A narrow space mechanical arm return control method

Abstract

The application provides a narrow space mechanical arm return control method, which is applied to the technical field of mechanical arm control, obtains geometric shape information and initial position coordinates of key obstacles, plans an initial return path in an ideal case, then scans a region where the key obstacles in the gear box are located in real time to obtain environmental data information, combines the geometric shape information to calculate a position deviation of the gear box relative to a mechanical arm base, and obtains actual position coordinates. On this basis, through dynamic path optimization and intelligent speed planning, the collision risk, low efficiency and limitations of traditional compensation schemes faced by the mechanical arm when returning in a narrow space are solved. Therefore, the method of the application can dynamically adjust the return path and speed according to actual environmental changes, avoid collisions, improve the return efficiency under the premise of safety, and significantly improve the return safety, efficiency and intelligent level of the mechanical arm in a complex narrow space.

Description

Technical Field

[0001] This application relates to the field of robotic arm control technology, and in particular to a method for controlling the return of a robotic arm in a confined space. Background Technology

[0002] In the manufacturing process of wind power equipment, robotic arms are used to perform precision quality monitoring tasks inside the gearbox housing. After completing the task, the robotic arm needs to safely retract from inside the gearbox and return to its initial standby position outside the housing, i.e., perform a return operation.

[0003] Traditional robotic arm return mechanisms typically rely on pre-programmed fixed trajectories. These trajectories are calculated based on standard 3D design drawings of the gearbox housing, theoretically ensuring that the robotic arm will not interfere with the housing's inner wall, gear system, or piping during return. However, in actual manufacturing environments, due to limitations in the positioning accuracy of large lifting equipment, thermal expansion and contraction of tooling fixtures, and casting and machining tolerances of the gearbox housing itself, there will be minute, random translational and rotational errors between the actual position and orientation of each gearbox upon arrival at the inspection station and its theoretical standard position and orientation. These millimeter- to centimeter-level errors are sufficient to pose a collision risk for gearboxes, whose internal space is severely constrained by gears, shafts, and piping. If the fixed, pre-programmed return trajectory is continued, the robotic arm's links, joints, and other components may collide with the housing's inner wall, gear teeth, or cooling pipes during movement, leading to damage to precision equipment or production interruption.

[0004] To address this issue, one direct solution is to increase safety redundancy, i.e., to reserve a larger safety distance when planning the fixed return trajectory. However, this significantly increases the return path length and time. On production lines that prioritize high efficiency, the additional detection cycle can become a bottleneck in the entire manufacturing process, reducing productivity per unit time. Another solution is to introduce an auxiliary sensing module. This module uses a laser rangefinder to scan key reference surfaces inside the gearbox, calculates the precise position and attitude deviation of the gearbox relative to the robotic arm base, and then compensates for the overall translation and rotation of the preset fixed trajectory. However, the robotic arm is a multi-link linkage mechanism with highly nonlinear kinematics. When the trajectory of the end effector is translated or rotated as a whole, the motion trajectories of the joints in the middle of the robotic arm undergo complex and non-intuitive changes. Under certain gearbox pose and attitude deviations, even if the compensated end effector trajectory is safe, its elbow or upper arm may swing out an unexpected arc during the return process, potentially colliding with the side wall of the gearbox, gear surfaces, or high-pressure pipelines. Furthermore, a fixed return speed cannot balance efficiency and safety. In open areas where the robotic arm is far from the inner wall of the gearbox or obstacles, its movement speed can be set to a higher value; while in narrow areas where it is close to gears, bearings or pipes, the movement speed needs to be reduced to ensure safety and control accuracy.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] In view of the shortcomings of the prior art, this application provides a method for controlling the return of a robotic arm in a confined space, which aims to solve the technical problems such as the risk of collision, low efficiency and limitations of traditional compensation schemes that may occur during the return process of a robotic arm in a confined and complex working space.

[0007] A first aspect is a method for controlling the return of a robotic arm in a confined space, the method comprising the following steps:

[0008] S1: Obtain the geometric shape information and initial position coordinates of key obstacles inside the gearbox, and plan the initial return path of the robotic arm under ideal conditions based on the geometric shape information and the initial position coordinates;

[0009] S2: Real-time scanning of the area containing key obstacles inside the gearbox to obtain environmental data information;

[0010] S3: Based on the environmental data and geometric information, calculate the positional deviation of the gearbox relative to the robotic arm base, and adjust the initial position coordinates of the key obstacle according to the positional deviation to obtain the actual position coordinates of the key obstacle;

[0011] S4: Based on the actual position coordinates, perform local optimization on the initial return path to generate a dynamic return path;

[0012] S5: Obtain the minimum distance between the actual position coordinates of the robotic arm and the key obstacle on the dynamic return path;

[0013] S6: Based on the minimum distance, plan the movement speed of the robotic arm, and integrate the dynamic return path with the movement speed to generate the return command of the robotic arm.

[0014] This technical solution effectively addresses the collision risks, low efficiency, and limitations of traditional compensation schemes faced by robotic arms when returning to their original position in confined spaces. Through real-time sensing, dynamic path optimization, and speed planning, it ensures the robotic arm can safely and efficiently complete its return operation in complex environments, significantly improving the automation level and equipment safety in industrial production.

[0015] Furthermore, step S1 includes:

[0016] S11: Obtain a pre-designed gearbox CAD model, and obtain the geometric shape information and initial position coordinates of the key obstacle from the gearbox CAD model;

[0017] S12: Calculate the distance between each link and joint of the robotic arm and the key obstacle based on the geometric shape information and initial position coordinates of the key obstacle;

[0018] S13: Plan the initial return path according to the distance, so that when the robotic arm runs along the initial return path, the distance between it and the key obstacle is minimized.

[0019] Through this technical solution, this application can perform fine-grained planning of the return path of the robotic arm in the initial stage based on accurate CAD model data, ensuring that the minimum safe distance between the robotic arm and obstacles is maintained under ideal conditions, laying a safe foundation for subsequent dynamic adjustments, thereby improving the initial safety and efficiency of the return path.

[0020] Furthermore, step S2 includes:

[0021] S21: Based on the initial return path and the initial position coordinates of the key obstacle, determine a first region where the minimum distance between the robotic arm and the key obstacle during the return process is less than a preset threshold, and a second region where the minimum distance is not less than the preset threshold;

[0022] S22: Scan the first region and the second region at a first frequency and a second frequency respectively to obtain the detailed geometric features of key obstacles in the first region and the approximate geometric features of key obstacles in the second region, wherein the first frequency is greater than the second frequency.

[0023] S23: Integrate the detailed geometric features with the general geometric features to obtain environmental data information.

[0024] Through this technical solution, this application can realize a differentiated scanning strategy for different regions, obtain high-precision detailed geometric features in critical regions, and obtain general geometric features in non-critical regions. This effectively reduces the computational burden of data acquisition and processing while ensuring the safety of the return to position, and improves the real-time performance and efficiency of the system.

