Robot teaching programming method and system based on global input

Abstract

The invention relates to the technical field of industrial robot intelligent programming and man-machine interaction, in particular to a robot teaching programming method and system based on global input. The system comprises a task intention interaction module, a semantic constraint matching module, an equipment modeling module, a trajectory optimization module, an interaction verification module and a cross-platform deployment module. According to the method, through global task modeling and semantic abstraction, a user does not need point-by-point teaching or manual segmentation, and the high-quality track can be generated through one key only by defining the key frame at a time and adding the behavior semantic label. According to the method, normal form upgrading from traditional'segmented input, local interpolation and post-processing smoothing 'to'global input, unified optimization and essential smoothing' is achieved, and high-level task semantics are automatically converted into a bottom-layer smooth track through a three-layer decoupling framework of a task intention layer, a semantic planning layer and a physical execution layer. A semantic planning layer is deeply fused with a multi-modal constraint template and a high-performance optimization algorithm, on the premise of strictly meeting dynamic constraints, the total time of generated tracks can be shortened by 15%-25%, the acceleration peak value is reduced by 40%-50%, ultra-smooth control is achieved from a motion instruction source, and mechanical vibration and impact are effectively restrained. According to the method, the complexity of teaching programming is remarkably reduced, the track smoothness and the system transportability are improved, and the method is particularly suitable for rapid deployment and efficient programming of middle-end and low-end industrial robots in typical process scenes such as assembly, carrying and detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial robot intelligent programming and human-computer interaction, and particularly relates to a robot teaching programming method and system based on global input. BACKGROUND

[0002] In the current field of teaching programming of low-end industrial robots, the traditional "segmented input, local interpolation, post-processing smoothing" trajectory generation mode is generally used. This paradigm requires the operator to manually decompose the continuous production task into a large number of discrete path segments, record the trajectory point by point, independently set the posture and speed, and then connect them by interpolation algorithm, and finally realize smoothing through post-processing. This method is simple in structure and strong in universality, but has the following defects: First, the teaching process is tedious and low in intelligence. Whether it depends on the coordinate acquisition of a specific hardware device (such as CN110587579A based on Kinect teaching point positioning) or extracts path points through a simulation environment (such as CN113836702B in a virtual environment), its essence still does not deviate from the category of "point-level programming". The operator still needs to pay attention to and define the spatial position of each path point, and cannot express and input the overall intention from the "task semantic" level, resulting in low programming efficiency and serious dependence on manual experience.

[0003] Second, the input information and motion planning are seriously separated. Existing methods (such as CN110757447A) record the trajectory through a movable device and convert the coordinates, which simplifies the point acquisition process, but only generates discrete point sequences or preliminary paths. These input information itself does not contain any high-level constraints such as motion smoothness, time optimization or dynamics feasibility, resulting in the trajectory planning link having to start from a "blind" or "poor" starting point, making it difficult to generate a globally optimized high-quality trajectory.

[0004] Finally, the existing architecture lacks global optimization and cross-platform collaboration. Most teaching systems regard "path input", "trajectory optimization" and "code generation" as independent serial links. For example, CN113836702B realizes program cross-platform through syntax tree conversion, but its trajectory planning still relies on local interpolation and post-processing in the simulation environment, and does not deeply integrate with high-performance global optimization algorithms. This architectural separation makes it difficult for the final program to achieve global optimization in terms of motion accuracy, efficiency and smoothness, and to adapt to different performance requirements of robot hardware.

[0005] In summary, the prior art is plagued by the fundamental limitations of the traditional paradigm, and there is an urgent need to realize the fundamental paradigm shift from "segmented input, local interpolation, post-processing smoothing" to "global input, unified optimization, and essential smoothing". The existing scheme is either plagued by the cumbersome operation of point-level input or cannot generate globally optimized high-performance trajectories due to architectural limitations, and none of them can fundamentally realize the intelligent teaching of "directly driving the robot to generate the optimal motion according to the user's intention".

