Atomic subtask reasoning method and device

Abstract

The invention provides an atomic subtask reasoning method and device. The method comprises the following steps: acquiring a full-field corpus; screening out atomic skill verbs from the full-field corpus, and constructing an atomic skill verb table; performing standardization processing on each atomic skill verb in the atomic skill verb table to generate a general atomic skill set; based on the general atomic skill set, subtask reasoning is carried out on the received to-be-executed task, atomic subtasks are generated, a specific scene and a robot platform can be decoupled, the general atomic skill is constructed, cross-domain and cross-platform action standardized description is achieved, and the generalization ability and semantic uniformity of a robot task planning system are improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to an atomic subtask reasoning method and apparatus. Background Technology

[0002] With the deepening research into large language models and embodied agents, robots' ability to perform multi-step tasks in complex environments has been significantly enhanced. Task planning, as a core component of robot intelligent decision-making, typically relies on the division and invocation of "atomic skills," that is, decomposing high-level tasks layer by layer into basic action units that the robot can execute. This method has become an important foundation for autonomous operation and collaborative control of robots, and has shown broad application prospects in service robots, industrial automation, intelligent warehousing, and many other fields.

[0003] Currently, existing methods for defining atomic skills are generally highly coupled with specific task scenarios and robot hardware. For example, skills such as "opening the refrigerator door" and "picking up a plate" are often defined in a kitchen environment, while operations such as "moving boxes" and "operating a forklift" are more common in a warehouse scenario. Furthermore, for different types of robot platforms—such as single-arm robotic arms, bipedal humanoid robots, or drones—skill libraries that match their structure and execution methods must be constructed separately. While these methods can achieve reliable task execution under specific configurations, they still have significant limitations: First, skill definitions heavily rely on prior scenario knowledge and robot configuration; each expansion to a new environment or platform requires a redesign of the skill set, making it difficult to build a unified skill representation system across scenarios and robots. Second, the tight binding of skills to specific environments hinders the transfer and reuse of skills across different tasks, limiting the overall scalability and adaptability of the system. Third, similar actions are often given different skill names in different contexts, lacking semantic consistency and increasing the complexity of task reasoning and program generation.

[0004] In summary, existing atomic skill definition methods based on scenarios and robot ontology suffer from three major drawbacks: weak universality, poor scalability, and semantic inconsistency. These limitations restrict the flexible deployment and efficient collaboration of intelligent agents in open environments. Summary of the Invention

[0005] One objective of this invention is to provide an atomic subtask reasoning method that decouples specific scenarios from robot platforms, constructs generalized atomic skills, and enables standardized action descriptions across domains and platforms, thereby improving the generalization ability and semantic uniformity of robot task planning systems. Another objective of this invention is to provide an atomic subtask reasoning device. A further objective is to provide a computer-readable medium. A final objective is to provide a computer device.

[0006] To achieve the above objectives, this invention discloses an atomic subtask reasoning method, comprising:

[0007] Obtain a corpus covering all domains;

[0008] Atomic skill verbs were selected from a full-domain corpus and a table of atomic skill verbs was constructed.

[0009] Standardize each atomic skill verb in the atomic skill verb list to generate a universal atomic skill set;

[0010] Based on a general set of atomic skills, subtask reasoning is performed on the received tasks to be executed to generate atomic subtasks.

[0011] Preferably, atomic skill verbs are selected from a full-domain corpus to construct an atomic skill verb table, including:

[0012] Extract all verbs from a comprehensive corpus;

[0013] Atomic skill verbs are selected from the full set of verbs to generate an atomic skill verb list.

[0014] Preferably, atomic skill verbs are selected from the full set of verbs to generate an atomic skill verb list, including:

[0015] By using a pre-built library of non-atomic verb decomposition patterns, rules are matched on each verb in the full set of verbs to filter out atomic verbs;

[0016] Perform similarity matching between atomic verbs and action tags in a pre-built executable action primitive library to filter out executable verbs;

[0017] In a full-domain corpus, frequency statistics were performed on performable verbs according to different domains, and generalized verbs were selected.

[0018] Generalized verbs are identified as atomic skill verbs, and a table of atomic skill verbs is generated.

[0019] Preferably, each atomic skill verb in the atomic skill verb list is standardized to generate a universal atomic skill set, including:

[0020] Add corresponding semantic tags and parameterization interfaces to each atomic skill verb to generate a general atomic skill;

[0021] Generate a set of general atomic skills based on multiple general atomic skills.

