Task processing method and device and agent with body

Abstract

The invention provides a task processing method and device and an agent with a body, and can be applied to the technical field of artificial intelligence. The method comprises the steps of obtaining current environment information, performing task planning based on the current environment information and a target task, and generating a multi-step action instruction sequence; outputting the multi-step action instruction sequence so as to enable an intelligent agent to execute a first-step action instruction in the multi-step action instruction sequence; the following steps from A to C are repeatedly executed until it is judged that the target task is completed based on the current environment information: step A, in response to a completion signal of receiving the first-step action instruction, the current environment information is obtained again; b, task planning is conducted again based on the current environment information obtained again and the target task, and a new multi-step action instruction sequence is generated; and step C, outputting the new multi-step action instruction sequence, so that the body intelligent agent executes the first-step action instruction in the new multi-step action instruction sequence.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to a task processing method, apparatus, and embodied intelligent agent. Background Technology

[0002] With the rapid development of embodied intelligent agent technology, the autonomous decision-making ability of robots in multi-step, complex task scenarios has received increasing attention. Embodied task planning—that is, translating natural language instructions into a series of executable actions to complete tasks in complex and dynamic environments—is a core technical problem in building intelligent robots. Existing robot control methods typically employ static task planning, where, after acquiring environmental perception, the model outputs a complete sequence of actions at once, which are then executed sequentially. This type of method models the task planning problem as a static prediction task, assuming that the environment remains unchanged or that no dynamic adjustments are needed during task execution. However, real-world environments are often highly dynamic and uncertain, causing static task planning to fail to adjust its execution strategy in a timely manner when faced with environmental changes or perception errors, exhibiting poor robustness and adaptability. Summary of the Invention

[0003] In view of the problems in the prior art, the embodiments of this application provide a task processing method, apparatus and embodied intelligent agent, which can at least partially solve the problems existing in the prior art.

[0004] According to a first aspect of this application, a task processing method is provided, comprising: acquiring current environmental information; performing task planning based on the current environmental information and a target task; generating a multi-step action instruction sequence; outputting the multi-step action instruction sequence to enable an embodied intelligent agent to execute the first action instruction in the multi-step action instruction sequence; and repeating steps A to C until the target task is determined to be completed based on the current environmental information: Step A: In response to receiving a completion signal of the first action instruction, reacquiring the current environmental information; Step B: Re-performing task planning based on the re-acquiring current environmental information and the target task to generate a new multi-step action instruction sequence; Step C: Outputting the new multi-step action instruction sequence to enable the embodied intelligent agent to execute the first action instruction in the new multi-step action instruction sequence.

[0005] In some embodiments, the environmental information is collected by the perception module of the embodied intelligent agent, and the multi-step action instruction sequence is provided to the control module of the embodied intelligent agent. The control module controls the actuator of the embodied intelligent agent to perform the corresponding action according to the first action instruction in the multi-step action instruction sequence.

[0006] In some embodiments, the task planning is implemented through a decision model, which is constructed based on a large language model or a visual language model.

[0007] In some embodiments, each action instruction is obtained by instantiating an atomic skill of the embodied agent.

[0008] In some embodiments, the task planning process of the decision model is as follows: receiving current environmental information and target task; selecting multiple atomic skills from the atomic skill set of the embodied agent based on the current environmental information and target task, and determining the parameters required to execute each atomic skill; combining the selected atomic skills with the corresponding parameters to generate a multi-step action instruction sequence.

[0009] In some embodiments, the action instruction for each step is one of the following: action instruction described in natural language, structured programming instruction, or JSON format instruction.

