Curved surface intelligent spraying model training method, spraying method and system thereof

Abstract

The invention provides a curved surface intelligent spraying model training method, a spraying method and a system thereof, and relates to the technical field of robot path planning. The intelligent curved surface spraying method comprises the steps that point cloud data of an object are collected through a depth camera; carrying out voxel filtering, statistical filtering and RANSAC plane segmentation preprocessing on the point cloud data in sequence; calculating the center of mass and the PCA principal component direction of the preprocessed point cloud, and carrying out rasterizing division on the space where the point cloud is located according to a preset grid size and the number of rows and columns to generate a grid map; performing path planning by using a DQN model; and the robot carries out spraying operation according to the planned path. The invention further provides a curved surface intelligent spraying model training method and a curved surface intelligent spraying system so as to cover all spraying areas with the optimal path and avoid invalid movement.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot path planning, and in particular relates to a curved surface intelligent spraying model training method, a spraying method and a system thereof. Background Art

[0002] In industrial automation, robot path planning is a critical issue, particularly in scenarios requiring comprehensive surface coverage, such as painting and inspection. Traditional path planning methods, often based on pre-set rules or models, struggle to adapt to complex and changing environments and task requirements.

[0003] At present, most spraying robots use teaching spraying, spraying according to the trajectory pre-set by technical staff. This spraying method is effective based on the experience of predecessors, but the initial programming work is large, and the trajectory optimization is not obvious, the efficiency is low, and the waste of paint is relatively large.

[0004] With the development of deep reinforcement learning technology, it has shown great potential in solving complex decision-making problems. However, the research on applying deep reinforcement learning to point cloud processing and path planning is still in its infancy, and there is currently a lack of effective systems and methods.

[0005] Therefore, in this technical field, when faced with unknown objects, existing technologies have not yet achieved intelligent solution of effective spraying paths to achieve comprehensive coverage of the object surface by spraying. Summary of the Invention

[0006] To this end, the technical problem to be solved by the present invention is to provide a surface intelligent spraying model training method, a spraying method and a system thereof, which, based on the reinforcement learning strategy, covers all sprayable areas with the optimal path to avoid invalid movement.

[0007] In a first aspect, the present invention provides a method for training a curved surface intelligent spraying model, comprising the following steps:

[0008] Step S11, collecting workpiece point cloud data:

[0009] Collect point cloud data of the workpiece through a depth camera;

[0010] Step S12: point cloud data preprocessing and rasterization:

[0011] Preprocess the collected workpiece point cloud data to remove noise points and outliers; rasterize the preprocessed point cloud data to generate a raster map;

[0012] Step S13: Training the DQN (Deep Q-Network) model:

[0013] Define the environment state as a grid map;

[0014] The robot's action space includes moving up, down, left, right, and jumping;

[0015] The reward mechanism is to encourage the robot to cover all spraying areas and reduce repeated visits;

[0016] A multi-layer fully connected neural network is used as DQN. The experience replay buffer is collected by interacting with the environment, and the network is trained using experience replay to learn the optimal path planning strategy.

[0017] Furthermore, step S12 specifically includes:

[0018] Step S121, sequentially performing voxel filtering and statistical filtering on the collected workpiece point cloud data;

[0019] Step S122, using RANSAC (Random Sample Consensus) to perform plane segmentation;

[0020] Step S123, calculating the centroid of the point cloud and performing feature extraction using PCA (Principal Component Analysis);

[0021] Step S124, projecting the point cloud data onto a two-dimensional plane based on the PCA results;

[0022] Step S125, dividing the grid into grids according to the preset grid size and number of rows and columns;

[0023] Step S126: construct a grid map based on the point cloud density within the grid.

[0024] Furthermore, in step S13, the action space A is:

[0025] A={a0,a1,a2,a3,a4,a5}

[0026] Among them, a0 moves up, a1 moves down, a2 moves left, a3 moves right, a4 jumps to the top of the next column, and a5 jumps to the bottom of the next column;

[0027] The rewards are set as:

[0028]

[0029] in,

[0030] r new : Rewards for unvisited areas;

[0031] r direction : The movement direction of the unvisited area is rewarded vertically and penalized horizontally;

[0032] r jump : Penalty for executing jump action;

[0033] r repeat : Penalty for repeatedly visiting covered areas or staying put;

[0034] r boundary : There is no additional reward or penalty for moving outside the boundary.