[0022] Preferably, based on a general atomic skill set, subtask reasoning is performed on the received task to be executed to generate atomic subtasks, including:

[0023] Natural language processing technology is used to decompose the task to be executed into task structure information;

[0024] Based on a pre-set common sense knowledge base, the missing action reasoning is performed on the task structure information, and the task structure information is supplemented.

[0025] The supplemented task structure information is mapped to the general atomic skills in the general atomic skill set to generate atomic subtasks.

[0026] Preferably, after generating atomic subtasks by performing subtask reasoning on the received task to be executed based on a general atomic skill set, the method further includes:

[0027] Using a pre-defined robot adapter, atomic subtasks are mapped to low-level robot instructions, which then drive the robot to execute those instructions.

[0028] The present invention also discloses an atomic sub-task inference device, comprising:

[0029] The corpus acquisition unit is used to acquire a full-domain corpus.

[0030] The atomic skill verb list construction unit is used to filter atomic skill verbs from a full-domain corpus and construct an atomic skill verb list;

[0031] The standardization processing unit is used to standardize each atomic skill verb in the atomic skill verb list to generate a general atomic skill set;

[0032] The subtask reasoning unit is used to perform subtask reasoning on the received tasks to be executed based on a general set of atomic skills, and generate atomic subtasks.

[0033] The present invention also discloses a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0034] The present invention also discloses a computer device, including a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions, wherein the processor executes the program to implement the method described above.

[0035] The present invention also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the method described above.

[0036] This invention acquires a full-domain corpus; filters atomic skill verbs from the full-domain corpus to construct an atomic skill verb list; standardizes each atomic skill verb in the atomic skill verb list to generate a general atomic skill set; and, based on the general atomic skill set, performs subtask reasoning on the received task to be executed to generate atomic subtasks. This decouples specific scenarios from robot platforms, constructs general atomic skills, and achieves standardized action descriptions across domains and platforms, thereby improving the generalization ability and semantic uniformity of the robot task planning system. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 A flowchart of an atomic subtask reasoning method provided in an embodiment of the present invention;

[0039] Figure 2 A flowchart illustrating another atomic subtask reasoning method provided in this embodiment of the invention;

[0040] Figure 3 This is a schematic diagram of the structure of an atomic subtask inference device provided in an embodiment of the present invention;

[0041] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] To facilitate understanding of the technical solution provided in this application, the relevant content of the technical solution will be explained below. This application is particularly applicable to realizing cross-platform and cross-scenario complex task decomposition and closed-loop control in large language models and embodied intelligent agent systems. The core of this invention lies in proposing an atomic skill modeling system that is highly decoupled from specific robot hardware and task scenarios. Its basic idea is to use the most granular general verbs in human natural language, such as: Grasp, Rotate, Pick, Push, etc., as the semantic description basis of atomic skills, and to parse any high-level task into a sequence of atomic skills composed of these verbs, and to dynamically map them to the low-level control instructions of different robots in the final execution stage.

[0044] This invention proposes using a set of general atomic skills as the modeling foundation for a unified robot action space, and generates instructions for robot-executable atomic subtasks, thereby constructing a standardized, reusable, and scalable task representation architecture. An atomic subtask is the smallest semantically indivisible task unit with standardized parameter interfaces and a clear execution triggering mechanism. Each general atomic skill, such as `Grasp(object)` and `Navigate_to(position)`, has a clear, unambiguous semantic definition and a structured parameter format. These general atomic skills constitute a unified, finite action space. The task planning model only needs to perform subtask reasoning within the set of general atomic skills based on the task to be executed, selecting or generating the next atomic subtask to be executed, greatly reducing reasoning complexity.

[0045] The advantage of this invention is that the input to the task to be performed is modally independent, the construction of the action space does not depend on a specific input form, and the system can receive and process environmental and state information of multiple modalities, including but not limited to natural language instructions, images, point cloud data, structured code descriptions, etc., showing strong compatibility and flexibility.

[0046] At the dynamic execution level, this invention constructs an efficient closed-loop control process of perception-decision-execution-feedback. Specifically, after each atomic subtask is executed, the system collects the latest environmental observations and state updates (such as sensor data, scene images, pose information, etc.), and uses this information to perform atomic subtask inference for the next time step, thereby achieving real-time and adaptive adjustment of the task execution process.

