An intelligent robot system and method for grasping small industrial parts

Abstract

The present invention provides an intelligent robot system and method for grasping small industrial parts. The intelligent robot system for grasping small industrial parts includes: an operation module, a mobile platform module, a perception module, and a control module; the operation module and the control module are both installed on the mobile platform module, the perception module is installed on the operation module, the operation module is used to provide the function of grasping objects, the mobile platform module is used to perform the movement function of the intelligent robot, and the perception module is used for the positioning and mapping of the intelligent robot, as well as the positioning of the target area and the visual grasping algorithm. The intelligent robot system for grasping small industrial parts provided by the present invention has a high degree of movement flexibility and fine operation capabilities, can adapt to complex and changing industrial scenarios, can complete complex mobile operation operations, and can be effectively applied to small industrial parts grasping tasks.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to an intelligent robot system and method for grasping small industrial parts. Background Art

[0002] Currently, most robotic grasping systems utilize a fixed design. While this design is stable in specific environments, it lacks the flexibility required to handle complex situations. The operating range of fixed robots is limited by their mechanical structure and mounting location, making it impossible to grasp objects that are not within the pre-defined workspace.

[0003] In addition, although existing vision-guided grasping methods can improve the grasping success rate in cluttered scenes to a certain extent, most of them are aimed at objects with regular shapes and obvious textures, and are mainly used in scenes with relatively simple object arrangements. For industrial parts with complex shapes, faint textures and small sizes, these methods often perform poorly, especially in complex scenes where these parts are densely stacked and occlude each other. On the one hand, objects such as parts often have irregular geometric appearances, most of which are small in size, complex in shape, and similar in color, lacking obvious texture features. On the other hand, in actual scenes, objects are often randomly distributed in the material frame and are prone to piling up, resulting in mutual occlusion, which increases the complexity of the scene. The above problems can easily confuse the robot's visual perception, making the current grasping algorithm unusable when directly deployed in industrial scenes.

[0004] Therefore, it is necessary to provide an intelligent robot system and method for grasping small industrial parts to solve the above technical problems. Summary of the Invention

[0005] The present invention provides an intelligent robot system and method for grasping small industrial parts, which solves the problem of lack of flexibility required to cope with complex environments.

[0006] To solve the above technical problems, the present invention provides an intelligent robot system for grasping small industrial parts, comprising: an operation module, a mobile platform module, a perception module, and a control module;

[0007] The operation module and the control module are both installed on the mobile platform module, and the perception module is installed on the operation module. The operation module is used to provide the function of grasping objects, and the mobile platform module is used to execute the movement function of the intelligent robot. The perception module is used for the positioning and mapping of the intelligent robot, as well as the target area positioning and visual grasping algorithm. The control module is used to process and manage related tasks of the intelligent robot operation.

[0008] Preferably, the operating module includes a robotic arm and an electric gripper, and the mobile platform module includes a quadruped bionic robot with twelve leg and foot joints.

[0009] Preferably, the perception module includes a lidar, an inertial measurement unit and a depth camera; wherein the lidar and the inertial measurement unit are used for positioning and mapping of the intelligent robot, and the depth camera is used for target area positioning and visual capture algorithm.

[0010] Preferably, the control module is used to coordinate the intelligent robot to perform various tasks, and is responsible for complex tasks such as visual processing, navigation decision-making, and high-level motion control. It also dispatches the motion control and signal acquisition of the underlying hardware through control instructions, and acts as a bridge for coordination and information transmission between various modules.

[0011] Preferably, based on the functional architecture, the intelligent robot system for grasping small industrial parts is divided into environmental perception, data processing, and planning execution.

[0012] Preferably, the operating module is rotatably connected to a rotating block, the rotating block is rotatably connected to a rotating sleeve, the rotating sleeve is equipped with a mounting block, the mounting block is equipped with a toggle piece, the toggle piece is used to toggle the workpiece in the material frame to change the clamping posture of the workpiece, the operating module is equipped with a driving piece for driving the rotating sleeve to rotate, the operating module is internally slidably connected to a sliding block, the sliding block is fixedly connected to a connecting rod, one end of the connecting rod passes through the middle of the rotating sleeve and is equipped with a grabbing assembly, the grabbing assembly is used to grab the workpiece, and the operating module is internally equipped with a pushing piece, the pushing piece is fixedly connected to the sliding block, and is used to drive the sliding block to move linearly to achieve feeding of the grabbing assembly.