[0035] Furthermore, the reward settings are as follows:

[0036] r new =15, r repeat =-20,

[0037]

[0038]

[0039] In a second aspect, the present invention provides a method for intelligent spraying of curved surfaces, comprising the following steps:

[0040] Step S21, collecting point cloud data of the object through a depth camera;

[0041] Step S22, performing voxel filtering, statistical filtering, and RANSAC plane segmentation preprocessing on the point cloud data in sequence;

[0042] Calculate the centroid and PCA principal component direction of the preprocessed point cloud, rasterize the space where the point cloud is located according to the preset grid size and number of rows and columns, and generate a grid map;

[0043] Step S23, performing path planning using the DQN model trained by any of the training methods described in the first aspect above;

[0044] In step S24, the robot performs spraying operations according to the planned path.

[0045] Furthermore, step S24 specifically includes:

[0046] Obtain the point cloud data of the grid; calculate the position of the center point coordinates of the grid point cloud in the robot arm base coordinate system, calculate the point cloud normal vector of the grid through PCA to determine the spraying posture and form a posture sequence;

[0047] The pose sequence is converted to generate joint space instructions for the robot arm, and the robot performs spraying operations according to the joint space instructions.

[0048] Furthermore, the transformation pose sequence generates the manipulator joint space instruction including the transformation matrix T from the camera coordinate system to the base coordinate system. camera_to_base Calculated by the following formula:

[0049] T camera_to_base =T end_effect0r ×T camera_to_end_effector

[0050] Among them, T end_effector is the position of the end effector of the robot arm, T camera_to_end_effector is the pose of the camera relative to the end effector;

[0051] The coordinates of the center point of the grid point cloud in the base coordinate system p base for:

[0052]

[0053] Among them, (x, y, z) is the coordinate of the center point of the grid point cloud in the camera coordinate system.

[0054] Furthermore, converting the pose sequence to generate the manipulator joint space instructions also includes:

[0055] 1) Determine the target Z-axis direction:

[0056]

[0057] Among them, normal is the point cloud normal vector in the grid, targetZ is the target Z axis;

[0058] If targetZ>0, reverse direction:

[0059] targetZ=-targetZ

[0060] 2) Project the X-axis of the current robot arm posture onto a plane perpendicular to the target Z-axis. The calculation formula is as follows:

[0061] newX=currentX-(currentX·targetZ)targetZ

[0062] Among them, newX is the projected X-axis, currentX is the X-axis of the current robot arm posture;

[0063] 3) Calculate the orthogonal Y axis

[0064] Calculate the orthogonal Y axis by cross multiplication:

[0065] newY=targetZ×newX

[0066] Normalize newY:

[0067]

[0068] 4) Construct a new rotation matrix

[0069] Combine the calculated newX, newY and targetZ into a new rotation matrix:

[0070] Rnew=(newX,newY,newZ)

[0071] 5) Convert to ZYZ Euler angles

[0072] Convert the new rotation matrix to ZYZ Euler angles to control the posture of the robotic arm:

[0073] α,β,γ=rotationMatrixToZYZ(Rnew)

[0074] Among them, α, β, and γ are the ZYZ Euler angles respectively.

[0075] In a third aspect, the present invention further provides a curved surface intelligent spraying system, comprising:

[0076] Point cloud data acquisition module, which collects point cloud data of objects through depth cameras;

[0077] The point cloud data preprocessing and rasterization module performs voxel filtering, statistical filtering, and RANSAC plane segmentation preprocessing on the point cloud data in sequence;

[0078] Calculate the centroid and PCA principal component direction of the preprocessed point cloud, rasterize the space where the point cloud is located according to the preset grid size and number of rows and columns, and generate a grid map;

[0079] Path planning module, which uses the trained DQN model for path planning;

[0080] In the spraying operation module, the robot performs spraying operations according to the planned path.

[0081] Furthermore, the system is based on the ROS framework, and each module communicates through topic publishing and subscription; the robot control communicates with the Modbus server with the IP address and port through the Modbus TCP protocol.

