Assembly Method, Device, Assembly Robot and Storage Medium of Server Components

Abstract

The present invention discloses an assembly method, device, assembly robot and storage medium for server components, which relates to the technical field of manipulators. By converting the target scenario involved in the assembly task into a semantic map to construct prompt words for action decision-making, and inputting the prompt words into an action decision-making model, the language understanding and reasoning capabilities of the action decision-making model are utilized to output a behavior tree of the assembly robot applicable to the assembly task, so as to achieve flexible control and dynamic adjustment of the robotic arm of the assembly robot in the server component assembly task. Therefore, the technical problem in the related art that the assembly actions based on the preset trajectory planning are difficult to be applied to complex and changeable assembly tasks and difficult to cope with emergencies, thereby affecting the assembly effect, can be solved, and the technical effect of determining the assembly actions according to the actual scene images, so as to realize the adaptive action trajectory decision-making under complex tasks and have higher flexibility on the premise of ensuring the assembly efficiency can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of manipulators, and in particular to an assembly method, device, assembly robot and storage medium for server components. Background Art

[0002] As an important link in server production, the assembly of server components directly affects the overall performance and production cost of the server. In related technologies, automated assembly technology can solve the problems of low work efficiency and large errors in manual assembly and can be used for large-scale production. As the core device of automated assembly technology, the control strategy of the robotic arm is directly related to the assembly efficiency and accuracy. However, in related technologies, most of the robotic arm control strategies are based on preset trajectory planning, and for complex and changeable assembly tasks, their adaptability is poor. When facing unexpected situations, it is difficult to make flexible responses, resulting in a relatively high probability of failures during the assembly process, which urgently needs to be improved. Summary of the Invention

[0003] The present invention provides an assembly method, device, assembly robot and storage medium for server components to at least solve the problem that in related technologies, assembly actions based on preset trajectory planning are difficult to apply to complex and changeable assembly tasks and are difficult to handle unexpected situations, thereby affecting the assembly effect.

[0004] The present invention provides an assembly method for server components, which is applied to an assembly robot. The method includes: obtaining an assembly task of a target server and collecting corresponding target scenario information based on the assembly task; converting the target scenario information into a semantic map to construct at least one prompt word based on the assembly task and the semantic map; inputting the at least one prompt word into a pre-constructed action decision model to output a behavior tree of the assembly robot, and controlling the assembly robot to perform corresponding assembly actions based on the behavior tree until the assembly task is completed.

[0005] The present invention also provides an assembly device for server components, which is applied to an assembly robot. The device includes: a first acquisition module, configured to obtain an assembly task of a target server and collect corresponding target scenario information based on the assembly task; a construction module, configured to convert the target scenario information into a semantic map to construct at least one prompt word based on the assembly task and the semantic map; an assembly module, configured to input the at least one prompt word into a pre-constructed action decision model to output a behavior tree of the assembly robot, and control the assembly robot to perform corresponding assembly actions based on the behavior tree until the assembly task is completed.

[0006] The present invention also provides an assembly server, including: a memory, configured to store a computer program; a processor, configured to implement the steps of any of the above-mentioned assembly methods for server components when executing the computer program.

[0007] The present invention also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the steps of any of the above-mentioned server component assembly methods are implemented.

[0008] The present invention also provides a computer program product including a computer program, wherein when the computer program is executed by a processor, the steps of any of the above-mentioned server component assembly methods are implemented.

[0009] Through the present invention, by transforming the target scenario involved in the assembly task into a semantic map to construct prompt words for action decision-making, and inputting the prompt words into the action decision-making model, the language understanding and reasoning capabilities of the action decision-making model are utilized to output the behavior tree of the assembly robot applicable to the assembly task. Based on the modular and efficient control characteristics of the behavior tree, flexible control and dynamic adjustment of the robotic arm of the assembly robot in the server component assembly task are realized. Therefore, the technical problem in the related art that the assembly actions based on the preset trajectory planning are difficult to be applied to complex and changeable assembly tasks and difficult to cope with emergencies, thereby affecting the assembly effect, can be solved, and the determination of the assembly actions according to the actual scenario images can be achieved, so as to realize the adaptive action trajectory decision-making under complex tasks and achieve a higher flexibility technical effect on the premise of ensuring the assembly efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0011] Figure 1 Schematic diagram of the principle of the server component assembly method provided by an embodiment of the present invention;

[0012] Figure 2 Flowchart of a server component assembly method provided by an embodiment of the present invention;

[0013] Figure 3 Flowchart of a server component assembly method provided by an embodiment of the present invention;

[0014] Figure 4 Schematic diagram of the structure of a server component assembly device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with 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. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0016] It should be noted that in the description of the present invention, the terms "include", "comprise" or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or device including a series of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article or device. The terms "first", "second", etc. in the present invention are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0017] In order to enable those skilled in the art of the present technology to better understand the solution of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0018] In combination with the specific application environment architecture or specific hardware architecture on which the execution of the assembly method of the server component depends, the specific application environment architecture or specific hardware architecture will be described herein.

[0019] Taking the large language model as the basic model of the action decision model as an example, the architecture on which the assembly method of the server component in the embodiment of the present invention depends can be as Figure 1 shown.

[0020] As Figure 1 shown, the architecture on which the embodiment of the present invention depends can use the target scene information collected visually, the task description of the assembly task, and the state of the robotic arm of the assembly robot as data of the large language model to output the behavior tree of the assembly robot, that is, convert the scene information into a semantic map through the visual module, and combine the user instructions as the input of the action decision model to generate a behavior plan in a specific format and convert it into an accurate behavior tree, so as to control the assembly robot to execute corresponding assembly actions based on the behavior tree to complete the assembly task.

