Double-arm welding robot collision judgment method based on point cloud

Abstract

The invention provides a double-arm welding robot collision judgment method based on point cloud. According to the method, the simplified model is constructed according to the easy-to-collide part of the physical model, and the workload of subsequent data processing is reduced. Point cloud model generation and collision judgment can be independent of a control system, and few PLC resources are occupied in the system operation process. Visualization of the point cloud model is realized based on an open C + + programming library PCL, so that a user can conveniently observe the predicted pose of the point cloud model. Data points of the point cloud model are directly calculated through an algorithm, and the timeliness and accuracy of collision judgment are improved. The robot carries out welding work, a space coordinate system is calibrated, and a visual positioning sensor obtains space coordinates of solid model feature targets; according to the current rotating speed of each shaft, calculating the spatial pose of an easy-to-collide part after n seconds, and performing spatial pose matching of the point cloud model; and carrying out collision judgment on the point cloud model and an external obstacle, namely dynamic-static collision judgment. If collision occurs, emergency braking is carried out, otherwise, the next step is continued; and carrying out collision judgment between the point cloud models of the two arms, namely dynamic-dynamic collision judgment. If collision occurs, emergency braking is carried out, otherwise, the next step is continued; and the robot moves to the next welding position and continues to conduct collision judgment. According to the double-arm welding robot collision judgment method based on the point cloud, collision judgment can be carried out with high accuracy and high timeliness, and the safety of the whole welding process is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of high-speed welding operations, and particularly to a collision determination method for a dual-arm welding robot based on point cloud. Background Art

[0002] A dual-arm welding robot is a new type of welding equipment mainly used for complex high-speed welding operations. It has extremely strong working capabilities and professional skills, and can complete welding tasks faster and more accurately. The dual-arm robot ensures efficient and safe operation in a complex environment through collaborative control and intelligent perception technologies, greatly improving the welding efficiency and quality.

[0003] During the welding process of a dual-arm welding robot, collisions are very likely to occur, and its collision problems mainly involve two aspects: one is collision with obstacles in the external working environment; the other is collision between the two arms themselves. Although the classic collision determination methods for dual-arm welding robots are effective, they cannot balance accuracy, timeliness, and economy: Sensor-based methods such as vision and sound waves can detect collision risks in real time, but their detection accuracy is limited by factors such as environmental light and equipment placement; The perception skin detection method can cover the entire robot, but it has complex wiring, weak anti-interference ability, and increases the computing workload of the processor; The force and torque sensor detection method is accurate but expensive.

[0004] Therefore, it is of great significance to develop a collision determination method for a dual-arm welding robot based on point cloud. Summary of the Invention

[0005] The purpose of the present invention is to provide a collision determination method for a dual-arm welding robot based on point cloud to solve the problems existing in the prior art.

[0006] The technical solution adopted to achieve the purpose of the present invention is as follows. A collision determination method for a dual-arm welding robot based on point cloud includes the following steps:

[0007] 1) Construct a simplified model based on the easily collidable parts of the physical model of the dual-arm welding robot and import it into the computer. Use the built-in algorithms of PCL to process the simplified model to obtain a point cloud model with a preset number of points. Among them, the dual-arm welding robot includes two bases and two multi-axis robotic arms. The multi-axis robotic arms are installed on the corresponding bases. The multi-axis robotic arm has 7 rotating shafts to achieve multi-pose positioning of the end of the robotic arm. The multi-axis robotic arm includes a two-axis joint, a three-axis joint, a three-axis extension joint, and an end joint. The two-axis joint is fixed on the rotating disk of the base through a first axis, the three-axis joint is fixedly connected to the two-axis drive shaft of the two-axis joint through a second axis, the three-axis extension joint is in interference fit with the rotating shaft of the three-axis extension motor of the three-axis joint through a three-axis extension drive shaft, and the end joint is fixedly connected to the three-axis extension of the three-axis extension joint through a flange. There is a work area guardrail around the dual-arm welding robot. The easily collidable part is the part where the working spaces of the two multi-axis robotic arms overlap. The simplified model uses a cube or a cylindrical bounding box to replace the connecting rod.

