High-speed sorting robot based on visual identification and grabbing control system thereof

Abstract

The invention discloses a high-speed sorting robot based on visual recognition and a grabbing control system thereof, relates to the technical field of robot application and control, and provides the following scheme that the high-speed sorting robot comprises a base used for supporting the high-speed sorting robot; the swinging mechanism is arranged on the base, is used for swinging the high-speed sorting robot back and forth, and comprises a first swinging arm used for adjusting the back and forth angle; the overturning mechanism is arranged at one end of the first swinging arm, is used for overturning the position of the high-speed sorting robot, and comprises a first connecting arm, a second connecting arm and a supporting frame; the recognition mechanism is arranged on one side of the supporting frame. Through integration of mechanical structure innovation, visual identification optimization, an intelligent control algorithm and a communication system, high-speed, accurate and self-adaptive material sorting is achieved, the intelligent material sorting device can be widely applied to the fields of logistics, industrial production and the like, the problems that in the prior art, sorting efficiency is low, adaptability is poor, and intelligence is insufficient are effectively solved, and the intelligent material sorting device is suitable for large-scale popularization and application. The method has remarkable economic value and practical significance.

Description

Technical Field

[0001] This invention relates to the field of robot application and control technology, and in particular to a high-speed sorting robot based on vision recognition and its grasping control system. Background Technology

[0002] With the rapid development of modern logistics and industrial production, increasingly higher demands are being placed on the efficiency and accuracy of goods sorting. Traditional sorting methods, such as manual sorting, are not only inefficient but also susceptible to human factors, resulting in a high error rate. Although some existing sorting equipment has improved sorting efficiency to some extent, many problems still exist. For example, some sorting equipment relies on specific item shapes, sizes, or pre-set rules, making it less adaptable to irregularly shaped and diverse items. Some sorting systems based on traditional sensors struggle to quickly and accurately acquire the location and feature information of items in complex environments and when dealing with high-speed moving items, resulting in sorting speed and accuracy failing to meet actual needs. In logistics warehouses, packages vary in size, shape, and material, and the speed of packages on conveyor lines is high. Existing sorting equipment struggles to ensure both high-speed and accurate sorting. Therefore, developing a sorting robot and its gripping control system that can quickly and accurately sort various items and adapt to complex environments and high-speed operations is of significant practical importance. To this end, we propose a high-speed sorting robot and its gripping control system based on vision recognition. Summary of the Invention

[0003] The high-speed sorting robot based on vision recognition and its grasping control system proposed in this invention solve the above-mentioned shortcomings of the prior art.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: High-speed sorting robots based on vision recognition include: Base, used to support high-speed sorting robots; A swing mechanism, located on a base, is used for the high-speed sorting robot to swing back and forth. It includes: a first swing arm for adjusting the forward and backward angle; A flipping mechanism, located at one end of the first swing arm, is used for flipping the position of the high-speed sorting robot. It includes: a first connecting arm, a second connecting arm, and a support frame. The identification mechanism, located on one side of the support frame, is used for visual recognition of the high-speed sorting robot and includes: a connecting sleeve and a high-speed recognition camera; The gripping mechanism, located below the connecting sleeve, is used to grip the identified material.

[0005] Furthermore, the swing mechanism also includes a first control box symmetrically and fixedly connected to the top of the base. A first motor is fixedly connected inside the first control box. A first drive gear is fixedly connected to the output shaft of the first motor. First connecting rods are fixedly connected to both sides of the first swing arm. The two first connecting rods are rotatably connected to the two first control boxes respectively. A first driven gear is fixedly connected to the first connecting rod located at the first drive gear. One side of the first driven gear meshes with the first drive gear for transmission.

[0006] Furthermore, the flipping mechanism also includes a second control box fixedly connected to one side of the first swing arm. A second motor is fixedly connected inside the second control box, and a second drive gear is fixedly connected to the output shaft of the second motor. The first connecting arm is rotatably connected to the first swing arm, and a second connecting rod is fixedly connected to the first connecting arm. A second driven gear is fixedly connected to one end of the second connecting rod at the location corresponding to the second drive gear. One side of the second driven gear meshes with the second drive gear for transmission. The second connecting arm is fixedly connected to the first connecting arm.

[0007] Furthermore, a third control box is fixedly connected to the top side of the second connecting arm, a third motor is fixedly connected inside the third control box, a third drive gear is fixedly connected to the output shaft of the third motor, a third connecting rod is fixedly connected to one end of the support frame, a third driven gear is fixedly connected to the third connecting rod, and one side of the third driven gear meshes with the third drive gear for transmission.