[0025] Furthermore, step S23 includes:

[0026] S231: Identify the overlapping area between the detailed geometric features and the approximate geometric features;

[0027] S232: Within the overlapping area, extract the geometric elements that are the same between the detailed geometric features and the approximate geometric features;

[0028] S233: Based on the same geometric elements, calculate the local translation and local rotation of the approximate set features relative to the detailed geometric features within the overlapping region;

[0029] S234: Based on the local translation and local rotation, adjust the approximate geometric features in the overlapping area, and fuse the adjusted approximate geometric features with the detailed geometric features and the approximate geometric features in the non-overlapping area to obtain the environmental data information.

[0030] Through this technical solution, this application can achieve seamless fusion of scanning data with different precision by matching and adjusting the geometric elements of overlapping areas, effectively eliminating inconsistencies between data, thereby generating more accurate and complete environmental data information, and providing reliable input for subsequent path optimization.

[0031] Furthermore, step S3 includes:

[0032] S31: Extract the set of geometric features of key obstacles inside the gearbox from the environmental data information;

[0033] S32: Match the set of geometric features with the corresponding preset set of geometric features in the geometric shape information, and calculate the pose transformation of the gearbox relative to the robot arm base;

[0034] S33: Based on the pose transformation, adjust the initial position coordinates of the key obstacle to obtain the actual position coordinates of the key obstacle.

[0035] Furthermore, step S32 includes:

[0036] S321: Identify at least one reference geometric element from the set of geometric features, the reference geometric element including but not limited to a specific plane, axis or feature point;

[0037] S322: Identify the ideal reference geometric element corresponding to the reference geometric element from the preset geometric feature set corresponding to the geometric shape information;

[0038] S323: Calculate the pose transformation of the gearbox relative to the robot arm base by aligning the reference geometric elements with the ideal reference geometric elements.

[0039] Furthermore, step S4 includes:

[0040] S411: Identify path segments on the initial return path that collide with the key obstacle or whose distance does not meet the requirements;

[0041] S412: For the path segment, based on the actual location coordinates, generate a local alternative path that avoids the key obstacle and meets the distance requirements;

[0042] S413: Connect the local substitution path with the remaining part of the initial return path to form the dynamic return path.

[0043] Furthermore, step S5 includes:

[0044] S51: Represent each link and joint of the robotic arm as a first set of geometric entities, and represent the key obstacle as a second set of geometric entities;

[0045] S52: At the sampling point of the dynamic return path, calculate the distance between the first geometric entity set and the second geometric entity set;

[0046] S53: The minimum distance at the sampling point is taken as the minimum distance between the robotic arm and the key obstacle.

[0047] Furthermore, step S6 includes:

[0048] S61: Take the minimum distance as the input value and substitute it into the preset speed adjustment function to calculate the movement speed of the robotic arm at each point on the dynamic return path;

[0049] S62: Integrate the motion speed with the dynamic return path in a timing sequence to generate a return command for the robotic arm;

[0050] The preset speed adjustment function includes:

[0051] S611: Pre-collect the minimum distance between the robotic arm and key obstacles at each sampling point on the dynamic return path;

[0052] S612: Define a maximum safe distance threshold and a maximum permissible speed corresponding to the maximum safe distance threshold; define a minimum safe distance threshold and a minimum permissible speed corresponding to the minimum safe distance threshold.

[0053] S613: Iterate through the minimum distance of each sampling point. When the minimum distance is greater than or equal to the maximum safe distance threshold, the movement speed is the maximum permissible speed; when the minimum distance is less than or equal to the minimum safe distance threshold, the movement speed is the minimum permissible speed.

[0054] S614: When the minimum distance is greater than the minimum safety threshold and less than the maximum safety threshold, the movement speed = minimum allowable speed + (maximum allowable speed - minimum allowable speed) * (minimum distance - minimum safety threshold) / (maximum safety threshold - minimum safety threshold)^2.

[0055] Secondly, a confined space robotic arm return control system is provided for implementing any of the methods described above, the system comprising:

[0056] First acquisition module: acquires the geometric shape information and initial position coordinates of key obstacles inside the gearbox, and plans the initial return path of the robotic arm under ideal conditions based on the geometric shape information and the initial position coordinates;

[0057] Scanning module: Scans the area inside the gearbox where key obstacles are located in real time to obtain environmental data information;

[0058] Calculation module: Based on the environmental data and geometric shape information, calculates the positional deviation of the gearbox relative to the robotic arm base, and adjusts the initial position coordinates of the key obstacle according to the positional deviation to obtain the actual position coordinates of the key obstacle;

[0059] Path optimization module: Based on the actual location coordinates, locally optimize the initial return path to generate a dynamic return path;

[0060] The second acquisition module acquires the minimum distance between the actual position coordinates of the robotic arm and the key obstacle on the dynamic return path.

[0061] Command generation module: Based on the minimum distance, it plans the movement speed of the robotic arm and integrates the dynamic return path with the movement speed to generate the return command of the robotic arm.

[0062] Beneficial Effects: This application proposes a method for controlling the return of a robotic arm in confined spaces. By acquiring the geometric shape information and initial position coordinates of key obstacles inside the gearbox and planning an ideal initial return path, it lays the foundation for the robotic arm's return operation. Subsequently, by scanning the area where the key obstacle is located inside the gearbox in real time, environmental data is acquired, and the positional deviation of the gearbox relative to the robotic arm base is calculated by combining the geometric shape information. This allows for the adjustment of the initial position coordinates of the key obstacle, resulting in the actual position coordinates. This step effectively solves the collision risk caused by deviations in the actual position and attitude of the gearbox in existing technologies. Through precise perception and compensation, the accuracy of obstacle position information is ensured. Based on this, this application locally optimizes the initial return path according to the actual position coordinates, generating a dynamic return path. This overcomes the limitations of traditional fixed trajectories that cannot adapt to changes in the actual environment, improving the adaptability and safety of the path. Next, the minimum distance between the robotic arm and the key obstacle on the dynamic return path is obtained, and the movement speed of the robotic arm is planned based on this minimum distance. Finally, the dynamic return path and movement speed are integrated to generate the return command for the robotic arm. This series of steps, through real-time perception, dynamic path optimization, and intelligent speed planning, effectively solves the problems of collision risk, low efficiency, and limitations of traditional compensation schemes faced by robotic arms when returning to their original position in confined spaces in existing technologies. Therefore, the method of this application can dynamically adjust the return path and speed according to changes in the actual environment, avoiding collisions, and improving return efficiency while ensuring safety. This significantly enhances the safety, efficiency, and intelligence level of the robotic arm's return operation in complex and confined spaces. Attached Figure Description

[0063] Figure 1 This is a flowchart of a method for controlling the return of a robotic arm in a confined space, as proposed in this application.

[0064] Figure 2 This is a structural diagram of a robotic arm return control system proposed in this application.

[0065] Figure 3 This is a simplified schematic diagram of a robotic arm return control system for confined spaces proposed in this application.