[0013] Preferably, the operating module is rotatably connected to a mounting sleeve, the sensing module is mounted on the mounting sleeve, a connecting block is mounted on the rotating block, and one end of the connecting block is fixed on the mounting sleeve to enable the mounting sleeve to rotate when the rotating block rotates.

[0014] Preferably, the outer surface of the connecting rod is rotatably connected to an auxiliary block, the auxiliary block is fixedly connected to a first docking piece, the rotating sleeve is fixedly connected to a second docking piece, the second docking piece is used to cooperate with the first docking piece to achieve docking between the rotating sleeve and the auxiliary block, the auxiliary block is fixedly connected to a guide rod, one end of the guide rod is installed with a third docking piece, the rotating block is installed with a fourth docking piece, the fourth docking piece is used to cooperate with the third docking piece to achieve docking between the guide rod and the rotating block.

[0015] A control method for an intelligent robot for grasping small industrial parts, including a robot push action pose estimation method and a multi-object grasping method combined with push operation assisted grasping;

[0016] The robot pushing motion posture estimation method comprises the following steps:

[0017] S1, obtain image data and filter target objects;

[0018] S2, grasping posture detection network predicts grasping parameters;

[0019] S3. Determine the primitive action to be executed;

[0020] S4, push action pose estimation;

[0021] S5. Execute primitive actions.

[0022] The multi-object grasping method combined with push operation assisted grasping comprises the following steps:

[0023] S1. Acquire image data and filter target objects according to the perception module to obtain color image blocks containing single objects and corresponding edge image blocks;

[0024] S2, sending it to the grasping posture detection network, predicting the planar grasping configuration composed of the grasping point and the grasping plane rotation angle parameters, and converting it into the grasping posture of the end gripper of the robotic arm in combination with the depth image;

[0025] S3. Perform collision detection on the grasp and determine the primitive action to be executed based on the result. If there is no grasp collision, control the robot movement and execute the grasp action. Otherwise, reuse the grasp detection network to make predictions. If no feasible grasp is found after three consecutive predictions, use the point cloud data converted from the depth image as input to estimate the reasonable push position of the robot arm and execute the push action to adjust the position relationship of objects in the scene, separate dense object clusters, and create a favorable space for subsequent grasping.

[0026] S4. The above process is repeated until there is no object in the material frame.

[0027] Compared with related technologies, the intelligent robot system and method for grasping small industrial parts provided by the present invention have the following beneficial effects:

[0028] (1) The intelligent grasping robot system proposed in the present invention, which is composed of a quadruped robot, a robotic arm, an electric gripper, various sensors, a control module, etc., has a high degree of motion flexibility and fine operation capabilities, can adapt to complex and changing industrial scenarios, can complete complex mobile operation operations, and can be effectively applied to small industrial parts grasping tasks;

[0029] (2) The robot push motion pose estimation method proposed in the present invention can effectively separate densely packed objects, improve or create a feasible grasping space, and reduce the probability of the end effector colliding with non-target objects or material frames during subsequent grasping of objects, thus providing more favorable conditions for successful grasping.

[0030] (3) The multi-object grasping method combined with push operation assisted grasping proposed in the present invention can greatly avoid the problem of grasping failure caused by collision during the grasping process due to the use of only grasping operations, such as multiple objects stacked and blocked, thereby improving the grasping success rate in unstructured scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic structural diagram of a first embodiment of an intelligent robot system for grasping small industrial parts provided by the present invention;

[0032] Figure 2 This is the overall framework diagram of the intelligent grasping robot system;

[0033] Figure 3 It is a flowchart of the present invention;

[0034] Figure 4 Schematic diagram of grasping key points (left) and grasping center points (right);

[0035] Figure 5 To capture the detection network architecture diagram;

[0036] Figure 6 This is a simplified schematic diagram of push strategy 1 when the camera is perpendicular to the plane of the frame;

[0037] Figure 7 This is a simplified schematic diagram of push strategy 2 when the camera is perpendicular to the plane of the frame;

[0038] Figure 8 This is a simplified diagram of push strategy 3 when the camera is perpendicular to the plane of the frame;

[0039] Figure 9 A schematic structural diagram of a second embodiment of the intelligent robot system for grasping small industrial parts provided by the present invention;

[0040] Figure 10 for Figure 9 A partial enlarged view of A shown;

[0041] Figure 11 for Figure 9 The schematic cross-sectional view of the operating module is shown.

