Digital twinning-based multi-specification robot palletizer autonomous palletizing method and system

Abstract

The invention discloses a multi-specification robot palletizer autonomous palletizing method and system based on digital twinning, and the method comprises the steps: building a robot palletizer twinning model, and building a digital twinning scene according to products, obstacles and surrounding environments in a robot palletizing operation production line; real space data are collected in real time, and a data real-time transmission channel is constructed; constructing a data storage layer for storing real space data and virtual space data; self-adaptive adjustment of stacking task points is carried out according to the planning requirements of the multi-specification products; training in a twin platform by using a deep reinforcement learning algorithm to obtain more training experience; and a planning result is transmitted to the entity palletizing robot to complete teaching point updating, and operation monitoring is conducted on the entity robot based on the virtual entity data two-way transmission technology and the virtual-real synchronization technology. By means of the method, autonomous planning of the palletizing robot for complex and diversified palletizing tasks can be achieved, and the method is suitable for continuously-changing production requirements.

Description

Technical Field

[0001] The present invention belongs to the field of palletizing operation planning of palletizing robots, and specifically relates to an autonomous palletizing method and system for multi-standard palletizing robots based on digital twins. Background Art

[0002] In modern manufacturing, as one of the core equipment of automated production lines, palletizing robots play an important role in improving production efficiency and reducing labor costs. Although traditional manual teaching programming has been widely used in history, its lack of flexibility in dealing with multiple specifications of products has become increasingly apparent. In the face of products with frequently changing shapes and specifications, palletizing robots often need to be programmed repeatedly, resulting in reduced production efficiency; in addition, the rapid changes in market demand require that production lines be able to adjust quickly to meet the requirements of different orders, which makes fixed palletizing solutions no longer applicable, further restricting the improvement of production efficiency.

[0003] The rise of digital twin technology provides new possibilities for solving this problem. By building a high-fidelity virtual model of a physical entity and achieving real-time synchronization with the real environment, the technology shows the potential to significantly improve the flexibility and accuracy of robots; however, its application in palletizing robots is still in the exploratory stage and has not yet fully utilized its advantages. Therefore, the present invention proposes an autonomous palletizing method and system for multi-standard palletizing robots based on digital twins, aiming to use the digital twin platform for intelligent programming, which can autonomously adapt to complex and changeable production tasks, thereby promoting the development of intelligent manufacturing and significantly improving overall production efficiency and flexibility. Summary of the invention

[0004] The present invention provides an autonomous palletizing method and system for multi-specification palletizing robots based on digital twins, which is used to plan the operations of palletizing robots that meet the planning requirements of multiple palletizing methods, multiple varieties, and multiple product specifications.

[0005] The technical solution of the present invention is: an autonomous palletizing method and system for multi-standard palletizing robots based on digital twins, comprising the following steps:

[0006] (1) Construct a twin model of the palletizing robot operation scene, and build a digital twin scene based on the products, obstacles, and surrounding environment in the palletizing robot's palletizing operation production line;

[0007] (2) Real-time collection of physical entity data in real space, such as palletizing robot operation data, product parameters, and obstacle information, to build a real-time data transmission channel;

[0008] (3) Construct a data storage layer to store physical space data and virtual space data, and use the data to drive the real space and virtual space entities accordingly, so that the robot system has bidirectional controllability;

[0009] (4) Make adaptive adjustments to the palletizing task points based on the planning requirements of various palletizing methods, various varieties, and various product specifications;

[0010] (5) Use reinforcement learning algorithms to train different palletizing task points and surrounding obstacle environment information in the twin platform to obtain more training experience, and then use the trained network model to calculate the theoretical optimal palletizing path;

[0011] (6) The planning result data is transmitted to the physical palletizing robot to complete the teaching point update, and the operation of the physical robot is monitored based on the virtual-physical data bidirectional transmission technology and the virtual-real synchronization technology to reduce the risk of collision.