[0066] Labeling Explanation: 201, First Acquisition Module; 202, Scanning Module; 203, Calculation Module; 204, Path Optimization Module; 205, Second Acquisition Module; 206, Instruction Generation Module. Detailed Implementation

[0067] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and marked in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0068] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0069] Please refer to Figure 1 A method for controlling the return of a robotic arm in a confined space, the method comprising the following steps:

[0070] S1: Obtain the geometric shape information and initial position coordinates of key obstacles inside the gearbox, and plan the initial return path of the robotic arm under ideal conditions based on the geometric shape information and initial position coordinates;

[0071] S2: Real-time scanning of the area containing key obstacles inside the gearbox to obtain environmental data information;

[0072] S3: Based on environmental data and geometric information, calculate the positional deviation of the gearbox relative to the robotic arm base, and adjust the initial position coordinates of the key obstacle according to the positional deviation to obtain the actual position coordinates of the key obstacle;

[0073] S4: Based on the actual position coordinates, perform local optimization on the initial return path to generate a dynamic return path;

[0074] S5: Obtain the minimum distance between the actual position coordinates of the robotic arm and the key obstacle on the dynamic return path;

[0075] S6: Based on the minimum distance, plan the movement speed of the robotic arm, integrate the dynamic return path and movement speed, and generate the return command of the robotic arm.

[0076] In this application, "robotic arm" specifically refers to an industrial robotic arm that operates in confined spaces (such as inside a gearbox), and its return process requires high-precision control to avoid collisions. A gearbox is a mechanical device used to transmit power and change speed and torque, containing complex structures such as gears, shafts, bearings, and pipes. These structures constitute key obstacles in the robotic arm's return process. Therefore, key obstacles refer to all structural components inside the gearbox that may collide with the robotic arm, including but not limited to gears, shafts, pipes, and the inner wall of the housing.

[0077] Geometric information refers to the three-dimensional geometric data of key obstacles, such as their size, shape, and surface features.

[0078] Initial position coordinates refer to the theoretical position and orientation information of a key obstacle relative to the robot arm base under ideal conditions (e.g., based on a gearbox CAD model).

[0079] The initial return path refers to the preset trajectory along which the robotic arm safely returns from the working position to the initial standby position under ideal conditions.

[0080] Environmental data information refers to the actual three-dimensional point cloud data or image data of the area where key obstacles are located inside the gearbox, obtained through real-time scanning.

[0081] Position deviation refers to the difference between the actual position and orientation of the gearbox and its theoretical initial position and orientation.

[0082] Actual position coordinates refer to the precise position and attitude information of a critical obstacle in the current actual environment after adjustment based on position deviation.

[0083] Dynamic return path refers to the robotic arm's return trajectory, which is generated by locally optimizing the initial return path based on the actual position coordinates and can adapt to changes in the current environment.

[0084] Minimum distance refers to the shortest distance between the robotic arm (including links and joints) and key obstacles when the robotic arm moves on the dynamic return path.

[0085] The return command refers to complete motion planning data containing the motion angles, speeds, and time sequences of each joint of the robotic arm, used to guide the robotic arm to complete the return operation safely and efficiently.

[0086] In step S1, the geometric information and initial position coordinates of key obstacles can be directly extracted from the pre-designed gearbox CAD model. The gearbox CAD model typically contains accurate three-dimensional geometric information of all internal components and their relative positions in the assembly.

[0087] When planning the initial return path, a collision detection-based path planning algorithm can be used. Commonly used path planning algorithms in the prior art include, but are not limited to, RRT (Rapid Exploratory Random Tree) and PRM (Probabilistic Route Graph) algorithms. This application preferentially selects the RRT algorithm. Specifically, the RRT algorithm is a sampling-based path planning method. Its core idea is to randomly sample points in the workspace of the robotic arm and gradually construct a tree-like path from the starting point to the target point. In this application, the application process of the RRT algorithm can be understood as follows:

[0088] First, the system acquires the geometric shape information and initial position coordinates of key obstacles inside the gearbox. This information typically comes from a pre-designed gearbox CAD model, providing a precise three-dimensional geometric description of the obstacles and their spatial positions under ideal conditions. Simultaneously, the robotic arm's own geometric model (including the shape and dimensions of each link and joint) and its kinematic model (describing the relationship between the angles of each joint and the position and orientation of the end effector) are also loaded into the planning system.

[0089] Next, the RRT algorithm starts from the robot arm's initial return position (i.e., its current position after the task is completed) and performs iterative random sampling within the robot arm's workspace. In each iteration, the algorithm generates a random point and attempts to connect the nearest node in the tree to that random point. During the connection process, a straight path segment from the tree node to the random point is generated.

[0090] Rigorous collision detection is performed when generating each path segment. The system uses the acquired geometric shape information and initial position coordinates of key obstacles, as well as the geometric model of the robotic arm itself, to determine whether the newly generated path segment interferes with any key obstacles inside the gearbox. If a collision is detected, the path segment is discarded and resampled. If the path segment has no collision, it is added to the RRT tree.

[0091] This process continues until a node in the RRT tree can be connected to the robot arm's target return position (i.e., the preset initial standby position), or the tree expansion reaches the preset number of iterations or time limit. Once a collision-free path from the starting point to the target point is found, this path is determined as the robot arm's initial return path.

[0092] The goal of the initial return path planned by the RRT algorithm is to ensure that the robotic arm maintains a safe distance from key obstacles while running along the path, and to minimize the path length. The advantage of the RRT algorithm is that it can effectively explore high-dimensional space and find feasible paths in complex environments. Even in the confined and obstacle-filled interior space of a gearbox, it can generate a theoretically safe initial return trajectory.

[0093] In step S2, an optical sensing unit mounted on the robotic arm can be used to scan the inside of the gearbox. The optical sensing unit can be a stereo vision system that uses two or more cameras to acquire images from different perspectives, thereby generating three-dimensional environmental data information.

[0094] In step S3, a set of geometric features of key obstacles inside the gearbox is extracted from the real-time acquired environmental data, such as identifying specific planes, axes, or feature points. Simultaneously, a corresponding set of ideal geometric features is extracted from preset geometric shape information (such as a gearbox CAD model). Then, a feature matching algorithm is used to match the real-time extracted set of geometric features with the set of ideal geometric features. As those skilled in the art will understand, the feature matching algorithm typically involves the following key steps:

[0095] The first step involves extracting a set of geometric features of key obstacles inside the gearbox from the environmental data obtained through real-time scanning. These features can be basic geometric elements such as points, lines, and surfaces, or more complex structural features such as holes, edges, corners, and specific curved surfaces. Similarly, a set of corresponding ideal geometric features can be extracted from the pre-designed gearbox CAD model. These ideal features are based on the precise geometric information of the design drawings.

[0096] The second step, in order to effectively match the extracted features, is to describe these features. For ease of description, existing technologies typically use feature descriptors to describe geometric features. A feature descriptor is a compact and discriminative mathematical representation that captures the local geometric or topological information of a feature. For example, for point features, feature descriptors can be used to describe them based on information such as the normal vectors and curvature of its surrounding points; for line features, feature descriptors can be used to describe them based on their length, direction, and angle with other lines; and for surface features, feature descriptors can be used to describe them based on their area, normal vectors, and shape.