[0006] Therefore, the present application proposes a robot teaching programming method and system based on global input. SUMMARY

[0007] In order to make up for the shortcomings of the prior art and solve the technical problems in the background art, the present application proposes a robot teaching programming method and system based on global input.

[0008] The present application is realized by the following technical solutions: A robot teaching programming method based on global input, comprising the following steps: S1: receiving a user input task key frame sequence and corresponding behavior semantic label, integrating the key frame sequence and the behavior semantic label to generate a structured task description file, the behavior semantic label including motion characteristic description, process requirement description and safety priority description; S2: according to the behavior semantic label in the task description file, matching and activating the corresponding constraint template from the preset multi-modal constraint template library, the constraint template at least covering kinematics smoothing template, dynamics safety template, process constraint template and geometric safety template; S3: loading the device parameter information of the target robot to construct a device abstract model, the device parameter information including robot link structure parameter, joint motion parameter, dynamics limit parameter and end tool calibration information; S4: based on the task description file, the activated constraint template and the device abstract model, calling a hybrid optimization algorithm for global trajectory solving to generate a globally optimal trajectory, the hybrid optimization algorithm including an upper layer fast initial solution generation stage and a lower layer limited neighborhood fine search stage; S5: visualizing the globally optimal trajectory, providing a semantic slider interactive interface, receiving a preference adjustment instruction input by the user through the semantic slider, and re-executing step S4 to generate an adjusted globally optimal trajectory according to the preference adjustment instruction and a preset constraint priority mechanism; S6: compiling the globally optimal trajectory that passes the verification into a device-independent standardized intermediate representation, and then adaptively mapping the standardized intermediate representation into controller recognizable native execution instructions according to the target robot controller type.

[0009] Preferably, in step S1, the task description file adopts a structured data format, and at least contains task identification information, key frame spatial pose data, key frame specific semantic labels, global task semantic labels, and trajectory optimization objectives, the key frame spatial pose data is recorded in the form of six-dimensional coordinates, and the trajectory optimization objectives include time optimization, smoothness optimization, or multi-objective balance.

[0010] Preferably, in step S2, the kinematics smoothing template is used to constrain the jerk continuity of the trajectory, and a 3-5-3 segmented polynomial interpolation algorithm is adopted; the dynamics safety template is used to limit the robot joint torque, speed and acceleration within a preset threshold range; the process constraint template is used to match the motion parameter requirements in specific process scenarios, including the low-speed passing requirement of the welding process and the positioning accuracy requirement of the assembly process; and the geometric safety template is used to construct an obstacle bounding box to constrain the minimum safety distance between the robot link and the obstacle.

[0011] Preferably, in step S3, after the device abstraction model is constructed, a trajectory physical feasibility feedforward verification step is further included: the generated initial trajectory is compared with the parameters in the device abstraction model, and if the trajectory has joint motion out of limits, dynamics parameters beyond limits, or tool collision risk, abnormal prompt information and correction suggestions are output.

[0012] Preferably, in step S5, the visualization display includes a three-dimensional trajectory path diagram, a speed-time curve, an acceleration-time curve, and a multi-performance indicator Pareto frontier diagram, the multi-performance indicator at least covers the total trajectory time, a comprehensive smoothness indicator, and a safety margin; and the decision order of the constraint priority mechanism is: geometric safety constraint > dynamics feasibility constraint > process quality constraint > time efficiency constraint.

[0013] Preferably, in step S6, the standardized intermediate representation records the position instructions and speed instructions of each joint of the robot at fixed time intervals; the adaptive mapping is realized through a device adapter, the device adapter has built-in instruction mapping tables of different brands of robot controllers, and supports conversion of the standardized intermediate representation into KRL, KAREL, RAPID, or URScript native language programs.

