A structured execution plan generation method, apparatus, device and medium

Abstract

The application discloses a structured execution plan generation method and device, equipment and medium, relates to the technical field of robots, and is used for solving the technical problems that the existing execution plan and memory system are disconnected and the execution plan accuracy is low. The method is applied to an end-side robot and includes the following steps: determining a target task template associated with a target task intention; the target task template contains a task intention slot, a parameter slot, an object slot and a time slot; performing slot completion on the target task template according to the target task intention, a preference memory, an article memory and a habit memory to obtain a target completed task; generating a structured execution plan according to the target completed task; and executing the structured execution plan to obtain an execution result. Therefore, the application can improve the accuracy of the execution plan by performing slot completion on the target task template according to the preference memory, the article memory and the habit memory.

Description

Technical Field

[0001] This application relates to the field of robotics technology, and provides a method, apparatus, device and medium for generating structured execution plans. Background Technology

[0002] As technology continues to advance, service robots, companion robots, and home robots are gradually shifting from a "one-time question-and-answer" interaction model to long-term companionship and continuous collaboration. The real challenge for these robot systems lies not in "understanding a single command," but in their ability to remember users' habits and preferences over the long term, be familiar with the placement of items in the home, master the execution process of common tasks, and proactively offer assistance at the appropriate time.

[0003] In existing technologies, memory optimization systems, multimodal spatiotemporal memory retrieval systems, and dialogue memory methods are commonly used for robot control. Memory optimization systems typically preserve key emotional cues and logical continuity through multimodal encoding, node importance assessment, association weight updates, and graph compression. Multimodal spatiotemporal memory retrieval systems achieve memory management and top-k retrieval through unified memory storage, keyword / semantic / spatiotemporal joint retrieval, exponential decay, and nonlinear lifecycles. Dialogue memory methods recover context through sentence clustering, memory trees / forests, node similarity, and path search. However, these existing technologies suffer from the following problems: (1) Existing solutions usually treat memory itself as the core object, focusing on solving "how to store, how to retrieve, and how to compress", but do not further organize memory into an executable task plan; there is a disconnect between task decomposition, execution planning and memory system.

[0004] (2) Some solutions support lifecycle, importance scoring or multi-mode retrieval, but are more for memory management rather than directly outputting robot execution path, object, parameters and timing.

[0005] (3) Some solutions are usually based on speech, image or multimodal models to make parameter decisions or generate action trajectories, lacking long-term personalized memory support. Summary of the Invention

[0006] This application provides a method, apparatus, device, and medium for generating structured execution plans, which addresses the technical problems in the prior art, such as the disconnect between execution planning and memory systems, and the low accuracy of execution plans.

[0007] On the one hand, a structured execution plan generation method is provided, applied to an end-side robot, the method comprising: Perform task intent recognition on received user commands to obtain the target task intent; A target task template associated with the target task intent is determined; wherein, the target task template includes a task intent slot, a parameter slot, an object slot, and a time slot; Based on the target task intent, preference memory, item memory, and habit memory, the target task template is filled in with slots to obtain the target completion task; Complete the task based on the objective and generate a structured execution plan; wherein, the structured execution plan includes the execution path, the set of items required for the task, execution parameters, and execution timing; The structured execution plan is executed to obtain the execution result.

[0008] Optionally, the step of performing task intent recognition on the received user instructions to obtain the target task intent includes: The semantic similarity matching method is used to calculate the semantic similarity between the user command and each candidate task intent, and to obtain the task candidate score corresponding to each candidate task intent. The candidate task intent with the highest task candidate score is determined as the target task intent.

[0009] Optionally, the step of completing the target task template by filling in the slots based on the target task intent, preference memory, item memory, and habit memory to obtain the target completion task includes: Based on the target task intent, complete the task intent slots in the target task template to obtain the first completion task; Based on preference memory, the parameter slots of the first completion task are completed to obtain the second completion task; Based on the item memory, complete the object slots for the second completion task to obtain the third completion task; Based on habitual memory, the time slots of the third completion task are completed to obtain the target completion task.

[0010] Optionally, the step of completing the parameter slots of the first completion task according to preference memory to obtain the second completion task includes: Based on preference memory, the default parameters related to the target task template are determined; The default parameters are filled into the parameter slots of the first completion task to obtain the second completion task.