[0021] During the assembly process, if unexpected situations occur, such as the overlapping of components used for assembly, making it difficult for the assembly robot to accurately grasp, or the components slipping during assembly, etc., at this time, the embodiments of the present invention can accurately detect and identify the unexpected situations, so as to adjust the behavior tree according to the unexpected situations. That is, the assembly robot executes tasks and performs actions according to the generated behavior tree. At this time, the scene information changes, and the subtree of the behavior tree is dynamically updated using the changed environmental information. In order to cope with environmental changes and operation errors, a change detection algorithm is introduced to update the behavior tree generated by the large model in real time, ensuring that the assembly robot can continuously execute tasks correctly. This strategy improves the flexibility and adaptability of the robotic arm in complex assembly tasks.

[0022] Furthermore, embodiments of the present invention provide an assembly method for server components. Combining with the execution process of the assembly method for server components, the method will be described in detail.

[0023] As Figure 2 shown, embodiments of the present invention may include the following steps:

[0024] In step S201, obtain the assembly task of the target server, and collect the corresponding target scene information based on the assembly task.

[0025] It can be understood that the embodiments of the present invention are applied to an assembly robot, and its purpose is to assemble server components. For this reason, the embodiments of the present invention can first obtain the target server that needs to be assembled and obtain the corresponding assembly task. Among them, the assembly task can be the assembly task of the entire target server or the assembly task of a certain component of the target server. Depending on the different assembly tasks, the target scene information collected by the embodiments of the present invention and the subsequent constructed behavior tree are different.

[0026] Furthermore, if the assembly task is the assembly task of a certain component, then collect the target scene information including the component and the target server, so as to facilitate the assembly robot to accurately obtain the position and state of the component; if the assembly task is the assembly task of the target server, then collect the scene information including all components and the target server.

[0027] Optionally, in an embodiment of the present invention, after obtaining the assembly task of the target server, it further includes: obtaining the current state data and parameter information of the assembly robot; judging whether the assembly robot meets the preset reset condition based on the current state data and parameter information; if it meets the preset reset condition, then reset the assembly robot to execute the assembly task using the reset assembly robot.

[0028] As a possible implementation, embodiments of the present invention can obtain the current state data and parameter information of the assembly robot. Among them, the current state data can be inferred by obtaining the actual operation data of the assembly robot, and the parameter information can be determined according to the current set parameters of the assembly robot.

[0029] Combining the current state data and parameter information, embodiments of the present invention can infer whether there is an ongoing assembly task for the assembly robot, or whether the reset has been completed after the last assembly task is executed, so as to avoid the assembly robot from being in a chaotic state during the execution of the assembly task due to failure to reset in advance, which affects the actual assembly operation.

[0030] For example, embodiments of the present invention can determine that the assembly robot meets the reset condition and then complete the reset of the assembly robot when the current state data is the action execution state and / or the parameter information is not the initial set parameter.

[0031] Optionally, in an embodiment of the present invention, corresponding target scene information is collected based on the assembly task, including: parsing the assembly task to determine the task object, operation action, and task target of the assembly task; collecting the target scene information including the task object to obtain the assembly environment of the assembly robot and the object information and state information of the task object in the assembly environment.

[0032] In some embodiments, the assembly task can be parsed to convert the content and scene information involved in the assembly task into a semantic map in a subsequent process.

[0033] Among them, by parsing the assembly task, the task object (such as the target server or assembly components: CPU, motherboard, etc.), operation action (such as the actions required for assembling the server: pick up, put down, install, plug in, etc.), and task target (such as ensuring correct plugging of the connector) can be determined.

[0034] For example, the assembly task is "Install the CPU on the motherboard and ensure that all connectors are correctly plugged in." At this time, after parsing, embodiments of the present invention can obtain the task objects as the CPU and the motherboard, the operation actions can be pick up, install, and plug in, and the task target can be to correctly install the CPU on the motherboard.

[0035] According to the task target, embodiments of the present invention can obtain the target scene information including the task target, so as to ensure that the state, pose, and other data of the task object can be accurately obtained, so that the assembly robot can accurately assemble the task object.

[0036] In step S202, the target scene information is converted into a semantic map to construct at least one prompt word based on the assembly task and the semantic map.

[0037] It can be understood that a semantic map is a map containing semantic information, not a separate type of map. It can use any type of map such as a geometric map, vector map, road network map, point cloud map, etc. as a carrier and project semantic information onto it. Semantic maps can help robots work according to rules, plan and execute high-level tasks, and communicate with humans at the conceptual level. In addition, semantic maps can also provide richer navigation information to help robots reach their destinations more quickly and safely.

[0038] Converting scene information into a semantic map provides a basis for constructing an initial behavior tree, updating, and expanding the behavior tree for an action decision-making model. Specifically, 3D vision detection technology can be used to process the point cloud data captured by a depth camera, identify, and label objects. The recognition results are saved in an XML file that details information such as the category, three-dimensional coordinates, color, volume, and shape of the objects.

[0039] Combined with the parsing of the assembly task, the embodiments of the present invention obtain target scene information containing task objects. Combining scene semantic perception technology, the embodiments of the present invention can obtain object information and states in the assembly environment of the target server. For example, through a 3D sensor and a vision recognition system, information such as the position of the motherboard, the model of the CPU, and the state of the connector can be obtained, and then used to construct the behavior navigation of the assembly robot.

