velocity parameters, confirming that the robot's current motion has terminated and the speed has decayed to a safe level. Then, anomaly alerts are generated based on the motion suppression results. These alerts include anomaly type identifier, anomaly trigger time, anomaly trigger parameter information, and the stop completion status identifier. The anomaly trigger parameter information includes the specific parameter values ​​that triggered the anomaly and preset safety boundary parameters. Finally, the anomaly alerts and operational status data are packaged into an anomaly feedback data packet and sent to the user terminal via a preset communication link. The user terminal displays the anomaly feedback data packet visually and issues audible and visual alarms, thus obtaining the anomaly handling result. This is done because after anomaly handling, the cause of the anomaly and the handling status need to be clearly communicated to the user, facilitating timely understanding of the fault information and subsequent measures by operators. Simultaneously, the complete operational status data is fed back to support user anomaly tracing and analysis, improving the maintainability of the teaching system and the efficiency of human-machine collaboration.

[0062] This embodiment collects real-time operational status data such as pose, joint angles, speed, and trajectory deviation during the robot's cyclical operation. This data is compared with safety parameters to determine operational anomalies from multiple dimensions. Upon triggering an anomaly, the device's movement is immediately terminated, and the movement speed is smoothly reduced. Finally, a prompt message is generated and data is transmitted back. This allows for comprehensive troubleshooting of various operational faults, timely handling of abnormal conditions, effective avoidance of sudden speed changes and equipment damage, and ensures the robot's safe and stable operation throughout the entire process.

[0063] Based on the first embodiment of this application, this application also provides a robot teaching and control device based on a large model. Please refer to... Figure 3 The device includes: The voice acquisition module 10 is used to acquire the user's natural language teaching instructions and convert the natural language teaching instructions into text instructions.

[0064] The semantic parsing module 20 is used to input the preset robot control knowledge base and text instructions into the large model for semantic parsing to obtain structured control instructions.

[0065] The trajectory planning module 30 is used to call a preset path planning function according to the structured control instructions to plan the trajectory of the target pose and obtain the planned trajectory.

[0066] The singularity recognition module 40 is used to identify singular poses of sampling points on the planned trajectory and obtain singular pose recognition results.

[0067] The inverse solution module 50 is used to call the corresponding inverse solution function to perform case-by-case inverse solution based on the singular pose recognition result, and obtain the joint angle inverse solution result.

[0068] The motion control module 60 is used to control the movement of the robot joints based on the inverse kinematics of the joint angles, and simultaneously store the current pose to the local memory module to form a point list.

[0069] The loop execution module 70 is used to control the robot to execute the point list in a loop according to the loop motion instructions in the structured control instructions, and obtain the loop execution result.

[0070] The status feedback module 80 is used to collect the robot's running status based on the results of the loop execution and feed it back to the user terminal. When an abnormal situation is detected, the preset stop control function and the preset speed change suppression function are called to obtain the abnormal handling result.

[0071] The robot teaching control device based on a large model provided in this application, employing the robot teaching control method based on a large model in the above embodiments, can solve the technical problem of how to rely on a large model to achieve natural language speech interaction and reduce the threshold of robot teaching operations. Compared with the prior art, the beneficial effects of the robot teaching control device based on a large model provided in this application are the same as the beneficial effects of the robot teaching control method based on a large model provided in the above embodiments, and other technical features in the robot teaching control device based on a large model are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0072] This application provides a robot teaching control device based on a large model. The robot teaching control device based on a large model includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the robot teaching control method based on the large model in the above embodiment 1.

[0073] The following is for reference. Figure 4 This document illustrates a structural schematic diagram of a large-model-based robot teaching and control device suitable for implementing embodiments of this application. The large-model-based robot teaching and control device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 4 The robot teaching and control device based on a large model shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0074] like Figure 4 As shown, the robot teaching control device based on a large model may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the robot teaching control device based on the large model. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the large-model-based robot teaching control device to wirelessly or wiredly communicate with other devices to exchange data. Although various large-model-based robot teaching control devices are shown in the figures, it should be understood that it is not required to implement or possess all of them. More or fewer may be implemented alternatively.

[0075] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0076] The robot teaching control device based on a large model provided in this application, employing the robot teaching control method based on a large model in the above embodiments, can solve the technical problem of how to rely on a large model to achieve natural language speech interaction and reduce the threshold of robot teaching operation. Compared with the prior art, the beneficial effects of the robot teaching control device based on a large model provided in this application are the same as the beneficial effects of the robot teaching control method based on a large model provided in the above embodiments, and other technical features in this robot teaching control device based on a large model are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0077] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0078] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0079] This application provides a computer-readable medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the robot teaching control method based on a large model in the above embodiments.