[0097] The third step, after feature description is complete, is feature matching. This is typically achieved by comparing the similarity between the descriptors of the extracted features in real time and the ideal feature descriptors. Specifically, cosine similarity can be used to calculate the similarity between the descriptors of each feature in the real-time feature set and all features in the ideal feature set, and the pair with the highest similarity is selected as the matching pair.

[0098] The fourth step, after obtaining a sufficient number of correct matching pairs, is to use these pairs to calculate the pose transformation of the gearbox relative to the robot arm base. The pose transformation typically consists of a rotation matrix and a translation matrix. This can be solved using the least squares method or other optimization algorithms, with the goal of minimizing the distance error between the matched feature points.

[0099] This matching process allows for the calculation of the precise translation and rotation transformation matrices of the gearbox relative to the robotic arm base in the actual environment, i.e., the position deviation. Once the position deviation is obtained, it is applied to the initial position coordinates of key obstacles to obtain their precise actual position coordinates in the current environment.

[0100] In step S4, path segments on the initial return path that pose a collision risk with critical obstacles or whose distance does not meet safety requirements are identified. This can be achieved by densely sampling the initial return path and performing collision detection or distance calculation at each sampling point. Once problematic path segments are identified, local alternative paths that avoid the obstacles and meet safety distance requirements are generated for these segments based on the actual position coordinates of the critical obstacles. The generation of local alternative paths can employ local path planning algorithms, such as a sample-based local RRT algorithm. Finally, these local alternative paths are connected to the remaining safe portion of the initial return path to form a complete dynamic return path. This dynamic return path can adapt to the actual positional deviation of the gearbox, ensuring that the robotic arm maintains a safe distance from obstacles throughout the return process.

[0101] In step S5, each link and joint of the robotic arm is represented as a first set of geometric entities, and the key obstacle is represented as a second set of geometric entities. Then, dense sampling is performed on the dynamic return path. For each sampling point, the shortest distance between the first set of geometric entities and the second set of geometric entities is calculated. The minimum value among the distances calculated at all sampling points is taken as the minimum distance between the robotic arm and the key obstacle.

[0102] In step S6, the minimum distance obtained in step S5 is used as the input value and substituted into a preset speed adjustment function to calculate the movement speed of the robotic arm at each point on the dynamic return path. This speed adjustment function can be segmented according to a safety distance threshold. For example, a maximum safety distance threshold and a minimum safety distance threshold can be defined. When the minimum distance is greater than or equal to the maximum safety distance threshold, the movement speed of the robotic arm can be set to the maximum allowable speed to improve efficiency. When the minimum distance is less than or equal to the minimum safety threshold, the movement speed is reduced to the minimum allowable speed to ensure safety and control accuracy. When the minimum distance is between the maximum and minimum safety thresholds, the movement speed can be calculated based on the proportional relationship between the minimum distance and the threshold. Finally, the planned movement speed and the dynamic return path are integrated in time to generate a complete robotic arm return command containing the movement angle, speed, and timestamp of each joint.

[0103] In summary, this application provides a more intelligent, safe, and efficient method for controlling the return of a robotic arm in confined spaces through the synergistic effect of real-time perception, dynamic path optimization, and adaptive speed planning, which significantly improves the operational reliability and production efficiency of the robotic arm in complex industrial environments.

[0104] Furthermore, step S1 includes:

[0105] S11: Obtain the pre-designed gearbox CAD model, and extract the geometric shape information and initial position coordinates of key obstacles from the gearbox CAD model;

[0106] S12: Calculate the distance between each link and joint of the robotic arm and the key obstacle based on the geometric shape information and initial position coordinates of the key obstacle;

[0107] S13: Plan the initial return path based on the spacing to minimize the distance between the robotic arm and key obstacles when the robotic arm runs along the initial return path.

[0108] The pre-designed gearbox CAD model can be understood as a digital 3D model generated during the actual manufacturing or design phase of the gearbox. It contains precise geometric descriptions and relative positional relationships of all structural components, parts, and potential obstacles inside the gearbox. By analyzing this gearbox CAD model, the geometric information (such as dimensions, contours, surface normals, etc.) of the key obstacles (e.g., gears, bearings, sensors, etc.) and their initial position coordinates in the gearbox coordinate system can be accurately extracted.

[0109] Specifically, after obtaining the precise geometric information and initial position of the key obstacles, a pre-analysis of the possible paths the robotic arm might traverse during its return process is required. This step involves using distance calculation algorithms, such as the GJK algorithm, to calculate the minimum Euclidean distance between each link and joint of the robotic arm (as geometric entities) and the key obstacles inside the gearbox (as another set of geometric entities) under different orientations. Calculating this distance is a prerequisite for ensuring that the robotic arm operates without collisions along the planned path.

[0110] The reason this application chose the GJK algorithm is that the GJK (Gilbert-Johnson-Keerthi) algorithm is an efficient algorithm for calculating the distance between convex bodies, and it is particularly suitable for calculating the minimum distance between two convex geometries.

[0111] In practical applications, to simplify calculations and improve efficiency, the links, joints, and key obstacles inside the gearbox of a robotic arm are often simplified as convex bodies. Therefore, the GJK algorithm can quickly and accurately calculate the minimum Euclidean distance between them. This algorithm iteratively finds the closest pair of points on two convex bodies, gradually narrowing the search range until it converges to the minimum distance.

[0112] After calculating the distance between the robotic arm and the key obstacle, a Fast Exploration Random Tree (RRT) can be used to generate an initial return path from the robotic arm's current position to the target return position. The steps for generating the initial return path using RRT have been explained above and will not be repeated here. During planning, this initial return path is designed to minimize the distance between the robotic arm and the key obstacle while satisfying collision-free constraints. The aim is to fully utilize the limited space, improve return efficiency, and reserve a certain margin for subsequent dynamic adjustments.

[0113] Furthermore, step S2 includes:

[0114] S21: Based on the initial return path and the initial position coordinates of the key obstacle, determine the first region where the minimum distance between the robotic arm and the key obstacle during the return process is less than a preset threshold, and the second region where the minimum distance is not less than the preset threshold.

[0115] S22: Scan the first region and the second region at the first frequency and the second frequency respectively to obtain the detailed geometric features of the key obstacles in the first region and the approximate geometric features of the key obstacles in the second region. The first frequency is greater than the second frequency.

[0116] S23: Integrate detailed geometric features with general geometric features to obtain environmental data information.

[0117] By calculating the minimum distances between the robotic arm's links and joints and key obstacles as it travels along its initial return path, and comparing this distance with a preset threshold, potential collision risk areas can be identified. When the minimum distance is less than the preset threshold, this area is defined as the first area, indicating a high collision risk between the robotic arm and the obstacle. More precise sensing is required, so the first area is scanned at a higher first frequency to ensure that detailed geometric features of the key obstacle, such as its surface texture, minor protrusions, or depressions, are captured. When the minimum distance is not less than the preset threshold, this area is defined as the second area, indicating a relatively low collision risk. Lower precision sensing is used in this second area, so a lower second frequency is used to obtain the approximate outline and main dimensions of the key obstacle, reducing the amount of scanned data and processing burden. The preset threshold can be set according to the safety requirements of the actual application scenario and the motion characteristics of the robotic arm; for example, it can be set as the minimum safe distance between the robotic arm's end effector or links.