[0040] For example, taking the example of assembling a CPU onto the motherboard of a target server, the embodiments of the present invention can use sensors such as RGB-D cameras and lidar to obtain 3D point clouds, object poses, textures, etc. of the scene, and synchronously record the real-time pose data of the robotic arm / tools to obtain multimodal data. Detect and capture entities such as the target component (i.e., the CPU) and the assembly reference plane (the target socket of the motherboard), and label attributes (dimensions, materials, assembly directions, etc.) to extract key semantic elements. Then, perform spatial-semantic relationship modeling to achieve hierarchical map construction: at the geometric layer, construct a 3D grid map of the scene and label functional areas such as the assembly table and the material area; at the object layer, map the detected components to nodes in the map and store information such as poses and states (assembled / to be assembled); at the relationship layer, establish topological relationships (such as a certain plug of the CPU to be inserted into a certain socket of the motherboard) and temporal constraints between nodes.

[0041] That is to say, the embodiments of the present invention parse the assembly task and align it with the visual detection results. That is, by combining the assembly task and the target scene information, a corresponding semantic map can be generated, and thus at least one prompt word can be constructed based on the assembly task and the semantic map.

[0042] Optionally, in an embodiment of the present invention, at least one prompt word is constructed based on the assembly task and the semantic map, including: constructing a current available object list based on the object information and status information of the task object; constructing a set of action primitives based on the operation actions; generating at least one prompt word by combining the current available object list, the set of action primitives, the task goal, and a preset example task.

[0043] In the embodiment of the present invention, the current available object list can be determined according to the object information and status information. For example, if the current assembly task is to assemble a target server, the embodiment of the present invention can obtain all the components involved and the tools required for assembling the components, and then construct the current available object list to determine which need to be installed and which need to be used, so as to avoid accidentally mixing in unnecessary components and affecting the assembly effect.

[0044] Furthermore, in the embodiment of the present invention, a corresponding set of action primitives can be constructed according to the operation actions parsed previously, where the actions can be extracted from an action primitive library.

[0045] The action primitive library is a set that defines the basic actions that a robot can perform. These actions are directly mapped to the physical operation capabilities of the robot and provide an action framework required for the action decision model. For example, in the installation of server components, the action primitive library can include actions such as picking up, putting down, pressing, rotating, and moving.

[0046] Filter in the action primitive library based on the operation actions to construct a set of action primitives required to complete the assembly task.

[0047] By combining the current available object list, the set of action sources, the task goal, and the example task, the embodiment of the present invention can construct prompt words for the user to form an assembly robot behavior tree.

[0048] Among them, the example task can show how to combine action primitives, object lists, and external components to execute a specific task. This part is defined by the technical personnel according to the task requirements, aiming to show the conversion process from the task description to the action execution, which shows the conversion process from the task description to the action execution. The task description provides the detailed information and expected results of the task to be executed, clarifies the specific execution goal of the behavior tree generation algorithm, and can be obtained by parsing the assembly task.

[0049] Optionally, in an embodiment of the present invention, constructing a set of action primitives based on the operation actions includes: extracting a plurality of action nodes from a preset action primitive library based on the operation actions; sorting the plurality of action nodes and inserting corresponding condition nodes between any two action nodes to execute the next action node when the condition nodes are satisfied; constructing a set of action primitives by combining the plurality of action nodes and the condition nodes.

[0050] In the actual execution process, embodiments of the present invention can ensure the accuracy of assembly actions by inserting conditional nodes between action nodes, and execute the corresponding action nodes only when the conditional nodes are satisfied. For example, for the task of "installing the CPU on the motherboard and ensuring that all connectors are correctly plugged in", the corresponding process is as follows:

[0051] Step 1. Check whether the position of the motherboard is correct;

[0052] Step 2. Grab the CPU;

[0053] Step 3. Place the CPU into the CPU socket on the motherboard;

[0054] Step 4. Check whether the CPU is correctly installed;

[0055] Step 5. Plug in the connectors.

[0056] Among them, "Check whether the position of the motherboard is correct" and "Check whether the CPU is correctly installed" are conditional nodes. Only when these nodes are executed correctly, the action nodes of "Grab the CPU, Place the CPU into the CPU socket on the motherboard, Plug in the connectors" will be executed.

[0057] Combining action nodes and conditional nodes, embodiments of the present invention can construct a set of action primitives, so that the assembly robot can verify actions during the assembly process and ensure the assembly effect.

[0058] Optionally, in an embodiment of the present invention, after constructing the set of action primitives, it further includes: extracting all conditional nodes from the set of action primitives; obtaining the previous action node before any conditional node and the next action node after any conditional node, and determining whether a preset logical conflict condition is satisfied between any conditional node and the previous action node and the next action node; if the preset logical conflict condition is satisfied, an action conflict reminder is generated.

[0059] In the actual execution process, embodiments of the present invention can determine whether there is a logical conflict between the action nodes before and after the conditional node and the conditional node. For example, if the previous action node is to grab the CPU, the conditional node is to judge whether the CPU has been installed, and the next action node is to install the CPU, there is obviously a logical conflict. At this time, embodiments of the present invention can generate an action conflict reminder to discover action errors, so as to ensure the assembly effect and prevent defective products from the failed assembly from flowing into the next process.

[0060] In step S203, at least one prompt word is input into a pre-constructed action decision model to output a behavior tree of the assembly robot, and the assembly robot is controlled based on the behavior tree to execute corresponding assembly actions until the assembly task is completed.

[0061] Furthermore, the embodiments of the present invention can output a corresponding behavior tree through prompt words and an action decision-making model (such as a large language model) to control the assembly robot to complete the assembly action.

