Autonomous processing method and system for intelligent robot with body

Abstract

The invention relates to the field of robot machining, and provides an autonomous machining method and system for an intelligent robot with a body, and the method comprises the steps: scanning the contour of a workpiece to form point cloud data, carrying out the noise filtering and geometric feature analysis of the point cloud data, and obtaining the contour of the workpiece; comparing the processed point cloud data with an input three-dimensional model, and generating a machining program by combining offline programming based on workpiece pose registration; and identifying the machining features of the workpiece model, applying the machining program, and calling the machining process library to generate a machining track. Through the method, the intelligent robot with the body can achieve autonomous machining and improve the machining precision and efficiency, noise points are effectively removed through a dual filtering mechanism in the point cloud data processing link, edge features are reserved, and the method can adapt to complex curved surfaces.

Description

Technical Field

[0001] This invention relates to the field of robotic processing, and more specifically to an autonomous processing method for an embodied intelligent robot. Background Technology

[0002] With the rapid development of industrial automation, robots are increasingly widely used in manufacturing, especially in the field of complex workpiece processing. Robot autonomous processing technology, by combining technologies such as visual perception, point cloud processing and intelligent path planning, has achieved accurate identification and efficient processing of workpieces.

[0003] In particular, in aircraft wing manufacturing, the drilling process is a key step in connecting the skin to the frame structure, which directly affects the safety and aerodynamic performance of the aircraft.

[0004] In traditional aircraft wing machining operations, the reliance on worker experience has long resulted in low drilling efficiency. Tools and equipment depend on manual operation, leading to low process flow and standardization, and making it difficult to guarantee drilling accuracy. In the field of automated machining, the complexity of the machining environment (such as dynamic changes, lighting conditions, and occlusion) makes it difficult to acquire aircraft wing point cloud data. The original point cloud is easily affected by the environment (shading, reflection), resulting in noise pollution and information loss. Traditional automated drilling trajectories for aircraft wings cannot follow the movement of the workpiece position. Summary of the Invention

[0005] To address the aforementioned shortcomings of existing technologies, this invention provides an embodied intelligent robot with autonomous acceleration... Construction methods and systems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: In the first aspect, an autonomous processing method for an embodied intelligent robot is provided, comprising: scanning and photographing the workpiece contour based on lidar and visual SLAM to form point cloud data; performing noise filtering and geometric feature analysis on the point cloud data; comparing the processed point cloud data with the input three-dimensional model; and generating a processing program based on workpiece pose registration and offline programming. Identify the machining features of the workpiece model, apply the machining program, and retrieve the machining process library to generate the machining trajectory.

[0007] The workpiece pose is updated in real time to form a new workpiece pose. The original trajectory is then subjected to rigid body transformation to generate a new machining trajectory.

[0008] In some embodiments, point cloud data noise is filtered out by density statistical filtering and bilateral temporal filtering; The density statistical filtering includes: constructing a neighborhood density calculation and determining a density threshold; The bilateral temporal filtering includes: normal vector calculation, spatiotemporal weight calculation, and point position update. The normal vector guides the spatiotemporal weight allocation, the spatiotemporal weight determines the point position update direction, and the point position update feeds back to affect the normal vector calculation of the next frame.

[0009] In some embodiments, the construction of the neighborhood density is as follows: constructing a spherical neighborhood centered on the target point, and determining the total number of points within the spherical neighborhood. and the average distance between this point and all points in its neighborhood. ; A density threshold is determined based on the number of points in the neighborhood, and the formula for the density threshold determination model is as follows: in, : Number of points in the neighborhood Global points.

[0010] In some embodiments, the normal vector is calculated by: for the normal vector of each local neighborhood, solving for the eigenvector corresponding to the minimum eigenvalue using PCA; The formula for the spatiotemporal weight calculation model is as follows: , in Spatial weights; Range weight; The point position update is based on the weighted average neighborhood point position, while preserving the sharpness of the interface. The point position update model formula is: .

[0011] In some embodiments, the processing trajectory achieves the optimal path through surface adaptive sampling, dynamic heuristic functions, local surface expansion and optimization, and dynamic obstacle handling.

[0012] In some embodiments, the surface adaptive sampling is: geodesic distance calculation and dynamic offset probability adjustment, wherein the geodesic distance calculation is: calculating the distance between nodes through local parameterization; The formula for the dynamically adjusted target bias probability model is: in The local complexity coefficient of the surface. To adjust the parameters.