[0012] Furthermore, in the above-mentioned multi-standard palletizing robot autonomous palletizing method and system based on digital twin, the step (1) specifically includes:

[0013] The twin model of the stacking robot operation scene is constructed by combining direct modeling with a general model file and surrounding fuzzy obstacle scanning modeling.

[0014] The general model file direct modeling mainly includes modeling through the unified robot description format (URDF) model, and standard part standardized file modeling.

[0015] The URDF can easily obtain a model of a multi-degree-of-freedom robot by defining the robot's (size parameters) SP, (position and posture parameters) PP, (kinematic parameters) KP, and (appearance parameters) AP, thereby describing the links and joints of the robot arm and their relative positions, and obtaining a twin palletizing robot that is 1:1 restored to the actual palletizing robot.

[0016] The standard parts are usually described in standardized files. The three-dimensional geometric models of standard parts are usually described in standardized CAD file formats such as STEP (.stp, .step), IGES (.iges, .igs), etc., which are used to share geometric data in different design software, or many manufacturers and suppliers of standard parts will provide specific product catalogs and technical specification files (such as PDF format), which describe the detailed information of these standard parts (including dimensions, materials, applications, etc.).

[0017] The peripheral fuzzy obstacle scanning modeling can use a variety of sensors to perceive its size information, position and posture information, kinematic relationship information, appearance information and other physical attribute information.

[0018] Furthermore, in the above-mentioned multi-standard palletizing robot autonomous palletizing method and system based on digital twin, the acquisition method in step (2) is Socket data transmission technology, which performs two-way data transmission to achieve data interconnection and interoperability during the robot's autonomous palletizing process.

[0019] Furthermore, in the above-mentioned multi-standard palletizing robot autonomous palletizing method based on digital twin, the data storage method in step (3) specifically includes:

[0020] Add ODBC driver in the OPCUA server node and configure Data Logger to connect to MySQL for data storage.

[0021] Furthermore, in the above-mentioned autonomous palletizing method and system for multi-specification palletizing robots based on digital twins, the adaptive adjustment method in step (4) includes: reading initial values ​​from the system input module to obtain product specifications and palletizing methods, initializing corresponding variables in the program and calculating the pallet layout; verifying the tool position of the robot, if it is not in the starting position, moving the tool to the starting position, and then executing the main loop; verifying whether the pallet and the object are in place, if they are in place, moving the robot's gripper to the picking approach point, and proceeding to the final picking point along the planned path; if an error occurs during the picking process, the loop is terminated; checking whether the loop should be terminated, if terminated, moving the robot to the starting position.

[0022] Furthermore, in the above-mentioned digital twin-based multi-standard palletizing robot autonomous palletizing method and system, the step (5) proposes a digital twin training process based on deep reinforcement learning, and then uses the trained network model to calculate the theoretical optimal palletizing path.

[0023] Furthermore, in the above-mentioned digital twin-based multi-standard palletizing robot autonomous palletizing method and system, the step (6) transmits the obtained path planning simulation results to the physical robot to update the teaching points, and monitors the operation of the physical robot through the synchronous operation of the robot twin model and the physical robot. When the robot twin model detects a collision, it immediately feeds back an emergency stop signal to stop the physical robot urgently, thereby preventing the physical robot from colliding due to setting errors.

[0024] According to another aspect of the present invention, a robot arm path planning simulation analysis system based on digital twin is provided:

[0025] (1) Twin scene construction module: Build a twin model of the palletizing robot operation scene, and build a digital twin scene based on the products, obstacles and surrounding environment in the palletizing robot's palletizing operation production line;

[0026] (2) Data interaction module: Socket data transmission technology is used for bidirectional data transmission. After obtaining the real-time data of the palletizing robot and production line in KEPServerEX 6 through Python, a Socket server is built on the Python side and a Socket client is built on the Unity 3D side. Bidirectional data transmission is performed through IP addresses and port numbers. In this way, the physical entity data in the real space, such as the operation data of the palletizing robot, product parameters, and obstacle information, is collected in real time, and a real-time data transmission channel is built.