[0014] A system suitable for any one of the above robot teaching programming methods based on global input, comprising a task intention interaction module, a semantic constraint matching module, a device modeling module, a trajectory optimization module, an interactive verification module, and a cross-platform deployment module. The task intention interaction module is used to receive a key frame sequence and behavior semantic labels input by a user, and generate a structured task description file. The semantic constraint matching module is connected with the task intention interaction module, and is used for matching and activating a corresponding constraint template from a multi-modal constraint template library according to a behavior semantic label. The device modeling module is used for loading target robot parameters, constructing a device abstract model, and performing trajectory physical feasibility verification. The trajectory optimization module is connected with the task intention interaction module, the semantic constraint matching module and the device modeling module respectively, and is used for calling a hybrid optimization algorithm to generate a globally optimal trajectory. The interaction verification module is connected with the trajectory optimization module, and is used for visualizing the globally optimal trajectory and receiving user preference adjustment instructions, and driving the trajectory optimization module to regenerate the trajectory. The cross-platform deployment module is connected with the trajectory optimization module and the device modeling module respectively, and is used for compiling the verified trajectory into a standardized intermediate representation, and mapping the trajectory into native execution instructions of a target robot controller.

[0015] Preferably, the task intention interaction module provides a graphical operation interface, supports a user to define a key frame through a drag-and-drop, click or coordinate input method, and provides a semantic label selection list, the semantic label selection list including preset options of finishing, high-speed traversal, high obstacle avoidance priority and S-shaped speed profile.

[0016] Preferably, the multi-modal constraint template library adopts an extensible architecture, supports a user to add a new constraint template according to a newly added process scene or robot type, and the new constraint template needs to include a semantic label association rule, a mathematical constraint expression and a parameter configuration interface.

[0017] Preferably, the modules of the system communicate through a standardized data interface, the standardized data interface defines a task description file, a constraint set, a device abstract model, a standardized intermediate representation and a native execution instruction between interlayer data exchange formats, and supports independent upgrading and replacement of the modules.

[0018] The present application has the following advantages: Compared with the traditional teaching programming mode of segmented input-local interpolation-post processing smoothing, the present application realizes a technical leap from task-level semantic input to global trajectory optimization and cross-platform deployment through a three-layer decoupling architecture of a task intention layer-a semantic planning layer-a physical execution layer, and has the following advantages: 1. Revolutionary improvement of teaching efficiency: through global task modeling and semantic abstraction, a user does not need to teach point by point or manually segment, but only needs to define a key frame and add a behavior semantic label once, so that a high-quality trajectory can be generated by one key; compared with traditional point-level programming, the teaching programming time is expected to be shortened by more than 60%, greatly reducing the operation threshold and dependence on professionals.

[0019] 2. Trajectory performance achieves global optimization and essential smoothness: The system deeply integrates multi-modal constraint templates and high-performance optimization algorithms through the semantic planning layer. The generated trajectory can shorten the total time by 15%-25% while strictly meeting the dynamic constraints. At the same time, the acceleration peak is reduced by 40%-50% due to the continuity of acceleration, achieving ultra-smooth control from the source of motion instructions and effectively suppressing mechanical vibration and impact.

[0020] 3. Cross-platform reusability and system expansibility are significantly enhanced: Relying on device abstraction and standardized intermediate representation, the same task description can be seamlessly mapped to different brand robot controllers such as KUKA, FANUC, and ABB. Cross-platform deployment time is reduced by more than 60%, and the three-layer decoupled modular architecture supports independent upgrade and expansion of algorithms, templates, and adapters, providing a solid foundation for system continuous evolution and process accumulation. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is the overall architecture diagram of the system; Figure 2 is the overall execution flowchart of the method of the present application; Figure 3 is the multi-modal constraint template activation mechanism diagram; Figure 4 is the hierarchical hybrid optimization algorithm flowchart; Figure 5 is the real-time verification and interactive correction closed-loop flowchart; Figure 6 is the cross-platform deployment data flowchart. DETAILED DESCRIPTION

[0022] The present application will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present application and not to limit the scope of the present application. The experimental methods in the following examples are not specified, and the general conditions or the conditions recommended by the manufacturer are usually used.