[0047] In summary, this invention constructs a unified framework for robot task planning that is highly general, easily scalable, and has strong practical application capabilities by building a universal atomic skill set and establishing a corresponding modality-independent perception and decision-making closed loop. This framework effectively overcomes the shortcomings of existing technologies where skill definitions are tied to specific scenarios and robots, providing a core technological foundation for truly realizing cross-platform and cross-domain robot intelligence.

[0048] The following uses an atomic subtask inference device as an example to illustrate the implementation process of the atomic subtask inference method provided in this embodiment of the invention. It is understood that the execution entity of the atomic subtask inference method provided in this embodiment of the invention includes, but is not limited to, an atomic subtask inference device.

[0049] Figure 1 A flowchart of an atomic subtask reasoning method provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method includes:

[0050] Step 101: Obtain the full-domain corpus.

[0051] In this embodiment of the invention, the comprehensive corpus can be collected from massive amounts of text data covering multiple fields such as daily operations, industrial processes, and service tasks. Data sources may include, but are not limited to, books, operation manuals, subtitles of instructional videos, web page text, and existing robot instruction datasets. By cleaning, deduplicating, and standardizing these multi-source heterogeneous data, a large-scale, high-quality, and widely covered natural language corpus is formed.

[0052] In this embodiment of the invention, the core function of the full-domain corpus is to ensure that the set of verbs extracted from it can fully represent the actions performed by humans in various environments and tasks, thereby ensuring the comprehensiveness and generalization ability of the constructed skill vocabulary.

[0053] Step 102: Select atomic skill verbs from the full-domain corpus and construct an atomic skill verb table.

[0054] In this embodiment of the invention, based on a full-domain corpus, natural language processing technology is used to extract and filter atomic skill verbs. Part-of-speech tagging and dependency parsing are used to automatically identify and extract all verbs representing physical actions from the full-domain corpus. Then, based on atomicity filtering rules, executability filtering rules, and generalization filtering rules, atomic skill verbs are filtered from the full set of verbs to construct an atomic skill verb table.

[0055] Furthermore, the atomic skill verbs in the atomic skill verb list are deduplicated and categorized to form a standardized atomic skill verb list, which serves as the basis for constructing a unified action space.

[0056] Step 103: Standardize each atomic skill verb in the atomic skill verb table to generate a general atomic skill set.

[0057] In this embodiment of the invention, a standardized semantic tag and parameterized interface are defined for each atomic skill verb in the atomic skill verb list to generate a general atomic skill set.

[0058] Furthermore, the successful execution conditions, possible outcome states, and common failure modes of general atomic skills can be described in a standardized manner. This process ensures that each atomic skill has a clear and unified machine-readable interface, thereby forming a set of general atomic skills that can be consistently understood, invoked, and executed on different software and hardware platforms.

[0059] In this embodiment of the invention, each general atomic skill is an atomic action unit with clear semantics, input parameters and control interface, which can be directly mapped to specific skill calls in the robot control system, forming a finite but complete and scalable set of action semantics.

[0060] Step 104: Based on the general atomic skill set, perform subtask reasoning on the received tasks to be executed to generate atomic subtasks.

[0061] In this embodiment of the invention, the input to the task to be executed can be structured or unstructured data, and the input modality includes, but is not limited to, natural language instructions, images, point cloud data, structured code descriptions, etc. This invention does not limit the input modality, thereby improving compatibility and flexibility, and enhancing the universality of the interface and the portability of the system.

[0062] In this embodiment of the invention, the task to be executed includes information that supports subtask reasoning. The input types include, but are not limited to, task target instructions, environmental state perception information, robot skill capability descriptions, and historical behavior trajectories.

[0063] In this embodiment of the invention, natural language processing technology and a preset common sense knowledge base are used to perform missing action reasoning and supplement task structure information for the task to be performed. Atomic skills are mapped from the general atomic skill set to generate atomic subtasks.

[0064] In this embodiment of the invention, at the output end, it is only necessary to select or generate the optimal next atomic subtask from the general atomic skill set. These atomic subtasks are encapsulated through standardized semantic interfaces, which can directly trigger the robot's underlying control module without additional parsing or translation, possessing inherent executability and system consistency.

[0065] Furthermore, this application supports not only single-step reasoning based on the current input, but also long-term closed-loop feedback control processes. After each subtask is executed, the robot updates the input state based on the latest observations and re-invokes the model to reason about the next subtask, realizing a rolling decision-making process of perception-reasoning-execution-feedback, which is suitable for complex and ever-changing real-world environments.