[0008] Furthermore, the identification mechanism also includes a fourth motor fixedly connected to one side of the support frame. The output end of the fourth motor is fixedly connected to a drive rod. The drive rod is fixedly connected to a connecting sleeve. The connecting sleeve is rotatably connected to the support frame. The high-speed identification camera is fixedly connected inside the connecting sleeve.

[0009] Furthermore, the gripping mechanism includes two gripping arms that are symmetrically rotatably connected to the lower part of the connecting sleeve. One end of each of the two gripping arms is rotatably connected to an abutment plate on the opposite side. A fifth motor is fixedly connected to the lower part of the interior of the connecting sleeve. The other ends of the two gripping arms are fixedly connected to a first half gear and a second half gear, respectively. The first half gear and the second half gear mesh and drive each other. The output shaft of the fifth motor is fixedly connected to the first half gear.

[0010] Furthermore, a control panel is fixedly connected inside the base. The control panel is electrically connected to the first motor, the second motor, the third motor, the fourth motor, the fifth motor, and the high-speed recognition camera. Multiple support arms are connected to the outer array of the base. A control board is fixedly connected to one side of the base and is electrically connected to the control panel.

[0011] The vision recognition-based high-speed sorting robot grasping control system is applicable to the aforementioned vision recognition-based high-speed sorting robot. The control panel mainly consists of a vision recognition unit, a data processing and analysis unit, a motion control unit, a force feedback adjustment unit, and a communication unit. These units cooperate with each other to form a closed-loop control system, thereby achieving precise control of the high-speed sorting robot's grasping action. The visual recognition unit includes a high-speed recognition camera that captures image information of an object from different angles. Combined with structured light or TOF technology, it obtains the three-dimensional spatial information of the object, accurately identifies the shape, size, position and posture of the object, and uses deep learning algorithms, such as the improved YOLO series algorithms or Mask R-CNN, to process the captured image data in real time, quickly identify the category and features of the object, and provide an accurate basis for subsequent grasping decisions. Deep learning algorithms utilize automatic feature extraction to extract fundamental and boundary features of image information based on the acquisition and processing of image data. This lays a solid foundation for automatic image recognition. The fundamental features of image information, including grayscale data, spectral data, and tri-color data of image pixels, can all be obtained from the initial image data acquisition. Therefore, to extract and preserve the boundary features of image information, based on the image's transform coefficients, Fourier coefficient theory is used for boundary feature extraction and recognition. The core of deep learning-based automatic feature extraction lies in learning a high-dimensional representation of the data through multi-layer nonlinear transformations, which can be mathematically expressed as: ; in, Indicates the first Layer feature representation, This is the weight matrix for this layer. To place an item with a bias of 100,000, It is a non-linear activation function (such as ReLU); The data processing and analysis unit receives image data and recognition results transmitted by the visual recognition unit, performs further analysis and processing on the data, and converts the object information in the image coordinate system into the actual position and posture information in the robot coordinate system through coordinate transformation and spatial mapping algorithms. Based on the characteristics of the items and the requirements of the sorting task, and combined with the robot's kinematic model, the optimal path and posture parameters for the robotic arm to grasp the items are calculated, and control commands are generated and sent to the motion control unit. At the same time, this module is also responsible for storing and managing the data during the system operation for subsequent analysis and optimization. The kinematic model is used to train and optimize the recognition model based on the automatic feature extraction of the deep learning algorithm. Model optimization is a key factor in improving the performance of computer image recognition systems, mainly involving regularization, learning rate adjustment, and hyperparameter optimization. Among these, regularization methods are used to suppress overfitting, and L2 regularization (weight decay) is a commonly used technique. Its mathematical expression is: ; in, For loss function, These are the model's predicted values. For the true value, The regularization coefficient is . These are the weight parameters.