[0013] In some embodiments, the dynamic heuristic function is: in, Distance cost Security cost, :node Curvature penalty term at the location.

[0014] In some embodiments, the method includes: the robot receiving an instruction, performing a power detection according to the instruction, and if the power is sufficient, the robot autonomously navigating to the corresponding workstation; if the power is insufficient, the robot calling upon another robot or charging the battery before executing the instruction.

[0015] In some embodiments, the method includes: the robot selecting a processing tool based on the processing program, the robot performing a processing operation based on the processing trajectory, and after the operation is completed, the robot returning the tool to its original position and autonomously navigating to a waiting area.

[0016] Secondly, an embodied intelligent robot autonomous processing system is provided, comprising: Autonomous navigation unit, used for robots to autonomously navigate to their corresponding workstations; The workpiece acquisition unit is used to photograph / scan the workpiece contour and acquire workpiece data for processing. A data processing unit is used to process the workpiece data. Offline programming unit for inputting workpiece models; The path planning unit is used to generate the optimal processing path and obstacle avoidance. The data processing unit compares the merged and cropped data with the model input by the offline programming unit, and generates a machining program based on workpiece pose registration. The workpiece acquisition unit identifies the processing features of the workpiece model, applies the processing program, retrieves data from the processing technology library, and generates a processing trajectory. The path planning unit optimizes the processing trajectory to generate the optimal processing path.

[0017] Thirdly, a method for autonomous processing by an embodied intelligent robot is provided, specifically including the following steps: S1. The robot receives the work instruction and performs a power self-check. If the power is sufficient, the robot moves to the work area. If the power is insufficient, it charges the robot or calls on another robot. S2. After the robot arrives at the work area, it starts the photo-taking program to scan the workpiece processing area and the workpiece outline to form point cloud data, and then merges and trims the point cloud data. S3. Compare the point cloud data formed in step 2 with the 3D model in offline programming, and form a machining program and machining trajectory based on workpiece pose registration. S4. Based on the machining program in S3, the composite robot selects the cutting tools. S5. Perform automatic processing based on the processing trajectory of S3; S6. After the robot completes the processing operation, it automatically returns the tool to its original position and moves to the waiting area to wait for the next operation.

[0018] Compared with the prior art, the beneficial effects of the present invention are: 1. By adopting the above technical solution, the present invention achieves batch automated precision punching, reducing the workload and intensity of manual punching; 2. The lidar performs 360° scanning and multi-angle shooting to generate dense point clouds. The fusion of lidar and visual SLAM improves the accuracy and robustness of environmental perception through multi-level data interaction, and better acquires point cloud data. 3. By processing point cloud data through noise filtering and geometric feature analysis, the accuracy and reliability of target recognition are improved; 4. Visual SLAM or LiDAR updates the workpiece pose in real time. Based on the new workpiece pose, the original trajectory is transformed into a rigid body to generate a new machining path. Only one trajectory generation is needed. After that, the trajectory will move with the workpiece position and does not need to be regenerated. Attached Figure Description

[0019] Figure 1 Flowchart of the autonomous processing method for an embodied intelligent robot of the present invention; Figure 2 Structural block diagram of the embodied intelligent robot autonomous processing system of the present invention; Detailed Implementation The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 An autonomous processing method for an embodied intelligent robot, which is achieved through the following steps: Step 1: Workpiece contour scanning and point cloud data processing First, the workpiece contour is scanned and photographed from multiple angles using LiDAR and visual SLAM to generate point cloud data. After acquiring the point cloud data, noise filtering is performed using a combination of density statistical filtering and bilateral temporal filtering.

[0021] Density statistical filtering includes calculating neighborhood density and determining density thresholds. When constructing the neighborhood density, a spherical neighborhood is built centered on the target point; for example, a radius-based neighborhood is constructed centered on the target point. A spherical neighborhood; calculate the total number of points within the neighborhood. and the average distance between this point and all points in its neighborhood. The density threshold is determined based on the number of points in the neighborhood. The formula for the density threshold determination model is as follows: ,in, : Number of points in the neighborhood : Global point count, if the percentage of points in the neighborhood is less than 0.1% of the global average density (i.e. ( ), which was identified as a noise point.