[0027] (3) Data storage module: Build a data storage layer, add an ODBC driver to the OPCUA server node, configure the Data Logger to connect to MySQL, and store physical space data and virtual space data. Use the data to drive the real space and virtual space entities, so that the robot system has bidirectional controllability.

[0028] (4) Palletizing task point adaptive adjustment module: reads the initial value from the system input module to obtain the product specifications and palletizing methods, generates a main loop, and then the palletizing robot executes the main loop to complete the autonomous palletizing task, and makes adaptive adjustments to the palletizing task points according to the planning requirements of multiple palletizing methods, multiple varieties, and multiple product specifications;

[0029] (5) Robot path planning reinforcement learning module: Use reinforcement learning algorithms to train different palletizing task points and surrounding obstacle environment information in the twin platform to obtain more training experience, and then use the trained network model to calculate the theoretical optimal palletizing path;

[0030] (6) Teaching point update and physical robot operation monitoring module: The obtained path planning simulation results are transmitted to the physical robot to update the teaching points. The physical robot operation is monitored through the synchronous operation of the robot twin model and the physical robot. When the robot twin model detects a collision, it immediately feeds back an emergency stop signal to stop the physical robot urgently, thereby preventing the physical robot from colliding due to setting errors.

[0031] The beneficial effects of the present invention are as follows: for the problem that the existing palletizing robots have poor flexibility and the frequent changes and diversified products of the production line make it difficult to adapt to the fixed palletizing scheme, resulting in low production efficiency, the present invention proposes a multi-standard palletizing robot autonomous palletizing method and system based on digital twins. The palletizing robot twin model is constructed by digital twin technology, and the digital twin scene is built according to the products, obstacles and surrounding environment in the palletizing operation production line of the palletizing robot. The real space data is collected in real time, a real-time data transmission channel is constructed, and a data storage layer is constructed to store real space data and virtual space data. Finally, based on the digital twin platform, the autonomous adjustment function module of the palletizing task point, the path planning reinforcement learning training function module, the automatic update of the teaching point and the operation monitoring function module are designed to form a complete set of autonomous palletizing methods and systems for multi-standard palletizing robots, so as to improve the flexibility of the palletizing robot and adapt to the ever-changing and diversified production needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flowchart of the autonomous palletizing method and system construction of a multi-standard palletizing robot based on digital twins according to the present invention;

[0033] Figure 2 URDF model and 3D geometry model of the robot;

[0034] Figure 3 It is a diagram of the Socket bidirectional interaction mechanism of the present invention;

[0035] Figure 4 This is a training framework diagram of the path planning reinforcement learning algorithm of the present invention;

[0036] Figure 5 A training graph for the path planning reinforcement learning algorithm of the present invention; Figure 6 Run monitoring graph for the robotic palletizing task. DETAILED DESCRIPTION

[0037] The invention is further described below in conjunction with the accompanying drawings and embodiments, but the content of the invention is not limited to the scope described. Embodiment 1: This embodiment mainly uses a six-axis palletizing robot in a packaging production line as an application object for on-site verification. The palletizing module in the packaging production line is mainly composed of a palletizing robot, a pallet, a material collection table, a product, a peripheral packaging mechanical structure, a control cabinet, etc. The production line automation signal is controlled by the Siemens S71500 series PLC in the control box. The PLC can read all equipment automation signals and related data including robots, packaging mechanical structure status, sensors, etc.

[0038] This embodiment provides a multi-standard palletizing robot autonomous palletizing method and system based on digital twin, as shown in the attached Figure 1 As shown, the following steps are included:

[0039] The first step is to build a digital twin scene for robot path planning based on the robot processing production line environment.