[0062] It can be understood that the powerful natural language understanding and generation capabilities of the large language model provide a new solution idea for the manipulator control strategy. By applying the large language model to the manipulator control, control instructions can be dynamically generated according to task requirements to achieve flexible control of the manipulator. At the same time, the large language model can also adjust the control strategy in real time according to environmental changes to improve the adaptability of the manipulator.

[0063] As an efficient task planning method, the behavior tree has been widely used in control fields such as games and automation systems. The behavior tree constructs a hierarchical task structure to decompose complex tasks into a series of simple subtasks, thereby achieving efficient execution of tasks. Combining the behavior tree with the large language model can generate a behavior tree that meets task requirements, and further realize the intelligent control of the manipulator.

[0064] In the behavior tree, multiple types of nodes can be included: behavior nodes (such as moving, waiting, etc., whose return values include success, failure, and in execution), control nodes (controlling the execution flow of the behavior tree, including sequence nodes, selection nodes, parallel nodes, etc. The sequence node executes the child nodes in sequence and returns failure if a child node fails; the selection node executes the child nodes in turn and returns success as long as one child node is successful; the parallel node executes all child nodes concurrently), condition nodes (judging whether a condition holds, with a return value of true or false, providing a judgment basis for the control node), and decoration nodes (executing specific logic, such as looping to execute child nodes, etc.).

[0065] In the execution process of the behavior tree, the behavior tree starts from the root node and determines the next node to execute according to the logic of the control node, and finally executes to the behavior node to make a behavior decision. During the execution process, the state of the node will be continuously updated, and the parent node determines its own state and subsequent execution process according to the state of the child node.

[0066] Optionally, in an embodiment of the present invention, after controlling the assembly robot to perform the corresponding assembly action based on the behavior tree, it further includes: obtaining the operation state of the manipulator of the assembly robot after completing the assembly action, and obtaining new target scenario information based on the operation state; judging whether the assembly environment meets the preset change condition based on the new target scenario information; if it meets the preset change condition, determining the actual change state of the assembly environment based on the new target scenario information; updating the behavior tree using the actual change state to control the assembly robot using the updated behavior tree until the assembly task is completed.

[0067] It is understandable that unexpected situations may occur during the assembly process. For example, the assembly robot's robotic arm fails to clamp a component, causing the component to fall, or other components are moved due to a large amplitude of movement during the assembly process, resulting in multiple components overlapping. These situations will affect the assembly process.

[0068] The embodiment of the present invention can obtain the information of the new target scene after completing an assembly action, so as to re-confirm the parts according to the operation state, and use the new target scene information to determine whether the assembly environment has undergone abnormal changes, such as overlap, drop, etc. If there is an abnormal change, the actual abnormal change is determined according to the new target scene information, and then the behavior tree is updated. Among them, the operation state of the robot arm can include the state information fed back by the robot arm itself, such as position, speed, torque and other information.

[0069] For example, if the abnormal change is overlapping, then add an action of separating and assembling parts in the behavior tree; for another example, if the abnormal change is falling, then add an action of picking up parts in the behavior tree, etc.

[0070] Through the above scheme, the embodiment of the present application can promptly discover emergencies and handle them in a timely manner to avoid assembly errors or assembly interruptions due to parts not being found, thereby ensuring the smoothness of the movements and the assembly effect during the assembly process.

[0071] Optionally, in one embodiment of the present invention, the actual change state of the assembly environment is determined based on the new target scene information, including: extracting the target component from the target scene information and obtaining the first height information of the target component; tracking the second height information of the target component in the new target scene information, so as to calculate the change rate of the target component by using the first height information, the second height information and the collection time interval between the target scene information and the new target scene information; and using the change rate to determine whether the actual change state is a component sliding state.

[0072] In some embodiments, the present invention can use a 2D visual detection algorithm to compare the position changes of the same components in adjacent images to determine whether the environment has changed. For example, when a component is captured and moved, if the height of the same component is detected to change rapidly, and the rate of change is greater than a set threshold, the surface component is slipping. The formula is defined as follows

[0073]

[0074] in, and Indicates the height of the corresponding components of adjacent images. Represents the time interval for sampling two scene information. If If it is greater than a certain threshold, it indicates that the component has slipped.

[0075] Optionally, in an embodiment of the present invention, determining the actual change state of the assembly environment based on the new target scenario information includes: confirming the first contour feature and the first quantity of the target component in the assembly environment based on the target scenario information; confirming the second contour feature and the second quantity of the target component in the assembly environment based on the new target scenario information; comparing the first quantity and the second quantity to obtain a comparison result; matching the first contour feature and the second contour feature to obtain a matching result; combining the comparison result and the matching result to determine whether the actual change state is a component overlapping state.

[0076] In some other embodiments, it is possible to determine whether there are overlapping components by comparing the acquired image data. For example, in an embodiment of the present invention, the target scenario information can be used as a sample. After the robotic arm moves, new target scenario information is acquired for comparison to determine whether the quantity (subtracting the data taken away during robotic arm assembly) and features can be matched. If they can be matched, it can be determined that there are no overlapping components. If they cannot be matched, abnormal features can be identified as the features of the overlapping part.

[0077] In addition, it can also be identified by calculating the center position.

[0078]

[0079] The formula represents the position deviation of components i and j in the x, y, and z coordinate systems. If the deviation is less than the set threshold, it indicates that the two components overlap. For these situations, the behavior tree needs to be updated.