[0080] The computer-readable medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor devices, or any combination thereof. More specific examples of computer-readable media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable medium may be any tangible medium containing or storing a program that can be executed by instructions, used by a device, or used in conjunction with it. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0081] The aforementioned computer-readable medium may be included in a large-scale robot teaching and control device; or it may exist independently and not be assembled into a large-scale robot teaching and control device.

[0082] The aforementioned computer-readable medium carries one or more programs that, when executed by a large-model-based robot teaching and control device, enable the large-model-based robot teaching and control device to write computer program code for performing the operations of this application in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0083] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, all blocks in the flowcharts or block diagrams may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that all blocks in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using dedicated hardware-based implementations that perform the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0084] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0085] The readable medium provided in this application is a computer-readable medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-described robot teaching control method based on a large model. This solves the technical problem of how to leverage a large model to achieve natural language speech interaction and reduce the barrier to robot teaching operations. Compared with the prior art, the beneficial effects of the computer-readable medium provided in this application are the same as those of the robot teaching control method based on a large model provided in the above embodiments, and will not be repeated here.

[0086] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the large-model-based robot teaching control method described above.

[0087] The computer program product provided in this application can solve the technical problem of how to reduce the threshold of robot teaching operations by relying on large models to realize natural language voice interaction. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the robot teaching control method based on large models provided in the above embodiments, and will not be repeated here.

[0088] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A robot teaching control method based on a large model, characterized in that, The method includes: Collect user natural language teaching instructions and convert the natural language teaching instructions into text instructions; The preset robot control knowledge base and the text instructions are input into the large model for semantic parsing to obtain structured control instructions; The target pose is trajectory planned by calling a preset path planning function according to the structured control command, and the planned trajectory is obtained. Singular pose identification is performed on the poses of the sampling points on the planned trajectory to obtain singular pose identification results; Based on the singular pose recognition result, the corresponding inverse solution function is called to perform inverse solution for each case, and the inverse solution result of the joint angle is obtained. The robot joints are controlled to move according to the inverse kinematics of the joint angles, and the current pose is simultaneously stored in the local memory module to form a point list. The robot is controlled to execute the point list cyclically according to the cyclic motion instructions in the structured control instructions, and the cyclic execution result is obtained. The robot's operating status is collected based on the results of the loop execution and fed back to the user terminal. When an abnormal situation is detected, the preset stop control function and the preset speed change suppression function are called to obtain the abnormal handling result.

2. The method as described in claim 1, characterized in that, The step of inputting the preset robot control knowledge base and the text instructions into the large model for semantic parsing to obtain structured control instructions includes: Obtain the action semantic information and parameter semantic information from the text instructions; Read action definition rules, parameter mapping rules, and safety boundary rules from the robot control knowledge base; Based on the action semantic information and the parameter semantic information, the corresponding control rules are matched to obtain the semantic matching result; The semantic matching result is converted according to the preset structured control instruction parsing format to obtain the initial structured control instruction; The initial structured control command is subjected to safety boundary verification to check whether the position parameters, attitude parameters, velocity parameters and acceleration parameters in the initial structured control command exceed the preset safety boundary parameters, thereby obtaining the verified structured control command.

3. The method as described in claim 1, characterized in that, The step of calling a preset path planning function according to the structured control command to plan the trajectory of the target pose and obtain the planned trajectory includes: The target pose parameters and motion parameters are extracted from the structured control commands to obtain trajectory planning input data; Based on the trajectory planning input data, a preset path planning function is called to perform interpolation calculation on the motion path between the current pose and the target pose to obtain the initial trajectory curve. The initial trajectory curve is smoothed to obtain a smoothed trajectory curve; Based on the smooth trajectory curve and the preset sampling step size, the pose of the sampling points is extracted on the motion path to obtain the planned trajectory.

4. The method as described in claim 1, characterized in that, The step of performing singular pose identification on the poses of sampling points on the planned trajectory to obtain singular pose identification results includes: The position coordinates and attitude angles of each sampling point are extracted from the planned trajectory to obtain the pose data to be identified. The determinant of the robot Jacobian matrix is ​​calculated based on the pose data to be identified, and the pose singularity metric is obtained. The pose singularity metric is compared with a preset singular pose threshold to obtain the comparison result; Based on the comparison results and the preset threshold near the singular point, a hierarchical judgment is made to obtain the singular pose recognition result.