[0066] The technical solution provided in this invention involves: acquiring a full-domain corpus; selecting atomic skill verbs from the full-domain corpus to construct an atomic skill verb list; standardizing each atomic skill verb in the atomic skill verb list to generate a general atomic skill set; and performing subtask reasoning on the received task to be executed based on the general atomic skill set to generate atomic subtasks. This approach decouples specific scenarios from the robot platform, constructs general atomic skills, and achieves standardized action descriptions across domains and platforms, thereby improving the generalization ability and semantic uniformity of the robot task planning system.

[0067] Figure 2 A flowchart of another atomic subtask reasoning method provided in an embodiment of the present invention is shown below. Figure 2 As shown, the method includes:

[0068] Step 201: Obtain a full-domain corpus.

[0069] In this embodiment of the invention, the comprehensive corpus can be collected from massive amounts of text data covering multiple fields such as daily operations, industrial processes, and service tasks. Data sources may include, but are not limited to, books, operation manuals, subtitles of instructional videos, web page text, and existing robot instruction datasets. By cleaning, deduplicating, and standardizing these multi-source heterogeneous data, a large-scale, high-quality, and widely covered natural language corpus is formed.

[0070] Step 202: Extract all verbs from the full domain corpus.

[0071] In the embodiments of the present invention, through词性标注and dependency syntactic analysis,词性标注is performed on each word in each corpus sentence in the full-domain corpus, and its dependency relationship in the sentence is determined; by traversing the词性of each word, the words with the词性of verb are extracted; a large number of abstract verbs are included in the initially extracted verb words, such as: be, like, think. Based on the dependency relationship and the词性of other words in the sentence, the words representing physical actions are identified from the initially extracted verb words, and the identification logic is: the words representing physical actions dominate a specific object in the sentence; all the identified words representing physical actions are summarized and determined as the full set of verbs.

[0072] Step 203: Screen out the atomic skill verbs from the full set of verbs and generate an atomic skill verb list.

[0073] In the embodiments of the present invention, step 203 specifically includes:

[0074] Step 2031: Through a pre-constructed non-atomic verb decomposition pattern library, perform rule matching on each verb in the full set of verbs to screen out the atomic verbs.

[0075] In the embodiments of the present invention, the non-atomic verb decomposition pattern library defines the mapping relationship from compound actions to atomic actions in the form of rules. The rules adopt the IF-THEN structure. For example: IF verb = assemble, THEN it can be decomposed into: pick up (A), place (B), fasten (C); IF verb = pack, THEN it can be decomposed into: put in, seal.

[0076] Specifically, each verb in the full set of verbs is matched with the rules in the non-atomic verb decomposition pattern library. If a verb can match a certain decomposition rule, it is determined as a non-atomic verb and filtered out; if it cannot match any rule, it is determined as an atomic verb, retained and entered into the next screening step.

[0077] The present invention ensures the indivisibility of skills at the semantic level, that is, atomicity, through the automated method of rule matching.

[0078] Step 2032: Perform similarity matching on the atomic verbs and the action labels in the pre-constructed executable action primitive library to screen out the executable verbs.

[0079] In the embodiments of the present invention, the executable action primitive library pre-stores the action labels that the robot control system can understand and execute, such as: Move, Rotate, Open_Gripper.

[0080] It should be noted that the "词性标注" in the original text seems to be a term that needs to be further defined or replaced with the correct specific content. Here it is directly retained in the translation for the sake of presenting the original structure.Specifically, using a word vector model, the semantic similarity between atomic verbs and all action tags in the executable action primitive library is calculated. If the semantic similarity between an atomic verb and any action tag is greater than a preset similarity threshold, the atomic verb is determined to be executable, and is retained for the next screening stage. If the semantic similarity between an atomic verb and all action tags is less than or equal to the preset similarity threshold, the atomic verb is determined to be non-executable, and is filtered out.

[0081] It is worth noting that the similarity threshold can be set according to actual needs, and this embodiment of the invention does not limit it. As an optional solution, the similarity threshold is 0.7.

[0082] This invention ensures that the selected verbs have a direct or highly approximate mapping at the robot control level, laying a feasible foundation for subsequent skill mapping.

[0083] Step 2033: Perform word frequency statistics on the exercisable verbs in the full-domain corpus according to different domains and screen out the generalized verbs.

[0084] Specifically, in the full-domain corpus, the word frequency of each executable verb is counted according to the pre-divided domains. If the word frequency of an executable verb is higher than the preset word frequency threshold in more than N domains, it indicates that the executable verb has high generalization and is determined to be a generalized verb, and is retained. Conversely, if the word frequency is lower than the preset threshold, it indicates that the executable verb has low generalization and is determined to be a non-generalized verb, and is filtered out.