[0080] Optionally, in an embodiment of the present invention, it further includes: setting at least one detection node during the assembly process of the target server based on the assembly task; predicting the expected assembly state of the target server at at least one detection node using the behavior tree, and when the assembly process of the target server reaches any detection node, acquiring the actual assembly state of the target server at any detection node; comparing the actual assembly state and the expected assembly state corresponding to any detection node to obtain a state comparison result; using the state comparison result to determine whether the target server meets the preset node qualification condition to obtain a node judgment result; generating a corresponding node assembly qualification reminder or node assembly error reminder based on the node judgment result.

[0081] As a possible implementation method, in addition to relying on the conditional nodes in the behavior tree to determine whether the assembly is correct, embodiments of the present invention can also set multiple detection nodes to avoid assembly failures caused by external factors.

[0082] For example, during the assembly process, when other assembly robots are assembling, a component accidentally bounces off and falls into the target server that the current assembly robot is assembling. At this time, it is difficult to detect assembly problems only relying on conditional nodes.

[0083] In an embodiment of the present invention, the actual assembly state can be compared with the expected assembly state through a plurality of detection nodes provided. For example, in the current detection node, the expected assembly state of the target server is that the CPU has been inserted and is waiting to connect the wires, while the actual assembly state is that the CPU has been inserted, but there are other items at the wire insertion position. At this time, it can be determined that the state comparison result is inconsistent, that is, the node qualification condition is not met. Thus, when the node qualification condition is not met, an assembly error reminder is generated to avoid affecting subsequent assembly operations.

[0084] Optionally, in an embodiment of the present invention, after completing the assembly task, it further includes: collecting the actual scenario state information including the target server; extracting at least one item feature from the actual scenario state information, and judging whether the target server meets the preset abnormal assembly condition based on the at least one item feature; if the abnormal assembly condition is met, a corresponding assembly task error reminder is generated.

[0085] During the actual execution process, there may be a situation where a certain component's assembly node has been completed, but the component has not been installed on the target server and remains on the assembly table. For the above situation, in an embodiment of the present invention, after the assembly is completed, the actual scenario state information can be obtained to judge the items in the actual scenario state information to determine whether it is a component required for the assembly of the target server. If the object is a component required for the assembly of the target server and the object is not a spare part, it can be determined that the abnormal assembly condition is met. At this time, an assembly task error reminder can be generated by the embodiment of the present invention so that subsequent technicians can reassemble the target server according to the reminder and the remaining items.

[0086] Optionally, in an embodiment of the present invention, it further includes: running the target server and obtaining the running data of the target server; judging whether the target server meets the preset running expectation condition based on the running data; if the preset running expectation condition is met, a corresponding verification qualified reminder is generated, otherwise, the assembly abnormal node of the target server is evaluated based on the running data, and a corresponding verification failed reminder is generated based on the assembly abnormal node.

[0087] In some embodiments, after assembling the target server, the embodiments of the present invention can perform a running test on the target server to compare the expected running state and the actual running state of the target server, so as to determine whether the actual running state is consistent with the expected running state. If they are consistent, it can be determined that the assembly is correct this time, and a qualified verification reminder can be generated to put the target server into the next process. If they are inconsistent, it can be determined that there is a problem with the assembly this time, and a verification failure reminder can be generated. At the same time, the embodiments of the present invention can also infer the abnormal nodes during the assembly process based on the running data of the target server, and send the information of the abnormal nodes together with the verification failure reminder for subsequent troubleshooting of the target server.

[0088] Optionally, in an embodiment of the present invention, after obtaining the running data of the target server, it further includes: recording the completion duration of the assembly task; calculating the task completion score of the assembly robot by combining the completion duration and the running data; and optimizing the action decision model using the task completion score.

[0089] When starting the assembly, the embodiments of the present invention can record the assembly time and monitor the assembly task to obtain the completion duration of the assembly task of the assembly robot.

[0090] Furthermore, the embodiments of the present invention can construct a multi-dimensional scoring index system. For example, using the standard average duration (which can be set by technicians or obtained from a large amount of experimental data), time weight, benchmark energy consumption (which can be set by technicians or obtained from a large amount of experimental data), energy consumption weight, error coefficient, running result (whether the operation is successful, whether there is an error during the operation), and result weight, construct an evaluation formula, and then calculate the score of the assembly robot for this assembly. Then, optimize the action decision model according to the score, so that the action decision model can be continuously optimized based on the initial large language model to be more suitable for the process of server assembly.

[0091] Combined Figure 3 As shown, the working principle of the server component assembly method of the embodiments of the present invention is elaborated in detail with an embodiment.

[0092] As Figure 3 As shown, the embodiments of the present invention need to obtain a task description through the assembly task, transform the semantic map through the target scenario information, and then generate a prompt word by combining the task description and the semantic map to input it into the action decision model (large language model), and then generate a behavior tree (the behavior tree is composed of action primitives, a list of available objects, example tasks, and task descriptions) to complete the assembly task.

[0093] Among them, the semantic map provides a basis for constructing the initial behavior tree of the action decision model, as well as updating and expanding the behavior tree. Specifically, through 3D vision detection technology, the point cloud data captured by the depth camera can be processed to identify and label objects. The recognition results are saved in an XML file, which details information such as the category, three-dimensional coordinates, color, volume, and shape of the objects.

[0094] The action primitives come from an action primitive library, which is a collection that defines the basic actions that the robot can perform. These actions are directly mapped to the physical operation capabilities of the robot and provide the action framework required for the action decision model to complete tasks. For example, in server component installation, the action primitive library can include actions such as picking up, putting down, pressing, rotating, and moving. Conditional nodes are added to each action primitive, and the corresponding action node is executed only when the conditional nodes are satisfied. For example, "Install the CPU on the motherboard and ensure that all connectors are correctly plugged in."

