A low-code editable task sequence generation system for robots based on large models

Abstract

The present invention provides a robot low-code editable task sequence generation system based on a large model. By collecting data information and task examples of the robot's key capability nodes, and obtaining the correspondence between tasks and optimized behavior trees based on the data information and task examples, a basis is provided for the low-code development of the robot. The correspondence between tasks and optimized behavior trees is learned based on the large model, and the task sequence editing logic is obtained according to the learning results, which provides a basis for the intelligent generation of editable task sequences. Based on the task sequence editing logic, a task sequence that meets the robot's current task is obtained, which realizes the intelligent generation of the robot's task sequence, ensures correctness, breaks through technical barriers, and adjusts the task sequence in real time based on the actual application information of the task sequence, thereby improving the flexibility of the robot in executing tasks.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to a large model-based robot low-code editable task sequence generation system. Background Art

[0002] Demand for intelligent robots is growing across a wide range of fields. They have already penetrated numerous sectors, including industry, healthcare, education, housekeeping, and autonomous driving. As technology matures and costs decrease, intelligent robots are becoming increasingly integrated into our lives, particularly in the service and home services, medical rehabilitation, education, and entertainment sectors, where they will play an increasingly important role.

[0003] The intelligent mobile work of robots benefits from robot programming technology. The current pain points of traditional robot programming are:

[0004] High technical barriers: Traditional robot task configuration relies on professional programming and requires writing complex code logic, which is difficult for non-technical personnel to participate in. As a result, each task change requires the reliance on programmers, which is time-consuming and costly.

[0005] Lack of flexibility: To meet dynamic production needs, traditional code must be modified line by line, lacking intuitive visual adjustment methods and making it difficult to quickly respond to scenario changes;

[0006] High error rate: Manual coding is prone to logical errors and lacks intelligent detection mechanisms. Configuration errors require repeated debugging, affecting production efficiency.

[0007] Therefore, the present invention provides a low-code editable task sequence generation system for robots based on large models. Summary of the Invention

[0008] The present invention provides a large-model-based robot low-code editable task sequence generation system to solve the problems raised in the background technology.

[0009] A large-model-based low-code editable task sequence generation system for robots, including:

[0010] A relationship determination module is used to collect data information and task examples of key capability nodes of the robot, and obtain the corresponding relationship between tasks and optimized behavior trees based on the data information and task examples;

[0011] The logic learning module is used to learn the correspondence between tasks and optimized behavior trees based on the large model, and obtain the task sequence editing logic based on the learning results;

[0012] A sequence determination module is used to obtain a task sequence that satisfies the current task of the robot based on the task sequence editing logic;

[0013] The sequence adjustment module is used to adjust the task sequence in real time based on the actual application information of the task sequence.

[0014] Preferably, the relationship determination module includes:

[0015] an analyzing unit, configured to classify the key capability nodes in the robot according to node types to obtain type capability nodes, and determine execution characteristics of type capability nodes of the same node type based on position analysis of the type capability nodes in the robot;

[0016] An information acquisition unit, configured to acquire attribute information of the key capability node and obtain data information in combination with execution characteristics;

[0017] The example acquisition unit is used to acquire information about the robot's tasks and robot task sequences in historical execution tasks as task examples.

[0018] Preferably, the relationship determination module further includes:

[0019] A behavior tree building unit, configured to build an optimized behavior tree based on the task example and combined with data information;

[0020] The relationship determination unit is used to match tasks with optimized behavior trees to obtain the corresponding relationship between tasks and optimized behavior trees.

[0021] Preferably, the logic learning module includes:

[0022] a feature determination unit, configured to obtain corresponding features between task keywords and node behaviors from the corresponding relationship between the task and the optimized behavior tree;

[0023] The feature learning unit is used to learn all corresponding features based on the large model and obtain the task sequence editing logic based on the learning results.

[0024] Preferably, the sequence determination module includes:

[0025] A task analysis unit, configured to segment and parse the current task based on the large model language rules to obtain multiple subtasks and extract task keywords from each subtask;

[0026] a sequence determination unit, configured to determine the task operation corresponding to each task keyword using the task sequence editing logic, and obtain an initial task sequence based on all task operations;

[0027] The sequence optimization unit is used to optimize the initial task sequence based on the connection between adjacent task operations in the initial task sequence to obtain the final task sequence.