[0022] Bilateral temporal filtering includes normal vector calculation, spatiotemporal weight calculation, and point position update. Normal vector calculation is performed by using PCA to solve for the eigenvector corresponding to the minimum eigenvalue of each local neighborhood normal vector.

[0023] The spatiotemporal weight calculation model is as follows: Spatial weight : Range weight : Temporal constraints: In dynamic sequences, add a time continuity term; Point position updates are based on weighted average neighborhood point positions, preserving the sharpness of the interface. The point position update model formula is as follows: ; in: Better and more strictly limit the scope of spatial influence, avoiding excessive smoothing.

[0024] Better control of normal sensitivity ensures that the weight of the material interface (where the normal changes abruptly) drops sharply and the feature is not smoothed. The normal vector guides the allocation of spatiotemporal domain weights, which in turn determine the direction of point position updates. The updated point position then feeds back to affect the calculation of the normal vector in the next frame.

[0025] Based on geometric feature analysis, feature extraction and segmentation of point cloud data are performed. First, a local geometric reference system is constructed through KNN neighborhood (k=20). Then, the fatigue critical area is located through the standard deviation of principal curvature. When the standard deviation of curvature is >0.002mm⁻¹, it is a fatigue critical area. At the same time, outliers with normal direction deviation >15° are removed.

[0026] Step 2: Comparison of point cloud data and 3D model The point cloud data, after noise filtering and geometric feature analysis, is compared with the input 3D model. Based on workpiece pose registration, a machining program is generated using offline programming.

[0027] Step 3: Recognition of Machining Features of Workpiece Model The machining features of the workpiece model are identified, a machining program is applied, and a machining trajectory is generated by retrieving the machining process library. The machining trajectory achieves the optimal path through surface adaptive sampling, dynamic heuristic functions, local surface expansion and optimization, and dynamic obstacle handling.

[0028] Adaptive surface sampling includes geodesic distance calculation and dynamic offset probability adjustment. The geodesic distance calculation calculates the distance between nodes through local parameterization, and the target offset probability is dynamically adjusted according to the surface curvature and obstacle density. : in The local complexity coefficient of the surface. To adjust the parameters.

[0029] The dynamic heuristic function is: Multi-objective heuristic weighting, combined with distance cost and security costs (e.g., minimum distance from obstacles) For nodes Curvature penalty at certain points is used to prevent the path from becoming too convoluted.

[0030] Adaptive adjustment mechanism: dynamically updates weights based on real-time environmental feedback (such as the addition of obstacles). For example, increasing the size of dense obstacle areas .

[0031] Local expansion and optimization of surfaces: ensuring the feasibility and local optimum of the path on the surface.

[0032] Variable step size expansion: Reduce step size in regions of high curvature. (like This improves path accuracy.

[0033] Spline path smoothing: Apply three layers of the initial path. Spline interpolation ensures curvature continuity and satisfies kinematic constraints.

[0034] Dynamic obstacle handling: Ensuring path safety and timeliness in the spatiotemporal dimensions by acquiring the position, speed, and direction of dynamic obstacles in real time through sensors (such as LiDAR and vision) and predicting their trajectory over a period of time (such as a set of positions within 10 seconds).

[0035] Real-time collision detection: Dynamic obstacles are modeled as time-varying spherical envelopes, and obstacles are avoided in advance by predicting motion trajectories (such as linear extrapolation).

[0036] Local replanning: If a collision risk is detected, start a local RRT* subtree with the current node as the root to quickly generate an alternative path.

[0037] It is worth noting that the specific algorithm flow and implementation of the optimal path includes the following steps: (1) Initialization: Input surface parameters (such as triangular mesh or NURBS model), start point, end point and obstacle information.

[0038] (2) Main loop: Sampling phase: by dynamic Generate target bias or random sampling points.

[0039] Node expansion: Select the nearest neighboring node with the optimal heuristic function and expand new nodes with variable step size along the geodesic direction.

[0040] Reconnection optimization: Reselect the parent node within the geodesic distance neighborhood to shorten the path cost.

[0041] (3) Post-processing: The spline smooths the path and adjusts the endpoint heading using a kinematic model (such as a Reeds-Shepp curve).