[0023] Unless otherwise defined, all professional and scientific terms used herein have the same meaning as familiar to those skilled in the art. The reagents or raw materials used in the present application can be purchased through conventional channels. Unless otherwise specified, the reagents or raw materials used in the present application are used in accordance with the conventional methods in the art or in accordance with the product instructions. In addition, any method and material similar or equivalent to the described content can be applied to the method of the present application. The preferred implementation method and material described in the present application are only used for demonstration according to the drawings and specific embodiments of the present application.

[0024] A robot teaching programming method and system based on global input are as follows: 1. Overall architecture An intelligent teaching programming architecture with clear responsibilities, high cohesion and low coupling is constructed, which divides the complex robot programming task into three logical levels: Task intention layer: as a user interface, it is responsible for receiving and translating the user's job intention.

[0025] Semantic planning layer: as the brain of the system, it is responsible for thinking and planning the optimal motion scheme.

[0026] Physical execution layer: as the hands and feet of the system, it is responsible for executing the scheme and adapting it to the specific physical device.

[0027] This architecture takes the chain information flow of task semantics→constraint analysis→optimization solution→visualization verification→control mapping as the main line, forms a complete closed loop from human-machine intention input to physical execution, and fundamentally solves the semantic gap between user intention and robot execution in traditional methods, completely liberating the user from tedious and repetitive bottom-level parameter debugging.

[0028] 2. Function and module composition of each layer To achieve the above idea, the system of the application includes the following six core modules, which belong to the above three levels respectively and work cooperatively through standardized data interfaces.

[0029] First layer: task intention layer a) Global task modeling and semantic abstraction module: This module is the entrance for user and system interaction. The user inputs the key frame sequence (i.e. path point) of the task through the graphical interface (GUI) once, and adds behavior semantic labels to each key frame or the whole task. These labels are not bottom-level parameters, but high-level task descriptions, such as: ① motion smoothness: finishing (the system will analyze this as a smooth trajectory with continuous jerk); ② obstacle avoidance priority: high (the system will analyze this as a larger safety margin); speed profile: S-type (the system will analyze this as an S-type speed curve).

[0030] This module compiles the user's input into a structured, machine-readable task description file. This file not only contains geometric path information, but also carries the user's job intention and performance expectation, laying a solid foundation for subsequent intelligent planning.

[0031] Second layer: semantic planning layer This layer is the intelligent core of the system, which is responsible for converting the task intention of the upper layer into an executable optimal motion scheme.

[0032] a) Multi-modal constraint template library module: This module is an extensible knowledge base, which internally stores a variety of predefined constraint sets, including kinematic templates (such as acceleration continuity, Jacobian singularity avoidance, etc.), dynamic templates (such as torque saturation protection, etc.), process templates (such as specific requirements of welding, gluing, handling, etc.), and safety templates (such as collision avoidance rules, obstacle buffer zones, joint limit early warning, etc.). The system automatically matches and activates the corresponding constraint templates according to the semantic tags in the task description file, and translates the abstract user language into specific mathematical constraints.

[0033] b) Trajectory generation and compilation engine module: This module is the computing kernel of the system. It receives the task description file from the upper layer and the activated constraint set from the template library, and calls the built-in high-performance global trajectory optimization algorithm. This algorithm performs global search and optimization on the trajectory under strict mathematical constraints, and its output is a time-optimal and essentially super-smooth trajectory scheme file for the execution layer to call. In this invention, the engine is called as a black box, and its value lies in translating high-quality algorithm capabilities into simple user services.

[0034] c) Real-time verification and interactive correction module: This module provides visual feedback and closed-loop optimization capabilities. It presents the candidate trajectory output by the trajectory generation engine to the user in intuitive forms such as Pareto frontier charts (time vs. smoothness vs. safety), etc. The chart clearly labels the trade-off relationship between key performance indicators such as total trajectory time, comprehensive smoothness index, safety margin, etc. The user does not need to understand the underlying parameters, but only needs to express preferences through high-level semantic sliders (such as drag the faster ↔ more stable slider). The system re-calls the trajectory generation engine in real time based on the forced constraint priority mechanism (safety > dynamic feasibility > process quality > time efficiency), generates a trajectory that meets the new preferences, and forms an intelligent closed loop of planning-verification-adjustment.