[0011] Optionally, the step of completing the object slots of the second completion task based on item memory to obtain the third completion task includes: Based on the object's memory, multiple candidate locations for the object are determined; For any candidate location, the location support score of the candidate location is determined based on the degree of successful retrieval of the item at the candidate location, the historical frequency of the candidate location being confirmed as the item storage location, the confidence level of the candidate location, and the distance from the end-side robot to the candidate location. The candidate position with the highest positional support score is filled into the object slot of the second completion task to obtain the third completion task.

[0012] Optionally, the step of completing the time slot of the third completion task based on habitual memory to obtain the target completion task includes: Based on habitual memory, determine the time frame of the target task and the degree of habit stability; Generate habit items based on the time frame of the target task and the stability of the habit; The habitual items are filled into the time slot of the third completion task to obtain the target completion task.

[0013] Optionally, after executing the structured execution plan and obtaining the execution result, the method further includes: Based on the feedback results, the preference memory, object memory, and habit memory are updated.

[0014] On the one hand, a structured execution plan generation device is provided for use in an end-user robot, the device comprising: The intent recognition unit is used to recognize the task intent of the received user instructions and obtain the target task intent. The task template determination unit is used to determine the target task template associated with the target task intent; wherein, the target task template includes a task intent slot, a parameter slot, an object slot, and a time slot; The slot completion unit is used to complete the slots of the target task template based on the target task intent, preference memory, item memory, and habit memory to obtain the target completed task; The plan generation unit is used to complete the task based on the objective and generate a structured execution plan; wherein, the structured execution plan includes the execution path, the set of items required for the task, execution parameters, and execution timing; The plan execution unit is used to execute the structured execution plan and obtain the execution result.

[0015] On one hand, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the methods described above.

[0016] On the one hand, a storage medium is provided that stores computer program instructions thereon, which, when executed by a processor, implement any of the methods described above.

[0017] Compared with the prior art, the beneficial effects of this application are as follows: In this application, the structured execution plan generation method is applied to an edge robot. First, the received user instructions can be processed to identify the target task intent. Then, a target task template associated with the target task intent can be determined. The target task template includes task intent slots, parameter slots, object slots, and time slots. Next, the target task template can be filled in with slots based on the target task intent, preference memory, item memory, and habit memory to obtain a target completion task. Then, a structured execution plan can be generated based on the target completion task. The structured execution plan includes an execution path, a set of items required for the task, execution parameters, and execution timing. Finally, the structured execution plan can be executed to obtain the execution result.

[0018] Based on this, this application divides task execution-related information into three supporting domains: preference memory, object memory, and habit memory. Using a task template as the core, a structured execution plan is generated through a "memory-driven slot completion method." Therefore, compared to existing technologies, this application not only solves the problem of the disconnect between execution planning and the memory system, improving the accuracy of the execution plan, but also covers the multi-dimensional needs of tasks and facilitates verification and completion. Furthermore, since an "edge-side robot" is used to generate and execute the structured execution plan, user data does not need to be transmitted to the cloud, avoiding the risk of privacy leaks. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of an application scenario provided by an embodiment of this application; Figure 2 A schematic flowchart of a structured execution plan generation method provided in this application embodiment; Figure 3 This is a schematic diagram of a structured execution plan generation apparatus provided in an embodiment of this application.

[0021] The diagram is labeled as follows: 10-Structured execution plan generation device, 101-Processor, 102-Memory, 103-I / O interface, 104-Database, 30-Structured execution plan generation device, 301-Intent recognition unit, 302-Task template determination unit, 303-Slot completion unit, 304-Plan generation unit, 305-Plan execution unit, 306-Feedback update unit. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0023] As technology continues to advance, service robots, companion robots, and home robots are gradually shifting from a "one-time question-and-answer" interaction model to long-term companionship and continuous collaboration. The real challenge for these robot systems lies not in "understanding a single command," but in their ability to remember users' habits and preferences over the long term, be familiar with the placement of items in the home, master the execution process of common tasks, and proactively offer assistance at the appropriate time.

[0024] In existing technologies, memory optimization systems, multimodal spatiotemporal memory retrieval systems, and dialogue memory methods are commonly used for robot control. Memory optimization systems typically preserve key emotional cues and logical continuity through multimodal encoding, node importance assessment, association weight updates, and graph compression. Multimodal spatiotemporal memory retrieval systems achieve memory management and top-k retrieval through unified memory storage, keyword / semantic / spatiotemporal joint retrieval, exponential decay, and nonlinear lifecycles. Dialogue memory methods recover context through sentence clustering, memory trees / forests, node similarity, and path search. However, these existing technologies suffer from the following problems: (1) Existing solutions usually treat memory itself as the core object, focusing on solving "how to store, how to retrieve, and how to compress", but do not further organize memory into an executable task plan; there is a disconnect between task decomposition, execution planning and memory system.