[0010] According to a second aspect of this application, a task processing apparatus is provided, comprising: an acquisition module for acquiring current environmental information; a task planning module for performing task planning based on the current environmental information and a target task, generating a multi-step action instruction sequence; and an output module for outputting the multi-step action instruction sequence to enable an embodied intelligent agent to execute the first action instruction in the multi-step action instruction sequence. The acquisition module, task planning module, and output module are further configured to repeatedly execute steps A to C until the target task is determined to be completed based on the current environmental information: Step A: In response to receiving a completion signal of the first action instruction, reacquire the current environmental information; Step B: Based on the reacquired current environmental information and the target task, re-perform task planning to generate a new multi-step action instruction sequence; Step C: Output the new multi-step action instruction sequence to enable the embodied intelligent agent to execute the first action instruction in the new multi-step action instruction sequence.

[0011] According to a third aspect of this application, an embodied intelligent agent is provided, comprising: a processor configured to implement the method described in any of the above embodiments; a perception module connected to the processor and configured to collect environmental information; a control module connected to the processor and configured to control the actuator of the embodied intelligent agent to perform a corresponding action according to the first action instruction in the multi-step action instruction sequence; and an actuator connected to the control module and configured to perform a corresponding action under the drive of the control module.

[0012] According to a fourth aspect of this application, an electronic device is provided, including a processor and a memory, wherein the memory stores a computer program, and the processor executes the program to implement the method described in any of the above embodiments.

[0013] According to a fifth aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the methods described in any of the above embodiments.

[0014] According to a sixth aspect of this application, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the method described in any of the above embodiments.

[0015] The task processing method, apparatus, and embodied agent provided in this application introduce a closed-loop mechanism of dynamic perception and continuous decision-making in the task processing flow, overcoming the problems of poor robustness and weak adjustment capability of traditional static task planning in dynamic environments. It significantly improves the embodied agent's responsiveness to environmental changes. Each task decision is based on the latest observation information after the action is executed, rather than relying on a one-time plan in the initial state. This mechanism ensures that the robot can capture changes in dynamic elements in the scene in real time (such as obstacle movement, target position deviation, task context changes, etc.) and adjust subsequent action commands accordingly. This achieves high environmental adaptability and decision-making flexibility. Furthermore, through a complete "perception-decision-execution" closed-loop process, environmental feedback is embedded after each action, ensuring the stability and reliability of the task execution process. Unlike existing static pipeline models, this closed-loop mechanism can continuously correct the action path during execution, avoiding error accumulation and execution deviation, enhancing the system's robustness in complex or uncertain tasks, and improving closed-loop control capabilities. In addition, although only one action is executed at a time, task planning is still guided by the global task objective, outputting multi-step action instructions to maintain the overall coherence and direction of task planning. Simultaneously, each prediction can dynamically adjust the subsequent action sequence, enabling the system to possess global reasoning capabilities while responding to local environmental changes, achieving an organic combination of "global planning + local adaptation." This capability is particularly crucial for long-term, multi-step tasks. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate examples according to this application and, together with this description, serve to explain the principles of this application.

[0017] Figure 1 A flowchart illustrating an example task processing method of this application is shown.

[0018] Figure 2A flowchart illustrating the task planning method of the example decision model of this application is shown.

[0019] Figure 3 This is a flowchart of a robot closed-loop task control method with rolling multi-step planning and single-step execution provided in an embodiment of this application.

[0020] Figure 4 This is a structural block diagram of a task processing device provided in an embodiment of this application.

[0021] Figure 5 This is a structural block diagram of an embodied intelligent agent provided in an embodiment of this application.

[0022] Figure 6 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0023] Detailed reference is made to exemplary embodiments, examples of which are illustrated in the accompanying drawings. The following description refers to the accompanying drawings, wherein the same reference numerals in different drawings denote the same or similar elements unless otherwise stated. In the following description of exemplary embodiments, the described embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with several aspects of the invention as described in the appended claims.

[0024] The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. In this invention and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein is intended to represent and include any or all possible combinations of one or more of the associated listed items.