[0012] Furthermore, the motion control unit, based on a high-performance motion controller, receives control commands sent by the data processing and analysis unit to precisely control the robotic arm of the high-speed sorting robot. It employs advanced trajectory planning algorithms, such as quintic polynomial trajectory planning and spline curve trajectory planning, to ensure that the robotic arm can smoothly and quickly reach the target position during high-speed movement, while avoiding collisions with surrounding objects. Combined with a servo drive system, high-precision position and speed control of the robotic arm joints is achieved, enabling the robotic arm to accurately grasp objects according to a predetermined trajectory and posture. The motion control unit also has real-time monitoring and feedback functions, which can promptly obtain the actual motion state of the robotic arm and compare it with the preset motion trajectory. Error correction is performed through a closed-loop control algorithm to ensure the accuracy of the grasping action. The trajectory planning algorithm uses Cartesian space trajectory preprocessing, which directly affects subsequent trajectory planning. There are various path construction methods, such as straight lines, circular arc fitting, and advanced curves such as B-splines and NURBS. The choice depends on task requirements and performance considerations, such as smoothness, accuracy, flexibility, and efficiency. Complex trajectories are often composed of a combination of circular arcs and straight lines. In practical applications, comprehensive evaluation and testing are required to ensure that the path meets the application goals and adapts to the actual environment, laying a solid foundation for efficient and accurate sorting operations. Hermit curves are used for smooth transitions, effectively reducing speed abrupt changes, improving motion stability, and ensuring efficient and smooth sorting operations. On the tangents of two adjacent segments of the trajectory, according to the given length Select key points and ,here, The selection of the transition curve needs to comprehensively consider the smoothness of the transition curve, the continuity of the path, and the kinematic constraints of the robot. To ensure the rationality of the transition curve, the transition point also needs to be considered. and length and maximum distance parameter Relationship, ; The first derivative of the above curve equation is: ; The force feedback adjustment unit is equipped with a high-sensitivity force sensor on the contact plate to monitor the magnitude and direction of the force between the grasping device and the object in real time during the grasping process. The force sensor converts the collected force signal into an electrical signal and transmits it to the force feedback adjustment unit. The sorting robot's trajectory planning employs a backtracking-lookahead strategy to optimize multi-segment trajectory control, avoid frequent starts and stops, and after torque adaptation, adjust joint acceleration for excessive torque, followed by backtracking-lookahead planning to ensure the end-effector speed smoothly drops to zero, improving trajectory execution efficiency and stability. Given n discrete points, each path point corresponding to a speed value, a variable is set as the number of segments K = n-1 for subsequent planning. The process consists of the following four steps: Step 1: Determine the start and end points of the k-th path segment. Based on the robot's dynamic model and joint torque constraints, calculate the range of linear acceleration that the start and end points may generate under the joint torque constraints. Step 2: Calculate the acceleration corresponding to segment K. ,if If the acceleration is within the limit, proceed to step 4; otherwise, proceed to step 3. Step 3: If ,set up ; if ,set up ; calculate And replace the initial point velocity with ; Step 4: If K=1, the backtracking and look-ahead speed planning ends; Conversely, K = K - 1, run step 1.

[0013] Furthermore, the force feedback adjustment unit adjusts the gripping force of the gripping arm in real time based on the feedback information from the force sensor and parameters such as the material, shape, and weight of the item through an adaptive control algorithm. For example, for fragile items, a smaller gripping force is used and kept constant, while for heavier items, the gripping force is automatically increased to ensure a firm grip. At the same time, the force feedback adjustment unit can also identify and process abnormal force signals. When an abnormal change in gripping force is detected, an alarm is issued in time and corresponding protective measures are taken to prevent damage to the item or mechanical failure. The communication unit adopts a combination of wireless and wired communication to realize data transmission and communication between modules. The communication unit is responsible for the rapid and stable transmission of image data from the vision recognition unit, control commands from the data processing and analysis unit, status information from the motion control unit, and force signals from the force feedback adjustment unit within the system. The communication unit also provides a communication interface for connecting with external devices, enabling remote monitoring, parameter setting, and collaborative operation of the system. Through the communication unit, operators can understand the system's operating status in real time, perform remote debugging and optimization, and improve the system's intelligent management level.