[0025] (2) Some solutions support lifecycle, importance scoring or multi-mode retrieval, but are more for memory management rather than directly outputting robot execution path, object, parameters and timing.

[0026] (3) Some solutions are usually based on speech, image or multimodal models to make parameter decisions or generate action trajectories, lacking long-term personalized memory support.

[0027] Based on this, this application provides a structured execution plan generation method, which is applied to an edge robot. Specifically, firstly, the received user instructions can be used to identify the task intent to obtain the target task intent; then, a target task template associated with the target task intent can be determined; wherein, the target task template includes task intent slots, parameter slots, object slots, and time slots; next, the target task template can be filled in with slots according to the target task intent, preference memory, item memory, and habit memory to obtain a target completion task; then, a structured execution plan can be generated according to the target completion task; wherein, the structured execution plan includes an execution path, a set of items required for the task, execution parameters, and execution timing; finally, the structured execution plan can be executed to obtain the execution result. Based on this, this application divides task execution-related information into three supporting domains: preference memory, object memory, and habit memory. Using a task template as the core, a structured execution plan is generated through a "memory-driven slot completion method." Therefore, compared to existing technologies, this application not only solves the problem of the disconnect between execution planning and the memory system, improving the accuracy of the execution plan, but also covers the multi-dimensional needs of tasks and facilitates verification and completion. Furthermore, since an "edge-side robot" is used to generate and execute the structured execution plan, user data does not need to be transmitted to the cloud, avoiding the risk of privacy leaks.

[0028] After introducing the design concept of the embodiments of this application, the following is a brief introduction to the application scenarios to which the technical solutions of the embodiments of this application can be applied. It should be noted that the application scenarios described below are only for illustrating the embodiments of this application and are not intended to limit the scope. In specific implementation, the technical solutions provided by the embodiments of this application can be flexibly applied according to actual needs.

[0029] like Figure 1 The diagram shown illustrates an application scenario provided by an embodiment of this application. This application scenario may include a structured execution plan generation device 10, which is mounted on an end-side robot.

[0030] The structured execution plan generation device 10 can be used to generate and execute structured execution plans, for example, it can be an in-vehicle computer or an industrial personal computer (PC). The structured execution plan generation device 10 may include one or more processors 101, memory 102, I / O interfaces 103, and a database 104. Specifically, the processor 101 can be a central processing unit (CPU) or a digital processing unit, etc. The memory 102 can be volatile memory, such as random-access memory (RAM); the memory 102 can also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or the memory 102 can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. The memory 102 can be a combination of the above-mentioned memories. The memory 102 can store some program instructions of the structured execution plan generation method provided in this application embodiment. When these program instructions are executed by the processor 101, they can be used to implement the steps of the structured execution plan generation method provided in this application embodiment, so as to solve the technical problems in the prior art such as the separation between execution planning and memory system and the low accuracy of execution plans. The database 104 can be used to store data such as task templates, target completion tasks, structured execution plans and execution results involved in the solution provided in this application embodiment.

[0031] In this embodiment, the structured execution plan generation device 10 can obtain user instructions through the I / O interface 103. Then, the processor 101 of the structured execution plan generation device 10 will solve the technical problems existing in the prior art, such as the disconnect between execution planning and the memory system and the low accuracy of execution plans, according to the program instructions of the structured execution plan generation method provided in this embodiment in the memory 102. In addition, data such as task templates, target completion tasks, structured execution plans, and execution results can be stored in the database 104.

[0032] Of course, the methods provided in the embodiments of this application are not limited to... Figure 1 The application scenarios shown can also be used in other possible scenarios, and this application embodiment does not impose any limitations. Figure 1The functions that the various devices in the application scenarios shown can achieve will be described in subsequent method embodiments, and will not be elaborated on here. Below, the methods of the embodiments of this application will be described in conjunction with the accompanying drawings.

[0033] like Figure 2 The diagram shown is a flowchart illustrating a structured execution plan generation method provided in this application. This method can be applied to end-user robots and, through... Figure 1 The structured execution plan is generated by device 10 and executed. The specific process of this method is described below.

