High-precision automatic carrying system based on intelligent technology
Through a high-precision automatic handling system with embodied intelligent technology, combined with deep learning and natural language processing, the robot's object recognition and positioning accuracy is improved, intelligent path planning and decision-making is realized, and the existing handling robots' coordination and optimization problems in complex environments are solved. It is suitable for multi-device collaborative operations.
Patent Information
- Application Number
- CN202510378677.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-28
- Publication Date
- 2025-07-25
AI Technical Summary
The lack of learning and decision-making capabilities of existing handling robots leads to mechanization and immobility in complex handling operations, making it difficult to adapt to multi-equipment collaboration and optimization.
It adopts a high-precision automatic handling system with embodied intelligent technology, combining large-scale pre-trained convolutional neural networks and Transformer models, deep reinforcement learning algorithms, natural language processing technology and multimodal data fusion to realize robot visual recognition, path planning and adaptive learning, improve object recognition and positioning accuracy, and realize coordinated scheduling between robots and equipment.
It improves the object recognition and positioning accuracy of the transport robot, realizes intelligent path planning and decision-making, improves the adaptability and coordination efficiency of the automatic transport system, and is suitable for various types of transport operations.
Smart Images
Figure CN120364341A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of handling equipment and automatic handling systems, and more specifically, to a high-precision automatic handling system based on embodied intelligence technology. Background Art
[0002] A handling robot is an automated product that uses the motion trajectory of a robot to replace manual handling. And different end effectors can be installed on the handling robot to complete the handling work of workpieces with various shapes and states, greatly reducing the heavy physical labor of humans.
[0003] Currently, when the commonly used handling robots are working, their working modes are relatively single, and the handling robots themselves lack learning ability and decision-making ability. They are relatively mechanized and fixed in the handling operations of the handling robots, which is not conducive to the performance of more complex handling operations.
[0004] Therefore, in view of this, research and improvement are carried out on the existing structure, and a high-precision automatic handling system based on embodied intelligence technology is provided, with the expectation of achieving a more practical value. Summary of the Invention
[0005] 1. Technical Problems to be Solved
[0006] Aiming at the problems existing in the prior art, the purpose of the present invention is to provide a high-precision automatic handling system based on embodied intelligence technology, which can improve the object recognition and positioning accuracy of the handling robot, complete the intelligent path planning and decision-making of the robot, realize the coordination and intelligent scheduling between the handling robot and multiple devices, make the handling robot applicable to various different types of handling operations, and at the same time, improve the adaptability and optimization ability of the automatic handling system.
[0007] 2. Technical Solutions
[0008] To solve the above problems, the present invention adopts the following technical solutions.
[0009] A high-precision automatic handling system based on embodied intelligence technology includes a handling robot arranged beside a production line. There are several objects to be handled on the conveyor belt of the production line. The handling robot is composed of a base, a column, a large arm, a small arm and a fixture, and any two adjacent of the above structures are movably connected. A docking pipe is fixedly installed on the side of the base, and a goods transfer vehicle is docked through the docking pipe. Electric wheels and motors for controlling the wheels are fixedly installed at the bottom of the base and the bottom of the goods transfer vehicle, and the motors are all controlled by electrical signals. A searchlight board is fixedly installed on the side of the small arm, and several induction probes are fixedly installed at the bottom of the searchlight board;
[0010] An automatic handling system formed by the above-mentioned handling robot. The handling automatic handling system includes a training and learning module, a path planning module, a language collaboration module, a data fusion module, and an adaptive training and learning module. All of the above modules are centrally processed by the internal CPU of the handling robot. Multiple sensors are provided on the handling robot, and a visual automatic handling system, a tactile automatic handling system, a force perception automatic handling system, and a temperature perception automatic handling system of the handling robot are formed through the multiple sensors.
[0011] Furthermore, a docking plug rod is fixedly arranged on one side of the goods transfer vehicle, and the docking plug rod is inserted into the docking pipe.
[0012] Furthermore, a through groove hole is opened at the top of the docking pipe, a clamping groove hole is opened on the docking plug rod, and the aperture sizes of the through groove hole and the clamping groove hole are equal;
[0013] An installation seat is fixedly installed on the base, a small hydraulic cylinder is fixedly installed on the installation seat, and the output shaft of the small hydraulic cylinder passes through the through groove hole and is clamped into the clamping groove hole.