[0014] Compared with existing technologies, the beneficial effects of this invention are: 1. This invention, through the coordinated design of the swing mechanism and the flipping mechanism, enables the robotic arm to achieve multi-dimensional movements such as swinging back and forth and flipping positions. Combined with the precise linkage of the recognition mechanism and the gripping mechanism, it can quickly cover material sorting scenarios in different directions. Compared with traditional fixed trajectory sorting equipment, the working range is expanded by more than 30%, significantly improving sorting efficiency. 2. The present invention uses a structure in which motors and gears mesh to drive each joint (such as the first drive gear and the first driven gear), which has a transmission efficiency of over 90% and features smooth transmission and small error. This ensures that the robotic arm maintains accurate positioning during high-speed movement, with a repeatability accuracy of ≤0.5mm, meeting the accuracy requirements of high-speed sorting scenarios. 3. This invention uses a high-speed recognition camera combined with structured light or TOF technology to acquire the three-dimensional spatial information of items in real time. It achieves a recognition accuracy of 99.2% for items with irregular shapes and different materials (such as packages, irregularly shaped parts, etc.), which solves the limitation of traditional sorting equipment that depends on the shape of the items. It is especially suitable for complex scenarios in logistics warehouses where packages vary in size and material. Deep learning algorithm optimization: It adopts improved YOLO series or Mask R-CNN algorithms to process image data in real time, improving the feature extraction speed by 40%. It can complete the recognition of item category, position and posture within 100ms. Combined with Fourier coefficient theory to extract image boundary features, it can maintain high recognition accuracy even in complex environments such as changing light and stacked items. 4. This invention integrates visual recognition, data processing, and motion control units through a control panel, forming a closed-loop control of "recognition-planning-execution-feedback". For example, the data processing unit maps image information to the robot coordinate system through a coordinate transformation algorithm. The motion control unit uses fifth-order polynomial trajectory planning to improve the smoothness of the robotic arm's movement by 60%, avoiding impacts and collisions during high-speed movement. Force feedback adaptive adjustment: The high-sensitivity force sensor on the contact plate monitors the gripping force in real time. Combined with the adaptive control algorithm, the gripping force can be automatically adjusted according to the material of the item (such as fragile items or heavy objects). For fragile items, the gripping force fluctuation range is controlled within ±5%. For heavy objects, the gripping force can be automatically increased by 20%-30% to prevent falling. At the same time, through abnormal force signal recognition, an alarm can be triggered and the action can be stopped within 0.1 seconds, reducing the item damage rate and the risk of mechanical failure. 5. This invention utilizes a combination of wireless (5G, Wi-Fi) and wired (Ethernet) communication methods, achieving data transmission rates exceeding 100Mbps. This ensures real-time interaction of visual images, control commands, and other data, supporting collaborative operations between multiple sorting robots and host computers and warehouse management systems. It enables intelligent scheduling of the sorting process and provides data storage and optimization functions: the data processing and analysis unit stores and analyzes system operation data (such as grasping success rate and robotic arm movement trajectory), optimizing model parameters through algorithms such as L2 regularization. This allows the system to continuously improve sorting efficiency and accuracy over long-term operation, with sorting efficiency increasing by 15%-20% compared to the initial state. In summary, this equipment not only achieves high-speed, accurate, and adaptive material sorting through mechanical structure innovation, visual recognition optimization, intelligent control algorithms, and communication system integration, but also can be widely used in logistics, industrial production, and other fields. It effectively solves the problems of low sorting efficiency, poor adaptability, and insufficient intelligence in existing technologies, and has significant economic value and practical significance. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall top-view three-dimensional structure of the high-speed sorting robot based on vision recognition proposed in this invention; Figure 2 This is a schematic diagram of the overall top-down three-dimensional structure of the high-speed sorting robot based on vision recognition proposed in this invention; Figure 3 This is a schematic diagram of the overall top-down three-dimensional structure of the high-speed sorting robot based on vision recognition proposed in this invention; Figure 4 This is a top-view three-dimensional structural diagram of the swing mechanism of the high-speed sorting robot based on vision recognition proposed in this invention; Figure 5 This is a schematic diagram of the overall third top-view stereoscopic front view structure of the high-speed sorting robot based on vision recognition proposed in this invention; Figure 6 This is a top-down three-dimensional structural diagram of the base and control panel of the high-speed sorting robot based on vision recognition proposed in this invention. Figure 7 This is a top-down three-dimensional schematic diagram of the gripping mechanism of the high-speed sorting robot based on vision recognition proposed in this invention. Figure 8 This is a partial cross-sectional top view of the three-dimensional structure of the second connecting arm of the high-speed sorting robot based on vision recognition proposed in this invention. Figure 9 This is a system block diagram of the high-speed sorting robot grasping control system based on vision recognition proposed in this invention.