[0040] The Unified Robot Description Format (URDF) model is used to describe the links and joints of the robot arm and their relative positions. Then, the shapes and textures are added to the URD model through the OpenGL library and displayed in Unity3D. The URDF model and 3D geometric model of the 6-DOF ABB robot used in this system are shown in the attached figure. Figure 2 As shown, in order to simplify the model, the coordinates of the base joints are set to world coordinates.

[0041] As there is a lack of corresponding models for surrounding obstacles and environment, in order to reconstruct accurate and realistic solid geometric models, multiple sensors are used for perception. The EinScan HX&H scanner adopts a hybrid scanning mode of structured light and laser scanning to obtain high-quality 3D data. The accuracy of 0.04mm can be achieved in laser mode to construct geometric models of surrounding obstacles and environment, so as to build a digital model of physical entities.

[0042] In this way, physical property information such as size information, position and posture information, kinematic relationship information and appearance information of the palletizing robot body, end effector, sensor, product, surrounding obstacles and environment can be obtained.

[0043] The second step is to collect the physical entity data in the real space, such as the operation data of the palletizing robot, product parameters, obstacle information, etc., in real time, and build a real-time data transmission channel;

[0044] In order to make the robot system bidirectionally controllable, the communication between the real space and the virtual space uses Socket data transmission technology for bidirectional data transmission, realizing data interconnection and intercommunication during the robot's autonomous palletizing process. After obtaining the robot and production line related data in KEPServerEX 6 in real time through Python, a Socket server is built on the Python side, and a Socket client is built on the Unity 3D side to perform bidirectional data transmission through IP addresses and port numbers. The Socket bidirectional interaction mechanism is shown in the attached figure. Figure 3 shown.

[0045] The third step is to build a data storage layer to store physical space data and virtual space data, and use the data to drive the real space and virtual space entities accordingly, so that the robot system has bidirectional controllability;

[0046] Add ODBC driver: In the node configuration of the OPCUA (Open Platform Communications Unified Architecture) server, you first need to add the ODBC (Open Database Connectivity) driver. The ODBC driver is used to connect to the database and data source, allowing data exchange between different applications. Select the MySQL ODBC Connector driver version suitable for the target database, and make sure to install the latest driver version to ensure compatibility and performance. After installing and registering the ODBC driver, set the database connection information through the ODBC manager or system configuration interface, including key parameters such as the database IP address, port number, user name, password, etc.

[0047] Configure Data Logger: DataLogger is a data logging tool used to capture real-time data from a physical or virtual environment and store it in a database. When configuring DataLogger, you need to connect it to the OPCUA server to obtain real-time operation data of the palletizing robot. In the configuration interface, specify the frequency and data items of data collection. This data will be obtained from the OPCUA node regularly and transferred to the database through the ODBC driver. Through the configuration of the log group (LogGroup), create multiple log groups for different types of data and define the data storage method for each group. Record the robot's path points, product information, environmental data, etc. separately.

[0048] Connect to MySQL database: After configuration, DataLogger connects to MySQL database via ODBC driver. Create relevant data table structure in MySQL database to ensure that it can accommodate the data transmitted from OPCUA server. To ensure the reliability of data transmission, set connection parameters, including timeout, reconnection mechanism, etc. After configuration, DataLogger automatically stores the collected data in the table of MySQL database. The data table usually includes the following fields: timestamp, robot status data, sensor feedback data, path planning parameters.

[0049] The fourth step is to read the initial values ​​from the system input module to obtain the product specifications and palletizing methods, initialize the corresponding variables in the program, calculate the pallet layout, and determine the task point pose parameters in the palletizing task. Assuming that the robot base coordinate system coincides with the world coordinate system, the pose of the pallet base coordinate system OT-xyz relative to the robot base coordinate system OR-xyz can be expressed as R T T , where R is the symbol of the robot base coordinate system. After inputting the palletizing method and product specification information, the position and posture of the material's local coordinate system relative to the robot base coordinate system are output through the following formula: R T M , that is, the position and posture parameters of the task points in the palletizing task are determined.