[0014] Furthermore, the number of the induction probes is twelve, and each induction probe is arranged to incline outward by 30°.
[0015] Furthermore, in the training and learning module, a convolutional neural network and a Transformer model with large-scale pre-training are adopted, which are used for accurate recognition and pose estimation of objects such as wafers and chips in the robot visual automatic handling system;
[0016] In the training and learning module, through the powerful generalization ability of the large model, the robot can identify semiconductor products of different sizes, shapes, and materials, and perform stable recognition and positioning under different lighting and environmental conditions.
[0017] Furthermore, in the path planning module, a deep reinforcement learning algorithm is introduced, enabling the robot to learn the best path through interaction with the environment, respond to changes in the production line in real time, and improve the handling efficiency and flexibility.
[0018] Furthermore, in the language collaboration module, combined with natural language processing, the handling robot can conduct intelligent conversations and collaborations with other automated devices or operators on the production line;
[0019] The handling robot interacts with other devices through natural language commands, obtains production line information or conducts task scheduling, and improves the overall coordination efficiency of the automatic handling system.
[0020] Furthermore, in the data fusion module, the handling robot utilizes various perception data such as vision, touch, force sense, and temperature, and performs multi-modal data fusion through a large model to optimize the decision-making process.
[0021] Furthermore, in the adaptive training and learning module, a deep learning model is formed within the automatic handling system, which learns and updates in real time during the production process. By continuously collecting feedback data, the handling strategy and recognition ability are continuously optimized.
[0022] 3. Beneficial effects
[0023] Compared with the prior art, the advantages of the present invention are as follows:
[0024] In this solution, through the deep neural network model with large-scale pre-training, combined with the large model technology, the visual ability of the robot is significantly improved, and the object recognition and positioning accuracy of the handling robot is enhanced;
[0025] Through deep reinforcement learning, manual intervention is reduced and production efficiency is optimized, enabling the handling robot to not only carry out handling according to the target position, but also cope with unforeseen obstacles and production progress changes, realizing intelligent path planning and decision-making;
[0026] Through natural language processing technology, the handling robot can understand the scheduling instructions on the production line, cooperate with other devices, improve the overall cooperation efficiency of the automatic handling system, and the handling robot can communicate with manual operators through voice, enhancing the degree of factory automation and realizing the coordination and intelligent scheduling between the handling robot and multiple devices;
[0027] In summary, the handling robot is applicable to various different types of handling operations.
[0028] ② In this solution, the automatic handling system continuously optimizes modules such as recognition algorithms, path planning, and force feedback through real-time adaptive learning. As the experience accumulated by the automatic handling system becomes more and more rich, the handling robot realizes continuous self-improvement during the production process, gradually copes with more and more complex production tasks, and improves the adaptability and optimization ability of the automatic handling system. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 It is a schematic structural diagram of the production line and the handling robot in the present invention;
[0030] Figure 2 It is a side view schematic diagram of the handling robot in the present invention;
[0031] Figure 3 It is a schematic structural diagram of the docking pipe in the present invention;
[0032] Figure 4Schematic diagram of the structure of the goods transfer vehicle in the present invention;
[0033] Figure 5 Schematic diagram of the system of the automatic handling system in the present invention.