[0025] It should be understood that although the terms “first,” “second,” “third,” etc., may be used herein to describe various types of information, such information should not be limited by these terms. These terms are only used to distinguish the types of information. For example, without departing from the scope of the invention, first information may be referred to as second information, and similarly, second information may be referred to as first information. The term “if” as used herein may be understood, depending on the context, to mean “at the time,” “at this moment,” or “in response to a judgment.”

[0026] In recent years, robot task decision-making methods based on large models (such as large language models and visual language models) have gradually emerged. Leveraging their powerful reasoning and generalization capabilities, they can generate high-level task instructions based on multimodal inputs. Existing research typically integrates large models into artificially constructed decision pipelines to output action sequences and guide robot execution. However, this pipelined architecture still adheres to the "single observation-prediction" model, with the model participating in decision-making only as a high-level semantic reasoning module, lacking continuous and dynamic feedback capabilities.

[0027] In contrast, embodied task planning is essentially a dynamic closed-loop process of perception, decision-making, and execution. After each action, the agent should update its subsequent task plan based on new environmental observations to dynamically adapt to environmental changes. Therefore, it is necessary to adopt a more granular task decision-making mechanism to support continuous environmental perception and real-time updates to task execution.

[0028] Therefore, a task control method with closed-loop characteristics is urgently needed, enabling the robot to re-perceive the environment after each action and make independent action decisions based on the current observations. Specifically, a continuous rolling prediction and single-step execution strategy can be adopted. That is, the model predicts a sequence of actions to be performed to complete the task in the current state, but only executes the first step of the sequence. After executing the single step, a decision is made based on the latest environmental state, thus achieving a real-time feedback mechanism for continuous perception and decision-making. This fine-grained, dynamically updated control method helps enhance the robot's adaptability, stability, and task execution robustness in complex environments.

[0029] Figure 1 A flowchart illustrating the task processing method of this application is shown. The execution entity of the task processing method provided in this application embodiment includes, but is not limited to, a computer. The task processing method provided in this application embodiment can solve the problem of insufficient robustness of existing static task planning methods in dynamic environments in robot task control. By introducing a mechanism of continuous rolling prediction and single-step execution, dynamic perception, real-time decision-making, and efficient planning of multi-step tasks are achieved, improving the robot's adaptability and task completion ability in complex environments.

[0030] like Figure 1 As shown in the embodiments of this application, the task processing method includes:

[0031] S11. Obtain the current environmental information, perform task planning based on the current environmental information and the target task, and generate a multi-step action instruction sequence;

[0032] In step S11, obtaining current environmental information refers to collecting real-time data from the embodied agent's sensors (such as cameras, LiDAR, tactile sensors, etc.) or environmental models to obtain the dynamic state of the current environment. Environmental information may include the embodied agent's position, the layout of surrounding objects, obstacles, lighting conditions, etc. This information is the basis for task planning, ensuring that the planning is based on the latest situation and avoiding erroneous decisions caused by outdated data.

[0033] Task planning based on current environmental information and the target task refers to comprehensively analyzing the current environmental information and the target task (i.e., the ultimate goal that the embodied agent needs to achieve) to generate a preliminary task plan, which is a multi-step action instruction sequence. The multi-step action instruction sequence refers to the specific output result generated by the task planning process. It is an ordered list containing multiple (≥2) consecutive, low-level instructions that can be directly executed by the agent. For example: [Move to the table, identify the mug, grab the mug, move to the water dispenser].

[0034] S12. Output the multi-step action instruction sequence so that an embodied intelligent agent executes the first action instruction in the multi-step action instruction sequence;

[0035] In step S12, the embodied agent is not required to execute the entire sequence at once. Instead, only the first action instruction in the sequence (such as "move closer to the table") is extracted and executed. This reflects the method's prudence and real-time nature, leaving room for subsequent adjustments based on the actual execution results.

[0036] S13. Repeat steps A to C until the target task is determined to be completed based on the current environmental information: Step A: In response to receiving the completion signal of the first step action instruction, reacquire the current environmental information; Step B: Based on the reacquired current environmental information and the target task, re-plan the task and generate a new multi-step action instruction sequence; Step C: Output the new multi-step action instruction sequence so that the embodied intelligent agent executes the first step action instruction in the new multi-step action instruction sequence.