[0035] Third layer: Physical execution layer This layer ensures that the planned high-quality trajectory can be accurately and stably executed on real and diverse robot platforms.

[0036] a) Device abstraction and calibration module This module establishes a digital twin of the target robot. It is responsible for loading the physical parameters and tooling information of the target robot, including its kinematic structure (such as link length, joint type, DH parameter, etc.), dynamic upper limit (maximum speed, acceleration, torque), joint limit, and end tooling calibration information (TCP offset, fixture pose), which are encapsulated to form an abstract model of the device. Through this module, the system binds the abstract task description file with the specific device abstract model to form a task-device coupling description, ensuring that the generated trajectory meets the physical boundaries of the device. The system performs a feedforward feasibility verification at this stage: if there are joint limit exceedance, dynamic unattainability, or tooling collision risks in the trajectory, the system will mark them in red in the verification interface and return correction suggestions.

[0037] b) Cross-platform deployment and adaptive mapping module This module is the key to the system's versatility. It compiles the verified final trajectory scheme file into a standardized intermediate representation that is independent of the specific robot brand. Then, through the device adapter developed for different robot controllers (such as KUKA, FANUC, ABB, etc.), the intermediate representation is automatically recognized and translated into native instruction code that the target controller can recognize, and the adaptation process is achieved through semantic alignment (such as trajectory points, poses, interpolation commands, etc.) through the instruction mapping table. This achieves one planning, multiple running, greatly reducing the programming and deployment cost of multi-brand robot mixed production.

[0038] Table 1: Different robot brands and corresponding language types .

[0039] 3. Workflow In combination with the above modules, the workflow is as follows: S1 Task abstraction: the user defines keyframes and labels semantic tags, and the system generates a task description file; S2 Environment binding: the user selects or inputs the target robot model, and the system automatically loads the corresponding device model and environment parameters; S3 One-key compilation and optimization: the user triggers the generation of trajectory instructions, and the system automatically performs semantic analysis, constraint matching, global trajectory optimization, and outputs candidate trajectories; S4 Visual interactive verification: the user reviews the trajectory performance in the graphical interface and fine-tunes the preferences through semantic sliders, and the system responds in real time and regenerates the trajectory; S5 Deployment and execution: after the user confirms the trajectory, the system automatically compiles and delivers standardized control instructions to the target robot, completing the task.

[0040] In the following embodiments, the present embodiment takes the application of a Bronite six-axis industrial robot in a complex welding scenario as an example to explain the specific embodiments of the present application. The present embodiment will fully demonstrate the global process from one-key input to task definition, to trajectory generation, verification, and finally execution.

[0041] As shown in Figure 1 , the system of the present application generally adopts a three-layer decoupling architecture of task intention layer-semantic planning layer-physical execution layer.

[0042] The system mainly includes the following modules: global task modeling and semantic abstraction module (task intention layer), multi-modal constraint template library module, trajectory generation and compilation engine module, real-time verification and interactive correction module (semantic planning layer), device abstraction and calibration module, cross-platform deployment and adaptive mapping module (physical execution layer); The complete process of system operation is shown in Figure 2 , which includes the following steps: 1. System initialization and task intention layer input ① After the system starts, it is first initialized, and the user directly inputs to define several task key frames through the graphical interface of the global task modeling and semantic abstraction module or the teach pendant .