[0118] Finally, the detailed point cloud data acquired in the first region is integrated with the approximate model data acquired in the second region. For example, the detailed point cloud data can be embedded into the approximate model to obtain a unified environmental data information containing different levels of precision. This integrated environmental data information will be used for subsequent path optimization and collision detection to ensure that the robotic arm can operate efficiently and avoid obstacles safely during its return process.

[0119] Furthermore, step S23 includes:

[0120] S231: Identify the overlapping areas between detailed geometric features and general geometric features;

[0121] S232: Within the overlapping region, extract the geometric elements that are the same between the detailed geometric features and the approximate geometric features;

[0122] S233: Based on the same geometric elements, calculate the local translation and local rotation of the approximate set features relative to the detailed geometric features within the overlapping region;

[0123] S234: Based on the local translation and local rotation, adjust the approximate geometric features in the overlapping area, and then fuse the adjusted approximate geometric features with the detailed geometric features and the approximate geometric features of the non-overlapping area to obtain environmental data information.

[0124] Specifically, a feature matching algorithm can be used to obtain matching pairs between detailed and approximate geometric features; these matching pairs represent overlapping regions. The detailed principle of using the feature matching algorithm to obtain matching pairs between detailed and approximate geometric features has been explained in detail in the steps above. In this step, detailed and approximate geometric features are described separately using feature descriptors, and then feature matching is performed based on these descriptors to obtain the matching pairs.

[0125] Within the identified overlapping region, elements with the same physical meaning or geometric shape, such as identical corner points, edge segments, planar or curved surface features, are identified and extracted from both detailed and approximate geometric features. These identical geometric elements serve as the alignment reference. By overlapping the detailed and approximate geometric features with this alignment reference, the spatial translation and rotation deviations of the approximate geometric features relative to the detailed geometric features can be readily obtained. Calculating these translation and rotation deviations based on the alignment reference with identical geometric elements is simply a least-squares optimization problem. For example, suppose the outer contour of a pipe A inside a gearbox has a deviation during alignment and overlap, where a point in the detailed geometric features is P_i, and the corresponding point in the approximate geometric features is Q_i. The goal is to find a rotation matrix R and a translation vector t such that the transformed Q_i' = R * Q_i + t is as close as possible to P_i. In existing technologies, the least squares optimization problem is typically solved using Singular Value Decomposition (SVD) or iterative optimization algorithms (such as the Levenberg-Marquardt algorithm). To improve computational efficiency, this application employs the SVD method to solve this least squares optimization problem. The solution process is a mature technique in the prior art and is not the focus of this application; therefore, it will not be elaborated upon here. This application uses the SVD method to directly calculate the optimal rotation matrix R and translation vector t. This allows for the acquisition of the local translation and local rotation of the approximate set features relative to the detailed geometric features.

[0126] Finally, using local translation and local rotation, the approximate geometric features within the overlapping region are precisely spatially adjusted to achieve optimal alignment with the detailed geometric features. Subsequently, the adjusted approximate geometric features are seamlessly integrated with the detailed geometric features and the approximate geometric features of the non-overlapping regions, thereby generating a complete, consistent, and highly accurate environmental data.

[0127] Furthermore, step S3 includes:

[0128] S31: Extract the set of geometric features of key obstacles inside the gearbox from environmental data information;

[0129] S32: Match the set of geometric features with the corresponding preset set of geometric features in the geometric shape information, and calculate the pose transformation of the gearbox relative to the robot arm base;

[0130] S33: Based on the pose transformation, adjust the initial position coordinates of the key obstacle to obtain the actual position coordinates of the key obstacle.

[0131] The extraction of the geometric feature set can employ various point cloud processing or image processing algorithms. For example, methods such as segmentation, clustering, and feature point detection can be used to distinguish the geometric information of background noise from key obstacles, thereby obtaining accurate geometric data for subsequent matching. These segmentation, clustering, and feature point detection methods are considered conventional and mature techniques by those skilled in the art.

[0132] For ease of explanation, this application only exemplifies segmentation methods. The goal of segmentation algorithms is to divide point cloud or image data into different regions or objects. For example, in point cloud data, segmentation can be performed using a region growing method. A region growing algorithm starts with one or more seed points and merges adjacent points into the same region based on preset similarity criteria (such as normal direction, color, or curvature), thereby separating key obstacles from the background. This results in a set of geometric features of the key obstacles inside the gearbox.

[0133] The geometric information is derived from the gearbox's CAD model, which includes the precise geometry and initial position coordinates of key obstacles. The matching process aims to perform feature matching between the two geometric datasets and calculate the gearbox's pose transformation relative to the robotic arm base using these correspondences. This pose transformation typically includes three-dimensional translation and three-dimensional rotation, reflecting minor deviations that may occur during the actual installation or operation of the gearbox.

[0134] Since the actual installation position of the gearbox in the real environment may deviate from the ideal design position, the initial position coordinates of the critical obstacle can be converted into its precise position coordinates in the current actual environment through pose transformation. The actual position coordinates are an important basis for the robotic arm to perform subsequent path planning and collision detection, ensuring the safe return of the robotic arm in confined spaces.

[0135] Furthermore, step S32 includes:

[0136] S321: Identify at least one reference geometric element from a set of geometric features, the reference geometric element including but not limited to a specific plane, axis or feature point;

[0137] S322: Identify the ideal reference geometric element corresponding to the reference geometric element from the preset set of geometric features in the geometric shape information;

[0138] S323: Calculate the pose transformation of the gearbox relative to the robot arm base by aligning the reference geometry with the ideal reference geometry.

[0139] Reference geometric elements refer to geometric entities with clear geometric definitions and easily identifiable characteristics on key obstacles inside the gearbox. These could be a plane, an axis, the center point of a hole, or a corner point. These elements are typically assigned specific functions or structural meanings during the design phase, thus having clear definitions in the gearbox CAD model and being reliably extracted and identified from actual scanning data. Identifying these highly identifiable reference geometric elements provides a reliable benchmark for subsequent matching and pose calculations.

[0140] Ideal reference geometric elements refer to geometric elements in a pre-designed gearbox CAD model that correspond to reference geometric elements identified in actual scan data and possess ideal positions and orientations. These ideal reference geometric elements represent the geometric characteristics of the gearbox in its standard or design state. They are typically obtained by analyzing a pre-designed gearbox CAD model and extracting geometric information corresponding to the type and position of the actual reference geometric elements.

[0141] Specifically, aligning the reference geometric elements with the ideal reference geometric elements means using geometric transformations (such as translation and rotation) to make the reference geometric elements identified in the actual scan data coincide or achieve an optimal match with the ideal reference geometric elements in the CAD model in space. This alignment process can be achieved using a feature matching algorithm combined with the least squares method. The purpose of alignment is to determine the actual spatial position and orientation of the gearbox relative to the robot arm base, thereby calculating the pose transformation of the gearbox relative to the robot arm base. The pose transformation is usually represented in the form of a homogeneous transformation matrix, which contains translation vectors and rotation matrices, and can completely describe the relative relationship between the two coordinate systems.