[0028] Preferably, the sequence adjustment module includes:

[0029] Acquire actual application information of the task sequence, and determine whether the actual application information is consistent with the predicted information;

[0030] If so, determining that the task sequence can complete the current task;

[0031] Otherwise, it is determined that an exception occurs in the task sequence when completing the current task, and an exception sequence point is determined. Based on the real-time exception information of the exception sequence point and in combination with the task sequence editing logic, the exception sequence point is adjusted.

[0032] Preferably, the behavior tree building unit includes:

[0033] a parsing unit, configured to parse the task example to obtain a plurality of subtasks, and parse the data information to obtain node information of each key capability node;

[0034] An extraction unit is used to extract target subtasks with the same characteristics from multiple task examples, obtain node behavior data corresponding to the target subtask from the node information, and average the node behavior data to obtain the target node behavior corresponding to the target subtask;

[0035] A classification unit is used to classify the key capability nodes based on the node type to obtain multiple groups of type nodes, and to classify each group of type nodes into type-level nodes based on the position and execution order of the type nodes in the robot motion;

[0036] The generation unit is used to set the type level nodes of the same level to the same level, and generate the level behavior tree at the same level according to the level order and combined with the target node behavior;

[0037] An acquisition unit, used to integrate the hierarchical behavior trees in hierarchical order to obtain an initial behavior tree, and obtain the hierarchical connection features and inter-layer connection features of the initial behavior tree;

[0038] an adjustment unit, configured to perform an execution logic judgment on the inter-layer connection feature, determine that the hierarchical behavior tree is normal if the execution logic is satisfied, and otherwise determine that the hierarchical behavior tree is abnormal, and perform sequence adjustment on the hierarchical behavior tree to obtain a first adjustment result;

[0039] The adjustment unit is further configured to determine smoothness of the hierarchical connection characteristics. If smoothness is satisfied, the connection between the initial behavior tree layers is determined to be normal. Otherwise, the connection between the initial behavior tree layers is determined to be abnormal, and the connection between the initial behavior tree layers is adjusted to obtain a second adjustment result.

[0040] The adjustment unit is further configured to adjust the initial behavior tree based on the first adjustment result and the second adjustment result to obtain an optimized behavior tree.

[0041] Preferably, the adjustment unit includes:

[0042] a first adjustment unit, configured to perform a first adjustment on the initial behavior tree according to a first adjustment result to obtain an intermediate behavior tree;

[0043] The second adjustment unit is configured to perform a second adjustment on the intermediate behavior tree according to the second adjustment result to obtain an optimized behavior tree.

[0044] Preferably, the feature learning unit includes:

[0045] a standardization unit, configured to determine a task keyword feature and a node behavior feature from the corresponding features, and perform feature standardization on the task keyword feature and the node behavior feature to obtain a target keyword feature and a target behavior feature;

[0046] A feature acquisition unit, configured to acquire, based on the corresponding features, a set of behavioral features corresponding to the same target keyword features, and acquire a set of keyword features corresponding to the same target behavior features;

[0047] A model learning unit, configured to learn the corresponding features based on the large model and establish correlation features between keywords and behaviors based on initial learning results;

[0048] A specific learning unit is used to separately learn the behavioral feature set corresponding to the same target keyword feature based on the large model to obtain a second learning result, and to separately learn the keyword feature set corresponding to the same target behavior feature based on the large model to obtain a third learning result;

[0049] a difference determining unit, configured to obtain a first difference between the second learning result and the first learning result, and obtain a second difference between the third learning result and the first learning result;

[0050] A feature correction unit, configured to correct the associated feature based on the first difference and the second difference to obtain a target associated feature;

[0051] The logic establishing unit is used to establish the task sequence editing logic based on the association between the keywords and behaviors included in the target association feature.

[0052] Preferably, the correction unit includes:

[0053] a correlation acquisition unit, based on acquiring a first feature related to the first difference in the associated features and a second feature related to the second difference;

[0054] The correction unit is used to obtain the common features of the first feature and the second feature, correct the common features based on the average difference of the first difference and the second difference, and correct the remaining first features and second features based on the first difference and the second difference to obtain the target associated features.