[0050] R T M = R T T · T T M

[0051] Verify the robot's tool position. If it is not at the starting position, move the tool to the starting position. Then execute the main loop to verify whether the pallet and object are in place. If they are in place, move the robot's gripper to the pick approach point and follow the planned path; go to the final pick point; if an error occurs during the picking process, the loop is aborted; check whether the loop should terminate. If terminated, move the robot to the starting position.

[0052] The fifth step is to use the reinforcement learning algorithm to train different palletizing task points and surrounding obstacle environment information on the twin platform to obtain more training experience, and then use the trained network model to calculate the theoretical optimal palletizing path;

[0053] By combining direct modeling with general model files and modeling by scanning surrounding fuzzy obstacles, the twin model of the palletizing robot operation scene is established by keeping consistent with the physical property information such as size information, position and posture information, kinematic relationship information and appearance information in the physical environment.

[0054] State space design: Given the target position of the robot, its state vector s is defined as:

[0055] s=(x,y,z,x g ,y g ,z g ,t)

[0056] Among them, x, y, and z represent the three-dimensional position coordinates of the robot. g ,y g 、z g It represents the relative position coordinates of the robot relative to the target position, and t represents the status information of whether the robot has reached the target point.

[0057] The action space defines the operations that the robot can perform in each state. In 3D path planning, the action space can be discrete or continuous. The robot can perform discrete movement actions and move along the x, y, and z axes. The action set is as follows:

[0058] A={move forward, move backward, move left, move right, move up, move down}

[0059] The idea of ​​artificial potential field is applied to the design of reward function for reinforcement learning. By considering the target point as an "attraction field" and the obstacle as a "repulsion field", the robot is guided to move along the path with the least force. In order to combine reinforcement learning, the attraction and repulsion of the artificial potential field are converted into rewards. The following reward function can be designed:

[0060] The potential function of attraction is usually designed to be proportional to the quadratic distance to the target and is defined as follows:

[0061]

[0062] Among them: U att (s) is the attractive potential energy function, which represents the attractive potential energy in state s. att is the attraction coefficient, which controls the size of the attraction. d(s,g) is the Euclidean distance between the robot's current position s and the target point g.

[0063] Convert the attractive potential field into a reward:

[0064]

[0065] The farther the robot is from the target, the lower the reward; the closer it is to the target, the higher the reward. When it reaches the target point, the distance is 0 and the maximum reward is obtained.

[0066] Repulsion is used to prevent the robot from colliding with obstacles. When the robot approaches an obstacle, the repulsion increases rapidly, guiding the robot away from the obstacle. The repulsion potential energy is usually designed to be proportional to the inverse of the distance to the obstacle, and only takes effect when the robot is within a certain distance threshold d0:

[0067]

[0068] Among them: U rep (s) is the repulsive potential energy function, which represents the repulsive potential energy of state s. rep is the repulsion coefficient, which controls the strength of the repulsion. d(o,g) is the distance between the robot's current position s and the nearest obstacle o. d0 is the threshold of the repulsion. When the robot is farther away from the obstacle than this threshold, no repulsion is generated.

[0069] The repulsive field is converted into rewards:

[0070] R rep (s)=-U rep (s)

[0071] When the robot approaches an obstacle, the repulsion increases and the reward decreases, encouraging the robot to stay away from the obstacle. When the robot is far away from the obstacle, the repulsion is 0 and no penalty is incurred.