[0142] By employing the aforementioned technical solution, matching and pose transformation calculations using reference geometric elements can effectively improve the computational accuracy and robustness of the gearbox's pose transformation relative to the robotic arm base. Compared to directly matching the entire set of geometric features, the alignment method based on specific reference geometric elements can better address issues such as data noise and local occlusion, reducing the possibility of mismatches. This allows for accurate determination of the gearbox's actual pose even in confined spaces, where environmental data may be incomplete or inaccurate. This lays the foundation for precise adjustment of the actual coordinates of critical obstacles, ultimately ensuring the safety and reliability of the robotic arm's return path.

[0143] Furthermore, step S4 includes:

[0144] S41: Identify path segments on the initial return path that collide with key obstacles or whose distance does not meet requirements;

[0145] S42: For a path segment, generate a local alternative path that avoids key obstacles and meets distance requirements based on the actual location coordinates;

[0146] S43: Connect the local substitution path with the remaining part of the initial return path to form a dynamic return path.

[0147] The identification of path segments on the initial return path that collide with key obstacles or whose distance does not meet requirements involves analyzing each point or path segment on the initial return path using path planning or collision detection algorithms to calculate the distance between the robotic arm and the key obstacle at that point or path segment. If this distance is less than a preset safe distance threshold, or if a potential collision is detected, the path segment is considered to not meet the requirements. This safe distance threshold can be set based on the robotic arm's motion accuracy, the geometry of the key obstacle, and the safety requirements of the actual application scenario.

[0148] Furthermore, based on the path planning algorithm of Fast Exploratory Random Tree (RRT), a new path can be replanned in the local space around the unsatisfactory path segments to bypass key obstacles while maintaining a safe distance. The generation of this local alternative path utilizes the actual position coordinates of the key obstacles to ensure the effectiveness and safety of the path. The generated local alternative path is then connected to the remainder of the initial return path to form a dynamic return path. The connection process must ensure the continuity and smoothness of the path to avoid abrupt changes or jitters in the robotic arm during path switching.

[0149] Furthermore, step S5 includes:

[0150] S51: Represent each link and joint of the robotic arm as a first set of geometric entities, and represent the key obstacles as a second set of geometric entities;

[0151] S52: At the sampling point of the dynamic return path, calculate the distance between the first geometric entity set and the second geometric entity set;

[0152] S53: Take the minimum distance at the sampling point as the minimum distance between the robotic arm and the key obstacle.

[0153] Specifically, the links and joints of the robotic arm can be abstracted as a series of geometric entities, such as cylinders, cuboids, and spheres, which together constitute a first set of geometric entities. Similarly, the key obstacle can also be represented as a second set of geometric entities consisting of one or more geometric entities. This representation aims to simplify complex physical shapes, enabling them to be efficiently processed and calculated by computer programs. The first and second sets of geometric entities can be obtained based on pre-acquired gearbox CAD model data or geometric data obtained through 3D scanning.

[0154] Furthermore, at each sampling point, the posture and position of the robotic arm are determined. Based on this, the distance between the first set of geometric entities (representing the robotic arm) and the second set of geometric entities (representing key obstacles) can be calculated using the GJK algorithm.

[0155] Finally, by traversing all sampling points along the dynamic return path and calculating the distance between the first set of geometric entities and the second set of geometric entities at each sampling point, the minimum value among these distances is determined as the minimum distance between the robotic arm and the critical obstacle. This minimum distance is a key indicator for assessing whether there is a collision risk and safety margin during the robotic arm's return process.

[0156] Furthermore, step S6 includes:

[0157] S61: Take the minimum distance as the input value and substitute it into the preset speed adjustment function to calculate the movement speed of the robotic arm at each point on the dynamic return path.

[0158] S62: Integrates the motion speed and dynamic return path in a timing sequence to generate the return command for the robotic arm;

[0159] The preset speed adjustment functions include:

[0160] S611: Pre-collect the minimum distance between the robotic arm and key obstacles at each sampling point on the dynamic return path;

[0161] S612: Define the maximum safe distance threshold and the maximum permissible speed corresponding to the maximum safe distance threshold, and define the minimum safe distance threshold and the minimum permissible speed corresponding to the minimum safe distance threshold;

[0162] S613: Traverse the minimum distance of each sampling point. When the minimum distance is greater than or equal to the maximum safe distance threshold, the movement speed is the maximum allowable speed; when the minimum distance is less than or equal to the minimum safe distance threshold, the movement speed is the minimum allowable speed.

[0163] S614: When the minimum distance is greater than the minimum safety threshold and less than the maximum safety threshold, the movement speed = minimum permissible speed + (maximum permissible speed - minimum permissible speed) * (minimum distance - minimum safety threshold) / (maximum safety threshold - minimum safety threshold)^2.

[0164] This application's solution achieves adaptive planning of the robotic arm's movement speed by introducing a preset speed adjustment function. Specifically, when the minimum distance between the robotic arm and a key obstacle is large, the robotic arm can operate at a higher speed, thereby improving return efficiency. As the minimum distance decreases, the robotic arm's movement speed gradually decreases according to a preset nonlinear function, especially when approaching the minimum safety threshold, where the speed decrease is more significant. This distance-based adaptive speed adjustment mechanism enables the robotic arm to maintain stable and safe movement in confined spaces, effectively avoiding the risk of sudden deceleration or collision due to excessively close distances. By integrating the dynamic return path with the finely planned movement speed in a time sequence, the final generated return command ensures that the robotic arm is both efficient and safe throughout the return process. Specifically, integrating the movement speed with the dynamic return path in a time sequence to generate the robotic arm's return command means combining the calculated movement speeds at each point with the position information on the dynamic return path to form a complete motion sequence containing time, position, and speed information—that is, the robotic arm's return command.

[0165] As can be seen from the above, the robotic arm return control method proposed in this application can help the robotic arm safely and efficiently complete the return task in specific environments such as wind turbine gearboxes, which have complex internal structures, narrow spaces, and random positional deviations of the workpiece itself. Its core concept is as follows:

[0166] Before the robotic arm begins operation, the standard design drawings of the wind turbine gearbox are studied to identify the areas inside the gearbox that have the greatest impact on the robotic arm's safe return, such as the edges of the main bearing housing, bends in the cooling pipes, and gear tooth tips. These areas are referred to as key geometric features in this application. Those skilled in the art will establish a simplified spatial relationship description of these key geometric features, recording their relative positions and shape information. Simultaneously, based on the gearbox's state in its standard position, an initial return path is planned—the path the robotic arm should ideally follow.

[0167] Before the robotic arm returns to its original position within the gearbox being inspected, the end effector-mounted optical sensing unit rapidly scans the identified key geometric feature areas inside the gearbox. Using this scan data, the system identifies the actual position and shape of these key geometric features within the gearbox in real time.