[0055] Compared with the prior art, the present invention has achieved the following beneficial effects:

[0056] By collecting data information and task examples of the robot's key capability nodes, and obtaining the correspondence between tasks and optimized behavior trees based on the data information and task examples, a foundation is provided for low-code development of robots. The correspondence between tasks and optimized behavior trees is learned based on the large model, and the task sequence editing logic is obtained according to the learning results, providing a basis for the intelligent generation of editable task sequences. Based on the task sequence editing logic, a task sequence that meets the robot's current task is obtained, realizing the intelligent generation of the robot's task sequence, ensuring correctness, breaking through technical barriers, and making real-time adjustments to the task sequence based on the actual application information of the task sequence, thereby improving the flexibility of the robot in executing tasks.

[0057] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in this application document.

[0058] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0060] Figure 1 This is a structural diagram of a large-model-based robot low-code editable task sequence generation system in an embodiment of the present invention;

[0061] Figure 2 is a structural diagram of the logic learning module described in an embodiment of the present invention;

[0062] Figure 3 4 is a structural diagram of the sequence determination module described in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0064] Example 1:

[0065] The embodiment of the present invention provides a low-code editable task sequence generation system for robots based on a large model, such as Figure 1 As shown, including:

[0066] Relationship determination module: used to collect data information and task examples of the robot's key capability nodes, and obtain the corresponding relationship between tasks and optimized behavior trees based on the data information and task examples;

[0067] The logic learning module is used to learn the correspondence between tasks and optimized behavior trees based on the large model, and obtain the task sequence editing logic based on the learning results;

[0068] A sequence determination module is used to obtain a task sequence that satisfies the current task of the robot based on the task sequence editing logic;

[0069] The sequence adjustment module is used to adjust the task sequence in real time based on the actual application information of the task sequence.

[0070] In this embodiment, the data information of the robot's key capability nodes includes interface information, capability description, etc., and the task examples are tasks previously performed by the robot and their corresponding task sequences.

[0071] In this embodiment, the task sequence editing logic includes the association logic between the robot's movement, visual recognition, grasping and other actions and the robot node actions.

[0072] In this embodiment, the task sequence is adjusted in real time based on the actual application information of the task sequence, and the incorrect task sequence is adjusted through feedback and description.

[0073] In this embodiment, the correspondence between tasks and optimized behavior trees emphasizes the overall relationship between tasks and optimized behavior trees, and the task sequence editing logic emphasizes the logic between a robot's action and node execution.

[0074] In this embodiment, a task sequence is a set of steps that the robot needs to perform to complete a task.

[0075] The beneficial effects of the above design scheme are: by collecting data information and task examples of the robot's key capability nodes, and obtaining the correspondence between tasks and optimized behavior trees based on the data information and task examples, a basis is provided for low-code development of robots, and the correspondence between tasks and optimized behavior trees is learned based on a large model. The task sequence editing logic is obtained according to the learning results, which provides a basis for the intelligent generation of editable task sequences, and a task sequence that meets the robot's current task is obtained based on the task sequence editing logic, thereby realizing the intelligent generation of the robot's task sequence, ensuring correctness, breaking through technical barriers, and adjusting the task sequence in real time based on the actual application information of the task sequence, thereby improving the flexibility of the robot in executing tasks.

[0076] Example 2:

[0077] Based on Example 1, this embodiment of the present invention provides a system for generating a low-code editable task sequence for a robot based on a large model, wherein the relationship determination module includes:

[0078] an analyzing unit, configured to classify the key capability nodes in the robot according to node types to obtain type capability nodes, and determine execution characteristics of type capability nodes of the same node type based on position analysis of the type capability nodes in the robot;

[0079] An information acquisition unit, configured to acquire attribute information of the key capability node and obtain data information in combination with execution characteristics;

[0080] The example acquisition unit is used to acquire information about the robot's tasks and robot task sequences in historical execution tasks as task examples.

[0081] In this embodiment, the node types include root nodes, task nodes, control nodes, and decoration nodes.

[0082] In this embodiment, the execution features include, for example, movement, attack, control, and repeat.