[0085] In this embodiment of the invention, the value of N and the word frequency threshold can be set according to actual needs, and this embodiment of the invention does not limit them.

[0086] The present invention aims to select verbs that are not limited to specific fields and have strong versatility, thereby ensuring that the constructed skill vocabulary can cover diverse task scenarios.

[0087] Step 2034: Identify generalized verbs as atomic skill verbs and generate an atomic skill verb table.

[0088] In this embodiment of the invention, the set of generalizable verbs that are ultimately retained after being filtered for atomicity, executability, and generalization is determined as atomic skill verbs. Atomic skill verbs have atomicity, executability, and generalization.

[0089] Furthermore, multiple atomic skill verbs are structurally stored as an atomic skill verb table. This atomic skill verb table serves as the core action space defined in this invention, providing a unified semantic foundation for subsequent task planning, reasoning, and skill mapping.

[0090] As an alternative, the atomic skill verb list includes, but is not limited to: Grasp, Twist, Push, Pull, Place, Navigate.

[0091] It should be noted that the atomic skill verb list is open and extensible. With the expansion of the corpus and the adaptation of new robots, new generalized atomic verbs can be continuously incorporated through the above process.

[0092] Step 204: Add corresponding semantic tags and parameterized interfaces to each atomic skill verb to generate general atomic skills.

[0093] In the embodiments of the present invention, each verb in the atomic skill verb list is standardized and encapsulated, and a robot-readable semantic description and a structured parameter list are added to it, thereby generating a complete general atomic skill. Each general atomic skill is an independent and callable module, and its definition includes: a unique skill identifier, usually in English, for internal system calls; a natural language semantic tag for human-computer interaction and interpretation; a parameterized interface that clearly specifies the key parameters required to execute the skill and their data types (such as object, position, direction, etc.). This process transforms the verbs in human language into the basic execution units in robot task planning.

[0094] As an alternative, add the semantic tag: grasp, and the parameterized interface: object, that is, the target object, to the atomic skill verb Grasp. The general atomic skill is: Grasp(object), which means to perform a grasping action on a specific object.

[0095] As an alternative, add the semantic tag: twist, and the parameterized interfaces: object, rotate_direction, that is, the target object, the rotation direction, to the atomic skill verb Twist. The general atomic skill is: Twist(object,rotate_direction), which means to rotate a certain object in a specific direction.

[0096] As an alternative, add the semantic tag: push, and the parameterized interfaces: direction, object, that is, the direction, the target object, to the atomic skill verb Push. The general atomic skill is: Push (direction,object), which means to apply a pushing force to a certain object in a specific direction.

[0097] As an alternative, the atomic skill verb Pull is given a semantic tag: pull, and a parameterized interface: direction, object, meaning direction and target object. The general atomic skill is: Pull(direction, object), which means applying a pulling force to an object in a specific direction.

[0098] As an alternative, a semantic tag, "place," is added to the atomic skill verb "Place," along with parameterized interfaces: "object" and "position," representing the target object and target position, respectively. The general atomic skill is: "Place(object, position)," which means placing the currently grabbed object at the specified position.

[0099] As an alternative, the atomic skill verb Navigate can be given a semantic tag: move / navigate, and a parameterized interface: goal, i.e., spatial target. The general atomic skill is: Navigate(goal), which means moving the robot body to the location indicated by the spatial target.

[0100] Step 205: Generate a set of general atomic skills based on multiple general atomic skills.

[0101] In this embodiment of the invention, all generated general atomic skills are collected, stored, and managed to form a unified and standardized set of general atomic skills. This set of general atomic skills constitutes a finite yet complete action space, serving as the foundation for all higher-level task planning and reasoning. The set of general atomic skills exists in the form of a structured database or configuration file, with each skill as a record containing its complete interface definition and semantic information.

[0102] The design of this invention's universal atomic skill set ensures its ease of expansion and maintenance; new atomic skills can be added through the same standardized process. A single skill definition can cover different robots and scenarios, avoiding redundant development. New tasks only require combining existing verb skills, without needing to add new skill definitions. The same task sequence can be quickly migrated across different robots; only the underlying adapter needs to be replaced.

[0103] Step 206: Decompose the task to be executed into task structure information using natural language processing technology.