[0042] Step 4: The robot performs autonomous processing. After receiving the command, the robot first checks its battery level. If the battery is sufficient, the robot autonomously navigates to the corresponding workstation; if the battery is insufficient, it calls upon another robot or charges the battery before executing the command.

[0043] The robot selects processing tools based on the processing program and performs processing tasks according to the processing trajectory. After the processing is completed, the robot returns the tools to their original positions and autonomously navigates to the waiting area.

[0044] It is worth noting that once the machining trajectory is determined, only one trajectory generation is needed. After that, the trajectory will move with the changes in the workpiece position and does not need to be regenerated.

[0045] The robot updates the workpiece pose in real time using visual SLAM or laser odometry. For example, after the workpiece is translated or rotated, the feature points are realigned using the ICP algorithm.

[0046] Based on the new workpiece pose, a rigid body transformation (translation or rotation) is performed on the original trajectory to generate a new machining path.

[0047] By calculating inverse kinematics, the joint angles of the robot arm are planned to ensure that the machining tool is always aligned with the target position. Based on visual feedback, the machining position is monitored in real time, and the drilling force is detected by sensors during the machining process to dynamically adjust the machining speed.

[0048] Through the above steps, the embodied intelligent robot can achieve autonomous processing, improving processing accuracy and efficiency. The dual filtering mechanism in the point cloud data processing stage effectively removes noise points while preserving edge features; the application of surface adaptive sampling and dynamic heuristic functions makes the processing trajectory more reasonable and adaptable to complex surfaces; and the robot's power management and tool selection mechanisms ensure the continuity and reliability of the processing process.

[0049] Example 2 In this embodiment, an embodied intelligent robot autonomous processing system is provided. This system can realize functions such as robot autonomous navigation to the working position, collection of workpiece data, data processing, offline programming, and path planning, thereby realizing an automated processing process.

[0050] The embodied intelligent robot autonomous processing system includes an autonomous navigation unit, a workpiece acquisition unit, a data processing unit, an offline programming unit, and a path planning unit.

[0051] The autonomous navigation unit controls the robot to navigate autonomously to its designated workstation. Equipped with LiDAR and vision sensors, this unit can perceive its surroundings in real time, build an environmental map, and plan the optimal navigation path based on preset workstation coordinates. Employing SLAM (Simultaneous Localization and Mapping) technology, the unit achieves centimeter-level positioning accuracy, ensuring the robot accurately reaches its designated workstation. During navigation, the unit can detect and avoid obstacles in real time, guaranteeing navigation safety.

[0052] The workpiece acquisition unit is used to photograph / scan the workpiece contour and collect data for processing. This unit is equipped with a high-precision industrial camera and a 3D laser scanner, enabling omnidirectional scanning of the workpiece from multiple angles. The industrial camera has a resolution of 4K, capturing minute surface features; the 3D laser scanner has a scanning accuracy of 0.05mm, accurately acquiring the workpiece's 3D geometric information. The workpiece acquisition unit also features an automatic light source adjustment function, automatically adjusting the brightness and angle of the light source according to the workpiece material and ambient lighting conditions to ensure the quality and accuracy of the acquired data.

[0053] The data processing unit is used to process workpiece data. Equipped with a high-performance computing processor, this unit can perform filtering, noise reduction, and point cloud registration on the acquired raw data. Employing advanced image processing algorithms and point cloud processing technology, the unit can extract key geometric features of the workpiece, such as edges, planes, and holes, from the raw data. Furthermore, the data processing unit also features data compression and optimization capabilities, transforming large amounts of raw data into a more compact and efficient data structure, facilitating subsequent processing and analysis.

[0054] The offline programming unit is used to input workpiece models for machining. This unit provides a user-friendly interface and supports importing various 3D model file formats, such as STEP, IGES, and STL. Users can define machining parameters through this unit, such as machining accuracy, machining process, and machining sequence. The offline programming unit also features model editing capabilities, allowing users to modify and adjust the imported model to suit actual machining needs.

[0055] The path planning unit is used to generate the optimal machining path and avoid obstacles. Based on the workpiece geometry and machining parameters, this unit plans the optimal toolpath. Employing efficient algorithms, the path planning unit comprehensively considers factors such as machining quality, machining efficiency, and energy consumption to generate the most optimized machining path. Simultaneously, the path planning unit also features collision detection and obstacle avoidance capabilities, automatically detecting and avoiding potential collisions and interferences during machining to ensure machining safety.