[0034] Step 201: Perform task intent recognition on the received user instructions to obtain the target task intent.

[0035] Specifically, firstly, semantic similarity matching can be used to calculate the semantic similarity between the user's instruction and each candidate task intent to obtain the task candidate score corresponding to each candidate task intent; among which, semantic similarity can be calculated by a lightweight text encoding model, rule matching, or a combination of both.

[0036] Assuming the received user instruction is q and the candidate task intent is m, in this application, the task candidate score S_task can be calculated using the following formula: S_task(q,m)=Sim(q,m) Where q is the user instruction; m is the candidate task intent; Sim(q,m) is the semantic similarity between the user instruction and the candidate task intent, and the larger the value of Sim(q,m), the better the match between the user instruction and the candidate task intent.

[0037] Therefore, the candidate task intent with the highest task candidate score can be determined as the target task intent.

[0038] For example, taking "making coffee" as an example, when q = "give me a cup of coffee", the candidate task intents m1 = "make coffee", m2 = "prepare tea", and m3 = "get a cup of warm water". Based on this, using the above S_task(q,m) calculation formula, we know that Sim(q,m1) has the highest task candidate score. Therefore, the system can select candidate task intent m1 as the target task intent.

[0039] Step 202: Determine the target task template associated with the target task intent.

[0040] The target task template includes task intent slots, parameter slots, object slots, and time slots.

[0041] In this application, the task template is used to carry the slot information required for task execution, and can be specifically defined using the following task template: TaskTemplate={intent, slot_param, slot_object, slot_time} In this context, TaskTemplate is the task template; intent is the task intent slot; slot_param is the parameter slot; slot_object is the object slot; and slot_time is the time slot. That is, the task template contains task intent slots, parameter slots, object slots, and time slots.

[0042] Furthermore, when determining the target task template, the target task template associated with the target task intent can be determined from the preset task template table. The preset task template table is used to represent the mapping relationship between various task intents and various task templates.

[0043] For example, if the target task intent is "making coffee", the following task template can be directly matched from the preset task template table: TaskTemplate(MakeCoffee) = {intent: MakeCoffee, slot_param: {Sweetness, Milk Amount}, slot_object: {Coffee Beans, Milk, Cup}, slot_time: {Execution Timing}} Step 203: Based on the target task intent, preference memory, item memory, and habit memory, fill in the slots of the target task template to obtain the target completion task.

[0044] Specifically, firstly, based on the target task intent, the task intent slots of the target task template can be completed to obtain the first completion task; then, based on preference memory, the parameter slots of the first completion task can be completed to obtain the second completion task; next, based on item memory, the object slots of the second completion task can be completed to obtain the third completion task; finally, based on habit memory, the time slots of the third completion task can be completed to obtain the target completion task.

[0045] In one possible implementation, when completing the parameter slots of the first completion task based on preference memory to obtain the second completion task, firstly, the default parameters related to the target task template can be determined based on preference memory; that is, after determining the target task template, the default parameters or recent preferences (e.g., sugar content, milk volume, temperature, flavor preference, etc.) related to the target task intent can be read from preference memory.

[0046] In this application, the preference memory Pref(task) can be represented by the following formula: Pref(task)={slot_param_i ->value_i} Where slot_param_i is the i-th parameter slot in the task template, and value_i is the parameter value corresponding to the i-th parameter slot.

[0047] Then, the default parameters can be filled into the parameter slots of the first completion task to obtain the second completion task.

[0048] For example, taking "making coffee" as an example, if the system remembers that the user's preference is "no sugar, less milk", then after selecting the "making coffee" template, the parameter slots can be filled in to "sweetness = no sugar, milk = less". The purpose of this step is to transform "parameter decision-making" into "filling in the parameter slots from local preference memory".

[0049] In one possible implementation, when completing the object slots for the second completion task based on item memory to obtain the third completion task, firstly, multiple candidate locations for the item can be determined based on item memory. For example, assuming the item is O (object) and the candidate locations are L (letters taken from the first letter of location), then taking "making coffee" as an example, item O = coffee beans, and the candidate locations L, based on item memory, have the following three: L1 = next to the coffee machine, L2 = wall cabinet, and L3 = storage room.