[0037] Step S13 is a cyclical process that repeats steps A through C until the target task is determined to be completed based on the latest environmental information. During this cyclical process, the embodied agent is allowed to adapt to environmental changes.

[0038] In step A, once the embodied agent completes the first action command (e.g., confirming arrival at the target point via sensors), it sends a completion signal (such as a Boolean flag or event message) to the system. This signal may originate from the agent's state monitoring module, ensuring the system is aware that the first action command has been completed. Upon receiving the completion signal, environmental information is reacquired (e.g., updating the map or object states). This is because the embodied agent's actions may have altered the environment (e.g., moving objects or changing its own position), and the environment itself may have changed (e.g., objects have moved, or obstacles have appeared), requiring fresh environmental data to ensure the accuracy of subsequent planning. Reacquiring this data avoids planning errors based on outdated information.

[0039] In step B, the task planning is re-performed using the updated environmental information; that is, the complete planning process is run again, but this time with the latest, possibly changed, environmental information. The re-planning generates a new sequence of multi-step action instructions. This re-planning process demonstrates the adaptability and flexibility of task planning.

[0040] In step C, similar to step S12, a new multi-step action instruction sequence is output so that the embodied agent executes the first action instruction in the new multi-step action instruction sequence. After execution, it waits for a completion signal again and re-enters step A.

[0041] Loop termination condition: In each step B, determine whether the target task has been completed based on the current environmental information (e.g., the target object has been acquired or the target state has been achieved). If completed, the loop exits; otherwise, iterates. This ensures that the agent robustly achieves its goal in a dynamic environment.

[0042] The task processing method provided in this application introduces a closed-loop mechanism of dynamic perception and continuous decision-making into the task processing flow, overcoming the problems of poor robustness and weak adjustment ability of traditional static task planning in dynamic environments. It significantly improves the responsiveness of the embodied agent to environmental changes. Each task decision is based on the latest observation information after the action is executed, rather than relying on a one-time plan in the initial state. This mechanism ensures that the robot can capture changes in dynamic elements in the scene in real time (such as obstacle movement, target position deviation, task context changes, etc.) and adjust subsequent action commands accordingly, thereby achieving a high degree of [efforts / capabilities]. The system offers enhanced environmental adaptability and decision-making flexibility. Furthermore, through a complete "perception-decision-execution" closed-loop process, environmental feedback is embedded after each action, ensuring the stability and reliability of the task execution process. Unlike existing static pipeline models, this closed-loop mechanism continuously corrects the action path during execution, avoiding error accumulation and execution deviation, enhancing the system's robustness in complex or uncertain tasks, and improving closed-loop control capabilities. In addition, although only one action is executed at a time, task planning is still guided by the global task objective, outputting multi-step action instructions to maintain the overall coherence and direction of task planning. Simultaneously, each prediction can dynamically adjust the subsequent action sequence, enabling the system to possess global reasoning capabilities while responding to local environmental changes, achieving an organic combination of "global planning + local adaptation." This capability is particularly crucial for long-term, multi-step tasks.

[0043] In some embodiments, the environmental information is collected by the perception module of the embodied intelligent agent, and the multi-step action instruction sequence is provided to the control module of the embodied intelligent agent. The control module controls the actuator of the embodied intelligent agent to perform the corresponding action according to the first action instruction in the multi-step action instruction sequence.