[0008] 2) Select the characteristic target points of the model, and use the built-in algorithms of PCL to extract the coordinate data of the target points and insert them into the point cloud model.

[0009] 3) Calibrate the spatial coordinate system. During the welding operation of the dual-arm welding robot, the visual positioning sensor obtains the spatial coordinates of the characteristic target points of the physical model.

[0010] 4) Calculate the spatial pose of the easily collidable part after n seconds according to the current rotational speed of each axis, and perform spatial pose matching of the point cloud model.

[0011] 5) Perform collision determination between the point cloud model and external obstacles. If a collision occurs, the robot will brake emergently, otherwise continue to the next step. The external obstacle is the work area guardrail. The collision determination is achieved by judging whether there are points beyond the boundary.

[0012] 6) Perform collision determination between the dual-arm point cloud models. If a collision occurs, the robot will brake emergently, otherwise continue to the next step. The collision determination is achieved by judging whether the number of data points satisfying the density condition increases by 5 or more.

[0013] 7) The robot moves to the next welding position, and then jumps to step 4) to continue the collision determination.

[0014] Further, in step 1), the pcl_mesh_sampling algorithm is used to generate a point cloud model with a preset number of points. In step 2), the obj2pcd algorithm is used to extract the coordinate data of the target points.

[0015] Further, in step 2), 3 groups or more of the edge points of the physical model are selected as the characteristic target points.

[0016] Further, in step 4), through the point cloud registration algorithm, using the coordinate relationships of at least three pairs of points, the matching of the point cloud model to the predicted pose is achieved. The calculation process of the point cloud model matching is as follows:

[0017] A' = TA (1)

[0018] In the formula, A is the set of point cloud coordinates before coordinate transformation, A′ is the set of point cloud coordinates after coordinate transformation, and T is the coordinate transformation matrix.

[0019] Further, in step 5), the collision determination formula is as follows:

[0020] P ∈ N (2)

[0021] In the formula, P is the point in the point cloud model, and N is the collision range between the robotic arm and the environmental obstacle. If there is a point in the point cloud model that satisfies formula (2), it is regarded that the robotic arm collides with the environmental obstacle. Otherwise, there is no collision.

[0022] Further, in step 5), the collision determination is based on the nearest neighbor search algorithm within the R radius to evaluate the local density of the point cloud model, and then the collision determination is realized. Taking the point P in the point cloud model as the center point and R as the radius, the calculation formula for the number of points W within the radius is as follows:

[0023] W = num(P, R) (3)

[0024] In step 6), a threshold q is set. After the nearest neighbor search within the radius R, if the number of inner neighbor points of the data points greater than q increases by 5 or more compared with the original number of points, it is determined that a collision will occur. Otherwise, it is determined that no collision will occur. The calculation formula for the number of points Q that meet the threshold condition is as follows:

[0025] Q = num(W, q) (4)

[0026] The collision determination formula is as follows:

[0027] Q' - Q ≥ 5 (5)

[0028] In the formula, Q′ is the number of inner neighbor points within the R of the next pose, and Q is the number of inner neighbor points within the R of the current pose. If the point cloud model satisfies formula (3), it is regarded that a collision occurs between the robotic arms. Otherwise, there is no collision.