[0083] The beneficial effects of the above design scheme are: by classifying the key capability nodes in the robot according to the node type, the type capability nodes are obtained, and based on the position analysis of the type capability nodes in the robot, the execution characteristics of the type capability nodes of the same node type are determined, the attribute information of the key capability nodes is obtained, and the data information is obtained in combination with the execution characteristics, and the information of the robot's tasks in historical execution tasks and the robot task sequence is obtained as task examples, providing an information basis for low-code development based on large models.

[0084] Example 3:

[0085] Based on Example 1, an embodiment of the present invention provides a system for generating a low-code editable task sequence of a robot based on a large model, wherein the relationship determination module further includes:

[0086] A behavior tree building unit, configured to build an optimized behavior tree based on the task example and combined with data information;

[0087] The relationship determination unit is used to match tasks with optimized behavior trees to obtain the corresponding relationship between tasks and optimized behavior trees.

[0088] In this embodiment, the behavior tree manages and executes the robot's key capability nodes to meet task requirements.

[0089] In this embodiment, the optimized behavior tree is the best behavior path to complete the task.

[0090] The beneficial effect of the above design scheme is: by establishing an optimized behavior tree based on the task example and combining data information, matching the task with the optimized behavior tree, and obtaining the corresponding relationship between the task and the optimized behavior tree, a basis is provided for low-code development of robots.

[0091] Example 4:

[0092] Based on Example 1, the present invention provides a low-code editable task sequence generation system for robots based on a large model, such as Figure 2 As shown, the logic learning module includes:

[0093] a feature determination unit, configured to obtain corresponding features between task keywords and node behaviors from the corresponding relationship between the task and the optimized behavior tree;

[0094] The feature learning unit is used to learn all corresponding features based on the large model and obtain the task sequence editing logic based on the learning results.

[0095] In this embodiment, for example, the node behavior corresponding to "taking goods" is grabbing a certain node; the node behavior corresponding to "image acquisition" is taking a visual photo of a certain node.

[0096] The beneficial effect of the above design scheme is: by obtaining the corresponding features between task keywords and node behaviors from the correspondence between the tasks and the optimized behavior tree, learning is performed based on all the corresponding features of the large model, and the task sequence editing logic is obtained according to the learning results, providing a basis for the intelligent generation of editable task sequences.

[0097] Example 5:

[0098] Based on Example 1, the present invention provides a low-code editable task sequence generation system for robots based on a large model, such as Figure 3 As shown, the sequence determination module includes:

[0099] A task analysis unit, configured to segment and parse the current task based on the large model language rules to obtain multiple subtasks and extract task keywords from each subtask;

[0100] a sequence determination unit, configured to determine the task operation corresponding to each task keyword using the task sequence editing logic, and obtain an initial task sequence based on all task operations;

[0101] The sequence optimization unit is used to optimize the initial task sequence based on the connection between adjacent task operations in the initial task sequence to obtain the final task sequence.

[0102] The beneficial effects of the above design scheme are: by segmenting and parsing the current task according to the large model language rules, multiple sub-tasks are obtained, the task keywords in each sub-task are extracted, and the task sequence editing logic is used to determine the task operation corresponding to each task keyword, and the initial task sequence is obtained based on all task operations. The initial task sequence is optimized based on the connection between adjacent task operations in the initial task sequence to obtain the final task sequence, thereby realizing the intelligent generation of robot task sequences, ensuring correctness, and breaking through technical barriers.

[0103] Example 6:

[0104] Based on Example 1, a large model-based low-code editable task sequence generation system for robots, the sequence adjustment module includes:

[0105] Acquire actual application information of the task sequence, and determine whether the actual application information is consistent with the predicted information;

[0106] If so, determining that the task sequence can complete the current task;

[0107] Otherwise, it is determined that an exception occurs in the task sequence when completing the current task, and an exception sequence point is determined. Based on the real-time exception information of the exception sequence point and in combination with the task sequence editing logic, the exception sequence point is adjusted.