[0044] Specifically, the perception module can include various hardware sensors (such as cameras, LiDAR, depth cameras, microphones, inertial measurement units, etc.) to directly collect raw signals from the physical world, such as RGB images captured by cameras and point cloud data from LiDAR. The control module converts the first action command in the multi-step action command sequence into precise control commands. For example, if the first action command is "navigate to the table," based on this task, the control module first calculates a safe path from the current position to the table, avoiding known obstacles. Then, it decomposes this path into a series of subtle, time-sequential low-level control commands. The actuator receives the control commands from the control module and converts them into physical motion. For example, a motor drives the wheels or robotic arm joints to rotate, hydraulic / pneumatic cylinders provide pushing / pull actions, and mechanical grippers perform grasping or releasing actions.

[0045] In some embodiments, the task planning is implemented through a decision model, which is constructed based on a large language model or a visual language model. Specifically, constructing a decision model based on a large language model or a visual language model does not involve having the LLM / VLM directly output low-level control instructions, but rather utilizes their powerful semantic understanding, common-sense reasoning, and (for VLMs) visual perception capabilities to transform abstract human instructions into structured, executable action plans.

[0046] In some embodiments, each action instruction is obtained by instantiating an atomic skill of the embodied agent. Specifically, the atomic skill of the embodied agent refers to the most basic, lowest-level action unit that the embodied agent can perform; these skills cannot be further decomposed into smaller, meaningful independent actions. Different robots may possess different atomic skills. In this application, the robot plans the task based on the current environmental state and task objectives through a decision model, generating a multi-step action instruction sequence, that is, instantiating and ordering multiple atomic skill actions of the embodied agent.

[0047] like Figure 2 As shown, in some embodiments, the task planning process of the decision model is as follows:

[0048] S21. Receive current environmental information and target tasks;

[0049] In step S21, the current environmental information forms the basis for decision-making and may include at least one of the following: real-time images or video streams captured by a camera, text descriptions processed by the perception module, and sensor data. The target task may be a high-level command issued by the user, such as "Give me the apple."

[0050] S22. Based on the current environmental information and the target task, select multiple atomic skills from the atomic skill set of the embodied intelligent agent, and determine the parameters required to execute each atomic skill;

[0051] In step S22, the decision model infers, based on the target task and current environmental information, which basic skills need to be invoked in sequence to achieve the goal. For example, for the target task "Give me the apple on the table," the model might need to select the series of skills [MoveTo(table), PickUp(apple), MoveTo(user), Place(apple)]. Determining the parameters required to execute each atomic skill refers to the decision model filling in specific parameter values ​​for each selected atomic skill based on the current environmental information. For example:

[0052] The atomic skill MoveTo requires the parameter "position" (such as coordinates (x,y,z) or the semantic tag "next to the table").

[0053] The Atomic Skill PickUp requires the parameter "object" (such as an identifier for a "red apple" obtained through visual detection).

[0054] S23. Combine the selected atomic skills with the corresponding parameters to generate a multi-step action instruction sequence.

[0055] In step S23, each atomic skill is combined with its corresponding parameters to obtain an action instruction. These action instructions are then sorted in sequence to obtain a multi-step action instruction sequence. For example, for the target task "Give me the apple on the table," the multi-step action instruction sequence generated by the model might be: [MoveTo(table), PickUp(apple), MoveTo(user), Place(apple)].

[0056] In some embodiments, the action instruction for each step is one of the following: action instruction described in natural language, structured programming instruction, or JSON format instruction.

[0057] Specifically, action instructions described using natural language are expressed using structured, standardized natural language sentences. For example, "Move the robotic arm to the position at coordinates (10.5, 20.3, 5.0)". In this case, the embodied agent's control module requires a natural language parser. This parser identifies key elements in the sentence, such as verbs (e.g., "move"), objects (e.g., "robotic arm"), and parameters (e.g., coordinate values, force magnitude). The parser maps these elements to corresponding internal atomic skill functions and parameters.

[0058] Structured instructions in programming mimic the format of function calls or command statements in programming languages. Example:

[0059] move_to(target_position=[10.5, 20.3, 5.0])

[0060] This format almost directly corresponds to function calls in the code.

[0061] move_to is the function name, which directly corresponds to the atomic skill.