[0168] The system then compares the real-time identified key geometric features with pre-prepared standard spatial relationship descriptions. Through this comparison, the system can calculate the actual position and attitude deviation of the gearbox relative to the robot arm's base with great precision. This deviation is not generalized to the entire gearbox, but focuses on the relative position changes of key obstacles that have the greatest impact on the robot arm's movement.

[0169] Next, based on the calculated deviation, instead of redrawing a detailed 3D map of the entire gearbox, the system cleverly adjusts its perception of the spatial location of the key obstacles stored inside.

[0170] Specific implementation method:

[0171] Deviation Calculation: Using the results of feature matching, calculate the overall pose deviation of the current gearbox relative to the standard position. This can be achieved using a registration algorithm based on the least squares method. Assume N key feature points or geometric centers are identified, and their positions in the standard coordinate system are... The position in the current scanning coordinate system is... , where i is the index of the key feature point. A transformation matrix needs to be found. (R is the rotation matrix, t is the translation vector), such that Minimum. This transformation matrix T represents the overall pose deviation of the current gearbox relative to the standard position.

[0172] The system does not rebuild the entire environment model. Instead, it updates the coordinates of the simplified geometric elements representing key obstacles (e.g., the simplified envelope of a gear, the cylindrical model of a pipe) stored in the first step based on the calculated overall pose deviation. Specifically, if the transformation matrix of a key obstacle in the standard coordinate system is... Then its transformation matrix in the current actual environment is It will be updated to In this way, the system's internal perception of the spatial location of key obstacles remains consistent with the actual situation, without having to process the complex geometric information of the entire gearbox.

[0173] After the system updates its spatial understanding of key obstacles, it will make local fine-tuning adjustments based on the initial return path preset in the first step, thereby generating a dynamic return path that can effectively avoid these obstacles in their actual positions. This path will ensure that all links and joints of the robotic arm will not collide with any structure inside the gearbox throughout the entire return process.

[0174] At the same time, the system also measures the safe distance between various parts of the robotic arm and these updated critical obstacles in real time. Based on these distances, the system intelligently generates a speed curve: when the robotic arm is away from obstacles and in a relatively open area, it allows the robotic arm to move at a faster speed to save time; while when the robotic arm approaches narrow or critical components such as gears or pipes, it automatically reduces its speed to ensure precise control and absolute safety.

[0175] Ultimately, the system will integrate this dynamic return path and the corresponding velocity curve to generate a complete sequence of joint motion commands, ready to be sent to the robotic arm.

[0176] The internal structure of wind turbine gearboxes is complex, and each gearbox has a random deviation in position from millimeters to centimeters at its workstation. Traditional robotic arm control methods may simply translate or rotate the entire pre-set return path to accommodate the overall deviation of the gearbox. However, in narrow bends or areas with numerous obstacles, a link or joint of the robotic arm may accidentally collide with the internal structure due to this simple overall compensation. This is because the movement of the robotic arm is not a simple linear translation; its various joints and links form complex spatial trajectories during movement. While this solution is simple, it cannot address the risk of localized collisions caused by the complex movements of the robotic arm, nor can it dynamically adjust the speed according to the actual distance, resulting in low return efficiency or insufficient safety.

[0177] The proposed solution first memorizes the key geometric features and spatial relationships of the most critical obstacles inside the gearbox (such as bearing housing edges and cooling pipes). When the actual gearbox is in position, sensors at the end of the robotic arm act like "scouts," quickly scanning the actual positions of these critical obstacles. The system compares the scanned positions with the pre-memorized key geometric features to accurately calculate the positional deviations of these critical obstacles relative to the robotic arm. With this precise deviation information, the system does not need to redraw a detailed 3D map of the entire gearbox; instead, it only adjusts the spatial positions of these critical obstacles. Then, based on a preset return path, it makes local fine-tuning to ensure that all parts of the robotic arm can avoid these obstacles in their actual positions. Simultaneously, it intelligently adjusts its movement speed according to the actual distance between the robotic arm and these obstacles: in open areas far from obstacles, the robotic arm can move quickly; while in narrow areas close to obstacles, it automatically slows down, ensuring precise control and absolute safety. This focused perception and adjustment enables the robotic arm to safely and efficiently complete the return-to-position task in complex and confined spaces such as wind turbine gearboxes, where the internal structure is known but the overall position is off.

[0178] Please refer to Figure 2 , Figure 3A confined space robotic arm return control system, used to implement any of the above methods, the system comprising:

[0179] First acquisition module 201: Acquires the geometric shape information and initial position coordinates of key obstacles inside the gearbox, and plans the initial return path of the robotic arm under ideal conditions based on the geometric shape information and initial position coordinates;

[0180] Scanning module 202: Scans the area inside the gearbox where key obstacles are located in real time to obtain environmental data information;

[0181] Calculation module 203: Based on environmental data and geometric information, calculates the positional deviation of the gearbox relative to the robotic arm base, and adjusts the initial position coordinates of the key obstacle according to the positional deviation to obtain the actual position coordinates of the key obstacle;

[0182] Path optimization module 204: Based on the actual position coordinates, locally optimize the initial return path to generate a dynamic return path;

[0183] Second acquisition module 205: Acquires the minimum distance between the actual position coordinates of the robotic arm and the key obstacle on the dynamic return path;

[0184] Instruction generation module 206: Based on the minimum distance, it plans the movement speed of the robotic arm, integrates the dynamic return path and movement speed, and generates the return command of the robotic arm.

[0185] Specifically, the first acquisition module 201 is responsible for executing step S1 in the above method, acquiring basic information about obstacles through a data interface or a preset database, and performing preliminary path planning based on this information.

[0186] The scanning module 202 corresponds to step S2 in the above method. It uses integrated sensors (such as LiDAR, vision sensors, etc.) to dynamically perceive the working environment in order to obtain the latest environmental status.

[0187] The function of the calculation module 203 corresponds to step S3 in the above method. Its core is to accurately determine the current actual position of the obstacle through data fusion and coordinate transformation.

[0188] The path optimization module 204 executes step S4 of the above method, and adjusts the preset initial path in real time according to the actual obstacle positions provided by the calculation module to avoid potential collisions.

[0189] The second acquisition module 205 corresponds to step S5 in the above method. It is responsible for performing collision detection and distance calculation on the optimized path to assess the safety margin between the robotic arm and obstacles.

[0190] The instruction generation module 206 executes step S6 of the above method, dynamically adjusts the movement speed of the robotic arm according to the minimum distance, and combines the speed information with the path information to form an instruction that can directly drive the robotic arm to perform a return operation.

[0191] The solution in this application achieves systematization and automation by assigning each key step in the aforementioned confined space robotic arm return control method to an independent, functionally defined module. Specifically, the first acquisition module 201 provides basic obstacle information and initial path planning for the entire return process, laying the foundation for subsequent dynamic adjustments. The scanning module 202 continuously provides real-time environmental perception data, ensuring the system can promptly grasp environmental changes. The calculation module 203 compares this real-time data with preset information to accurately calibrate the actual position of obstacles, solving the problem of planning failure due to positional deviations in traditional methods. Based on this, the path optimization module 204 can locally adjust the initial path according to the latest obstacle position, ensuring the robotic arm can safely avoid obstacles in dynamic environments. The second acquisition module 205 performs a safety assessment of the optimized path, quantifying the minimum distance between the robotic arm and obstacles, providing a basis for speed planning. Finally, the command generation module 203 dynamically adjusts the robotic arm's movement speed according to the safe distance and generates precise return commands based on the optimized path, enabling the robotic arm to achieve safe and efficient return in confined and dynamic environments. This modular design ensures clear responsibilities for each functional unit and efficient data flow, effectively solving integration and execution challenges that may be encountered in actual deployment.