[0042] Markings in the figure: 1. Operation module, 2. Mobile platform module, 3. Sensing module, 4. Control module, 5. Rotating block, 6. Rotating sleeve, 7. Mounting block, 8. Toggle member, 9. Driving member, 10. Sliding block, 11. Connecting rod, 12. Grasping assembly, 13. Pushing member, 14. Mounting sleeve, 15. Connecting block, 16. Auxiliary block, 17. First docking member, 18. Second docking member, 19. Guide rod, 20. Third docking member, 21. Fourth docking member. DETAILED DESCRIPTION

[0043] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0044] First embodiment

[0045] Please refer to Figure 1-8 ,in, Figure 1 A schematic structural diagram of a first embodiment of an intelligent robot system for grasping small industrial parts provided by the present invention; Figure 2 This is the overall framework diagram of the intelligent grasping robot system; Figure 3 It is a flowchart of the present invention; Figure 4 Schematic diagram of grasping key points (left) and grasping center points (right); Figure 5 To capture the detection network architecture diagram; Figure 6 This is a simplified schematic diagram of push strategy 1 when the camera is perpendicular to the plane of the frame; Figure 7 This is a simplified schematic diagram of push strategy 2 when the camera is perpendicular to the plane of the frame; Figure 8 This is a simplified schematic diagram of push strategy 3 when the camera is perpendicular to the plane of the material frame. The intelligent robot system for grasping small industrial parts includes: an operation module 1, a mobile platform module 2, a perception module 3, and a control module 4;

[0046] The operation module 1 and the control module 4 are both installed on the mobile platform module 2, and the perception module 3 is installed on the operation module 1. The operation module 1 is used to provide the function of grasping objects, and the mobile platform module 2 is used to execute the movement function of the intelligent robot. The perception module 3 is used for the positioning and mapping of the intelligent robot and the target area positioning and visual grasping algorithm. The control module 4 is used to process and manage related tasks of the intelligent robot operation.

[0047] The operation module 1 includes a mechanical arm and an electric gripper, and the mobile platform module 2 includes a quadruped bionic robot with twelve leg and foot joints.

[0048] The perception module 3 includes a laser radar, an inertial measurement unit and a depth camera; among them, the laser radar and the inertial measurement unit are used for positioning and mapping of the intelligent robot, and the depth camera is used for target area positioning and visual capture algorithm.

[0049] The control module 4 is used to coordinate the intelligent robot to perform various tasks. It is responsible for complex tasks such as visual processing, navigation decision-making, and high-level motion control. It also dispatches the motion control and signal acquisition of the underlying hardware through control instructions, and plays the role of a bridge for coordination and information transmission between various modules.

[0050] According to the functional architecture, the intelligent robot system for grasping small industrial parts is divided into environmental perception, data processing, and planning execution.

[0051] In this embodiment, if Figure 2 As shown in the figure, in the environmental perception part, the lidar collects scene point cloud data, the IMU (inertial measurement unit) obtains robot state information including acceleration and angular velocity data, and the depth camera obtains color images and depth images.

[0052] In the data processing part, it is mainly divided into two parts. One part is used to calculate the motion information of the mobile platform. It is necessary to process the scene data obtained by the lidar and fuse the IMU (inertial measurement unit) data to generate a three-dimensional point cloud map of the environment and convert it into a two-dimensional grid map. After that, the AprilTag code is combined to assist in locating the area where the parts to be grasped are located and determine the target posture to which the mobile platform needs to move. The other part is used to calculate the motion information of the robotic arm. It is necessary to process the original image obtained by the depth camera and use the visual grasping algorithm to obtain the grasping configuration information. Combined with the camera's internal parameter information, the obtained grasping configuration information is converted to obtain the gripper target posture in the camera coordinate system. Then, the target posture is converted through the results of the hand-eye calibration to obtain the target posture in the robotic arm base coordinate system.

[0053] In the planning and execution part, it is responsible for the motion planning and low-level control of the intelligent grasping robot. According to the position information of the mobile platform and the operating device that need to be moved to obtained in the previous layer, a reasonable planning path is formed, and the planning instructions are converted into the robot's low-level control instructions, which are sent to the controllers corresponding to the quadruped robot, robotic arm and gripper to perform motion control.