[0095] Example tasks demonstrate how to combine action primitives, object lists, and external components to perform specific tasks. This part is defined by the user according to task requirements and aims to show the conversion process from task description to action execution.

[0096] Task description provides detailed information about the task to be executed and the expected results, and clarifies the specific execution objectives of the behavior tree generation algorithm.

[0097] Furthermore, by combining the operating states (position, speed) of the robotic arm during the assembly process of the assembly robot and the real-time change detection of the target scene to determine whether the operation is incorrect and environmental changes, the execution order and node parameters of the behavior tree are dynamically adjusted, and the control strategy of the robotic arm is optimized in real time to ensure that the robotic arm can adapt to the changes in the dynamic environment. There are mainly two parts: the semantic map and environmental change detection.

[0098] The transformation of the semantic map can be carried out through visual information and task description in the same way as constructing the behavior tree before.

[0099] In terms of environmental change detection, the embodiments of the present invention can adopt a 2D vision detection algorithm to compare the position changes of the same components in adjacent images to determine whether the environment has changed. For example, during the process of a component being grasped and moved, if it is detected that the position height of the same component changes rapidly, and the change rate is greater than the set threshold, it indicates that the component is slipping. The formula is defined as follows

[0100]

[0101] Among them, and represent the heights of the corresponding components in adjacent pictures, represents the time interval for sampling the information of two scenarios. If If it is greater than a certain threshold, it indicates that the component has slipped.

[0102] In the embodiments of the present invention, recognition can also be performed by calculating the central position.

[0103]

[0104] The formula represents the position deviation of components i and j in the three coordinate systems of x, y, and z. If the deviation is less than the set threshold, it indicates that the two components overlap. For these situations, the behavior tree needs to be updated.

[0105] During the assembly process, in the embodiments of the present invention, the assembly process of the server can also be monitored by setting detection nodes, and after the assembly is completed, the target server can be inspected to determine the correctness of the assembly.

[0106] In summary, the embodiments of the present invention can generate a behavior tree applicable to the server component assembly task through an action decision model constructed based on a large language model, realizing flexible control and dynamic adjustment of the robotic arm. Only the scene information and the state of the robotic arm are required, and the model can autonomously generate the behavior tree without manual intervention. During the execution of the behavior tree, the state of the robotic arm, environmental changes, and operation conditions are monitored in real time, and the execution order and node parameters of the behavior tree are dynamically adjusted to optimize the control strategy of the robotic arm.

[0107] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases, the former is a better implementation method.

[0108] The embodiments of the present invention also provide an assembly device for server components, which is applied to an assembly robot. Among them, device 10 includes: a first acquisition module 100, a construction module 200, and an assembly module 300.

[0109] Specifically, the first acquisition module 100 is used to obtain the assembly task of the target server and collect corresponding target scene information based on the assembly task.

[0110] The construction module 200 is used to convert the target scene information into a semantic map to construct at least one prompt word based on the assembly task and the semantic map.

[0111] The assembly module 300 is used to input at least one prompt word into a pre-constructed action decision model to output the behavior tree of the assembly robot, and control the assembly robot to perform corresponding assembly actions based on the behavior tree until the assembly task is completed.

[0112] Optionally, in an embodiment of the present invention, the first acquisition module 100 includes: a parsing unit and an acquisition unit.

[0113] Among them, the parsing unit is used to parse the assembly task to determine the task object, operation action, and task target of the assembly task.

[0114] The acquisition unit is used to acquire target scene information containing the task object to obtain the assembly environment of the assembly robot and the object information and status information of the task object in the assembly environment.

[0115] Optionally, in an embodiment of the present invention, the assembly device 10 of the server component further includes: an acquisition module, a judgment module, a determination module, and an update module.

[0116] Among them, the acquisition module is used to acquire the operation state of the robotic arm of the assembly robot after completing the assembly action, and acquire new target scene information based on the operation state.

[0117] The judgment module is used to judge whether the assembly environment meets the preset change condition based on the new target scene information.

[0118] The determination module is used to determine the actual change state of the assembly environment based on the new target scene information when the preset change condition is met.

[0119] The update module is used to update the behavior tree with the actual change state to control the assembly robot with the updated behavior tree until the assembly task is completed.

[0120] Optionally, in an embodiment of the present invention, the determination module includes: a first extraction unit, a calculation unit, and a first determination unit.

[0121] Among them, the first extraction unit is used to extract the target component from the target scene information and obtain the first height information of the target component.

[0122] The calculation unit is used to track the second height information of the target component in the new target scene information to calculate the change rate of the target component using the first height information, the second height information, and the acquisition time interval between the target scene information and the new target scene information.

[0123] The first determination unit is used to determine whether the actual change state is a component slipping state using the change rate.

[0124] Optionally, in an embodiment of the present invention, the determination module includes: a first confirmation unit, a second confirmation unit, a comparison unit, a matching unit, and a second determination unit.

[0125] Among them, the first confirmation unit is used to confirm the first contour feature and the first quantity of the target component in the assembly environment based on the target scenario information.

[0126] The second confirmation unit is used to confirm the second contour feature and the second quantity of the target component in the assembly environment based on the new target scenario information.

[0127] The comparison unit is used to compare the first quantity and the second quantity to obtain a comparison result.

[0128] The matching unit is used to match the first contour feature and the second contour feature to obtain a matching result.

[0129] The second determination unit is used to combine the comparison result and the matching result to determine whether the actual change state is the component overlapping state.