[0016] In the diagram: 1. Base; 2. Support arm; 3. Control panel; 4. Swing mechanism; 401. First control box; 402. First motor; 403. First swing arm; 404. First connecting rod; 405. First driven gear; 406. First drive gear; 5. Tilting mechanism; 501. First connecting arm; 502. Second connecting arm; 503. Support frame; 504. Second motor; 505. Second drive gear; 506. Second driven gear; 507. Second control box; 508. Third control box; 509. Third motor; 510. Third drive gear; 511. Third driven gear; 6. Recognition mechanism; 601. Connecting sleeve; 602. High-speed recognition camera; 603. Fourth motor; 7. Grabbing mechanism; 701. Grabbing arm; 702. Abutment plate; 703. Fifth motor; 704. First half gear; 705. Second half gear; 8. Control panel. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0019] Example 1 Reference Figure 1-8 A high-speed sorting robot based on vision recognition includes: a base 1, a swing mechanism 4, a flipping mechanism 5, a recognition mechanism 6, and a gripping mechanism 7. The swing mechanism 4 includes a first swing arm 403 and a first control box 401 symmetrically and fixedly connected to the top of the base 1. A first motor 402 is fixedly connected inside the first control box 401. A first drive gear 406 is fixedly connected to the output shaft of the first motor 402. A first connecting rod 404 is fixedly connected to both sides of the first swing arm 403. The two first connecting rods 404 are rotatably connected to the two first control boxes 401 respectively. A first driven gear 405 is fixedly connected to the first connecting rod 404 located at the first drive gear 406. One side of the first driven gear 405 meshes with the first drive gear 406. The rotation of the first drive gear 406 synchronously drives the meshing of the first driven gear 405. The flipping mechanism 5 includes: a first connecting arm 501, a second connecting arm 502, a support frame 503, and a second control box 507 fixedly connected to one side of the first swing arm 403. A second motor 504 is fixedly connected inside the second control box 507. A second drive gear 505 is fixedly connected to the output shaft of the second motor 504. The first connecting arm 501 is rotatably connected to the first swing arm 403. A second connecting rod is fixedly connected to the first connecting arm 501. A second driven gear 506 is fixedly connected to one end of the second connecting rod at the location corresponding to the second drive gear 505. One side of the second driven gear 506 meshes with the second drive gear 505 for transmission. The second connecting arm 502 is fixedly connected to the first connecting arm 501. The identification mechanism 6 includes: a connecting sleeve 601 and a high-speed identification camera 602, and also includes a fourth motor 603 fixedly connected to one side of the support frame 503. The output end of the fourth motor 603 is fixedly connected to a drive rod, which is fixedly connected to the connecting sleeve 601. The connecting sleeve 601 is rotatably connected to the support frame 503. The high-speed identification camera 602 is fixedly connected inside the connecting sleeve 601. The gripping mechanism 7 includes two gripping arms 701 symmetrically rotatably connected to the lower part of the connecting sleeve 601. One end of each gripping arm 701 is rotatably connected to an abutment plate 702 on opposite sides. A fifth motor 703 is fixedly connected to the lower part of the connecting sleeve 601. The other ends of the two gripping arms 701 are fixedly connected to a first half gear 704 and a second half gear 705, respectively. The first half gear 704 and the second half gear 705 mesh and drive each other. The output shaft of the fifth motor 703 is fixedly connected to the first half gear 704. The start of the fifth motor 703 drives the first half gear 704 to rotate through the output shaft. The rotation of the first half gear 704 drives the second half gear 705 to rotate synchronously. The synchronous rotation of the first half gear 704 and the second half gear 705 drives the two gripping arms 701 to unfold and close, thereby gripping and holding objects.

[0020] In this invention, a third control box 508 is fixedly connected to one side of the top of the second connecting arm 502. A third motor 509 is fixedly connected inside the third control box 508. A third drive gear 510 is fixedly connected to the output shaft of the third motor 509. A third connecting rod is fixedly connected to one end of the support frame 503. A third driven gear 511 is fixedly connected to the third connecting rod. One side of the third driven gear 511 meshes with the third drive gear 510 for transmission. The start of the third motor 509 drives the third drive gear 510 to rotate and meshes with the third driven gear 511, thereby driving the rotation of the third connecting rod. The rotation of the third connecting rod drives the rotation of the support frame 503. The rotation of the support frame 503 drives the synchronous rotation of the connecting sleeve 601.

[0021] In this invention, a control panel 8 is fixedly connected inside the base 1. The control panel 8 is electrically connected to the first motor 402, the second motor 504, the third motor 509, the fourth motor 603, the fifth motor 703, and the high-speed recognition camera 602. Multiple support arms 2 are arrayed on the outside of the base 1. A control board 3 is fixedly connected to one side of the outside of the base 1. The control board 3 is electrically connected to the control panel 8. The control panel 8 is used to control the start and stop of the first motor 402, the second motor 504, the third motor 509, the fourth motor 603, the fifth motor 703, and the high-speed recognition camera 602.