[0072] In order to obtain the final reward function, we can combine attraction and repulsion to design the following reward function:

[0073]

[0074] Second, set the goal-reaching reward: if the robot reaches the goal point, give a large positive reward:

[0075] R(s)=+100, if reaching the target point

[0076] Set the collision penalty: If the robot collides with an obstacle, give a large negative reward:

[0077] R(s)=-100, if collision with obstacles

[0078] Set time penalty: To prevent the robot from wandering around ineffectively in the environment, a small time penalty can be given for each action performed:

[0079] R(s)=-0.1, if the target point is reached

[0080] Combining all the elements together, we get the complete reward function design:

[0081]

[0082] Interaction and data collection in virtual environment: Interaction between the agent and the digital twin: The agent performs actions in the digital twin and observes the results. The digital twin simulates the real physical environment and generates virtual interaction data. These data include: the current state of the agent, the actions performed by the agent, the rewards given by the environment, and the new state after the action is performed; Data collection: The interaction data between the agent and the virtual environment is stored in the experience playback buffer. The agent randomly extracts data from the buffer for training, breaking the time correlation between the data and improving the stability of the training.

[0083] Deep neural network training: Using the twin platform combined with the Dueling DQN algorithm for intensive training, the Q value is first decomposed into the state value function V(s) and the advantage function A(s, a). The state value function V(s) represents the overall return of all actions in state s, and the advantage function A(s, a) represents the superiority of a specific action over other actions in state s. The Q value calculation formula is as follows:

[0084]

[0085] Where V(s) represents the state value function; A(s,a) represents the advantage function, which evaluates the superiority of an action relative to other actions; Α represents the size of the action space.

[0086] The update formula is as follows:

[0087]

[0088] In the formula, Q network (s,a) is the Q value approximated by a deep neural network; Q target (s′, a′) is the Q value calculated using the target network, which is used to stabilize the update.

[0089] Strategy update in virtual environment: In each round of training, the agent performs actions according to the current strategy and obtains feedback (rewards), and then updates the strategy so that the agent can make better decisions in the future state. During the training process, the agent selects actions according to the ∈-Greedy exploration strategy. The agent selects random actions for exploration with probability ∈ and selects the current optimal action for use with probability 1-∈. The value of ∈ is high at the beginning, and ∈ gradually decreases as training progresses, gradually reducing the exploration behavior.

[0090] Strategy verification and evaluation: In order to evaluate the effectiveness of training, the performance of the agent is tested in a virtual environment to verify whether it can successfully complete the palletizing task. Commonly used performance evaluation indicators include cumulative rewards, task completion time, number of collisions, etc.

[0091] Application and optimization in physical systems: When the training of the agent in the digital twin achieves the expected results, the trained control strategy can be applied to the physical system. Through real-time interaction with the physical system, the agent can perform optimized actions to complete the actual task. When the physical environment changes, the digital twin system can obtain the change data of the physical system in real time, transmit it to the virtual environment, and continue to train deep reinforcement learning based on the new data. This online update and adaptive training can enable the agent to cope with the changing environment and achieve continuous optimization.

[0092] The sixth step is to transmit the planning result data to the physical palletizing robot to complete the teaching point update, and monitor the operation of the physical robot based on the virtual-physical data bidirectional transmission technology and the virtual-real synchronization technology.

[0093] The robot twin model is driven by the real-time operation data of the physical robot to achieve synchronous operation of the robot twin model and the physical robot, and synchronously complete the entire processing operation. The consistency of the robot twin model operation process and the physical robot operation process is verified. If the robot twin model operation process is inconsistent with the physical robot, adjustments are made until they are consistent.

[0094] Based on the path planning simulation results obtained by the robot path planning reinforcement learning algorithm, data segmentation is performed with each mechanical joint angle as a group. The segmented path data is transmitted to the physical robot and bound to the relevant variables of the robot motion control module to complete the teaching point update; after the teaching point update is completed, the collision volume of the robot twin model and the relevant parts of the obstacle three-dimensional model is appropriately expanded, and the physical robot is started to run according to the latest teaching point. The robot twin model is driven to run based on the real-time operation data of the physical robot, and the physical robot is monitored through the synchronous operation of the robot twin model and the physical robot. The robot twin model is used for collision detection. When the robot twin model detects a collision, it immediately sends an emergency stop signal to the physical robot for emergency stop to prevent collision accidents.