[0108] The beneficial effects of the above design scheme are: by obtaining the actual application information of the task sequence, it is determined whether the actual application information is consistent with the predicted information; if so, it is determined that the task sequence can complete the current task; otherwise, it is determined that an abnormality occurs in the task sequence when completing the current task, and the abnormal sequence point is determined. Based on the real-time abnormality information of the abnormal sequence point, the abnormal sequence point is adjusted in combination with the task sequence editing logic to improve the flexibility of the robot in executing tasks.

[0109] Example 7:

[0110] Based on Example 3, this embodiment of the present invention provides a low-code editable task sequence generation system for robots based on a large model, wherein the behavior tree building unit includes:

[0111] a parsing unit, configured to parse the task example to obtain a plurality of subtasks, and parse the data information to obtain node information of each key capability node;

[0112] An extraction unit is used to extract target subtasks with the same characteristics from multiple task examples, obtain node behavior data corresponding to the target subtask from the node information, and average the node behavior data to obtain the target node behavior corresponding to the target subtask;

[0113] A classification unit is used to classify the key capability nodes based on the node type to obtain multiple groups of type nodes, and to classify each group of type nodes into type-level nodes based on the position and execution order of the type nodes in the robot motion;

[0114] The generation unit is used to set the type level nodes of the same level to the same level, and generate the level behavior tree at the same level according to the level order and combined with the target node behavior;

[0115] An acquisition unit, used to integrate the hierarchical behavior trees in hierarchical order to obtain an initial behavior tree, and obtain the hierarchical connection features and inter-layer connection features of the initial behavior tree;

[0116] an adjustment unit, configured to perform an execution logic judgment on the inter-layer connection feature, determine that the hierarchical behavior tree is normal if the execution logic is satisfied, and otherwise determine that the hierarchical behavior tree is abnormal, and perform sequence adjustment on the hierarchical behavior tree to obtain a first adjustment result;

[0117] The adjustment unit is further configured to determine smoothness of the hierarchical connection characteristics. If smoothness is satisfied, the connection between the initial behavior tree layers is determined to be normal. Otherwise, the connection between the initial behavior tree layers is determined to be abnormal, and the connection between the initial behavior tree layers is adjusted to obtain a second adjustment result.

[0118] The adjustment unit is further configured to adjust the initial behavior tree based on the first adjustment result and the second adjustment result to obtain an optimized behavior tree.

[0119] The beneficial effect of the above design is that by performing logical judgment on the inter-layer connection features and smoothness judgment on the hierarchical connection features, the accuracy and feasibility of the resulting behavior tree at the same level and between different levels are guaranteed, and the superiority and accuracy of the resulting optimized behavior tree are guaranteed, providing a foundation for the accurate design of editing logic.

[0120] Example 8:

[0121] Based on Example 7, this embodiment of the present invention provides a system for generating a low-code editable task sequence for a robot based on a large model, wherein the adjustment unit includes:

[0122] a first adjustment unit, configured to perform a first adjustment on the initial behavior tree according to a first adjustment result to obtain an intermediate behavior tree;

[0123] The second adjustment unit is configured to perform a second adjustment on the intermediate behavior tree according to the second adjustment result to obtain an optimized behavior tree.

[0124] The beneficial effect of this design is that by performing a first adjustment on the initial behavior tree based on the first adjustment result, an intermediate behavior tree is obtained. Then, by performing a second adjustment on the intermediate behavior tree based on the second adjustment result, an optimized behavior tree is obtained. This ensures the superiority and accuracy of the resulting optimized behavior tree, providing a foundation for accurate design of editing logic.

[0125] Example 9:

[0126] Based on Example 4, this embodiment of the present invention provides a system for generating a low-code editable task sequence for a robot based on a large model, wherein the feature learning unit includes:

[0127] a standardization unit, configured to determine a task keyword feature and a node behavior feature from the corresponding features, and perform feature standardization on the task keyword feature and the node behavior feature to obtain a target keyword feature and a target behavior feature;

[0128] A feature acquisition unit, configured to acquire, based on the corresponding features, a set of behavioral features corresponding to the same target keyword features, and acquire a set of keyword features corresponding to the same target behavior features;

[0129] A model learning unit, configured to learn the corresponding features based on the large model and establish correlation features between keywords and behaviors based on initial learning results;

[0130] A specific learning unit is used to separately learn the behavioral feature set corresponding to the same target keyword feature based on the large model to obtain a second learning result, and to separately learn the keyword feature set corresponding to the same target behavior feature based on the large model to obtain a third learning result;

[0131] a difference determining unit, configured to obtain a first difference between the second learning result and the first learning result, and obtain a second difference between the third learning result and the first learning result;

[0132] A feature correction unit, configured to correct the associated feature based on the first difference and the second difference to obtain a target associated feature;

[0133] The logic establishing unit is used to establish the task sequence editing logic based on the association between the keywords and behaviors included in the target association feature.