[0022] Example 2 Reference Figure 9 The high-speed sorting robot grasping control system based on vision recognition is applicable to the aforementioned high-speed sorting robot based on vision recognition. The control panel 8 mainly consists of a vision recognition unit, a data processing and analysis unit, a motion control unit, a force feedback adjustment unit, and a communication unit. These units cooperate to form a closed-loop control system, achieving precise control of the high-speed sorting robot's grasping action. Deep learning algorithms utilize automatic feature extraction. Based on the acquisition and processing of image information data, basic and boundary features of the image information are extracted, laying a good foundation for automatic image recognition. The basic features of the image information, including the grayscale data, spectral data, and tri-color data of image pixels, can all be obtained from the initial image information acquisition. Therefore, in order to extract and retain the boundary features of the image information, based on the image transformation coefficients, Fourier coefficient theory is used to extract and recognize the boundary features of the image information. The core of deep learning automatic feature extraction lies in learning the high-dimensional representation of the data through multi-layer nonlinear transformations, which can be mathematically expressed as: ; in, Indicates the first Layer feature representation, This is the weight matrix for this layer. To place an item with a bias of 100,000, It is a non-linear activation function (such as ReLU); The visual recognition unit includes a high-speed recognition camera 602 that acquires image information of objects from different angles. Combined with structured light or TOF (Time-of-Flight) technology, it obtains the three-dimensional spatial information of the objects, accurately identifies the shape, size, position and posture of the objects, and uses deep learning algorithms, such as the improved YOLO (You Only Look Once) series of algorithms or MaskR-CNN (Mask Region-based Convolutional Neural Networks), to process the acquired image data in real time, quickly identify the category and features of the objects, and provide accurate basis for subsequent grasping decisions. The data processing and analysis unit receives image data and recognition results transmitted by the vision recognition unit, performs further analysis and processing on the data, and converts the object information in the image coordinate system into the actual position and posture information in the robot coordinate system through coordinate transformation and spatial mapping algorithms. Based on the characteristics of the items and the requirements of the sorting task, and combined with the robot's kinematic model, the optimal path and posture parameters for the robotic arm to grasp the items are calculated, and control commands are generated and sent to the motion control unit. At the same time, this module is also responsible for storing and managing the data during the system operation for subsequent analysis and optimization. The kinematic model is used to train and optimize the recognition model based on the automatic feature extraction of the deep learning algorithm. Model optimization is a key factor in improving the performance of computer image recognition systems, mainly involving regularization, learning rate adjustment, and hyperparameter optimization. Among these, regularization methods are used to suppress overfitting, and L2 regularization (weight decay) is a commonly used technique. Its mathematical expression is: ; in, For loss function, These are the model's predicted values. For the true value, The regularization coefficient is . These are the weight parameters.

[0023] In this invention, the motion control unit is based on a high-performance motion controller and receives control commands sent by the data processing and analysis unit to precisely control the robotic arm of the high-speed sorting robot. It adopts advanced trajectory planning algorithms, such as quintic polynomial trajectory planning and spline curve trajectory planning, to ensure that the robotic arm can reach the target position smoothly and quickly during high-speed movement, while avoiding collisions with surrounding objects. By combining a servo drive system, high-precision position and speed control of the robotic arm joints is achieved, enabling the robotic arm to accurately grasp objects according to a predetermined trajectory and posture. The motion control unit also has real-time monitoring and feedback functions, which can promptly acquire the actual motion state of the robotic arm and compare it with the preset motion trajectory. Error correction is performed through a closed-loop control algorithm to ensure the accuracy of the grasping action. The trajectory planning algorithm uses Cartesian space trajectory preprocessing, which directly affects subsequent trajectory planning. There are various path construction methods, such as straight lines, circular arc fitting, and advanced curves such as B-splines and NURBS. The choice depends on task requirements and performance considerations, such as smoothness, accuracy, flexibility, and efficiency. Complex trajectories are often composed of a combination of circular arcs and straight lines. In practical applications, comprehensive evaluation and testing are required to ensure that the path meets the application goals and adapts to the actual environment, laying a solid foundation for efficient and accurate sorting operations. Hermit curves are used for smooth transitions, effectively reducing speed abrupt changes, improving motion stability, and ensuring efficient and smooth sorting operations. On the tangents of two adjacent segments of the trajectory, according to the given length Select key points and ,here, The selection of the transition curve needs to comprehensively consider the smoothness of the transition curve, the continuity of the path, and the kinematic constraints of the robot. To ensure the rationality of the transition curve, the transition point also needs to be considered. and length and maximum distance parameter Relationship, ; The first derivative of the above curve equation is: ; The force feedback adjustment unit is equipped with a high-sensitivity force sensor on the contact plate 702 to monitor the magnitude and direction of the force between the gripping device and the object in real time during the gripping process. The force sensor converts the collected force signal into an electrical signal and transmits it to the force feedback adjustment unit. The sorting robot's trajectory planning employs a backtracking-lookahead strategy to optimize multi-segment trajectory control, avoid frequent starts and stops, and after torque adaptation, adjust joint acceleration for excessive torque, followed by backtracking-lookahead planning to ensure the end-effector speed smoothly drops to zero, improving trajectory execution efficiency and stability. Given n discrete points, each path point corresponding to a speed value, a variable is set as the number of segments K = n-1 for subsequent planning. The process consists of the following four steps: Step 1: Determine the start and end points of the k-th path segment. Based on the robot's dynamic model and joint torque constraints, calculate the range of linear acceleration that the start and end points may generate under the joint torque constraints. Step 2: Calculate the acceleration corresponding to segment K. ,if If the acceleration is within the limit, proceed to step 4; otherwise, proceed to step 3. Step 3: If ,set up ; if ,set up ; calculate And replace the initial point velocity with ; Step 4: If K=1, the backtracking and look-ahead speed planning ends; Conversely, K = K - 1, run step 1.