[0043] In this example, the coordinates of 6 spatial key points are input at one time: Point 1: (0.51, 0.32, 0.45, 90, 0, 0); Point 2: (0.55, 0.30, 0.42, 90, 0, 0); Point 3: (0.62, 0.28, 0.38, 90, 0, 0); Point 4: (0.58, 0.25, 0.41, 90, 0, 0); Point 5: (0.53, 0.23, 0.43, 90, 0, 0); Point 6: (0.48, 0.22, 0.44, 90, 0, 0).

[0044] ② Add behavior semantic labels to the key frames: Label the entire trajectory: motion smoothness: finishing (requires continuous jerk); label point 3: obstacle avoidance priority: high (there is a device obstacle near this point); label point 4: process requirement: low-speed accurate passing (requires the speed at this point to be ≤0.2m / s) ③ The system compiles the above input into a structured task description file, the core content of which in JSON format is as follows: { "task_id":"Complex_Path_Task", "keypoints" : [ {"id" : "P1", "pose" : [0.51, 0.32, 0.45, 90, 0, 0], "tags" : ["start_point"]}, {"id" : "P2", "pose" : [0.55, 0.30, 0.42, 90, 0, 0], "tags" : []}, {"id" : "P3", "pose" : [0.62, 0.28, 0.38, 90, 0, 0], "tags" : ["obstacle_avoidance: :high"]}, {"id" : "P4", "pose" : [0.58, 0.25, 0.41, 90, 0, 0], "tags" : ["low_speed_pass: :0.2m / s"]}, {"id" : "P5", "pose" : [0.53, 0.23, 0.43, 90, 0, 0], "tags" : []}, {"id" : "P6", "pose" : [0.48, 0.22, 0.44, 90, 0, 0], "tags" : ["end_point"]} ], "global_semantics" : ["motion_smoothness: :high", "speed_profile: :s_curve"], "optimization_objective" : "minimize_time" } 2. Intelligent decision and optimization of semantic planning layer After receiving the task description file, the semantic planning layer expands the work according to the flow shown in the figure. Figure 3

[0045] (1) Multi-modal constraint template activation and modeling The multi-modal constraint template library module automatically activates the following constraint templates according to the semantic label: ① Kinematics smoothing template (responding to motion smoothness: finishing): A 3-5-3 segmented polynomial is used for trajectory interpolation, and a jerk continuity constraint is applied at the interpolation points: ; This constraint fundamentally guarantees the super-smoothness of the trajectory.

[0046] ② Dynamics safety template:​ Considering the upper limit of each joint torque, the constraint condition is: ; Wherein, the dynamics parameters are obtained through system identification.

[0047] ③Geometric safety template (response to obstacle avoidance priority: high): Establish an obstacle bounding box to constrain the distance between the robot arm link and the obstacle: ; ④Process constraint template (response to process requirements: low-speed accurate passing): Apply an acceleration constraint near point P4: ; (2) Trajectory generation and global optimization solution The trajectory generation and compilation engine module calls the mixed optimization algorithm (its flow is shown in Figure 4 ) for solution.

[0048] The upper layer analysis guide adopts an S-shaped velocity curve for initial time allocation, generating 5 initial solutions with a target function value distribution of ; The lower layer intelligent fine search uses an improved whale optimization algorithm to perform fine search in the initial solution neighborhood. The optimization objective function is: ; Wherein, the adaptive weight [λ1, λ2, λ3, λ4, λ5] is initialized according to the semantic label, and is dynamically adjusted according to the constraint violation during the optimization process.

[0049] After 157 generations of iteration, the algorithm converges to the optimal solution, with a total trajectory time =7.2s, which is 19% shorter than the initial best solution time.

[0050] It should be particularly noted that the above description of the layered optimization algorithm is based on its commonly known general principles in the art. Figure 4 A typical execution flowchart of such an algorithm is shown. The protection focus of the present application is to integrate such a general optimization algorithm into the three-layer architecture of the present application, and drive its work through the semantic template and interaction mechanism, rather than the specific implementation details of the algorithm flow itself.