[0104] In this embodiment of the invention, the natural language processing technology can be a pre-trained large language model. By utilizing the integrated large language model and combining it with the verb system defined by the general atomic skill set, deep semantic analysis and preliminary decomposition of the task to be performed are carried out, and task structure information is output.

[0105] It's worth noting that task structure information is an intermediate representation, initially outlining the sequence of steps for the task, but may lack complete details or implicit conditions required for execution. This decomposition process consistently uses general verbs as the basic unit, completely avoiding the generation of specific action descriptions bound to a particular robot or scene.

[0106] For example, the task to be performed is: Please unscrew the water bottle on the table and hand it to me. By using the integrated large language model and the set of general atomic skills, the task to be performed is decomposed, and the task structure information obtained is: grab the water bottle, unscrew the water bottle, move it to the owner.

[0107] All task planning in this invention is based on natural language verbs, improving consistency and interpretability across different scenarios. This method achieves systematic standardization of the action space, making the model's inference results both interpretable and able to directly drive robot behavior.

[0108] Step 207: Based on the preset common sense knowledge base, perform missing action reasoning on the task structure information and supplement the task structure information.

[0109] In this embodiment of the invention, the common sense knowledge base includes knowledge about object attributes, preconditions and postconditions for actions, and daily operating conventions. Logical reasoning based on the common sense knowledge base identifies and automatically supplements any missing necessary steps or parameters in the task structure information.

[0110] For example: The task to be performed is: Please unscrew the water bottle on the table and hand it to me; the task structure information is: grab the water bottle, unscrew the water bottle, move to the owner; based on common sense knowledge, it is deduced that unscrewing the water bottle requires first grabbing the bottle body, then unscrewing the cap, and rotating the cap counterclockwise to open it; while handing it to me requires first grabbing the bottle body, then moving it to my vicinity, and placing the water bottle at my location. Therefore, the supplemented task structure information is: grab the water bottle, unscrew the water bottle cap counterclockwise, grab the water bottle, move to the owner, place the water bottle to the owner.

[0111] Step 208: Map the supplemented task structure information to the general atomic skills in the general atomic skill set to generate atomic subtasks.

[0112] In this embodiment of the invention, each action in the supplemented task structure information is mapped to a general atomic skill in the general atomic skill set, and the parameters of the mapped general atomic skill are replaced with the current action to generate atomic subtasks, and finally an executable atomic subtask sequence is generated.

[0113] For example, the supplemented task structure information is: Grab the water bottle, twist the water bottle cap counterclockwise, grab the water bottle, move to the owner, place the water bottle to the owner. Specifically, the "Grab the water bottle" task is mapped to a general atomic skill in the general atomic skill set, mapping out the corresponding general atomic skill "Grasp(object)," and replacing "object" with "water bottle" to obtain the atomic subtask "Grasp(water bottle)". The "Twist the water bottle cap counterclockwise" task is mapped to a general atomic skill in the general atomic skill set, mapping out the corresponding general atomic skill "Twist(object, rotate_direction)," replacing "object" with "water bottle cap" and "rotate_direction" with "counterclockwise" to obtain the atomic subtask "Twist(water bottle cap, counterclockwise)". The "Grab the water bottle" task is mapped to a general atomic skill in the general atomic skill set, mapping out the corresponding general atomic skill "Twist(object, rotate_direction)," and replacing "object" with "water bottle cap" and "rotate_direction" with "counterclockwise" to obtain the atomic subtask "Twist(water bottle cap, counterclockwise)". The general atomic skill `Grasp(object)` is used to generate the atomic subtask `Grasp(Water Bottle)` by replacing `object` with `Water Bottle`. Moving the object to the owner is mapped to a general atomic skill in the set of general atomic skills, resulting in the general atomic skill `Navigate(goal)`. Replacing `goal` with `owner` yields the atomic subtask `Navigate(Owner)`. Placing the water bottle to the owner is mapped to a general atomic skill in the set of general atomic skills, resulting in the general atomic skill `Place(object, position)`. Replacing `object` with `water bottle` and `position` with `owner` yields the atomic subtask `Place(Water Bottle, Owner)`. The final executable sequence of atomic subtasks is: `Grasp(object)`, `Twist(Water Bottle Cap, Counter-clockwise)`, `Grasp(object)`, `Navigate(Owner)`, `Place(Water Bottle, Owner)`.

[0114] Step 209: Using a preset robot adapter, map the atomic subtasks to robot low-level instructions and drive the robot to execute the robot low-level instructions.