[0082] Beneficial effects:

[0083] The surface intelligent spraying model training method provided by the present invention is based on the reinforcement learning strategy. The data collection, data processing and training process steps are clear and easy to operate. When the trained model is used for surface intelligent spraying, it can calculate an effective path when facing unknown objects, making the spraying more intelligent.

[0084] The reward mechanism encourages the robot to cover all sprayable areas and reduce repeated visits. The action design adds jumping to the top and bottom sprayable points in the next column to adapt to the complex grid map structure and improve the flexibility and efficiency of path planning.

[0085] The curved surface intelligent spraying method provided by the present invention does not require programming and teaching, and does not require pre-production of point cloud templates. The model trained by the curved surface intelligent spraying model training method realizes intelligent spraying.

[0086] The curved surface intelligent spraying system provided by the present invention has the same technical effects as above. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings.

[0088] Figure 1 This is a flow chart of Example 1 of the present invention;

[0089] Figure 2 This is a flowchart of Example 2 of the present invention;

[0090] Figure 3 Collecting point cloud data of an object for embodiment 2 of the present invention;

[0091] Figure 4 A grid map of an object in Example 2 of the present invention;

[0092] Figure 5 This is the spraying path of the object in Example 2 of the present invention. DETAILED DESCRIPTION

[0093] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. The principles and features of the present invention will be described below in conjunction with the accompanying drawings. It should be noted that the embodiments and features of the embodiments in this application may be combined with each other unless there is a conflict. The embodiments are provided only to illustrate the present invention and are not intended to limit the scope of the invention.

[0094] Example 1

[0095] like Figure 1 As shown, this embodiment is a surface intelligent spraying model training method based on reinforcement learning strategy, including:

[0096] Hardware environment construction: In the industrial spray robot system, 3D sensors (such as machine vision cameras) are equipped to collect workpiece point cloud data. The industrial robot is equipped with a spray end effector, and communication and control between modules are achieved through the ROS (Robot Operating System) framework.

[0097] Software system initialization: Start the point cloud data processing node (pcd_deal_node), subscribe to the point cloud topic published by the 3D sensor, load the deep reinforcement learning path planning service (DQNpathplan_node), and initialize the DQN model parameters. Start the robotic arm control client and establish a communication connection with the robotic arm controller.

[0098] Point Cloud Data Processing and Rasterization: 3D sensors collect workpiece point cloud data and publish it to a ROS topic. The pcd_deal_node receives the point cloud data and performs preprocessing operations, including voxel filtering, statistical filtering, and RANSAC plane segmentation. The point cloud's centroid and PCA principal component directions are calculated, and then rasterized according to the preset grid size and number of rows and columns. A grid map is generated and published to the Raster topic.

[0099] Path planning request and response: The DQNpathplan_node listens to the grid topic and receives grid map data. It builds a GridEnv (grid environment simulator) environment, uses the grid map as the initial state, and calls the trained DQN model for path planning. The DQN model outputs the optimal action sequence based on the current state, maps it to a grid index sequence as the planned path, and publishes it to the path topic.

[0100] Point cloud data processing specifically includes:

[0101] 1) Perform voxel filtering and statistical filtering on the point cloud data in sequence to remove noise points and outliers;

[0102] 2) Use the Random Sample Consensus (RANSAC) algorithm to separate the background plane and extract the workpiece point cloud data;

[0103] 3) Feature extraction, calculation of the point cloud centroid and principal component analysis (PCA) to provide a basis for subsequent rasterization processing;

[0104] 4) Based on the PCA results, the point cloud data is projected onto a two-dimensional plane;

[0105] 5) Divide the grid according to the preset grid size and number of rows and columns;

[0106] 6) Analyze the point cloud density within the grid and construct a grid map.

[0107] Deep reinforcement learning path planning specifically includes:

[0108] A grid environment simulator (GridEnv) is used, with a grid map as the environment state. The robot's action space is defined, including up, down, left, right, and jumping actions. A reward mechanism is set up to encourage the robot to cover all sprayable areas and reduce repeated visits.