[0134] The beneficial effects of the above design scheme are: initial association features are obtained through overall learning based on the corresponding features of the large model, and then specific learning is performed with keywords and behaviors respectively, and then the initial association results are corrected to ensure the correctness of the obtained target association features, providing a basis for the accurate establishment of the task sequence editing logic.

[0135] Example 10:

[0136] Based on Example 9, an embodiment of the present invention provides a system for generating a low-code editable task sequence of a robot based on a large model, wherein the feature correction unit includes:

[0137] a correlation acquisition unit, based on acquiring a first feature related to the first difference in the associated features and a second feature related to the second difference;

[0138] The correction unit is used to obtain the common features of the first feature and the second feature, correct the common features based on the average difference of the first difference and the second difference, and correct the remaining first features and second features based on the first difference and the second difference to obtain the target associated features.

[0139] In this embodiment, the remaining first features and second features are corrected based on the first difference and the second difference. Specifically, the remaining first features are corrected based on the first difference, and the remaining second features are corrected based on the second difference.

[0140] The beneficial effects of the above design scheme are: by obtaining the first feature related to the first difference in the associated features, obtaining the common features of the first feature and the second feature based on the second feature related to the second difference, correcting the common features based on the average difference of the first difference and the second difference, and correcting the remaining first features and second features based on the first difference and the second difference, obtaining the target associated features, ensuring the correctness of the obtained target associated features, and providing a basis for accurately establishing the task sequence editing logic.

[0141] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of this application document and its equivalents, the present invention is intended to include these modifications and variations.

Claims

1. A low-code editable task sequence generation system for robots based on large models, characterized by: include: The relationship determination module is used to collect data information and task examples of the robot's key capability nodes, and obtain the corresponding relationship between tasks and optimized behavior trees based on the data information and task examples, including: A behavior tree building unit is used to build an optimized behavior tree based on the task example and combined with data information, including: a parsing unit, configured to parse the task example to obtain a plurality of subtasks, and parse the data information to obtain node information of each key capability node; An extraction unit is used to extract target subtasks with the same characteristics from multiple task examples, obtain node behavior data corresponding to the target subtask from the node information, and average the node behavior data to obtain the target node behavior corresponding to the target subtask; A classification unit is used to classify the key capability nodes based on the node type to obtain multiple groups of type nodes, and to classify each group of type nodes into type-level nodes based on the position and execution order of the type nodes in the robot motion; The generation unit is used to set the type level nodes of the same level to the same level, and generate the level behavior tree at the same level according to the level order and combined with the target node behavior; An acquisition unit, used to integrate the hierarchical behavior trees in hierarchical order to obtain an initial behavior tree, and obtain the hierarchical connection features and inter-layer connection features of the initial behavior tree; an adjustment unit, configured to perform an execution logic judgment on the inter-layer connection feature, determine that the hierarchical behavior tree is normal if the execution logic is satisfied, and otherwise determine that the hierarchical behavior tree is abnormal, and perform sequence adjustment on the hierarchical behavior tree to obtain a first adjustment result; The adjustment unit is further configured to determine smoothness of the hierarchical connection characteristics. If smoothness is satisfied, the connection between the initial behavior tree layers is determined to be normal. Otherwise, the connection between the initial behavior tree layers is determined to be abnormal, and the connection between the initial behavior tree layers is adjusted to obtain a second adjustment result. An adjustment unit is further configured to adjust the initial behavior tree based on the first adjustment result and the second adjustment result to obtain an optimized behavior tree; A relationship determination unit is used to match tasks with optimized behavior trees to obtain the corresponding relationship between tasks and optimized behavior trees; The logic learning module is used to learn the correspondence between tasks and optimized behavior trees based on the large model, and obtain the task sequence editing logic based on the learning results; A sequence determination module is used to obtain a task sequence that satisfies the current task of the robot based on the task sequence editing logic; The sequence adjustment module is used to adjust the task sequence in real time based on the actual application information of the task sequence.