[0051] (3) Real-time verification and interactive correction The real-time verification and interactive correction module presents the optimization results to the user in the Figure 5 flow shown. The interface includes: a three-dimensional trajectory graph (displaying the path curve and key frame position); a speed / acceleration curve; a safety envelope graph; and a multi-objective Pareto frontier graph.

[0052] When the user adjusts the semantic preference through the slider (e.g. faster -> increase , more stable -> increase ), the system immediately recalculates the objective function and outputs the updated trajectory, realizing the semantic-optimization closed loop.

[0053] 3. Adaptation and deployment of the physical execution layer The physical execution layer ensures the accurate and safe execution of high-quality trajectories on physical robots.

[0054] (1) Device abstraction and feasibility verification The device abstraction and calibration module loads the robot model parameters (see Table 2) and tool parameters, constructing a device abstraction model.

[0055] Table 2: Key parameters of a six-axis robot ; The system uses this model for feedforward verification to confirm that each point of the trajectory satisfies: joint position and joint torque (reserve 15% safety margin).

[0056] (2) Cross-platform compilation and instruction generation As shown in the cross-platform deployment data flow diagram of Figure 6 , the block compiles the verified trajectory into an intermediate representation, which records the position and speed instructions of each joint at intervals of 0.01s.

[0057] Subsequently, the system calls the device adapter specially developed for the Bournet controller to convert the intermediate representation into native control instructions, and the conversion process maintains the consistency of motion semantics, such as converting continuous S-shaped speed curves into dense point position instructions.

[0058] 4. Implementation effect and performance comparison To quantify the advantages of the invention, comparative experiments were conducted on the same hardware platform as the traditional teaching programming method, with the following results: Table 3: Performance comparison results .

[0059] The experimental results show that the invention has achieved significant improvement in teaching efficiency, motion performance, stability, and precision, etc. Especially in vibration suppression, the super-smooth trajectory reduces the end vibration by more than 57%, which is of great significance for high-quality machining operations.

[0060] The unmentioned parts of the invention apply to the prior art.

Claims

1. A robot teach programming method based on global input, characterized in that, The method comprises the following steps: S1: receiving a user-inputted task key frame sequence and a corresponding behavior semantic label, integrating the key frame sequence and the behavior semantic label, and generating a structured task description file, wherein the behavior semantic label comprises motion characteristic description, process requirement description, and safety priority description; S2: matching and activating a corresponding constraint template from a preset multi-modal constraint template library according to the behavior semantic label in the task description file, wherein the constraint template at least covers a kinematics smoothing template, a dynamics safety template, a process constraint template, and a geometric safety template; S3: loading device parameter information of a target robot, and constructing a device abstraction model, wherein the device parameter information comprises robot link structure parameters, joint motion parameters, dynamics limit parameters, and end tool calibration information; S4: calling a hybrid optimization algorithm for global trajectory solving based on the task description file, the activated constraint template, and the device abstraction model, and generating a global optimal trajectory, wherein the hybrid optimization algorithm comprises an upper-layer fast initial solution generation stage and a lower-layer limited neighborhood fine search stage; S5: visualizing the global optimal trajectory, providing a semantic slider interactive interface, receiving a preference adjustment instruction input by a user through the semantic slider, and re-executing step S4 to generate an adjusted global optimal trajectory according to the preference adjustment instruction and a preset constraint priority mechanism; S6: compiling the verified global optimal trajectory into a device-independent standardized intermediate representation, and adaptively mapping the standardized intermediate representation into a controller-recognizable native execution instruction according to a target robot controller type.

2. The robot teach programming method based on global input according to claim 1, characterized in that, In step S1, the task description file adopts a structured data format, and at least comprises task identification information, key frame spatial posture data, key frame exclusive semantic label, global task semantic label, and trajectory optimization target, wherein the key frame spatial posture data is recorded in the form of six-dimensional coordinates, and the trajectory optimization target comprises time optimization, smoothness optimization, or multi-target balance.