[0115] In this embodiment of the invention, the robot adapter maintains a mapping table for each robot platform, translating atomic subtasks into robot-specific, executable low-level control instructions. Different robot models only need to provide their respective adapters to execute the same set of atomic subtasks, thereby achieving complete decoupling between task planning and the robot hardware platform. Finally, the generated low-level instructions are sent to the robot controller to drive the robot to physically execute the entire task.

[0116] In this embodiment of the invention, the robot's behavior is always constrained within a well-defined set of subtasks, ensuring that each action has clear semantics and an execution interface. This mechanism effectively supports multi-step execution, dynamic adaptation, and semantic controllability of tasks, avoiding the uncertainty and non-executability issues of traditional natural language generation.

[0117] This invention encapsulates the action space, input representation, and output format into a standardized interface, possessing high modularity and system compatibility. It can be integrated into existing large models to enhance reasoning controllability and is also suitable for the standardized interface design of novel task planning systems.

[0118] It is worth noting that the acquisition, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. The user information in the embodiments of this application was obtained through legal and compliant means, and the acquisition, storage, use, and processing of user information have been authorized and agreed upon by the client.

[0119] It is worth noting that the information collected in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0120] It is worth noting that the technical solution provided in this application provides users with a corresponding operation entry point, allowing users to choose to agree to or reject the automated decision-making results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0121] The technical solution of the atomic subtask reasoning method provided in this invention involves: acquiring a full-domain corpus; selecting atomic skill verbs from the full-domain corpus to construct an atomic skill verb list; standardizing each atomic skill verb in the atomic skill verb list to generate a general atomic skill set; and performing subtask reasoning on the received task to be executed based on the general atomic skill set to generate atomic subtasks. This method can decouple specific scenarios from robot platforms, construct general atomic skills, achieve cross-domain and cross-platform standardized action descriptions, and improve the generalization ability and semantic uniformity of the robot task planning system.

[0122] Figure 3 This is a schematic diagram of an atomic subtask inference device provided in an embodiment of the present invention. This device is used to execute the aforementioned atomic subtask inference method, such as... Figure 3 As shown, the device includes: a corpus acquisition unit 11, an atomic skill verb table construction unit 12, a standardization processing unit 13, and a subtask reasoning unit 14.

[0123] Corpus acquisition unit 11 is used to acquire a full-domain corpus.

[0124] The atomic skill verb list construction unit 12 is used to filter atomic skill verbs from the full-domain corpus and construct the atomic skill verb list.

[0125] The standardization processing unit 13 is used to standardize each atomic skill verb in the atomic skill verb list to generate a general atomic skill set.

[0126] The subtask reasoning unit 14 is used to perform subtask reasoning on the received task to be executed based on the general atomic skill set, and generate atomic subtasks.

[0127] In this embodiment of the invention, the atomic skill verb table construction unit 12 is specifically used to extract all verbs from the full domain corpus; filter out atomic skill verbs from the full verbs, and generate an atomic skill verb table.

[0128] In this embodiment of the invention, the atomic skill verb table construction unit 12 is specifically used to perform rule matching on each verb in the full set of verbs through a pre-built non-atomic verb decomposition pattern library to filter out atomic verbs; to perform similarity matching between atomic verbs and action tags in a pre-built executable action primitive library to filter out executable verbs; to perform word frequency statistics on executable verbs in the full-domain corpus according to different domains and filter out generalized verbs; to determine the generalized verbs as atomic skill verbs and generate an atomic skill verb table.

[0129] In this embodiment of the invention, the standardization processing unit 13 is specifically used to add corresponding semantic tags and parameterized interfaces to each atomic skill verb to generate a general atomic skill; and to generate a set of general atomic skills based on multiple general atomic skills.

[0130] In this embodiment of the invention, the subtask reasoning unit 14 is specifically used to decompose the task to be executed into task structure information through natural language processing technology; perform missing action reasoning on the task structure information based on a preset common sense knowledge base, and supplement the task structure information; map the supplemented task structure information to the general atomic skills in the general atomic skill set to generate atomic subtasks.

[0131] In this embodiment of the invention, the device further includes an instruction mapping unit 15.

[0132] The instruction mapping unit 15 is used to map atomic subtasks into robot low-level instructions through a preset robot adapter, and drive the robot to execute the robot low-level instructions.