[0062] `target_position` is the parameter name, and the value that follows is the parameter value.

[0063] The control module of an embodied intelligent agent can directly extract function names and parameters through a script engine or simple string splitting, and then call the corresponding underlying control functions. This demonstrates that structured programming instructions can be highly compatible with the implementation code, resulting in high conversion efficiency.

[0064] JSON format instructions refer to instructions organized using the lightweight data-interchange format JSON. Example:

[0065] json

[0066] {

[0067] "action": "move_to",

[0068] "parameters": {

[0069] "target_position": [10.5, 20.3, 5.0],

[0070] "speed": 0.8

[0071] }

[0072] }

[0073] The control module of the embodied agent can parse this string using a built-in or standard JSON library in any programming language (such as Python or C++). After parsing, it can determine which skill to execute by accessing the `action` field and then read all necessary parameters from the `parameters` object.

[0074] The method proposed in this application aims to improve the continuity and robustness of robot task execution in dynamic environments. This method introduces the concept of Markov decision processes into a task-level control framework and employs a sequential closed-loop control mechanism of "multi-step prediction – single-step execution – environmental feedback – re-prediction". In this method, within each decision cycle, the agent, based on the current state s and the task objective, predicts multiple subsequent executable actions {S1, S2, S3, ..., S} simultaneously using a decision model. nThis constitutes a temporary sequence of actions. However, during the execution phase, the system only executes the first action, s1, in this sequence. Subsequently, it acquires a new environmental state, s', and re-predicts multiple steps based on this state. The entire process continues in a "rolling" manner, forming a closed-loop control chain of "prediction-execution-feedback-re-prediction." This mechanism essentially follows the state-action-state transition logic in the Markov decision framework, but enhances local task reasoning capabilities at each step by predicting action sequences, while simultaneously achieving real-time adaptation to dynamic changes by executing only the first action. Unlike traditional static path planning or one-time sequence execution, the method provided in this application effectively combines sequence reasoning with real-time feedback, balancing global coherence and local responsiveness. Therefore, the innovations of this application include at least: introducing multi-step task reasoning capabilities while maintaining the execution granularity of each decision action, achieving a unity of decision depth and execution dynamism in task control. This method improves the stability and adaptive adjustment capabilities of embodied agents in complex environments, and is particularly suitable for application scenarios with clear task objectives but frequently changing environments.

[0075] Figure 3 This is a flowchart of a robot closed-loop task control method with rolling multi-step planning and single-step execution provided in an embodiment of this application. Figure 3 As shown, during each task execution, the decision model generates a multi-step sequence of action instructions based on the current environmental state, and the system then executes the first action in that sequence. After each action is executed, the AI ​​re-observes the environment and, based on the new observation information, calls upon the decision model again to perform continuous rolling predictions, generating a new sequence of action instructions. During execution, the system executes only the first step of the current prediction sequence at a time, and after the action is completed, it adjusts the task based on the new state, forming a dynamic closed loop of perception-decision-execution.

[0076] To further illustrate the closed-loop control method of continuous rolling prediction and single-step execution proposed in this application, the application process of this application is described below in conjunction with a specific task.

[0077] Mission objective: The robot needs to enter the conference room, find and grab the body BB cream from the tiered display shelf, and finally hand it over to its owner.

[0078] 1. First model call

[0079] The decision model receives current perceived information as input and outputs a predicted sequence of multi-step action instructions, such as: Navigate_to (meeting room).

[0080] Navigate_to (tiered display rack)

[0081] Scan (Body Makeup Base)

[0082] Grasp (Body Balm Cream)

[0083] Navigate_to(owner)

[0084] Place (Body Balm Cream, Owner)

[0085] The output structure can take the form of a programmatic structure (such as Navigate_to(meeting room)), JSON format (such as ["Navigate_to", "meeting room"]), or natural language phrases.