[0054] A control method for an intelligent robot for grasping small industrial parts, including a robot push action pose estimation method and a multi-object grasping method combined with push operation assisted grasping;

[0055] The robot pushing motion posture estimation method comprises the following steps:

[0056] S1, obtain image data and filter target objects;

[0057] First, a color image of the scene and an aligned depth image are acquired from the depth camera. Next, the color image is fed into an object detection network such as YOLOV5 to predict the approximate area where each object resides, generating multiple bounding boxes containing individual objects and their confidence scores. The center point coordinates of each bounding box are then used as the center of mass of the corresponding object, and the kernel density estimation (KDE) method is used to estimate the density of the area surrounding each object. The formula is as follows:

[0058]

[0059] Among them, x is the coordinate of the center of mass of the target object, x j is the centroid coordinate of the jth object around the target object, K is the Gaussian kernel function, and h is the smoothing parameter.

[0060] Calculate the D value of each object to represent the density of the area around the object and record it as a set T. Select the object corresponding to the minimum value in T as the target object to be grasped (if there are multiple identical minimum values, select the one with the highest bounding box confidence score), that is, filter out possible isolated objects with the lowest degree of grasping space restriction. Crop the original color image according to the bounding box coordinates of the target object to obtain a color image block containing only a single object. Use the Sobel operator to perform edge detection on the color image block to obtain the corresponding edge image block. Finally, resize the two image blocks to a predefined fixed width and height size and splice them in the channel dimension as input data for the grasping pose detection network.

[0061] S2, grasping posture detection network predicts grasping parameters;

[0062] Referring to the grasping parameter configuration used by Kumra et al., the grasp is characterized as G = (x, y, w, θ, q), where (x, y) represents the coordinates of the grasping point in the pixel coordinate system, w represents the gripper width during grasping, θ represents the plane rotation angle of the grasp, and q is the quality score of each grasp. In addition, according to Sun et al.

[16] The proposed method uses the grasping center point P c = (x, y) and the grasp key points P1 = (x1, y1) and P2 = (x2, y2) to determine the grasp representation g, and these points are predicted by the heat map predicted by the grasp pose detection network, such as Figure 4 shown.

[0063] The entire grasping posture detection network adopts an encoder-decoder architecture, such as Figure 5As shown in the figure, the model structure refers to the DeeplabV3+ model commonly used for semantic segmentation tasks, and is modified to add an additional output head to achieve pixel-level prediction. The input of the encoder is the color image block and edge image block obtained in the previous step. Through the residual network module (ResidualNetwork, ResNet) with void convolution and the spatial pyramid pooling module (AtrousSpatialPyramidPooling, ASPP), low-level features and high-level features are extracted respectively and sent to the decoder. In the decoder part, the low-level features and high-level features are further processed by convolution, combined with upsampling and splicing operations, and finally the predicted capture center point heat map and capture key point heat map are output, achieving the pixel-level regression task.

[0064] The model parameters are optimized according to the following loss function:

[0065] L=αL c +βL k

[0066] Among them, L c and L k are the SmoothL1 losses for grasping the center point and grasping the key points, respectively, and α and β are their respective weight factors.

[0067] The grasp center point P can be determined based on the maximum confidence value of the grasp center point heat map and the grasp key point heat map. c And the key points P1 and P2 are captured. Based on these points, the capture representation G can be constructed. The coordinates (x, y) of the capture point in G in the pixel coordinate system are the corresponding points P c The grasping width is predetermined according to the type of object to be grasped, and the plane rotation angle θ can be obtained based on points P1 and P2 using the following formula:

[0068] θ=arctan2(-(y2-y1),(x2-x1))+π / 2

[0069] Among them, arctan2 is a function that calculates the inverse tangent.

[0070] Combining the above-obtained grasp representation G with the depth image information of the scene can obtain grasp parameters containing position information and posture information.

[0071] S3. Determine the primitive action to be executed;

[0072] The grasping parameters obtained in the previous step are collided with the gripper to ensure that the gripper does not collide with the material frame or other objects when grasping the target object. The collision detection method is to convert the simplified gripper model point cloud to the specific location of the scene point cloud based on the grasping parameters, and check whether there is a point cloud near the gripper model area to determine whether a collision has occurred. If the detection result shows no collision, it means that the grasping is feasible; if there is a collision, the grasping posture detection network is reused for prediction. If no feasible grasping solution is found after three consecutive predictions, the appropriate pushing action posture is estimated to adjust the positional relationship between objects in the scene, separate dense objects, and create a feasible operating space for subsequent grasping.