[0029] The technical effects of the present invention are beyond doubt:

[0030] A. Construct a simplified model according to the easily collidable parts of the physical model, reducing the workload of subsequent data processing;

[0031] B. The generation of the point cloud model and the collision determination can be independent of the control system, occupying less PLC resources during the operation of the system;

[0032] C. Visualize the point cloud model based on the open C++ programming library PCL, facilitating users to observe the predicted pose of the point cloud model;

[0033] D. Directly calculate the data points of the point cloud model through the algorithm, improving the timeliness and accuracy of collision determination. Description of the Drawings

[0034] Figure 1 Schematic diagrams of the physical model, simplified model, and point cloud model in the dynamic-static case; 1a is the physical model of the welding robot; 1b is the simplified model corresponding to the easily colliding part of the physical model; 1c is the pose of the point cloud model predicted after n seconds;

[0035] Figure 2 Schematic diagrams of the physical model, simplified model, and point cloud model in the dynamic-dynamic case; 2a is the physical model of the welding robot; 2b is the simplified model corresponding to the easily colliding part of the physical model; 2c is the pose of the point cloud model predicted after n seconds;

[0036] Figure 3 Is the collision determination flowchart in the dynamic-static case;

[0037] Figure 4 Is the collision determination flowchart in the dynamic-dynamic case;

[0038] Figure 5 Is the collision determination flowchart.

[0039] In the figure: Work area guardrail 1, multi-axis robotic arm 2, base 3. Detailed Implementation Modes

[0040] The present invention will be further described below in conjunction with embodiments, but it should not be understood that the above-mentioned subject scope of the present invention is limited to the following embodiments. Without departing from the above-mentioned technical idea of the present invention, various substitutions and changes made according to ordinary technical knowledge and customary means in the art should be included within the protection scope of the present invention.

[0041] Embodiment 1:

[0042] A point cloud is a set of points representing the surface characteristics of an object, including three-dimensional coordinates and reflection intensity. PCL (Point Cloud Library) is an open-source C++ programming library available for multiple platforms, supporting various point cloud processing functions such as filtering, segmentation, registration, etc., and is widely used in multiple fields.

[0043] See Figure 5 , this embodiment provides a collision determination method for a dual-arm welding robot based on point cloud, including the following steps:

[0044] 1) Construct a simplified model based on the easily collidable parts of the physical model of the dual-arm welding robot and import it into the computer. Use the built-in algorithms of PCL to process the simplified model to obtain a point cloud model with a preset number of points. Among them, the dual-arm welding robot includes two bases 3 and two multi-axis robotic arms 2. The multi-axis robotic arms 2 are installed on the corresponding bases 3. The multi-axis robotic arm 2 has 7 rotating shafts to achieve multi-posture positioning of the end of the robotic arm. The multi-axis robotic arm 2 includes a two-axis joint, a three-axis joint, a three-axis extension joint, and an end joint. The two-axis joint is fixed on the rotating disk of the base 3 through a first axis, the three-axis joint is fixedly connected to the second-axis drive shaft of the two-axis joint through a second axis, the three-axis extension joint is in interference fit with the rotating shaft of the three-axis extension motor of the three-axis joint through a three-axis extension drive shaft, and the end joint is fixedly connected to the three-axis extension of the three-axis extension joint through a flange. There is a working area guardrail 1 beside the dual-arm welding robot. The easily collidable part is the part where the working spaces of the two multi-axis robotic arms 2 overlap. The simplified model uses a cube or a cylindrical bounding box to replace the connecting rods.

[0045] 2) Select the target points of the model features, use the built-in algorithms of PCL to extract the coordinate data of the target points and insert them into the point cloud model.

[0046] 3) Calibrate the space coordinate system. During the welding operation of the dual-arm welding robot, the visual positioning sensor obtains the space coordinates of the target points of the physical model features.

[0047] 4) Calculate the spatial pose of the easily collidable part after n seconds based on the current rotational speeds of each axis, and perform spatial pose matching of the point cloud model.

[0048] 5) Refer to Figure 1 and Figure 3 , perform the collision determination between the point cloud model and external obstacles, that is, dynamic-static collision determination. If a collision occurs, the robot brakes urgently; otherwise, continue to the next step. The external obstacle is the working area guardrail 1. The collision determination is achieved by judging whether there are points exceeding the boundary.

[0049] 6) Refer to Figure 2 and Figure 4 , perform the collision determination between the dual-arm point cloud models, that is, dynamic-dynamic collision determination. If a collision occurs, the robot brakes urgently; otherwise, continue to the next step. The collision determination is achieved by judging whether the number of data points satisfying the density condition increases by 5 or more.