[0086] 2. Single-step execution

[0087] Based on the action sequence predicted by the model, the robot first executes the action Navigate_to (meeting room). After this action is completed, the robot updates its own observation, historical action and other perception information.

[0088] 3. Second model call

[0089] The robot inputs the updated information back into the decision-making model, which then outputs a new sequence of multi-step action instructions based on the current environmental state.

[0090] Navigate_to (tiered display rack)

[0091] Scan (Body Makeup Base)

[0092] Grasp (Body Balm Cream)

[0093] Navigate_to(owner)

[0094] Place (Body Balm Cream, Owner)

[0095] 4. Execute in a loop

[0096] The robot then executes only the first step, Navigate_to (the shelf display), and repeats the process of steps 3 and 4 above. That is, after each step, the robot re-perceives the environment, calls up the model, and outputs new multi-step pre-action instructions until the task is completed.

[0097] Therefore, the innovative aspects of this application include at least the following:

[0098] The continuous rolling prediction task planning mechanism uses a decision model to generate a sequence of action instructions for subsequent multiple steps based on the current environmental observation information. After each action is executed, the environment is re-perceived, the input information is updated, and the model is called again to carry out a new round of multi-step task planning, forming a closed-loop planning mechanism for continuous rolling prediction.

[0099] The closed-loop control process of multi-step prediction and single-step execution: The generated sequence of action instructions is predicted based on the current information and represents all the action instructions required to complete the task, with the actions arranged sequentially. Only the first action in the current prediction sequence is executed at a time. After execution, the environmental state is re-perceived and a new model inference is triggered, thus forming a closed-loop process of multi-step prediction and single-step execution.

[0100] Based on the same inventive concept, this application also provides a task processing apparatus that can be used to implement the methods described in the above embodiments, as shown in the following embodiments. Since the principle by which the task processing apparatus solves the problem is similar to the above methods, the implementation of the task processing apparatus can refer to the implementation of the above methods, and repeated details will not be elaborated further. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0101] Figure 4 This is a structural block diagram of a task processing device provided in an embodiment of this application, as shown below. Figure 4 As shown, the task processing apparatus provided in this application embodiment includes:

[0102] Module 31 is used to obtain current environmental information;

[0103] Task planning module 32 is used to perform task planning based on the current environmental information and target task, and generate a multi-step action instruction sequence;

[0104] Output module 33 is used to output the multi-step action instruction sequence so that an embodied intelligent agent executes the first action instruction in the multi-step action instruction sequence;

[0105] The acquisition module 31, task planning module 32, and output module 33 are also used to repeatedly execute steps A to C until it is determined that the target task has been completed based on the current environmental information:

[0106] Step A: In response to receiving the completion signal of the first step action command, reacquire the current environmental information;

[0107] Step B: Based on the reacquired current environmental information and the target task, re-plan the task and generate a new multi-step action instruction sequence;

[0108] Step C: Output the new multi-step action instruction sequence so that the embodied intelligent agent executes the first action instruction in the new multi-step action instruction sequence.

[0109] Based on the same inventive concept, this application also provides an embodied intelligent agent, which can be used to implement the methods described in the above embodiments, as shown in the following embodiments. Since the principle of the embodied intelligent agent in solving the problem is similar to that of the above methods, the implementation of the embodied intelligent agent can refer to the implementation of the above methods, and repeated details will not be described again.

[0110] Figure 5 A structural block diagram of an embodied intelligent agent provided in an embodiment of this application, such as... Figure 5 As shown, the embodied intelligent agent provided in this application embodiment includes:

[0111] Processor 41 is configured to implement the method described in any of the above embodiments;

[0112] The sensing module 42, connected to the processor 41, is configured to collect environmental information;

[0113] Control module 43, connected to processor 41, is configured to control the actuator 44 of the embodied intelligent agent to perform corresponding actions according to the first action instruction in the multi-step action instruction sequence;

[0114] Actuator 44, connected to the control module 43, is configured to perform corresponding actions under the drive of the control module 43.