[0109] For DQN training, a multi-layer fully connected neural network is used as DQN, grid state is input, action value is output, experience replay buffer is collected through interaction with the environment, and the network is trained using experience replay to learn the optimal path planning strategy.

[0110] A novel action and reward mechanism is designed in deep reinforcement learning path planning, including:

[0111] Action Design: In addition to the traditional up, down, left, and right movement actions, jumping actions have been added, including jumping to the top sprayable point in the next column and jumping to the bottom sprayable point in the next column, to adapt to the complex grid map structure and improve the flexibility and efficiency of path planning;

[0112] For example, the action space is:

[0113] A={a0,a1,a2,a3,a4,a5}

[0114] Among them, a0 is moving upward, a1 is moving downward, a2 is moving left, a3 is moving right, a4 jumps to the top sprayable point in the next column, and a5 jumps to the bottom sprayable point in the next column;

[0115] The reward mechanism includes rewards for covering unvisited areas, additional rewards for vertical movement, and penalties for repeatedly visiting covered areas and lateral movement, guiding the robot to cover all sprayable areas with the optimal path while avoiding invalid movements.

[0116] The rewards are set as:

[0117]

[0118] Among them, r new =15, r repeat =-20, r boundary = No additional reward or penalty (position remains unchanged)

[0119]

[0120] Example 2

[0121] This embodiment is a method for intelligent spraying of curved surfaces based on a reinforcement learning strategy. The method is based on the model trained in Example 1 and performs intelligent spraying on the curved surface. Figures 2 to 5 As shown, the method includes the following steps:

[0122] Step S21, collecting point cloud data of the object through a depth camera;

[0123] Step S22, performing voxel filtering, statistical filtering, and RANSAC plane segmentation preprocessing on the point cloud data in sequence;

[0124] Calculate the centroid and PCA principal component direction of the preprocessed point cloud, rasterize the space where the point cloud is located according to the preset grid size and number of rows and columns, and generate a grid map;

[0125] Step S23, using the DQN model trained by the training method to perform path planning;

[0126] In step S24, the robot performs spraying operations according to the planned path.

[0127] In step S24, the robot arm control and spraying execution involves: pcd_deal_node receives the planned path and retrieves the corresponding grid center point cloud data based on the path index. The pose of the grid point cloud center coordinates in the robot arm base coordinate system is calculated and the spraying posture is determined by combining the PCA normal vector. The pose sequence is converted into robot arm joint space instructions, and the robot control service is called to drive the robot to perform the spraying operation according to the planned path.

[0128] Robotic arm control requires transformation from the camera coordinate system to the robot arm base coordinate system.

[0129] The pose of the robot's end effector (end_effector_pose) and the pose of the camera relative to the end effector (camera_to_end_effector_pose) are known. We need to convert the point cloud data in the camera coordinate system to the robot base coordinate system.

[0130] Assumptions: The pose of the end effector of the robot is Tend_effector, which is expressed as a homogeneous transformation matrix. The pose of the camera relative to the end effector is Tcamera_to_end_effector, which is expressed as a homogeneous transformation matrix.

[0131] The camera to base transformation matrix T camera_to_base Calculated by the following formula:

[0132] T camera_to_base =T end_effector ×T camera_to_end_effector

[0133] Among them, T end_effector is the position of the end effector of the robot arm, T camera_to_end_effector is the pose of the camera relative to the end effector;

[0134] The coordinates of the center point of the point cloud in the base coordinate system p base for:

[0135]

[0136] Among them, (x, y, z) is the coordinate of the center point of the point cloud in the camera coordinate system.

[0137] Then adjust the normal vector:

[0138] Generating the robot arm joint space instructions includes:

[0139] 1) Determine the target Z-axis direction:

[0140] The target Z-axis direction is determined by the point cloud's normal vector, and ensures that it points downwards:

[0141]

[0142] If targetZ>0, reverse direction:

[0143] targetZ=-targetZ

[0144] 2) Project the current X axis to the vertical plane of the target Z axis:

[0145] Project the X-axis of the current posture onto a plane perpendicular to the target Z-axis. The calculation formula is as follows:

[0146] newX=currentX-(currentX·targetZ)targetZ

[0147] Among them, newX is the projected X-axis, currentX is the X-axis of the current robot arm posture;