[0073] S4, push action pose estimation;

[0074] 1) Point cloud preprocessing and selection of target area

[0075] First, the scene depth image acquired from the depth camera is converted into a point cloud based on the camera's intrinsic parameters, and the point cloud portion of the material frame area is extracted. Next, a statistical filtering method is used to remove isolated points, and the Random Sample Consensus (RANSAC) method is used to detect and segment the bottom surface of the material frame, filtering out a point cloud containing only the object. The object point cloud is then clustered using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and the cluster with the most points is selected as the target area for the push operation.

[0076] 2) Get the push operation position

[0077] Referring to the definition of push operation by Danielczuk et al., a single push action is represented as Where q = (x1, y1, z1) is the starting point of the push operation, and r = (x2, y2, z2) is the end point of the push operation, both located in the camera coordinate system. The position of the two-finger gripper's fingertips is the origin of the push operation. Unless otherwise specified, the two-finger gripper remains closed during the push operation.

[0078] Depending on the number of points in the target region, different heuristic operations for pushing and separating objects are selected. When the target region contains a large number of points (greater than 4000 points), push operation 1 is used to separate the target object clusters. When the target region contains a small number of points (less than 1000 points), push operation 2 is used to separate the target object clusters. In other cases, push operation 3 is used to separate the target object clusters.

[0079] See also Figure 6Push strategy 1: zigzag two-step push operation. Specifically, calculate the axis-aligned bounding box of the cluster point cloud, obtain the coordinates of the four vertices on the upper side of the axis-aligned bounding box and the coordinates of the center point of the axis-aligned bounding box, replace the z-axis value of the four vertex coordinates with the z-axis value of the center point coordinate, and obtain a new set of four vertices. The rectangular area formed by these four points is used as a candidate area for the push operation. Randomly select one of the vertices A, and you can determine the other vertex B on the long side where it is located and the midpoint E on the opposite side of the long side. Take point A as the first starting position of the push operation, point E as the first end position and the second starting position of the push operation, and point B as the second end position of the push operation. Move from A to E, and then move from E to B. Perform two push operations in quick succession to push the object apart.

[0080] See also Figure 7 Push strategy 2: Combine the opening and closing degrees of freedom of the gripper to perform the push operation. Specifically, calculate the center of mass O of the cluster point cloud and the maximum distance D from the points contained in the cluster point cloud to the center of mass. Construct a circle with O as the center and radius D, uniformly sample 40 points at equal intervals on the circumference, then arbitrarily select two points to connect to form a line segment and calculate the number of points passed through the line segment, find the line segment that passes through the most points, and draw a perpendicular line segment AB with a length of 2D through the midpoint A of the line segment. Take the endpoints B and A of the perpendicular line segment as the starting and ending positions of the push operation respectively, and quickly open the gripper at the end position A to separate the object.

[0081] See also Figure 8 Push strategy 3: Push along the direction that passes through the most objects. Specifically, calculate the center of mass O of the cluster point cloud and the maximum distance D from the center of mass of the points in the cluster point cloud. Construct a 2D line segment with O as the midpoint and rotate the line segment clockwise 9 degrees around point O. Ultimately, 20 line segments are obtained as candidate push operation directions. Traverse these line segments and find the line segment AB that passes through the most points. Take the two endpoints A and B of this line segment as the starting and ending positions of the push operation, respectively, to push the objects apart.

[0082] S5. Execute primitive actions.

[0083] The target position of the gripper in the camera coordinate system corresponding to the feasible grasping operation or pushing operation obtained in the above steps is Convert to the robot base coordinate system to control the movement of the robot. Specifically, according to the relevant theory of robot coordinate transformation, the following formula is established:

[0084]

[0085] in, is the transformation matrix from the manipulator base coordinate system to the manipulator end coordinate system corresponding to the current camera view, which can be obtained by calculating the joint angle array q read by the manipulator controller and the robot forward kinematics; is the hand-eye calibration matrix.

[0086] right Performing the transformation yields:

[0087]

[0088] in, is the transformation matrix from the end coordinate system to the gripper coordinate system.

[0089] Combining the above formula, further transformation can be obtained:

[0090]

[0091] in, is the target pose command actually sent to the robot controller.