[0056] In practical applications, the data processing unit compares the merged and cropped data with the model input by the offline programming unit. During this process, the data processing unit first preprocesses the collected raw data, including denoising, filtering, and point cloud registration. Then, it compares the processed data with the ideal model to identify the differences between the actual workpiece and the ideal model. Based on workpiece pose registration, the data processing unit can accurately determine the workpiece's position and orientation in the workspace and generate a machining program adapted to the actual workpiece.

[0057] The workpiece acquisition unit identifies the machining features of the workpiece model, applies the machining program, retrieves data from the machining process library, and generates a machining trajectory. In this process, the workpiece acquisition unit first identifies machining features on the workpiece, such as planes, holes, and grooves. Then, based on the type and parameters of these features, it retrieves the corresponding machining parameters and processes from the machining process library. The machining process library contains optimal machining parameters and processes for various materials and machining features, and can automatically select the most suitable machining strategy based on the workpiece material and machining requirements. Based on this information, the workpiece acquisition unit generates a preliminary machining trajectory, including toolpath, feed rate, and cutting parameters.

[0058] The path planning unit optimizes the machining trajectory and generates the optimal machining path. The unit employs advanced optimization algorithms to refine the initially generated machining trajectory, including path smoothing, tool posture optimization, and cutting parameter adjustment. This optimization significantly improves machining efficiency, reduces machining time, and ensures machining quality and accuracy. The path planning unit also considers robot kinematic constraints and workspace limitations to ensure that the planned path is achievable within the robot's working range.

[0059] In a preferred embodiment, the autonomous navigation unit further includes a multi-sensor fusion module, which can fuse data from lidar, visual sensors, and inertial measurement units to improve the accuracy and reliability of navigation and positioning, especially in complex environments.

[0060] In a preferred embodiment, the workpiece acquisition unit further includes an adaptive scanning control module, which can automatically adjust the scanning density and scanning path according to the surface features and complexity of the workpiece, so as to improve scanning efficiency while ensuring data quality.

[0061] In a preferred embodiment, the data processing unit uses a deep learning algorithm for workpiece feature recognition, which can more accurately identify various processing features on the workpiece, especially significantly improving the recognition capability for complex shapes and non-standard features.

[0062] In a preferred embodiment, the path planning unit also has a real-time path adjustment function, which can dynamically adjust the processing path and parameters based on real-time feedback during the processing, such as cutting force and vibration, to further improve processing quality and efficiency.

[0063] Through the collaborative work of the above units, this embodied intelligent robot autonomous processing system can realize a fully automated process from workpiece data acquisition to processing path planning, which greatly improves processing efficiency and accuracy, reduces the need for manual intervention, and is particularly suitable for flexible processing scenarios with small batches and multiple varieties.

[0064] Example 3 An autonomous processing method for an embodied intelligent robot specifically includes the following steps: S1: The robot receives the work instruction and performs a battery self-check. The robot detects the current battery level through its internal battery monitoring module and compares it with a preset safe working battery level threshold. This threshold is typically set to 120% of the battery required to complete a full processing task, ensuring that the task will not be interrupted due to insufficient battery power. If the battery is sufficient (above the safe threshold), the robot activates its navigation system, plans the optimal path, and moves to the work area; if the battery is insufficient (below the safe threshold), the system automatically determines whether to recharge or call upon another robot. When the battery level is below the safe threshold but above a critical value (usually 60% of the safe threshold), the robot automatically moves to a charging station for fast charging; when the battery level is below the critical value, the system sends a request to the central control unit to call upon another robot that is in standby mode and has sufficient battery power to execute the work instruction.

[0065] S2: After the robot arrives at the work area, it initiates the photography program. The robot first uses its vision system to initially locate the workpiece, then precisely adjusts the position of the robot arm to ensure the vision sensor is at the optimal shooting angle. Once the photography program is started, the robot employs a multi-angle, multi-layer scanning method to perform a comprehensive scan of the workpiece's processing area and contour. During the scanning process, a laser rangefinder and a high-definition camera work together to collect depth and texture information from the workpiece surface, forming initial point cloud data. This raw point cloud data typically contains noise points and redundant information. The robot uses point cloud processing algorithms to denoise, filter, and thin the data. Then, it merges the point cloud data collected from multiple angles using a registration algorithm. Finally, based on the actual processing requirements of the workpiece, the point cloud data is cropped to remove point cloud information from non-processed areas, forming an accurate 3D point cloud model of the workpiece.