[0115] Figure 6 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of this application, as shown below. Figure 6 As shown, the electronic device 05 may include a processor 51, a communications interface 52, a memory 53, and a communication bus 54, wherein the processor 51, the communications interface 52, and the memory 53 communicate with each other via the communication bus 54. The processor 51 may call logical instructions in the memory 53 to execute the methods described in any of the above embodiments.

[0116] Furthermore, the logical instructions in the aforementioned memory 53 can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0117] This embodiment of the application provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer can perform the methods provided in the above-described method embodiments.

[0118] This embodiment provides a computer-readable storage medium storing a computer program that causes the computer to perform the methods provided in the above-described method embodiments.

[0119] 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.

[0120] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0121] 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.

[0122] 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.

[0123] In the description of this specification, the references to terms such as "an embodiment," "a specific embodiment," "some embodiments," "for example," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0124] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A task processing method, characterized in that, include: Obtain current environmental information, perform task planning based on the current environmental information and the target task, and generate a multi-step action instruction sequence; Output the multi-step action instruction sequence so that an embodied intelligent agent executes the first action instruction in the multi-step action instruction sequence; Repeat steps A through C until the target task is determined to be completed based on the current environmental information: Step A: In response to receiving the completion signal of the first step action command, reacquire the current environmental information; Step B: Based on the reacquired current environmental information and the target task, re-plan the task and generate a new multi-step action instruction sequence; Step C: Output the new multi-step action instruction sequence so that the embodied intelligent agent executes the first action instruction in the new multi-step action instruction sequence.

2. The method according to claim 1, characterized in that, The environmental information is collected by the perception module of the embodied intelligent agent, and the multi-step action instruction sequence is provided to the control module of the embodied intelligent agent. The control module controls the actuator of the embodied intelligent agent to perform the corresponding action according to the first action instruction in the multi-step action instruction sequence.

3. The method according to claim 2, characterized in that, The task planning is achieved through a decision model, which is constructed based on a large language model or a visual language model.

4. The method according to claim 3, characterized in that, The action instruction described in each step is obtained by instantiating an atomic skill of the embodied intelligent agent.

5. The method according to claim 4, characterized in that, The task planning process of the decision-making model is as follows: Receive current environmental information and target tasks; Based on the current environmental information and the target task, select multiple atomic skills from the atomic skill set of the embodied intelligent agent, and determine the parameters required to execute each atomic skill; The selected atomic skills are combined with the corresponding parameters to generate a multi-step action instruction sequence.

6. The method according to claim 1, characterized in that, The action instruction described in each step is one of the following: action instruction described in natural language, structured programming instruction, or JSON format instruction.

7. A task processing device, characterized in that, include: The acquisition module is used to obtain current environment information; The task planning module is used to plan tasks based on the current environmental information and target tasks, and generate a multi-step action instruction sequence. The output module is used to output the multi-step action instruction sequence so that an embodied intelligent agent executes the first action instruction in the multi-step action instruction sequence; The acquisition module, task planning module, and output module are also used to repeatedly execute steps A to C until it is determined that the target task has been completed based on the current environmental information: Step A: In response to receiving the completion signal of the first step action command, reacquire the current environmental information; Step B: Based on the reacquired current environmental information and the target task, re-plan the task and generate a new multi-step action instruction sequence; Step C: Output the new multi-step action instruction sequence so that the embodied intelligent agent executes the first action instruction in the new multi-step action instruction sequence.

8. An embodied intelligent agent, characterized in that, include: A processor configured to implement the method of any one of claims 1 to 6; A sensing module, connected to the processor, is configured to collect environmental information; The control module, connected to the processor, is configured to control the actuator of the embodied intelligent agent to perform corresponding actions according to the first action instruction in the multi-step action instruction sequence; An actuator, connected to the control module, is configured to perform corresponding actions under the drive of the control module.

9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the program to implement the method of any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1 to 6.

11. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images