[0130] Optionally, in an embodiment of the present invention, the construction module 200 includes: a first construction unit, a second construction unit, and a first generation unit.

[0131] Among them, the first construction unit is used to construct the current available object column based on the object information and status information of the task object.

[0132] The second construction unit is used to construct an action primitive set based on the operation action.

[0133] The first generation unit is used to generate at least one prompt word by combining the current available object column, the action primitive set, the task target, and the preset example task.

[0134] Optionally, in an embodiment of the present invention, the second construction unit includes: an extraction subunit, an insertion subunit, and a construction subunit.

[0135] Among them, the extraction subunit is used to extract multiple action nodes from the preset action primitive library based on the operation action.

[0136] The insertion subunit is used to sort the multiple action nodes and insert corresponding condition nodes between any two action nodes to execute the next action node when the condition node is satisfied.

[0137] The construction subunit is used to construct an action primitive set by combining the multiple action nodes and the condition nodes.

[0138] Optionally, in an embodiment of the present invention, the construction module 200 further includes: a second extraction unit, an acquisition unit, and a second generation unit.

[0139] Among them, the second extraction unit is used to extract all condition nodes from the action primitive set.

[0140] An acquisition unit is configured to acquire the previous action node before any conditional node and the next action node after any conditional node, and determine whether a preset logical conflict condition is satisfied between any conditional node and the previous action node and the next action node.

[0141] A second generation unit is configured to generate an action conflict reminder when the preset logical conflict condition is satisfied.

[0142] Optionally, in an embodiment of the present invention, the assembly device 10 of the server component further includes: a setting module, a prediction module, a comparison module, a first judgment module, and a first reminder module.

[0143] Among them, the setting module is configured to set at least one detection node during the assembly process of the target server based on the assembly task.

[0144] The prediction module is configured to predict the expected assembly state of the target server of the assembly task at at least one detection node by using a behavior tree, and acquire the actual assembly state of the target server at any detection node when the assembly process of the target server reaches any detection node.

[0145] The comparison module is configured to compare the actual assembly state with the expected assembly state corresponding to any detection node to obtain a state comparison result.

[0146] The first judgment module is configured to judge whether the target server meets the preset node qualification condition by using the state comparison result to obtain a node judgment result.

[0147] The first reminder module is configured to generate a corresponding node assembly qualification reminder or node assembly error reminder based on the node judgment result.

[0148] Optionally, in an embodiment of the present invention, the assembly device 10 of the server component further includes: a second acquisition module, a second judgment module, and a second reminder module.

[0149] Among them, the second acquisition module is configured to acquire the actual scene state information including the target server.

[0150] The second judgment module is configured to extract at least one item feature from the actual scene state information and judge whether the target server meets the preset abnormal assembly condition based on the at least one item feature.

[0151] The second reminder module is configured to generate a corresponding assembly task error reminder when the abnormal assembly condition is satisfied.

[0152] Optionally, in an embodiment of the present invention, the assembly device 10 of the server component further includes: an operation module, a third judgment module, and a third reminder module.

[0153] Among them, the running module is used to run the target server and obtain the running data of the target server.

[0154] The third judgment module is used to judge whether the target server meets the preset running expectation conditions based on the running data.

[0155] The third reminder module is used to generate a corresponding qualified verification reminder when the preset running expectation conditions are met; otherwise, it evaluates the assembly abnormal nodes of the target server based on the running data and generates a corresponding failed verification reminder based on the assembly abnormal nodes.

[0156] Optionally, in an embodiment of the present invention, the assembly device 10 of the server component further includes: a recording module, a scoring module, and an optimization module.

[0157] Among them, the recording module is used to record the completion duration of the assembly task.

[0158] The scoring module is used to calculate the task completion score of the assembly robot by combining the completion duration and the running data.

[0159] The optimization module is used to optimize the action decision model by using the task completion score.

[0160] For the description of the features in the corresponding embodiment of the assembly device of the server component, reference can be made to the relevant description in the corresponding embodiment of the assembly method of the server component, which will not be elaborated here one by one.

[0161] An embodiment of the present invention also provides an assembly robot, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any of the above-mentioned embodiments of the assembly method of the server component.

[0162] An embodiment of the present invention also provides a computer-readable storage medium, in which a computer program is stored. The computer program is configured to execute the steps in any of the above-mentioned embodiments of the assembly method of the server component when running.

[0163] In an exemplary embodiment, the above-mentioned computer-readable storage medium may include, but is not limited to: USB flash drives, read-only memories (ROM for short), random access memories (RAM for short), mobile hard disks, magnetic disks, or optical discs and other various media that can store computer programs.

[0164] An embodiment of the present invention also provides a computer program product. The above-mentioned computer program product includes a computer program, and when the computer program is executed by a processor, it implements the steps in any of the above-mentioned embodiments of the assembly method of the server component.

[0165] An embodiment of the present invention further provides another computer program product, including a non-volatile computer-readable storage medium. The non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above-described embodiments of the server component assembly method are implemented.

[0166] Those skilled in the art can further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Skilled professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0167] The above has introduced in detail a server component assembly method, device, assembly robot, and storage medium provided by the present invention. Specific examples are used herein to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention. It should be noted that for those of ordinary skill in the art in the technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A method for assembling a server component, characterized in that, Applied to an assembly robot, wherein the method comprises the following steps: Obtain an assembly task of a target server, and collect corresponding target scenario information based on the assembly task; Convert the target scenario information into a semantic map, and construct at least one prompt word based on the assembly task and the semantic map; Input the at least one prompt word into a pre-constructed action decision model to output a behavior tree of the assembly robot, and control the assembly robot to perform corresponding assembly actions based on the behavior tree, obtain the operation state of the robotic arm of the assembly robot after completing the assembly actions, and obtain new target scenario information based on the operation state. Determine whether the assembly environment meets a preset change condition based on the new target scenario information. If the preset change condition is met, determine the actual change state of the assembly environment based on the new target scenario information, and update the behavior tree using the actual change state to control the assembly robot using the updated behavior tree until the assembly task is completed.