[0050] 7) The robot moves to the next welding position, and then jumps to step 4) to continue the collision determination.

[0051] It should be noted that constructing a simplified model based on the easily collidable parts of the physical model can effectively reduce the computational workload of subsequent work. Select an appropriate number of point clouds according to the surface area of the test model. If the number of point clouds is large, the subsequent data processing time will increase; conversely, collision determination cannot be effectively performed. In step 3), the purpose of calibration is to ensure that the relative poses between the point cloud model and the models of each device at the welding station are exactly the same. To ensure obtaining the accurate spatial coordinates of the feature target points of the physical model, an appropriate number of visual positioning sensors and their arrangement methods should be considered. When the robot stops, the collision determination process automatically exits.

[0052] Example 2:

[0053] The main content of this example is the same as that of Example 1. Among them, in step 1), the pcl_mesh_sampling algorithm is used to generate a point cloud model with a preset number of points. In step 2), the obj2pcd algorithm is used to extract the target point coordinate data.

[0054] Example 3:

[0055] The main content of this example is the same as that of Example 1 or 2. Among them, in step 2), the edge points of 3 groups or more of physical models are selected as feature target points.

[0056] Example 4:

[0057] The main content of this example is the same as any one of Examples 1 to 3. Among them, in step 4), through the point cloud registration algorithm, using the coordinate relationships of at least three pairs of points, the matching of the point cloud model to the predicted pose is realized. The calculation process of point cloud model matching is as follows:

[0058] A' = TA (1)

[0059] In the formula, A is the set of point cloud coordinates before coordinate transformation, A′ is the set of point cloud coordinates after coordinate transformation, and T is the coordinate transformation matrix.

[0060] The collision determination formula is as follows:

[0061] P ∈ N (2)

[0062] In the formula, P is the point in the point cloud model, and N is the collision range between the robotic arm and the environmental obstacle. If there is a point in the point cloud model that satisfies formula (2), it is regarded as a collision between the robotic arm and the environmental obstacle. Conversely, there is no collision.

[0063] Collision determination is based on the nearest neighbor search algorithm within the R radius to evaluate the local density of the point cloud model, and then collision determination is realized. Taking the point P in the point cloud model as the center point and R as the radius, the formula for the number of points W within the radius is as follows:

[0064] W = num(P, R) (3)

[0065] In step 6), set the threshold q. After the nearest neighbor search within the radius R, if the number of data points with the number of inner nearest neighbors greater than q increases by 5 or more compared to the original number of points, it is determined that a collision will occur. Otherwise, it is determined that no collision will occur. The calculation formula for the number of points Q that meet the threshold condition is as follows:

[0066] Q = num(W, q) (4)

[0067] The collision determination formula is as follows:

[0068] Q' - Q ≥ 5 (5)

[0069] In the formula, Q′ is the number of inner nearest neighbors within the next pose R, and Q is the number of inner nearest neighbors within the current pose R. If the point cloud model satisfies formula (3), it is considered that a collision occurs between the robotic arms. Otherwise, no collision occurs.