[0024] In this invention, the force feedback adjustment unit adjusts the gripping force of the gripping arm 701 in real time based on the feedback information from the force sensor and parameters such as the material, shape, and weight of the item through an adaptive control algorithm. For example, for fragile items, a smaller gripping force is used and kept constant, while for heavier items, the gripping force is automatically increased to ensure a firm grip. At the same time, the force feedback adjustment unit can also identify and process abnormal force signals. When an abnormal change in gripping force is detected, an alarm is issued in time and corresponding protective measures are taken to prevent damage to the item or mechanical failure. The communication unit uses a combination of wireless communication (such as 5G, Wi-Fi) and wired communication (such as Ethernet) to realize data transmission and communication between modules. The communication unit is responsible for the fast and stable transmission of image data from the vision recognition unit, control commands from the data processing and analysis unit, status information from the motion control unit, and force signals from the force feedback adjustment unit within the system. The communication unit also provides a communication interface for connecting with external devices (such as host computers, monitoring systems, and other sorting equipment) to realize remote monitoring, parameter setting, and collaborative operation of the system. Through the communication unit, operators can understand the operating status of the system in real time, perform remote debugging and optimization of the system, and improve the intelligent management level of the system.

[0025] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A high-speed sorting robot based on vision recognition, characterized in that, include: Base (1), used for supporting high-speed sorting robots; The swing mechanism (4), located on the base (1), is used for the high-speed sorting robot to swing back and forth. It includes: a first swing arm (403) for adjusting the back and forth angle; The flipping mechanism (5), located at one end of the first swing arm (403), is used for flipping the position of the high-speed sorting robot. It includes: a first connecting arm (501), a second connecting arm (502), and a support frame (503). The identification mechanism (6), located on one side of the support frame (503), is used for visual identification of the high-speed sorting robot and includes: a connecting sleeve (601) and a high-speed identification camera (602). The gripping mechanism (7), located below the connecting sleeve (601), is used to grip the identified material.

2. The high-speed sorting robot based on vision recognition according to claim 1, characterized in that, The swing mechanism (4) further includes a first control box (401) symmetrically fixedly connected to the top of the base (1). A first motor (402) is fixedly connected inside the first control box (401). A first drive gear (406) is fixedly connected to the output shaft of the first motor (402). A first connecting rod (404) is fixedly connected to both sides of the first swing arm (403). The two first connecting rods (404) are rotatably connected to the two first control boxes (401) respectively. A first driven gear (405) is fixedly connected to the first connecting rod (404) located at the first drive gear (406). One side of the first driven gear (405) meshes with the first drive gear (406) for transmission.

3. The high-speed sorting robot based on vision recognition according to claim 1, characterized in that, The flipping mechanism (5) further includes a second control box (507) fixedly connected to one side of the first swing arm (403). A second motor (504) is fixedly connected inside the second control box (507). A second drive gear (505) is fixedly connected to the output shaft of the second motor (504). The first connecting arm (501) is rotatably connected to the first swing arm (403). A second connecting rod is fixedly connected to the first connecting arm (501). A second driven gear (506) is fixedly connected to one end of the second connecting rod at the location corresponding to the second drive gear (505). One side of the second driven gear (506) meshes with the second drive gear (505) for transmission. The second connecting arm (502) is fixedly connected to the first connecting arm (501).