2. The assembling method of the server component according to claim 1, wherein, The collecting corresponding target scenario information based on the assembly task includes: Analyze the assembly task to determine the task object, operation action, and task goal of the assembly task; Collect target scenario information including the task object to obtain the assembly environment of the assembly robot and the object information and state information of the task object in the assembly environment.

3. The assembling method of the server component according to claim 1, wherein, The determining the actual change state of the assembly environment based on the new target scenario information includes: Extract a target component from the target scenario information and obtain the first height information of the target component; Track the second height information of the target component in the new target scenario information to calculate the change rate of the target component using the first height information, the second height information, and the acquisition time interval between the target scenario information and the new target scenario information; Use the change rate to determine whether the actual change state is a component slipping state.

4. The assembling method of the server component according to claim 1, characterized in that, The determining the actual change state of the assembly environment based on the new target scenario information includes: Confirm the first contour feature and the first quantity of the target component in the assembly environment based on the target scenario information; Confirm the second contour feature and the second quantity of the target component in the assembly environment based on the new target scenario information; Compare the first quantity and the second quantity to obtain a comparison result; Match the first contour feature and the second contour feature to obtain a matching result; Combine the comparison result and the matching result to determine whether the actual change state is a component overlapping state.

5. The assembling method of the server component according to claim 2, characterized in that, The constructing at least one prompt word based on the assembly task and the semantic map includes: Construct a current available object list based on the object information and state information of the task object; Construct an action primitive set based on the operation action; Generate the at least one prompt word by combining the current available object list, the action primitive set, the task goal, and a preset example task.

6. The assembling method of the server component according to claim 5, wherein, The constructing an action primitive set based on the operation action includes: Extract multiple action nodes from a preset action primitive library based on the operation actions; Sort the multiple action nodes and insert corresponding condition nodes between any two action nodes to execute the next action node when the condition nodes are satisfied; Construct the action primitive set by combining the multiple action nodes and the condition nodes.

7. The assembling method of the server component according to claim 6, wherein, After constructing the action primitive set, it further includes: Extract all condition nodes from the action primitive set; Obtain the previous action node before any condition node and the next action node after the any condition node, and determine whether the any condition node satisfies a preset logical conflict condition with the previous action node and the next action node; If the preset logical conflict condition is satisfied, generate an action conflict reminder.

8. The assembling method of the server component according to claim 1, wherein, It also includes: Set at least one detection node during the assembly process of the target server based on the assembly task; Use the behavior tree to predict the expected assembly state of the target server at the at least one detection node during the assembly task, and when the assembly process of the target server reaches any detection node, obtain the actual assembly state of the target server at the any detection node; Compare the actual assembly state with the expected assembly state corresponding to the any detection node to obtain a state comparison result; Use the state comparison result to judge whether the target server meets the preset node qualification condition to obtain a node judgment result; Generate a corresponding node assembly qualification reminder or node assembly error reminder based on the node judgment result.

9. The assembly method of the server component according to claim 8, wherein After completing the assembly task, it further includes: Collect the actual scenario state information including the target server; Extract at least one item feature from the actual scenario state information and judge whether the target server meets the preset abnormal assembly condition based on the at least one item feature; If the abnormal assembly condition is satisfied, generate a corresponding assembly task error reminder.

10. The assembling method of the server component according to claim 1, wherein It also includes: Run the target server and obtain the running data of the target server; Judge whether the target server meets the preset running expectation condition based on the running data; If the preset running expectation condition is satisfied, generate a corresponding verification qualified reminder, otherwise, evaluate the assembly abnormal nodes of the target server based on the running data and generate a corresponding verification failure reminder based on the assembly abnormal nodes.

11. The assembling method of the server component according to claim 10, characterized in that, After obtaining the running data of the target server, it further includes: Record the completion duration of the assembly task; Calculate the task completion score of the assembly robot by combining the completion duration and the running data; Optimize the action decision model using the task completion score.

12. An assembling device for server components, characterized in that, Applied to an assembly robot, where the device includes: A first acquisition module for obtaining the assembly task of the target server and collecting corresponding target scenario information based on the assembly task; A construction module for converting the target scenario information into a semantic map to construct at least one prompt word based on the assembly task and the semantic map; An assembly module, configured to input the at least one prompt word into a pre-constructed action decision model to output a behavior tree of the assembly robot, and control the assembly robot to perform corresponding assembly actions based on the behavior tree, obtain the operation state of the robotic arm of the assembly robot after completing the assembly actions, and obtain new target scenario information based on the operation state, determine whether the assembly environment meets a preset change condition based on the new target scenario information, if the preset change condition is met, determine the actual change state of the assembly environment based on the new target scenario information, and update the behavior tree by using the actual change state to control the assembly robot by using the updated behavior tree until the assembly task is completed.

13. An assembly robot, characterized in that, Comprising: A memory, configured to store a computer program; A processor, configured to implement the steps of the assembly method of the server component according to any one of claims 1 to 11 when executing the computer program.

14. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, wherein the computer program, when executed by a processor, implements the steps of the assembly method of the server component according to any one of claims 1 to 11.

Images