2. A large model-based robot low-code editable task sequence generation system according to claim 1, characterized in that: The relationship determination module includes: an analyzing unit, configured to classify the key capability nodes in the robot according to node types to obtain type capability nodes, and determine execution characteristics of type capability nodes of the same node type based on position analysis of the type capability nodes in the robot; An information acquisition unit, configured to acquire attribute information of the key capability node and obtain data information in combination with execution characteristics; The example acquisition unit is used to acquire information about the robot's tasks and robot task sequences in historical execution tasks as task examples.

3. A large model-based robot low-code editable task sequence generation system according to claim 1, characterized in that: The logic learning module includes: a feature determination unit, configured to obtain corresponding features between task keywords and node behaviors from the corresponding relationship between the task and the optimized behavior tree; The feature learning unit is used to learn all corresponding features based on the large model and obtain the task sequence editing logic based on the learning results.

4. A large model-based robot low-code editable task sequence generation system according to claim 1, characterized in that: The sequence determination module comprises: A task analysis unit, configured to segment and parse the current task based on the large model language rules to obtain multiple subtasks and extract task keywords from each subtask; a sequence determination unit, configured to determine the task operation corresponding to each task keyword using the task sequence editing logic, and obtain an initial task sequence based on all task operations; The sequence optimization unit is used to optimize the initial task sequence based on the connection between adjacent task operations in the initial task sequence to obtain the final task sequence.

5. A large model-based robot low-code editable task sequence generation system according to claim 1, characterized in that: The sequence adjustment module includes: Acquire actual application information of the task sequence, and determine whether the actual application information is consistent with the predicted information; If so, determining that the task sequence can complete the current task; Otherwise, it is determined that an exception occurs in the task sequence when completing the current task, and an exception sequence point is determined. Based on the real-time exception information of the exception sequence point and in combination with the task sequence editing logic, the exception sequence point is adjusted.

6. A large model-based robot low-code editable task sequence generation system according to claim 1, characterized in that: The adjustment unit includes: a first adjustment unit, configured to perform a first adjustment on the initial behavior tree according to a first adjustment result to obtain an intermediate behavior tree; The second adjustment unit is configured to perform a second adjustment on the intermediate behavior tree according to the second adjustment result to obtain an optimized behavior tree.

7. A large model-based robot low-code editable task sequence generation system according to claim 3, characterized in that: The feature learning unit includes: a standardization unit, configured to determine a task keyword feature and a node behavior feature from the corresponding features, and perform feature standardization on the task keyword feature and the node behavior feature to obtain a target keyword feature and a target behavior feature; A feature acquisition unit, configured to acquire, based on the corresponding features, a set of behavioral features corresponding to the same target keyword features, and acquire a set of keyword features corresponding to the same target behavior features; A model learning unit, configured to learn the corresponding features based on the large model, and establish correlation features between the keywords and the behaviors based on the first learning result; A specific learning unit is used to separately learn the behavioral feature set corresponding to the same target keyword feature based on the large model to obtain a second learning result, and to separately learn the keyword feature set corresponding to the same target behavior feature based on the large model to obtain a third learning result; a difference determining unit, configured to obtain a first difference between the second learning result and the first learning result, and obtain a second difference between the third learning result and the first learning result; A feature correction unit, configured to correct the associated feature based on the first difference and the second difference to obtain a target associated feature; The logic establishing unit is used to establish the task sequence editing logic based on the association between the keywords and behaviors included in the target association feature.

8. A large model-based robot low-code editable task sequence generation system according to claim 7, characterized in that: The feature correction unit includes: a correlation acquisition unit, based on acquiring a first feature related to the first difference in the associated features and a second feature related to the second difference; The correction unit is used to obtain the common features of the first feature and the second feature, correct the common features based on the average difference of the first difference and the second difference, and correct the remaining first features and second features based on the first difference and the second difference to obtain the target associated features.

Images