[0095] The results show that the autonomous palletizing method and system of the multi-standard palletizing robot based on digital twin of the present invention can adapt to the calculation and acquisition of palletizing task points of various product specifications and palletizing methods, and conduct deep reinforcement learning training based on the digital twin platform, transmit the path planning results to the physical robot for teaching point update, and monitor the physical robot palletizing task based on the virtual-physical data bidirectional transmission technology and the virtual-real synchronization technology, as shown in the attached figure. Figure 6 After the path planning result of the Dueling DQN deep reinforcement learning algorithm is transmitted to the physical robot to complete the teaching point update, the robot twin model and the physical robot run synchronously with the same posture and no collision occurs, which provides a new solution for the physical robot teaching point update and operation monitoring.

[0096] Corresponding to the above-mentioned multi-standard palletizing robot autonomous palletizing method and system based on digital twin, this embodiment also provides a robot arm path planning simulation analysis system based on digital twin, including:

[0097] (1) Twin scene construction module: Build a twin model of the palletizing robot operation scene, and build a digital twin scene based on the products, obstacles and surrounding environment in the palletizing robot's palletizing operation production line;

[0098] (2) Data interaction module: Socket data transmission technology is used for bidirectional data transmission. After obtaining the real-time data of the palletizing robot and production line in KEPServerEX 6 through Python, a Socket server is built on the Python side and a Socket client is built on the Unity 3D side. Bidirectional data transmission is performed through IP addresses and port numbers. In this way, the physical entity data in the real space, such as the operation data of the palletizing robot, product parameters, and obstacle information, is collected in real time, and a real-time data transmission channel is built.

[0099] (3) Data storage module: Build a data storage layer, add an ODBC driver to the OPCUA server node, configure the Data Logger to connect to MySQL, and store physical space data and virtual space data. Use the data to drive the real space and virtual space entities, so that the robot system has bidirectional controllability.

[0100] (4) Palletizing task point adaptive adjustment module: reads the initial value from the system input module to obtain the product specifications and palletizing methods, generates a main loop, and then the palletizing robot executes the main loop to complete the autonomous palletizing task, and makes adaptive adjustments to the palletizing task points according to the planning requirements of multiple palletizing methods, multiple varieties, and multiple product specifications;

[0101] (5) Robot path planning reinforcement learning module: Use deep reinforcement learning algorithms to train different palletizing task points and surrounding obstacle environment information in the twin platform to obtain more training experience, and then use the trained network model to calculate the theoretical optimal palletizing path;

[0102] (6) Teaching point update and physical robot operation monitoring module: The obtained path planning simulation results are transmitted to the physical robot to update the teaching points. The physical robot operation is monitored through the synchronous operation of the robot twin model and the physical robot. When the robot twin model detects a collision, it immediately feeds back an emergency stop signal to stop the physical robot urgently, thereby preventing the physical robot from colliding due to setting errors.

[0103] The above are only preferred 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 specification and drawings of the present invention are also included in the patent protection scope of the present invention.

Claims

1. An autonomous palletizing method and system for multi-standard palletizing robots based on digital twins, characterized in that: The following steps are involved: (1) Construct a twin model of the palletizing robot operation scene, and build a digital twin scene based on the products, obstacles, and surrounding environment in the palletizing robot's palletizing operation production line; (2) Real-time collection of physical entity data in real space, such as palletizing robot operation data, product parameters, and obstacle information, to build a real-time data transmission channel; (3) Construct a data storage layer to store physical space data and virtual space data, and use the data to drive the real space and virtual space entities accordingly, so that the robot system has bidirectional controllability; (4) Make adaptive adjustments to the palletizing task points based on the planning requirements of various palletizing methods, multiple varieties, and multiple product specifications; (5) Use deep reinforcement learning algorithms to train different palletizing task points and surrounding obstacle environment information in the twin platform to obtain more training experience, and then use the trained network model to calculate the theoretical optimal palletizing path; (6) The planning result data is transmitted to the physical palletizing robot to complete the teaching point update, and the operation of the physical robot is monitored based on the virtual-physical data bidirectional transmission technology and the virtual-real synchronization technology to reduce the risk of collision.