[0034] Description of the reference numerals in the figure:
[0035] 1. Production line;
[0036] 2. Handling robot;
[0037] 201. Base; 2011. Mounting seat; 2012. Small hydraulic cylinder;
[0038] 202. Column; 203. Boom; 204. Forearm; 205. Fixture;
[0039] 3. Docking pipe; 301. Through slot hole;
[0040] 4. Goods transfer vehicle; 401. Docking plug; 4011. Card slot hole;
[0041] 5. Searchlight board;
[0042] 6. Inductive probe. Specific implementation manner
[0043] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention; obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0044] Embodiment 1:
[0045] Please refer to Figure 1 - Figure 4 , a high-precision automatic handling system based on embodied intelligence technology, including a handling robot 2 arranged beside a production line 1. There are several objects to be handled on the conveyor belt of the production line 1. The handling robot 2 is composed of a base 201, a column 202, a boom 203, a forearm 204 and a fixture 205, and any two adjacent of the above structures are movably connected. A docking pipe 3 is fixedly installed on the side of the base 201, and a goods transfer vehicle 4 is docked through the docking pipe 3. Casters and motors for controlling the casters are fixedly installed at the bottom ends of the base 201 and the goods transfer vehicle 4, and the motors are all controlled by electrical signals. A searchlight board 5 is fixedly installed on the side of the forearm 204, and several inductive probes 6 are fixedly installed at the bottom end of the searchlight board 5;
[0046] Refer to Figure 1 , Figure 4, specifically, a docking plug 401 is fixedly arranged on one side of the goods transfer vehicle 4, and the docking plug 401 is inserted into the docking pipe 3.
[0047] This facilitates the completion of the docking between the goods transfer vehicle 4 and the docking pipe 3.
[0048] Refer to Figure 1 、 Figure 3 、 Figure 4 , specifically, a through slot hole 301 is opened at the top of the docking pipe 3, a clamping slot hole 4011 is opened on the docking plug 401, and the aperture sizes of the through slot hole 301 and the clamping slot hole 4011 are equal;
[0049] An installation seat 2011 is fixedly installed on the base 201, a small hydraulic cylinder 2012 is fixedly installed on the installation seat 2011, the output shaft of the small hydraulic cylinder 2012 passes through the through slot hole 301 and is clamped into the clamping slot hole 4011.
[0050] When the output shaft of the small hydraulic cylinder 2012 extends, the output shaft passes through the through slot hole 301 and is clamped into the clamping slot hole 4011, which can lock the goods transfer vehicle 4 and the docking pipe 3 after docking;
[0051] When the output shaft of the small hydraulic cylinder 2012 fully retracts, the output shaft returns to the through slot hole 301 and disengages from the inside of the clamping slot hole 4011, which can release the lock between the goods transfer vehicle 4 and the docking pipe 3. At this time, the handling robot 2 and the goods transfer vehicle 4 can be separated, which is convenient for the independent use of either of them.
[0052] Refer to Figure 1 、 Figure 2 , specifically, the number of induction probes 6 is twelve, and each induction probe 6 is arranged to be inclined outward by 30°.
[0053] This can effectively ensure the detection range of the induction probe 6, and thus ensure the detection effect of the induction probe 6.
[0054] Working principle:
[0055] When the goods to be handled move on the production line 1, the induction probe 6 on the handling robot 2 detects the approach of the goods to be handled, and the base 201, the column 202, the boom 203, the forearm 204 and the fixture 205 cooperate. The fixture 205 can clamp the goods to be handled, move them onto the goods transfer vehicle 4, and release them. Then the handling robot 2 returns to the initial position to complete the handling operation.
[0056] Embodiment 2:
[0057] Based on the above Embodiment 1, a further description is made.
[0058] Refer toFigure 1 , Figure 5 , specifically, an automatic handling system formed by the handling robot 2, which includes a training and learning module, a path planning module, a language collaboration module, a data fusion module, and an adaptive training and learning module. All the above modules are centrally processed by the internal CPU of the handling robot 2, and multiple sensors are provided on the handling robot 2, and a visual automatic handling system, a tactile automatic handling system, a force perception automatic handling system, and a temperature perception automatic handling system of the handling robot 2 are formed through the multiple sensors.
[0059] Specifically, in the training and learning module, a large-scale pre-trained convolutional neural network CNN and a Transformer model are adopted, which are used for accurate recognition and pose estimation of objects such as wafers and chips in the robot visual automatic handling system.
[0060] In the training and learning module, through the powerful generalization ability of the large model, the robot can identify semiconductor products of different sizes, shapes, and materials, and perform stable recognition and positioning under different lighting and environmental conditions.
[0061] The training and learning module can perform large-scale pre-training to form a training model, and then use it for object recognition and pose estimation.
[0062] The combination of the large-scale pre-trained model and computer vision technology enables the present invention to provide high-precision object recognition and pose estimation in complex environments by using a large-scale deep learning model.