[0092] After calculating the target pose After that, the trajectory planning is completed using cubic spline interpolation and sent to the robot controller to control the movement of the manipulator and its end effector to perform the corresponding primitive action. After the movement is completed, the manipulator and its end effector are controlled to return to the initial position and return to step 1.

[0093] The multi-object grasping method combined with the push operation to assist grasping is specifically as follows: Figure 3 As shown, the following steps are included:

[0094] S1. Acquire image data and filter target objects according to the perception module to obtain color image blocks containing single objects and corresponding edge image blocks;

[0095] S2, sending it to the grasping posture detection network, predicting the planar grasping configuration composed of the grasping point and the grasping plane rotation angle parameters, and converting it into the grasping posture of the end gripper of the robotic arm in combination with the depth image;

[0096] S3. Perform collision detection on the grasp and determine the primitive action to be executed based on the result. If there is no grasp collision, control the robot movement and execute the grasp action. Otherwise, reuse the grasp detection network to make predictions. If no feasible grasp is found after three consecutive predictions, use the point cloud data converted from the depth image as input to estimate the reasonable push position of the robot arm and execute the push action to adjust the position relationship of objects in the scene, separate dense object clusters, and create a favorable space for subsequent grasping.

[0097] S4. The above process is repeated until there is no object in the material frame.

[0098] Compared with related technologies, the intelligent robot system and method for grasping small industrial parts provided by the present invention have the following beneficial effects:

[0099] (1) The intelligent grasping robot system proposed in the present invention, which is composed of a quadruped robot, a robotic arm, an electric gripper, various sensors, a control module, etc., has a high degree of motion flexibility and fine operation capabilities, can adapt to complex and changing industrial scenarios, can complete complex mobile operation operations, and can be effectively applied to small industrial parts grasping tasks;

[0100] (2) The robot push motion pose estimation method proposed in the present invention can effectively separate densely packed objects, improve or create a feasible grasping space, and reduce the probability of the end effector colliding with non-target objects or material frames during subsequent grasping of objects, thus providing more favorable conditions for successful grasping.

[0101] (3) The multi-object grasping method combined with push operation assisted grasping proposed in the present invention can greatly avoid the problem of grasping failure caused by collision during the grasping process due to the use of only grasping operations, such as multiple objects stacked and blocked, thereby improving the grasping success rate in unstructured scenarios.

[0102] Second embodiment

[0103] Please refer to Figure 9-11 Based on the intelligent robot system for grasping small industrial parts provided in the first embodiment of this application, the second embodiment of this application proposes another intelligent robot system for grasping small industrial parts. The second embodiment is merely a preferred embodiment of the first embodiment, and the implementation of the second embodiment will not affect the independent implementation of the first embodiment.

[0104] Specifically, the difference of the intelligent robot system for grasping small industrial parts provided in the second embodiment of the present application is that the operating module 1 is rotatably connected to a rotating block 5, the rotating block 5 is rotatably connected to a rotating sleeve 6, the rotating sleeve 6 is installed with a mounting block 7, the mounting block 7 is installed with a toggle member 8, the toggle member 8 is used to toggle the workpiece in the material frame to change the clamping posture of the workpiece, the operating module 1 is installed with a driving member 9 for driving the rotating sleeve 6 to rotate, the operating module 1 is internally slidably connected to a sliding block 10, the sliding block 10 is fixedly connected to a connecting rod 11, one end of the connecting rod 11 passes through the middle of the rotating sleeve 6 and is installed with a grasping assembly 12, the grasping assembly 12 is used to grasp the workpiece, and the operating module 1 is internally installed with a pushing member 13, the pushing member 13 is fixedly connected to the sliding block 10, and is used to drive the sliding block 10 to move linearly to realize the feeding of the grasping assembly 12.

[0105] The toggle member 8 includes but is not limited to a toggle rod, a toggle block or a special-shaped toggle bar. The installation method is fixed or detachable. It only needs to realize the toggle of the workpiece, and the size can be adjusted according to actual needs.

[0106] The driving member 9 includes a power member and a transmission member, wherein the power member includes but is not limited to an electric motor, a motor, and a transmission machine, and the transmission member includes but is not limited to a friction wheel, a belt and a pulley, a gear and a gear, a worm wheel and a worm, a sprocket and a chain, and can drive the rotating sleeve 6 when the driving member 9 rotates.

[0107] The grasping assembly 12 includes but is not limited to a bilateral clamp, a three-axis clamp, and a four-axis clamp, and any one of them can realize the grasping function and is used to grasp the workpiece.