5. The method as described in claim 1, characterized in that, The step of calling the corresponding inverse solution function to perform case-by-case inverse solution solving based on the singular pose recognition result to obtain the joint angle inverse solution result includes: The singular pose recognition results are analyzed to obtain pose classification information; When the pose classification information is a normal pose, the preset conventional inverse kinematics function is called to perform conventional inverse kinematics solution to obtain the first inverse kinematics result; When the pose classification information is a singular pose, the preset singular point inverse kinematics function is called to perform the singular pose inverse kinematics solution and obtain the second inverse solution result. When the pose classification information is a pose near a singular point, the preset velocity mutation suppression function is called to replan the planned trajectory to obtain the replanned trajectory, and the preset conventional inverse solution function is called to solve the inverse solution of the pose of the sampling point on the replanned trajectory to obtain the third inverse solution result. The joint angle inverse solution result is obtained by taking the first inverse solution result, the second inverse solution result, or the third inverse solution result.

6. The method as described in claim 1, characterized in that, The step of controlling the robot to cyclically execute according to the point list based on the cyclic motion instructions in the structured control instructions, and obtaining the cyclic execution result, includes: The cyclic motion instruction is parsed from the structured control instruction to obtain the cyclic control parameters, wherein the cyclic control parameters include the number of cycles parameter and the cyclic mode parameter; The point list is retrieved from the local memory module according to the loop control parameters, and the trajectory planning parameters associated with each point in the point list are extracted to obtain a set of reused trajectory parameters. Based on the set of reused trajectory parameters, a preset path planning function is called to perform trajectory reuse planning, resulting in a loop trajectory. Singular pose identification is performed on the poses of the sampling points on the cyclic trajectory to obtain the cyclic singular pose identification result. Based on the cyclic singular pose recognition result, the corresponding inverse solution function is called to perform inverse solution for each case, and the cyclic joint angle inverse solution result is obtained. Based on the inverse kinematics solution of the cyclic joint angles, the robot joints are controlled to move to the current cyclic target point to obtain the single-point execution result. When the single-point execution result satisfies the termination condition corresponding to the loop count parameter, the loop execution result is output.

7. The method as described in claim 1, characterized in that, The steps of collecting the robot's operating status based on the loop execution result and feeding it back to the user terminal, and calling the preset stop control function and the preset speed mutation suppression function when an abnormal situation is detected to obtain the abnormal handling result include: Based on the results of the loop execution, real-time operation status data is obtained, wherein the operation status data includes the robot's current pose parameters, joint angle parameters, motion speed parameters, and trajectory execution deviation parameters; The operational status data is compared with preset safety boundary parameters to obtain the status comparison result; When any of the following conditions are met: the state comparison result exceeds the safety boundary, the trajectory execution deviation is greater than the preset deviation threshold, the point list is empty, or a singular pose causes motion abnormality, an abnormal judgment result is obtained. Based on the anomaly type of the anomaly judgment result, a preset stop control function is invoked to terminate the robot's current motion, and a preset speed mutation suppression function is invoked to smooth the decay of the robot's current motion speed parameters, thereby obtaining the motion suppression result; An anomaly alert is generated based on the motion suppression result, and the anomaly alert and the running status data are fed back to the user terminal to obtain the anomaly handling result.

8. A robot teaching and control device based on a large model, characterized in that, The device includes: The voice acquisition module is used to acquire the user's natural language teaching instructions and convert the natural language teaching instructions into text instructions; The semantic parsing module is used to input the preset robot control knowledge base and the text instructions into the large model for semantic parsing to obtain structured control instructions; The trajectory planning module is used to call a preset path planning function according to the structured control instructions to plan the trajectory of the target pose and obtain the planned trajectory. The singularity identification module is used to identify singular poses of the sampling points on the planned trajectory and obtain singular pose identification results. The inverse solution module is used to call the corresponding inverse solution function to perform case-by-case inverse solution based on the singular pose recognition result, and obtain the joint angle inverse solution result. The motion control module is used to control the movement of the robot joints based on the inverse kinematics results of the joint angles, and simultaneously store the current pose to the local memory module to form a point list; The loop execution module is used to control the robot to execute the point list cyclically according to the cyclic motion instructions in the structured control instructions, and obtain the cyclic execution result; The status feedback module is used to collect the robot's running status based on the loop execution results and feed it back to the user terminal. When an abnormal situation is detected, the preset stop control function and the preset speed change suppression function are called to obtain the abnormal handling result.

9. A robot teaching and control device based on a large model, characterized in that, The device includes: a memory, a processor, and a large-model-based robot teaching control program stored in the memory and running on the processor, the large-model-based robot teaching control program being configured to implement the steps of the large-model-based robot teaching control method as described in any one of claims 1-7.

10. A storage medium, characterized in that, The storage medium stores a robot teaching control program based on a large model, which, when executed by a processor, implements the steps of the robot teaching control method based on a large model as described in any one of claims 1-7.

Images