4. The high-speed sorting robot based on vision recognition according to claim 1, characterized in that, A third control box (508) is fixedly connected to one side of the top of the second connecting arm (502). A third motor (509) is fixedly connected inside the third control box (508). A third drive gear (510) is fixedly connected to the output shaft of the third motor (509). A third connecting rod is fixedly connected to one end of the support frame (503). A third driven gear (511) is fixedly connected to the third connecting rod. One side of the third driven gear (511) meshes with the third drive gear (510) for transmission.

5. The high-speed sorting robot based on vision recognition according to claim 1, characterized in that, The identification mechanism (6) also includes a fourth motor (603) fixedly connected to one side of the support frame (503). The output end of the fourth motor (603) is fixedly connected to a drive rod. The drive rod is fixedly connected to the connecting sleeve (601). The connecting sleeve (601) is rotatably connected to the support frame (503). The high-speed identification camera (602) is fixedly connected inside the connecting sleeve (601).

6. The high-speed sorting robot based on vision recognition according to claim 1, characterized in that, The gripping mechanism (7) includes two gripping arms (701) symmetrically rotatably connected to the lower part of the connecting sleeve (601). One end of each gripping arm (701) is rotatably connected to an abutment plate (702) on opposite sides. A fifth motor (703) is fixedly connected to the lower part of the connecting sleeve (601). The other ends of the two gripping arms (701) are fixedly connected to a first half gear (704) and a second half gear (705). The first half gear (704) and the second half gear (705) mesh and drive each other. The output shaft of the fifth motor (703) is fixedly connected to the first half gear (704).

7. The high-speed sorting robot based on vision recognition according to claim 1, characterized in that, The base (1) is internally fixedly connected to a control panel (8), which is electrically connected to the first motor (402), the second motor (504), the third motor (509), the fourth motor (603), the fifth motor (703) and the high-speed recognition camera (602). The base (1) is externally connected to a plurality of support arms (2), and a control panel (3) is fixedly connected to one side of the base (1). The control panel (3) is electrically connected to the control panel (8).

8. A vision-based high-speed sorting robot gripping control system, applicable to the vision-based high-speed sorting robot described in any one of claims 1-7, characterized in that, The control panel (8) is mainly composed of a visual recognition unit, a data processing and analysis unit, a motion control unit, a force feedback adjustment unit and a communication unit. Each unit cooperates with the others to form a closed-loop control system, thereby achieving precise control of the high-speed sorting robot's grasping action. The visual recognition unit includes a high-speed recognition camera (602) that collects image information of the object from different angles, and combines structured light or TOF technology to obtain the three-dimensional spatial information of the object, accurately identify the shape, size, position and posture of the object, and uses deep learning algorithms to process the collected image data in real time, quickly identify the category and features of the object, and provide an accurate basis for subsequent grasping decisions. The data processing and analysis unit receives image data and recognition results transmitted by the visual recognition unit, performs further analysis and processing on the data, and converts the object information in the image coordinate system into the actual position and posture information in the robot coordinate system through coordinate transformation and spatial mapping algorithms.

9. The high-speed sorting robot based on vision recognition and its grasping control system according to claim 8, characterized in that, The motion control unit, based on a high-performance motion controller, receives control commands sent by the data processing and analysis unit to precisely control the robotic arm of the high-speed sorting robot. The force feedback adjustment unit is equipped with a high-sensitivity force sensor on the abutment plate (702) to monitor the magnitude and direction of the force between the grasping device and the object in real time during the grasping process. The force sensor converts the collected force signal into an electrical signal and transmits it to the force feedback adjustment unit.

10. The high-speed sorting robot based on vision recognition and its grasping control system according to claim 9, characterized in that, The force feedback adjustment unit adjusts the gripping force of the gripping arm (701) in real time through an adaptive control algorithm based on the feedback information from the force sensor and in combination with parameters such as the material, shape and weight of the item. The communication unit adopts a combination of wireless and wired communication to realize data transmission and communication between modules. The communication unit is responsible for the rapid and stable transmission of image data from the vision recognition unit, control commands from the data processing and analysis unit, status information from the motion control unit, and force signals from the force feedback adjustment unit within the system. The communication unit also provides a communication interface for connecting to external devices, enabling remote monitoring, parameter setting, and collaborative operation of the system.

Images