[0148] 3) Dealing with degradation:

[0149] If the projected X axis is close to the zero vector (i.e. ‖newX‖ < 1e-6), use the Y axis of the current posture for projection instead:

[0150] newX=currentY-(currentY·targetZ)targetZ

[0151] If it still degenerates, project using the global X axis:

[0152] newX=globalX-(globalX·targetZ)targetZ

[0153] Finally, normalize newX:

[0154]

[0155] 4) Calculate the orthogonal Y axis:

[0156] Calculate the orthogonal Y axis by cross multiplication: newY = targetZ × newX

[0157] and normalize newY:

[0158] 5) Construct a new rotation matrix:

[0159] Combine the calculated newX, newY, and targetZ into a new rotation matrix:

[0160] Rnew = (newX, newY, newZ)

[0161] 6) Convert to ZYZ Euler angles:

[0162] Convert the new rotation matrix to ZYZ Euler angles for controlling the pose of the robot arm:

[0163] α, β, γ = rotationMatrixToZYZ(Rnew)

[0164] α, β, and γ are the ZYZ Euler angles respectively.

[0165] After the above pose adjustment, the following purposes are achieved:

[0166] Z-axis parallel and downward: Through the above normal vector adjustment and rotation matrix construction, it is ensured that the Z-axis of the robot arm end is parallel to the normal vector and downward.

[0167] Reduce X and Y axis rotation: When constructing the rotation matrix, try to keep the X and Y axis directions consistent with the base coordinate system, so as to reduce the rotation angle of the robot arm end in X and Y axes. This avoids the change of camera direction with the rotation of robot arm, ensuring the stability of camera in the spraying process.

[0168] Fill the converted pose data and Euler angles into the message of the robot arm control instruction:

[0169] data = [xbase, ybase, zbase, α, β, γ]

[0170] Through the ROS service call, the control instruction is sent to the robot arm controller to drive the robot arm to perform the spraying operation according to the planned path and pose.

[0171] Example 3

[0172] The point cloud data acquisition module acquires the point cloud data of the object through the depth camera;

[0173] The point cloud data preprocessing and gridding module sequentially performs voxel filtering, statistical filtering, and RANSAC plane segmentation preprocessing on the point cloud data;

[0174] The centroid and the PCA principal component direction of the pre-processed point cloud are calculated, and the space where the point cloud is located is rasterized and divided according to a preset grid size and row and column number, to generate a grid map;

[0175] The path planning module uses the trained DQN model to perform path planning.

[0176] The spraying operation module sprays according to the planned path.

[0177] The system is based on the ROS framework, and each module communicates through topic publishing and subscription. The robot control communicates with the Modbus server with an IP address and a port through the Modbus TCP protocol.

[0178] Obviously, the above embodiments are only examples for clearly illustrating, but not limitation to the embodiments. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, all the embodiments need not and cannot be exhausted. The obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. A surface intelligent spraying model training method, characterized in that: The steps include: Step S11, collecting workpiece point cloud data: Collect point cloud data of the workpiece through a depth camera; Step S12: point cloud data preprocessing and rasterization: Preprocess the collected workpiece point cloud data to remove noise points and outliers; rasterize the preprocessed point cloud data to generate a raster map; Step S13: training the DQN model: Define the environment state as a grid map; The robot's action space includes moving up, down, left, right, and jumping; The reward mechanism is to encourage the robot to cover all spraying areas and reduce repeated visits; A multi-layer fully connected neural network is used as DQN. The experience replay buffer is collected by interacting with the environment, and the network is trained using experience replay to learn the optimal path planning strategy.

2. The training method according to claim 1, characterized in that Step S12 specifically includes: Step S121, sequentially performing voxel filtering and statistical filtering on the collected workpiece point cloud data; Step S122, using RANSAC to perform plane segmentation; Step S123, calculating the centroid of the point cloud and performing feature extraction using PCA; Step S124, projecting the point cloud data onto a two-dimensional plane based on the PCA results; Step S125, dividing the grid into grids according to the preset grid size and number of rows and columns; Step S126: construct a grid map based on the point cloud density within the grid.