[0066] S3: Compare the point cloud data generated in step 2 with the 3D model in offline programming. The robot imports the acquired workpiece point cloud data into its internal data processing unit and compares it with the pre-stored standard 3D model of the workpiece. First, the system uses an iterative nearest-point algorithm to perform coarse registration of the two sets of data, determining the approximate position and orientation of the workpiece. Then, a feature matching algorithm is used for precise registration, calculating the precise position, orientation, and orientation parameters of the workpiece in the robot coordinate system. Based on the actual pose information of the workpiece, the system automatically adjusts the preset machining program, recalculates the toolpath, and forms a machining program and machining trajectory adapted to the current workpiece position. The system also performs simulation verification of the machining trajectory to ensure no collision risk and optimizes the machining path to improve machining efficiency and quality.

[0067] S4: Based on the machining program in S3, the composite robot selects the cutting tool. The robot chooses a suitable tool from the tool library according to the material type, machining accuracy requirements, and machining process parameters in the program. The system considers factors such as workpiece material hardness, surface roughness requirements, and machining depth to select the most suitable tool type, size, and geometric parameters. The robot grasps the selected tool using its robotic arm and verifies the tool's integrity and wear condition using its built-in tool inspection system. If an abnormality is detected, the system automatically selects a spare tool. After tool installation, the robot performs tool center point calibration to ensure machining accuracy.

[0068] S5: Based on the machining trajectory in S3, the robot performs automated machining operations. Following the optimized machining trajectory, the robot controls the cutting tool for precise machining. During machining, the robot monitors machining parameters in real time, including spindle speed, feed rate, cutting force, and temperature. When abnormal parameters are detected, the system automatically adjusts the machining parameters or pauses machining. Simultaneously, the robot's vision system continuously monitors the machining area to ensure the safety and stability of the machining process. During machining, the robot also performs mid-process checks at preset detection points to verify machining quality and perform compensatory machining when necessary. The machining process employs an adaptive control strategy, dynamically adjusting machining parameters based on real-time feedback of the machining status to achieve optimal machining results.

[0069] S6: After completing the machining operation, the robot automatically returns the tool to its original position and repositions itself. Upon completion, the robot first checks the machining results to confirm that the task is complete and of acceptable quality. Then, the robot unloads the used tools and returns them to the designated location in the tool magazine, while simultaneously updating the tool usage record and wear data. Next, the robot restores the robotic arm to a safe posture, ensuring that all joints are under minimal load to reduce mechanical stress in a static state. Finally, the robot cleans the work area, removes debris generated during machining, and moves to a pre-set waiting area. In the waiting area, the robot enters a low-power mode but keeps the system monitoring functions active, ready to receive new work instructions and await the next operation.

[0070] In a preferred embodiment, the robot uses a binocular stereo vision system to scan the workpiece in step S2, which can acquire the three-dimensional information of the workpiece more efficiently, reduce scanning time, and improve the accuracy of point cloud data.

[0071] In another preferred embodiment, the robot uses a deep learning algorithm for workpiece pose registration in step S3, which can better handle the registration problem of workpieces with complex shapes and improve registration accuracy and speed.

[0072] In another preferred embodiment, the robot uses an adaptive machining control system in step S5 to monitor cutting force and vibration signals in real time, dynamically adjust machining parameters, and effectively improve machining quality and tool life.

[0073] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for autonomous processing by an embodied intelligent robot, characterized in that, The method includes: The workpiece contour is scanned and photographed using LiDAR and visual SLAM to generate point cloud data. The point cloud data is then subjected to noise filtering and geometric feature analysis. The processed point cloud data is compared with the input 3D model. Based on workpiece pose registration, a machining program is generated using offline programming. Identify the machining features of the workpiece model, apply the machining program, and retrieve the machining process library to generate a machining trajectory; The workpiece pose is updated in real time to form a new workpiece pose. The original trajectory is then subjected to rigid body transformation to generate a new machining trajectory.