[0133] In the solution of this invention embodiment, a full-domain corpus is obtained; atomic skill verbs are selected from the full-domain corpus to construct an atomic skill verb table; each atomic skill verb in the atomic skill verb table is standardized to generate a general atomic skill set; based on the general atomic skill set, subtask reasoning is performed on the received task to be executed to generate atomic subtasks. This can decouple specific scenarios and robot platforms, construct general atomic skills, realize cross-domain and cross-platform standardized action descriptions, and improve the generalization ability and semantic uniformity of the robot task planning system.

[0134] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer device, specifically, a computer device can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0135] This invention provides a computer device including a memory and a processor. The memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions. When the program instructions are loaded and executed by the processor, they implement the steps of the above-described atomic subtask reasoning method. For a detailed description, please refer to the above-described atomic subtask reasoning method embodiments.

[0136] The following is for reference. Figure 4 It shows a schematic diagram of the structure of a computer device 600 suitable for implementing the embodiments of this application.

[0137] like Figure 4 As shown, the computer device 600 includes a central processing unit (CPU) 601, which can perform various appropriate tasks and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the computer device 600. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0138] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal feedback (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 610 as needed so that computer programs read from it can be installed as needed in storage section 608.

[0139] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program including program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611.

[0140] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0141] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0142] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0143] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0144] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0145] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0146] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0147] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0148] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0149] This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0150] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0151] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An atomic subtask reasoning method, characterized in that, The method includes: Obtain a corpus covering all domains; Atomic skill verbs are selected from the aforementioned full-domain corpus to construct an atomic skill verb table; Each atomic skill verb in the atomic skill verb list is standardized to generate a general atomic skill set; Based on the aforementioned set of general atomic skills, subtask reasoning is performed on the received tasks to be executed to generate atomic subtasks.

2. The atomic subtask reasoning method according to claim 1, characterized in that, The step of filtering atomic skill verbs from the full-domain corpus and constructing an atomic skill verb table includes: Extract all verbs from the aforementioned full-domain corpus; Atomic skill verbs are selected from the full set of verbs to generate the atomic skill verb table.

3. The atomic subtask reasoning method according to claim 2, characterized in that, The step of filtering atomic skill verbs from the full set of verbs to generate the atomic skill verb table includes: By using a pre-built library of non-atomic verb decomposition patterns, each verb in the full set of verbs is matched against rules to filter out atomic verbs; The atomic verbs are matched with the action tags in the pre-built executable action primitive library to filter out executable verbs; In the comprehensive corpus, the frequency of the executable verbs is statistically analyzed according to different domains, and generalized verbs are selected. The generalized verbs are identified as atomic skill verbs, and the atomic skill verb table is generated.

4. The atomic subtask reasoning method according to claim 1, characterized in that, The standardization process for each atomic skill verb in the atomic skill verb list to generate a general atomic skill set includes: Add corresponding semantic tags and parameterized interfaces to each atomic skill verb to generate a general atomic skill; Generate a set of general atomic skills based on multiple general atomic skills.

5. The atomic subtask reasoning method according to claim 1, characterized in that, The step of performing subtask reasoning on the received task to be executed based on the general atomic skill set to generate atomic subtasks includes: The task to be executed is decomposed into task structure information using natural language processing technology; Based on a pre-set common sense knowledge base, the missing action reasoning is performed on the task structure information, and the task structure information is supplemented. The supplemented task structure information is mapped to the general atomic skills in the general atomic skill set to generate the atomic subtask.

6. The atomic subtask reasoning method according to claim 1, characterized in that, After generating atomic subtasks by performing subtask reasoning on the received task to be executed based on the general atomic skill set, the method further includes: The atomic subtasks are mapped to robot low-level instructions through a preset robot adapter, and the robot is driven to execute the robot low-level instructions.

7. An atomic subtask inference device, characterized in that, The device includes: The corpus acquisition unit is used to acquire a full-domain corpus. The atomic skill verb list construction unit is used to filter atomic skill verbs from the full-domain corpus and construct an atomic skill verb list; A standardization processing unit is used to standardize each atomic skill verb in the atomic skill verb list to generate a general atomic skill set; The subtask reasoning unit is used to perform subtask reasoning on the received task to be executed based on the general atomic skill set, and generate atomic subtasks.

8. A computer-readable medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the atomic subtask reasoning method as described in any one of claims 1 to 6.

9. A computer device comprising a memory and a processor, the memory for storing information including program instructions, and the processor for controlling the execution of the program instructions, characterized in that, When the program instructions are loaded and executed by the processor, they implement the atomic subtask reasoning method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the atomic subtask reasoning method according to any one of claims 1 to 6.

Images