3. The robot teach programming method based on global input according to claim 1, characterized in that, In step S2, the kinematics smoothing template is used to constrain the jerk continuity of the trajectory, and a 3-5-3 segmented polynomial interpolation algorithm is adopted to achieve the same; the dynamics safety template is used to limit the robot joint torque, speed, and acceleration within a preset threshold range; the process constraint template is used to match the motion parameter requirements in a specific process scene, including low-speed passing requirements of a welding process and positioning accuracy requirements of an assembly process; and the geometric safety template is used to construct an obstacle bounding box to constrain the minimum safety distance between the robot link and the obstacle.

4. The robot teach programming method based on global input according to claim 1, characterized in that, In step S3, after the device abstraction model is constructed, a trajectory physical feasibility feedforward verification step is further included: comparing the generated initial trajectory with the parameters in the device abstraction model, and if the trajectory has joint motion over-limit, dynamics parameter over-limit, or tool collision risk, outputting abnormal prompt information and correction suggestions.

5. The robot teach programming method based on global input according to claim 1, characterized in that, In step S5, the visualized display includes a three-dimensional trajectory path diagram, a speed-time curve, an acceleration-time curve, and a multi-performance indicator Pareto frontier diagram, the multi-performance indicator at least covering a total trajectory time, a comprehensive smoothness indicator, and a safety margin; and the decision-making order of the constraint priority mechanism is: geometric safety constraint > dynamic feasibility constraint > process quality constraint > time efficiency constraint.

6. The robot teach programming method based on global input according to claim 1, characterized in that, In step S6, the standardized intermediate representation records the position instructions and speed instructions of each joint of the robot at fixed time intervals; and the adaptive mapping is realized through a device adapter, the device adapter internally storing instruction mapping tables of different brands of robot controllers, and supporting conversion of the standardized intermediate representation into KRL, KAREL, RAPID, or URScript native language programs.

7. A system suitable for use in the robot teach programming method based on global input according to any one of claims 1 to 6, characterized in that, The system comprises a task intention interaction module, a semantic constraint matching module, a device modeling module, a trajectory optimization module, an interaction verification module, and a cross-platform deployment module. The task intention interaction module is configured to receive a key frame sequence and a behavior semantic label input by a user, and generate a structured task description file. The semantic constraint matching module is connected to the task intention interaction module, and is configured to match and activate a corresponding constraint template from a multi-modal constraint template library according to the behavior semantic label. The device modeling module is configured to load target robot parameters, construct a device abstract model, and perform trajectory physical feasibility verification. The trajectory optimization module is connected to the task intention interaction module, the semantic constraint matching module, and the device modeling module, and is configured to call a hybrid optimization algorithm to generate a globally optimal trajectory. The interaction verification module is connected to the trajectory optimization module, and is configured to visually display the globally optimal trajectory, receive a user preference adjustment instruction, and drive the trajectory optimization module to regenerate a trajectory. The cross-platform deployment module is connected to the trajectory optimization module and the device modeling module, and is configured to compile a verified trajectory into a standardized intermediate representation, and map the standardized intermediate representation into native execution instructions of a target robot controller.

8. The robot teach programming system based on global input according to claim 7, characterized in that, The task intention interaction module provides a graphical operation interface, supports a user to define a key frame through a drag-and-drop, click, or coordinate input method, and provides a semantic label selection list, the semantic label selection list including preset options of finishing, high-speed traversal, high obstacle avoidance priority, and S-shaped speed profile.

9. The robot teach programming system based on global input according to claim 7, characterized in that, The multi-modal constraint template library adopts an extensible architecture, supports a user to add a new constraint template according to a newly added process scene or robot type, and requires the new constraint template to include a semantic label association rule, a mathematical constraint expression, and a parameter configuration interface.

10. The robot teach programming system based on global input according to claim 7, characterized in that, The modules of the system communicate through standardized data interfaces, the standardized data interfaces define a layer-to-layer data exchange format between a task description file, a constraint set, a device abstract model, a standardized intermediate representation, and native execution instructions, and support independent upgrading and replacement of the modules.

Images