[0063] Specifically, in the path planning module, a deep reinforcement learning DRL algorithm is introduced, enabling the robot to learn the best path through interaction with the environment, and respond in real time to changes in the production line 1 such as obstacles and production progress, improving the handling efficiency and flexibility.
[0064] The path planning module forms a deep reinforcement learning algorithm for completing intelligent path planning.
[0065] The advantage of the DRL algorithm is that it can optimize decisions based on feedback and even self-learn and improve from experience.
[0066] When applying deep reinforcement learning in path planning, the automatic handling system introduces deep reinforcement learning technology, enabling the robot to autonomously learn and optimize the path planning strategy, and flexibly handle complex situations in a dynamic production environment.
[0067] Specifically, in the language collaboration module, combined with natural language processing NLP, the handling robot 2 conducts intelligent dialogue and collaboration with other automated devices or operators on the production line 1.
[0068] The handling robot 2 interacts with other devices through natural language instructions, obtains information about the production line 1 or performs task scheduling, improving the overall coordination efficiency of the automatic handling system.
[0069] In the language collaboration module, a large-scale language model is formed, which can effectively collaborate with the handling robot 2.
[0070] When natural language processing collaborates with the robot, combined with the natural language processing model, the robot can understand and interact intelligently with other devices and operators on the production line 1, improving the coordination efficiency.
[0071] Specifically, in the data fusion module, the handling robot 2 utilizes various perception data such as vision, touch, force sense, and temperature, and performs multi-modal data fusion through a large model to optimize the decision-making process.
[0072] The data fusion module can complete multi-modal data fusion.
[0073] For example, the handling robot 2 can make more accurate handling decisions based on the combination of visual recognition data and force sense sensor data. Through the training of the large model, the automatic handling system can dynamically adjust the grasping force for different objects, avoiding problems of over-force or under-force.
[0074] Specifically, in the adaptive training and learning module, a deep learning model is formed within the automatic handling system, which learns and updates in real time during the production process. By continuously collecting feedback data, the handling strategy and recognition ability are continuously optimized.
[0075] The adaptive training and learning module can complete adaptive learning and realize the update of the automatic handling system.
[0076] For example, through the edge computing platform, the robot can perform real-time training and fine-tuning at the production site without the need to transmit data back to the central server, which can greatly improve the adaptability and response speed of the automatic handling system.
[0077] When the handling robot 2 performs real-time adaptive learning and optimization through the automatic handling system, the automatic handling system can perform online learning based on the feedback data in the actual production process and continuously optimize the performance of the automatic handling system.
[0078] In summary, the high-precision automatic handling system of the present invention combining large model technology breaks through the limitations of traditional automated handling systems and has broad application prospects in the semiconductor production line 1. The introduction of the large model enables the automatic handling system to demonstrate stronger intelligent capabilities in aspects such as visual recognition, path planning, and object grasping, not only improving production efficiency but also greatly enhancing the adaptability and sustainability of the automatic handling system.
[0079] Embodiment 3:
[0080] Based on the above-mentioned Embodiment 1 and Embodiment 2, a further description is made.
[0081] When the handling robot 2 works in cooperation with the automatic handling system:
[0082] Specifically, in a semiconductor wafer automated production line 1, the automatic handling system is equipped with a computer vision model pre-trained by deep learning for precise positioning and pose estimation of wafers. The robot can autonomously plan the handling path in a dynamic environment through a deep reinforcement learning algorithm and adjust the grasping pose according to visual feedback. The automatic handling can also coordinate tasks with other robots on the production line 1 through natural language to improve the collaborative operation efficiency.
[0083] Specifically, when the handling robot 2 works in cooperation with the automatic handling system, for example, in another electronic component assembly line, the robot utilizes the visual perception ability of the large model to achieve automatic classification and handling of components with different shapes and materials. Through the fusion of multi-modal sensors (vision, force sense, touch, etc.), the robot can adjust the operation strategy in real time to avoid damaging sensitive electronic components. In addition, the automatic handling system continuously optimizes the grasping strategy of the robot according to the actual feedback of the production line 1 to gradually improve the production efficiency.