Claims

1. A collision determination method for a dual-arm welding robot based on point cloud, characterized in that Including the following steps: 1) Construct a simplified model based on the easily collidable parts of the physical model of the dual-arm welding robot and import it into the computer; use the built-in algorithm of PCL to process the simplified model to obtain a point cloud model with a preset number of points; wherein, the dual-arm welding robot includes two bases (3) and two multi-axis robotic arms (2); the multi-axis robotic arms (2) are installed on the corresponding bases (3); the multi-axis robotic arms (2) have 7 rotating shafts to achieve multi-pose positioning of the end of the robotic arm; the multi-axis robotic arms (2) include a two-axis joint, a three-axis joint, a three-axis extension joint and an end joint; the two-axis joint is fixed on the rotating disk of the base (3) through a first axis, the three-axis joint is fixedly connected to the two-axis drive shaft of the two-axis joint through a second axis, the three-axis extension joint is in interference fit with the rotating shaft of the three-axis extension motor of the three-axis joint through a three-axis extension drive shaft, and the end joint is fixedly connected to the three-axis extension of the three-axis extension joint through a flange; the working area is surrounded by a working area guardrail (1) beside the dual-arm welding robot; the easily collidable part is the part where the working spaces of the two multi-axis robotic arms (2) overlap; the simplified model uses a cube or a cylindrical bounding box to replace the connecting rod; 2) Select the model feature target points, use the built-in algorithm of PCL to extract the target point coordinate data and insert it into the point cloud model; 3) Calibrate the space coordinate system; during the welding work of the dual-arm welding robot, the visual positioning sensor obtains the space coordinates of the feature target points of the physical model; 4) Calculate the spatial pose of the easily collidable part after n seconds according to the current rotational speed of each axis, and perform the spatial pose matching of the point cloud model; 5) Perform the collision determination between the point cloud model and the external obstacle; if a collision occurs, the robot will brake emergently, otherwise continue to the next step; the external obstacle is the working area guardrail (1); the collision determination is achieved by judging whether there are points beyond the boundary; 6) Perform the collision determination between the dual-arm point cloud models; if a collision occurs, the robot will brake emergently, otherwise continue to the next step; the collision determination is achieved by judging whether the number of data points satisfying the density condition increases by 5 or more; 7) The robot moves to the next welding position, and then jumps to step 4) to continue the collision determination.

2. The method for collision determination of a dual-arm welding robot based on point cloud according to claim 1, wherein: In step 1), the pcl_mesh_sampling algorithm is used to generate a point cloud model with a preset number of points; in step 2), the obj2pcd algorithm is used to extract the target point coordinate data.

3. The method for judging the collision of a dual-arm welding robot based on point cloud according to claim 1, wherein: In step 2), select the edge points of 3 groups or more of the physical models as the feature target points.

4. The method for determining collision of a dual-arm welding robot based on point cloud according to claim 1, wherein: In step 4), through the point cloud registration algorithm, using the coordinate relationships of at least three pairs of points, the matching of the point cloud model to the predicted pose is realized; the calculation process of the point cloud model matching is as follows: A' = TA (1) In the formula, A is the set of point cloud coordinates before coordinate transformation, A′ is the set of point cloud coordinates after coordinate transformation, and T is the coordinate transformation matrix.

5. The method for determining collision of a dual-arm welding robot based on point cloud according to claim 1, wherein In step 5), the collision determination formula is as follows: P ∈ N (2) In the formula, P is the point in the point cloud model, and N is the collision range between the robotic arm and the environmental obstacle; if there is a point in the point cloud model that satisfies formula (2), it is considered that a collision occurs between the robotic arm and the environmental obstacle; otherwise, no collision occurs.

6. The method for determining collision of a dual-arm welding robot based on point cloud according to claim 1, wherein: In step 5), the collision determination is based on the local density of the point cloud model evaluated by the nearest neighbor search algorithm within the radius R, and then the collision determination is realized; with the point P in the point cloud model as the center point and R as the radius, the calculation formula for the number of points W within the radius is as follows: W = num(P, R) (3) In step 6), a threshold q is set. After the nearest neighbor search within the radius R, if the number of data points with the number of inner nearest neighbors greater than q increases by 5 or more compared to the original number of points, it is determined that a collision will occur; otherwise, it is determined that no collision will occur; the calculation formula for the number of points Q that meet the threshold condition is as follows: Q = num(W, q) (4) The collision determination formula is as follows: Q' - Q ≥ 5 (5) In the formula, Q′ is the number of inner nearest neighbors within the radius R of the next pose, and Q is the number of inner nearest neighbors within the radius R of the current pose; if the point cloud model satisfies formula (3), it is considered that a collision occurs between the robotic arms; otherwise, no collision occurs.

Images