[0192] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0193] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for return control of a narrow space robot arm, characterized by, The method comprises the steps of: S1: acquiring geometric shape information and initial position coordinates of key obstacles inside the gearbox, and planning an initial homing path of the mechanical arm in an ideal case according to the geometric shape information and the initial position coordinates; S2: real-time scanning of a region where the key obstacles inside the gearbox are located to obtain environmental data information; S3: calculating a position deviation of the gearbox relative to a base of the mechanical arm according to the environmental data information and the geometric shape information, and adjusting the initial position coordinates of the key obstacles according to the position deviation to obtain actual position coordinates of the key obstacles; S4: locally optimizing the initial homing path according to the actual position coordinates to generate a dynamic homing path; S5: acquiring a minimum distance between the mechanical arm and the actual position coordinates of the key obstacles on the dynamic homing path; S6: planning a movement speed of the mechanical arm according to the minimum distance, and integrating the dynamic homing path and the movement speed to generate a homing instruction of the mechanical arm.

2. The method of claim 1, wherein, Step S1 comprises: S11: acquiring a pre-designed gearbox CAD model, and acquiring the geometric shape information and the initial position coordinates of the key obstacles from the gearbox CAD model; S12: calculating distances between each link and joint of the mechanical arm and the key obstacles according to the geometric shape information and the initial position coordinates of the key obstacles; S13: planning the initial homing path according to the distances so that the distance between the mechanical arm and the key obstacles is minimum when the mechanical arm runs along the initial homing path.

3. The method of claim 1, wherein, Step S2 comprises: S21: determining a first region where a minimum distance between the mechanical arm and the key obstacles is less than a preset threshold and a second region where the minimum distance is not less than the preset threshold in the homing process of the mechanical arm according to the initial homing path and the initial position coordinates of the key obstacles; S22: scanning the first region and the second region at a first frequency and a second frequency respectively to obtain detailed geometric features of the key obstacles in the first region and sketchy geometric features of the key obstacles in the second region, the first frequency being greater than the second frequency; S23: integrating the detailed geometric features and the sketchy geometric features to obtain the environmental data information.

4. The method of claim 3, wherein, Step S23 comprises: S231: identifying an overlapping region between the detailed geometric features and the sketchy geometric features; S232: extracting the same geometric elements between the detailed geometric features and the sketchy geometric features in the overlapping region; S233: calculating a local translation amount and a local rotation amount of the sketchy geometric features relative to the detailed geometric features in the overlapping region according to the same geometric elements; S234: adjusting the sketchy geometric features in the overlapping region according to the local translation amount and the local rotation amount, and fusing the adjusted sketchy geometric features with the detailed geometric features and the sketchy geometric features in the non-overlapping region to obtain the environmental data information.

5. The method of claim 1, wherein, Step S3 comprises: S31: Extract a set of geometric features of the key obstacle inside the gearbox from the environment data information; S32: Match the set of geometric features with a corresponding preset set of geometric features in the geometric shape information, and calculate a pose transformation of the gearbox relative to the robot base; S33: According to the pose transformation, adjust the initial position coordinates of the key obstacle to obtain actual position coordinates of the key obstacle.

6. The method of claim 5, wherein, Step S32 includes: S321: Identify at least one reference geometric element from the set of geometric features, including but not limited to a specific plane, an axis, or a feature point; S322: Identify an ideal reference geometric element corresponding to the reference geometric element from the corresponding preset set of geometric features in the geometric shape information; S323: Calculate the pose transformation of the gearbox relative to the robot base by aligning the reference geometric element with the ideal reference geometric element.

7. The method of claim 1, wherein, Step S4 includes: S41: Identify a path segment on the initial homing path that collides with the key obstacle or does not meet the distance requirement; S42: For the path segment, generate a local replacement path that avoids the key obstacle and meets the distance requirement based on the actual position coordinates; S43: Connect the local replacement path with the remaining part of the initial homing path to form the dynamic homing path.

8. The method of claim 1, wherein, Step S5 includes: S51: Represent each link and joint of the robot arm as a first set of geometric entities, and represent the key obstacle as a second set of geometric entities; S52: Calculate the distance between the first set of geometric entities and the second set of geometric entities at the sampling points of the dynamic homing path; S53: Take the minimum value of the distances at the sampling points as the minimum distance between the robot arm and the key obstacle.

9. The method of claim 1, wherein, Step S6 includes: S61: Take the minimum distance as an input value and substitute it into a preset speed adjustment function to calculate the movement speed of the robot arm at each point on the dynamic homing path; S62: Time-integrate the movement speed with the dynamic homing path to generate a homing instruction for the robot arm; The preset speed adjustment function includes: S611: Pre-collect the minimum distance between the robot arm and the key obstacle at each sampling point on the dynamic homing path; S612: Define a maximum safe distance threshold and a maximum allowed speed corresponding to the maximum safe distance threshold, and define a minimum safe distance threshold and a minimum allowed speed corresponding to the minimum safe distance threshold; S613: Traverse the minimum distance of each sampling point, when the minimum distance is greater than or equal to the maximum safe distance threshold, the movement speed is the maximum allowed speed; when the minimum distance is less than or equal to the minimum safe distance threshold, the movement speed is the minimum allowed speed; S614: When the minimum distance is greater than the minimum safety distance threshold and less than the maximum safety distance threshold, the motion speed = minimum allowable speed + (maximum allowable speed - minimum allowable speed) * (minimum distance - minimum safety distance threshold) / (maximum safety distance threshold - minimum safety distance threshold) ^ 2.

10. A narrow space robot arm return control system, characterized by, The system for implementing the method of any one of claims 1-9 comprises: A first acquisition module: acquiring geometric shape information and initial position coordinates of key obstacles inside the gearbox, and planning an initial homing path of the mechanical arm in an ideal case according to the geometric shape information and the initial position coordinates; A scanning module: real-time scanning of an area where the key obstacles inside the gearbox are located to obtain environmental data information; A calculation module: calculating a position deviation of the gearbox relative to a base of the mechanical arm according to the environmental data information and the geometric shape information, and adjusting the initial position coordinates of the key obstacles to obtain actual position coordinates of the key obstacles according to the position deviation; A path optimization module: locally optimizing the initial homing path according to the actual position coordinates to generate a dynamic homing path; A second acquisition module: acquiring a minimum distance between the actual position coordinates of the mechanical arm and the key obstacles on the dynamic homing path; An instruction generation module: planning a motion speed of the mechanical arm according to the minimum distance, and integrating the dynamic homing path and the motion speed to generate a homing instruction of the mechanical arm.

Images