[0084] The above is only a preferred specific embodiment of the present invention; however, the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and its improved concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.
Claims
1. A high-precision automatic handling system based on embodied intelligence technology, comprising a handling robot (2) arranged beside a production line (1), wherein there are several objects to be handled on the conveyor belt of the production line (1), and it is characterized in that: The handling robot (2) is composed of a base (201), a column (202), a large arm (203), a small arm (204) and a fixture (205), and any two adjacent of the above structures are movably connected. A docking pipe (3) is fixedly installed on the side of the base (201), and a goods transfer vehicle (4) is docked through the docking pipe (3). Electrically controlled motors for controlling casters are fixedly installed at the bottom ends of the base (201) and the goods transfer vehicle (4), and the motors are all controlled by electrical signals. A searchlight plate (5) is fixedly installed on the side of the small arm (204), and a number of induction probes (6) are fixedly installed at the bottom end of the searchlight plate (5). An automatic handling system formed by the above handling robot (2), the handling automatic handling system includes a training and learning module, a path planning module, a language collaboration module, a data fusion module and an adaptive training and learning module. All of the above modules are centrally processed by the internal CPU of the handling robot (2). A plurality of sensors are provided on the handling robot (2), and a visual automatic handling system, a tactile automatic handling system, a force perception automatic handling system, and a temperature perception automatic handling system of the handling robot (2) are formed through the plurality of sensors.
2. The high-precision automatic handling system based on embodied intelligence technology according to claim 1, wherein: A docking plug (401) is fixedly provided on one side of the goods transfer vehicle (4), and the docking plug (401) is inserted into the inside of the docking pipe (3).
3. The high-precision automatic handling system based on embodied intelligence technology according to claim 2, wherein: A through slot hole (301) is opened at the top end of the docking pipe (3), and a clamping slot hole (4011) is opened on the docking plug (401), and the aperture sizes of the through slot hole (301) and the clamping slot hole (4011) are equal. An installation seat (2011) is fixedly installed on the base (201), a small hydraulic cylinder (2012) is fixedly installed on the installation seat (2011), and the output shaft of the small hydraulic cylinder (2012) passes through the through slot hole (301) and is clamped into the inside of the clamping slot hole (4011).
4. An automatic high-precision handling system based on embodied intelligence technology according to claim 1, characterized in that: The number of the induction probes (6) is twelve, and each induction probe (6) is arranged to be inclined outward by 30°.
5. The high-precision automatic handling system based on embodied intelligence technology according to claim 1, wherein: In the training and learning module, a convolutional neural network and a Transformer model with large-scale pre-training are adopted, and in the robot visual automatic handling system, they are used for accurate recognition and pose estimation of objects such as wafers and chips. In the training and learning module, through the powerful generalization ability of the large model, the robot can identify semiconductor products of different sizes, shapes and materials, and perform stable recognition and positioning under different lighting and environmental conditions.
6. The high-precision automatic handling system based on embodied intelligence technology according to claim 1, wherein: In the path planning module, a deep reinforcement learning algorithm is introduced, so that the robot can learn the best path through interaction with the environment, respond to changes in the production line (1) in real time, and improve the handling efficiency and flexibility.
7. An automatic high-precision handling system based on embodied intelligence technology according to claim 1, characterized in that: In the language collaboration module, combined with natural language processing, the handling robot (2) conducts intelligent dialogue and collaboration with other automated devices or operators on the production line (1). The handling robot (2) interacts with other devices through natural language instructions, obtains information of the production line (1) or conducts task scheduling, and improves the overall cooperation efficiency of the automatic handling system.
8. The high-precision automatic handling system based on embodied intelligence technology according to claim 1, wherein: In the data fusion module, the handling robot (2) utilizes various perception data such as vision, touch, force sense, and temperature, and performs multimodal data fusion through a large model to optimize the decision-making process.
9. The high-precision automatic handling system based on embodied intelligence technology according to claim 1, wherein: In the adaptive training and learning module, a deep learning model is formed within the automatic handling system, which learns and updates in real time during the production process. By continuously collecting feedback data, the handling strategy and recognition ability are continuously optimized.