2. The autonomous processing method for an embodied intelligent robot according to claim 1, characterized in that, Noise removal from point cloud data is achieved through density statistical filtering and bilateral temporal filtering. The density statistical filtering includes: constructing a neighborhood density calculation and determining a density threshold; The bilateral temporal filtering includes: normal vector calculation, spatiotemporal weight calculation, and point position update. The normal vector guides the spatiotemporal weight allocation, the spatiotemporal weight determines the point position update direction, and the point position update feeds back to affect the normal vector calculation of the next frame.

3. The autonomous processing method for an embodied intelligent robot according to claim 2, characterized in that, The neighborhood density is constructed as follows: a spherical neighborhood is constructed with the target point as the center, and the total number of points within the spherical neighborhood is determined. and the average distance between this point and all points in its neighborhood. ; A density threshold is determined based on the number of points in the neighborhood, and the formula for the density threshold determination model is as follows: in, : Number of points in the neighborhood Global points.

4. The autonomous processing method for an embodied intelligent robot according to claim 2, characterized in that, The normal vector is calculated as follows: for the normal vector of each local neighborhood, the eigenvector corresponding to the minimum eigenvalue is solved by PCA. The formula for the spatiotemporal weight calculation model is as follows: ， in Spatial weights; Range weight; The point position update is based on the weighted average neighborhood point position, while preserving the sharpness of the interface. The point position update model formula is: .

5. The autonomous processing method for an embodied intelligent robot according to claim 1, characterized in that, The processing trajectory achieves the optimal path through surface adaptive sampling, dynamic heuristic function, local surface expansion and optimization, and dynamic obstacle handling.

6. The autonomous processing method for an embodied intelligent robot according to claim 5, characterized in that, The surface adaptive sampling consists of geodesic distance calculation and dynamic bias probability adjustment. The geodesic distance is calculated by calculating the distance between nodes through local parameterization. The formula for the dynamically adjusted target bias probability model is: in The local complexity coefficient of the surface. To adjust the parameters.

7. The autonomous processing method for an embodied intelligent robot according to claim 5, characterized in that, The dynamic heuristic function is: in, Distance cost Security cost, :node Curvature penalty term at the location.

8. The autonomous processing method for an embodied intelligent robot according to claim 1, characterized in that, The method includes: the robot receiving an instruction, performing a power detection according to the instruction, and if the power is sufficient, the robot autonomously navigating to the corresponding workstation; if the power is insufficient, the robot calling upon another robot or charging the battery before executing the instruction.

9. The autonomous processing method for an embodied intelligent robot according to claim 1, characterized in that, The method includes: the robot selecting a processing tool based on the processing program, the robot performing processing operations based on the processing trajectory, and after the processing is completed, the robot returning the tool and autonomously navigating to the waiting area.

10. An embodied intelligent robot autonomous processing system, characterized in that, include: Autonomous navigation unit, used for robots to autonomously navigate to their corresponding workstations; The workpiece acquisition unit is used to photograph / scan the workpiece contour and acquire workpiece data for processing. A data processing unit is used to process the workpiece data. Offline programming unit for inputting workpiece models; The path planning unit is used to generate the optimal processing path and obstacle avoidance. The data processing unit compares the merged and cropped data with the model input by the offline programming unit, and generates a machining program based on workpiece pose registration. The workpiece acquisition unit identifies the processing features of the workpiece model, applies the processing program, retrieves data from the processing technology library, and generates a processing trajectory. The path planning unit optimizes the processing trajectory to generate the optimal processing path.

11. A method for autonomous processing by an embodied intelligent robot, characterized in that, Specifically, the following steps are included: S1. The robot receives the work instruction and performs a power self-check. If the power is sufficient, the robot moves to the work area. If the power is insufficient, it charges the robot or calls on another robot. S2. After the robot arrives at the work area, it starts the photo-taking program to scan the workpiece processing area and the workpiece outline to form point cloud data, and then merges and trims the point cloud data. S3. Compare the point cloud data formed in step 2 with the 3D model in offline programming, and form a machining program and machining trajectory based on workpiece pose registration. S4. Based on the machining program in S3, the composite robot selects the cutting tools. S5. Perform automatic processing based on the processing trajectory of S3; S6. After the robot completes the processing operation, it automatically returns the tool to its original position and moves to the waiting area to wait for the next operation.

Images