3. The training method according to claim 1, characterized in that In step S13, the action space A is: A={a0,a1,a2,a3,a4,a5} Among them, a0 moves up, a1 moves down, a2 moves left, a3 moves right, a4 jumps to the top of the next column, and a5 jumps to the bottom of the next column; The reward R is set as: in, r new : Rewards for unvisited areas; r direction : The movement direction of the unvisited area is rewarded vertically and penalized horizontally; r jump : Penalty for executing jump action; r repeat : Penalties for repeatedly visiting covered areas or staying put; r boundary : There is no additional reward or penalty for moving outside the boundary.

4. The training method according to claim 3, characterized in that The specific reward settings are as follows: r new ＝15，r repeat ＝-20， 5. A curved surface intelligent spraying method, characterized in that: The steps include: Step S21, collecting point cloud data of the object through a depth camera; Step S22, performing voxel filtering, statistical filtering, and RANSAC plane segmentation preprocessing on the point cloud data in sequence; Calculate the centroid and PCA principal component direction of the preprocessed point cloud, rasterize the space where the point cloud is located according to the preset grid size and number of rows and columns, and generate a grid map; Step S23, performing path planning using the DQN model trained by the training method according to any one of claims 1 to 4; In step S24, the robot performs spraying operations according to the planned path.

6. The spraying method according to claim 5, characterized in that: Step S24 specifically includes: Obtain the point cloud data of the grid; calculate the position of the center point coordinates of the grid point cloud in the robot arm base coordinate system, calculate the point cloud normal vector of the grid through PCA to determine the spraying posture and form a posture sequence; The pose sequence is converted to generate joint space instructions for the robot arm, and the robot performs spraying operations according to the joint space instructions.

7. The spraying method according to claim 6, characterized in that: The transformation pose sequence generates the robot arm joint space instructions, including the transformation matrix T from the camera coordinate system to the base coordinate system. camera_to_base Calculated by the following formula: T camera_to_base =T end_effect0r ×T camera_to_end_effector Among them, T end_effector is the position of the end effector of the robot arm, T camera_to_end_effector is the pose of the camera relative to the end effector; The coordinates of the center point of the grid point cloud in the base coordinate system p base for: Among them, (x, y, z) is the coordinate of the center point of the grid point cloud in the camera coordinate system.

8. The spraying method according to claim 7, characterized in that: Converting pose sequences to generate robot joint space instructions also includes: 1) Determine the target Z-axis direction: Among them, normal is the point cloud normal vector in the grid, tar getZ is the target Z axis; If targetZ>0, reverse direction: targetZ=-targetZ 2) Project the X-axis of the current robot arm posture onto a plane perpendicular to the target Z-axis. The calculation formula is as follows: newX=currentX-(currentX·targetX)targetX Among them, newX is the projected X-axis, currentX is the X-axis of the current robot arm posture; 3) Calculate the orthogonal Y axis Calculate the orthogonal Y axis by cross multiplication: newY=targetZ×newX Normalize newY: 4) Construct a new rotation matrix Combine the calculated newX, newY and tar getZ into a new rotation matrix: Rnew=(newX,newY,newZ) 5) Convert to ZYZ Euler angles Convert the new rotation matrix to ZYZ Euler angles to control the posture of the robotic arm: α,β,γ=rotationMatrixToZYZ(Rnew) Among them, α, β, and γ are the ZYZ Euler angles respectively.

9. A curved surface intelligent spraying system, characterized in that: include: Point cloud data acquisition module, which collects point cloud data of objects through depth cameras; The point cloud data preprocessing and rasterization module performs voxel filtering, statistical filtering, and RANSAC plane segmentation preprocessing on the point cloud data in sequence; Calculate the centroid and PCA principal component direction of the preprocessed point cloud, rasterize the space where the point cloud is located according to the preset grid size and number of rows and columns, and generate a grid map; Path planning module, which uses the trained DQN model for path planning; In the spraying operation module, the robot performs spraying operations according to the planned path.

10. The spraying system according to claim 9, characterized in that: The system is based on the ROS framework, and each module communicates through topic publishing and subscription; the robot control communicates with the Modbus server with the IP address and port through the Modbus TCP protocol.

Images