[0108] The pushing member 13 includes but is not limited to an electric push rod, a linear motor, a hydraulic rod, and a reciprocating motion mechanical component, which can drive the sliding block 10 to move linearly.

[0109] The operating module 1 is rotatably connected to a mounting sleeve 14, the sensing module 3 is mounted on the mounting sleeve 14, and a connecting block 15 is mounted on the rotating block 5. One end of the connecting block 15 is fixed to the mounting sleeve 14 to enable the mounting sleeve 14 to rotate when the rotating block rotates.

[0110] The outer surface of the connecting rod 11 is rotatably connected to an auxiliary block 16, and a first docking piece 17 is fixedly connected to the auxiliary block 16. The rotating sleeve 6 is fixedly connected to a second docking piece 18, and the second docking piece 18 is used to cooperate with the first docking piece 17 to achieve docking between the rotating sleeve 6 and the auxiliary block 16. A guide rod 19 is fixedly connected to the auxiliary block 16, and a third docking piece 20 is installed at one end of the guide rod 19. A fourth docking piece 21 is installed on the rotating block 5, and the fourth docking piece 21 is used to cooperate with the third docking piece 20 to achieve docking between the guide rod 19 and the rotating block 5.

[0111] The connection between the first connecting member 17 and the second connecting member 18 includes, but is not limited to, a snap-on connection and a magnetic connection.

[0112] When the card connection is adopted, the first docking member 17 is provided with a card slot, and the second docking member 18 is provided with a card block corresponding to the card slot, and the two can be connected to achieve transmission;

[0113] When the magnetic type is adopted, a magnetic block is provided on the first docking member 17, and another magnetic block is provided on the second docking member 18. Transmission can be achieved between the two docking. Preferably, the magnetic block is an electromagnetic block, which can be quickly separated by turning on and off the power, and the magnetic force meets the requirements.

[0114] The first docking member 17 , the second docking member 18 , the third docking member 20 , and the fourth docking member 21 may be of the same type.

[0115] During operation, the driving member 9 drives the rotating sleeve 6 to rotate, which can make the toggle member 8 move in a circular motion, and can toggle the workpiece in the material frame to adjust the position of the workpiece;

[0116] The push of the push member 13 can drive the sliding block 10 to move to the left, thereby enabling the connecting rod 11 to move to the left, driving the grabbing assembly 12 to extend out to grab the workpiece;

[0117] By rotating the rotating sleeve 6, the posture of the workpiece in the material frame can be adjusted individually; by pushing the pushing member 13, the feeding of the grasping component 12 can be realized individually, so that the grasping component 12 can grasp the workpiece; when the pushing member 13 is pushing, the driving member 9 is also driving the toggle member 8 to move in a circle. At this time, the two will cooperate with each other, indirectly driving the installation sleeve 14 to rotate, so as to adjust the position of the perception module 3, so that the perception module 3 can perceive in all directions, provide better scene data acquisition, and facilitate subsequent work.

[0118] Compared with related technologies, the intelligent robot system for grasping small industrial parts provided by the present invention has the following beneficial effects:

[0119] During the clamping operation, if the posture of the workpiece in the material frame is not good, making it difficult for the grasping component 12 to grasp it, the operating module 1 is stretched and deformed, so that the toggle member 8 is extended into the material frame, and the driving member 9 drives the rotating sleeve 6 to rotate, indirectly driving the toggle member 8 to move in a circle, so that the posture of the workpiece can be adjusted, and then the pushing member 13 is pushed to drive the connecting rod 11 to extend, thereby driving the grasping component 12 to grasp the workpiece. When the pushing member 13 pushes, the auxiliary block 16 will move to the left, so that the first docking member 17 docks with the second docking member 18, and the guide rod 19 will also move to the left, so that the third docking member 20 docks with the fourth docking member 21. At this time, in conjunction with the rotation of the rotating sleeve 6, the auxiliary block 16 will rotate, thereby driving the rotating block 5 to rotate, indirectly causing the mounting sleeve 14 to rotate and adjust the position of the sensing module 3, so that perception in different directions can be performed, thereby improving the comprehensiveness of perception.