2. The autonomous palletizing method and system of multi-standard palletizing robots based on digital twins according to claim 1 is characterized in that: The system refers to a robot path planning simulation analysis system based on digital twins, including: (1) Twin scene construction module: Build a twin model of the palletizing robot operation scene, and build a digital twin scene based on the products, obstacles and surrounding environment in the palletizing robot's palletizing operation production line; (2) Data interaction module: real-time collection of physical entity data in real space such as palletizing robot operation data, product parameters, obstacle information, etc., and construction of real-time data transmission channel; (3) Data storage module: Build a data storage layer to store physical space data and virtual space data, and use the data to drive the real space and virtual space entities accordingly, so that the robot system has bidirectional controllability; (4) Palletizing task point adaptive adjustment module: Make adaptive adjustments to palletizing task points according to the planning requirements of various palletizing methods, various varieties, and various product specifications; (5) Robot path planning reinforcement learning module: Use deep reinforcement learning algorithms to train different palletizing task points and surrounding obstacle environment information in the twin platform to obtain more training experience, and then use the trained network model to calculate the theoretical optimal palletizing path; (6) Teaching point update and physical robot operation monitoring module: The obtained path planning simulation results are transmitted to the physical robot to update the teaching points. The physical robot operation is monitored through the synchronous operation of the robot twin model and the physical robot. When the robot twin model detects a collision, it immediately feeds back an emergency stop signal to stop the physical robot urgently, thereby preventing the physical robot from colliding due to setting errors.

3. The autonomous palletizing method and system of multi-standard palletizing robots based on digital twins according to claim 1 is characterized in that: The twin model of the stacking robot operation scene described in step (1) is constructed by combining direct modeling of a general model file with surrounding fuzzy obstacle scanning modeling.

4. The autonomous palletizing method and system of multi-standard palletizing robots based on digital twins according to claim 1 is characterized in that: The acquisition method in step (2) is Socket data transmission technology, which performs bidirectional data transmission to achieve data interconnection and intercommunication during the robot's autonomous palletizing process.

5. The autonomous palletizing method and system of multi-standard palletizing robots based on digital twins according to claim 1 is characterized in that: The physical space data in step (3) includes physical entity data such as robot operation data, product parameters, obstacle information, etc. in the real space, and the virtual space data includes virtual model simulation data such as robot path point data, motion parameters, etc. obtained by simulation analysis in the virtual space.

6. The autonomous palletizing method and system of multi-standard palletizing robots based on digital twins according to claim 1 is characterized in that: The adaptive adjustment method described in step (4) is to read the initial value from the system input module to obtain the product specifications and the palletizing method, generate a main loop, and then the palletizing robot executes the main loop to complete the autonomous palletizing task.

7. The autonomous palletizing method and system of multi-standard palletizing robots based on digital twins according to claim 1 is characterized in that: As described in step (5), a digital twin training process based on deep reinforcement learning is proposed, and then the trained network model is used to calculate the theoretical optimal palletizing path.

8. The autonomous palletizing method and system of multi-standard palletizing robots based on digital twins according to claim 6 is characterized in that: The digital twin training process based on deep reinforcement learning includes the establishment of a twin model of the palletizing robot operation scenario, agent initialization, interaction and data collection in the virtual environment, deep neural network training, strategy update in the virtual environment, strategy verification and evaluation, and application and optimization in the physical system.

9. The autonomous palletizing method and system of multi-standard palletizing robots based on digital twins according to claim 1 is characterized in that: As described in step (6), the robot twin model is driven to run based on the real-time operation data of the physical robot, so as to realize the synchronous operation of the robot twin model and the physical robot, synchronously complete the whole process of palletizing operation, and update the teaching points and monitor the operation of the physical robot.

Images