[0120] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An intelligent robot system for grasping small industrial parts, characterized by: include: Operation module, mobile platform module, perception module and control module; The operation module and the control module are both installed on the mobile platform module. The perception module is installed on the operation module. The operation module is used to provide the function of grasping objects. The mobile platform module is used to perform the movement function of the intelligent robot. The perception module is used for positioning and mapping of the intelligent robot and the positioning and visual grasping algorithm of the target area. The control module is used to process and manage the relevant tasks of the operation of the intelligent robot. The operating module is rotatably connected to a rotating block, the rotating block is rotatably connected to a rotating sleeve, the rotating sleeve is installed with a mounting block, the mounting block is installed with a toggle piece, and the toggle piece is used to toggle the workpiece in the material frame to change the clamping posture of the workpiece, the operating module is installed with a driving member for driving the rotating sleeve to rotate, the operating module is internally slidably connected with a sliding block, the sliding block is fixedly connected with a connecting rod, one end of the connecting rod passes through the middle of the rotating sleeve and is installed with a grabbing assembly, the grabbing assembly is used to grab the workpiece, and a pushing member is installed inside the operating module, the pushing member is fixedly connected to the sliding block, and is used to drive the sliding block to move linearly to realize the feeding of the grabbing assembly; The operating module is rotatably connected to a mounting sleeve, the sensing module is mounted on the mounting sleeve, and a connecting block is mounted on the rotating block, one end of which is fixed to the mounting sleeve so as to drive the mounting sleeve to rotate when the rotating block rotates; The outer surface of the connecting rod is rotatably connected to an auxiliary block, the auxiliary block is fixedly connected to a first docking piece, the rotating sleeve is fixedly connected to a second docking piece, the second docking piece is used to cooperate with the first docking piece to achieve docking between the rotating sleeve and the auxiliary block, the auxiliary block is fixedly connected to a guide rod, one end of the guide rod is installed with a third docking piece, the rotating block is installed with a fourth docking piece, the fourth docking piece is used to cooperate with the third docking piece to achieve docking between the guide rod and the rotating block.

2. The intelligent robot system for grasping small industrial parts according to claim 1 is characterized in that: The operation module includes a mechanical arm and an electric gripper, and the mobile platform module includes a quadruped bionic robot with twelve leg and foot joints.

3. The intelligent robot system for grasping small industrial parts according to claim 2 is characterized in that: The perception module includes a lidar, an inertial measurement unit and a depth camera; among them, the lidar and inertial measurement unit are used for positioning and mapping of the intelligent robot, and the depth camera is used for target area positioning and visual capture algorithm.

4. The intelligent robot system for grasping small industrial parts according to claim 3 is characterized in that: The control module is used to coordinate the intelligent robot to perform various tasks. It is responsible for complex tasks such as visual processing, navigation decision-making, and high-level motion control. It also dispatches the motion control and signal acquisition of the underlying hardware through control instructions, acting as a bridge for coordination and information transmission between various modules.

5. The intelligent robot system for grasping small industrial parts according to claim 4 is characterized in that: According to the functional architecture, the intelligent robot system for grasping small industrial parts is divided into environmental perception, data processing, and planning execution.

6. The intelligent robot system for grasping small industrial parts according to claim 2, characterized in that: The control method of the intelligent robot system for grasping small industrial parts includes a robot pushing action posture estimation method and a multi-object grasping method combined with pushing operation assisted grasping; The robot pushing motion posture estimation method comprises the following steps: S1, obtain image data and filter target objects; S2, grasping posture detection network predicts grasping parameters; S3. Determine the primitive action to be executed; S4, push action pose estimation; S5. Execute primitive actions.

7. The intelligent robot system for grasping small industrial parts according to claim 6, characterized in that: The multi-object grasping method combined with push operation assisted grasping comprises the following steps: S1. Acquire image data and filter target objects according to the perception module to obtain color image blocks containing single objects and corresponding edge image blocks; S2, sending it to the grasping posture detection network, predicting the planar grasping configuration composed of the grasping point and the grasping plane rotation angle parameters, and converting it into the grasping posture of the end gripper of the robotic arm in combination with the depth image; S3. Perform collision detection on the grasp and determine the primitive action to be executed based on the result. If there is no grasp collision, control the robot movement and execute the grasp action. Otherwise, reuse the grasp detection network to make predictions. If no feasible grasp is found after three consecutive predictions, use the point cloud data converted from the depth image as input to estimate the reasonable push position of the robot arm and execute the push action to adjust the position relationship of objects in the scene, separate dense object clusters, and create a favorable space for subsequent grasping. S4. The above process is repeated until there is no object in the material frame.

Images