[0050] Then, for any candidate location, a location support score can be determined based on the success rate of item retrieval at that location, the historical frequency of that location being confirmed as an item storage location, the confidence level of that location, and the distance the end-side robot travels to that location. Based on this, in this application, the location support score S_pos(O,L) can be represented by the following formula: S_pos(O,L)=a·Rec(O,L)+b·Freq(O,L)+c·Conf(O,L)-d·Dist(L) In this formula, Rec(O,L) represents the success rate or recent occurrence of item O at candidate location L; Freq(O,L) represents the historical frequency of candidate location L being confirmed as the storage location of item O; Conf(O,L) represents the confidence level of candidate location L; Dist(L) represents the distance or cost for the end-user robot to reach candidate location L; and a, b, c, and d are all weights. The purpose of this formula is to guide the end-user robot to prioritize the "most likely, most reliable, and most time-efficient" item location, rather than blindly following a fixed path. In this application, the location support score in item localization can be calculated not only using the above-mentioned location support score formula, but also through rule-based methods, learning methods, or hybrid methods.

[0051] Next, the candidate position with the highest positional support score is filled into the object slot of the second completion task to obtain the third completion task.

[0052] For example, if the target task is "making coffee", assuming O = coffee beans, the candidate locations L have three according to the memory system: L1 = next to the coffee machine, L2 = wall cabinet, and L3 = storage room.

[0053] Based on this, for candidate position L1, assume the memory system records: The success rate of Rec(O,L1) is high because it is frequently found there recently; The historical frequency Freq(O,L1) is high because it has been successfully obtained multiple times in history; The confidence level Conf(O,L1) is high; Dist(L1) is moderate.

[0054] For candidate position L3, assume the memory system records: Success rate of retrieval Rec(O,L3) is low; Historical frequency Freq(O,L3) is low; The confidence level Conf(O,L3) is low; It may be further away from Dist(L3); In this application, the scores corresponding to the above low, medium, high, medium distance, and possibly farther distance can be determined by "querying the preset mapping table". Then, by substituting these scores into the above position support score formula, it can be seen that S_pos(O,L1) is much higher than S_pos(O,L3). Therefore, the edge robot can prioritize looking for coffee beans next to the coffee machine.

[0055] As can be seen, the item memory in this application can be used to complete the object slots in the task template and give priority to the candidate positions.

[0056] In one possible implementation, when completing the time slots of the third completion task based on habitual memory to obtain the target completion task, the time range and habit stability of the target task can first be determined based on habitual memory. Habitual memory can be used to record recurring patterns of the same type of task over time and proactively trigger suggestions or services at similar future time points. Furthermore, habit formation can employ fixed time windows, sliding time windows, or clustered time windows.

[0057] Then, habit terms are generated based on the time frame and habit stability of the target task. In this application, it is assumed that similar tasks occur at multiple time points, t_1, t_2, ..., t_n. If these time points are clustered and their frequency exceeds a threshold, the habit term Habit(task) can be generated using the following formula: Habit(task)={time_window, confidence} Wherein, time_window is the time range in which the target task frequently occurs; confidence is the degree of stability of the target task's habits.

[0058] For example, taking "making coffee" as an example, if it is recorded that users frequently perform this task around 8 a.m. on multiple weekdays, the following habit items can be formed: Habit (Making Coffee) = {time_window:08:00~08:15, confidence:high} Finally, you can fill the habit items into the time slot of the third completion task to obtain the target completion task.

[0059] Furthermore, if the current time falls within this time window, the robot can proactively suggest "whether coffee needs to be prepared," or check in advance whether the coffee beans, milk, and cups are ready.

[0060] It is evident that the habit memory of this application allows the edge robot to shift from passive response to proactive service, instead of passively waiting for instructions every time. Compared with solutions that only perform task planning or motion control, this application significantly improves the service level and user experience.

[0061] Step 204: Complete the tasks based on the objectives and generate a structured execution plan.

[0062] The structured execution plan includes the execution path, the set of items required for the task, execution parameters, and execution timing.

[0063] In this application, the structured execution plan can be specifically represented by the following formula: Plan={path, objects, parameters, timing} Wherein, path is the execution path or path point sequence (which can be obtained from the object slot in the target completion task); objects is the set of items required for the task (which can be obtained from the object slot in the target completion task); parameters are the execution parameters (which can be obtained from the parameter slot in the target completion task); and timing is the execution timing (which can be obtained from the time slot in the target completion task).

[0064] For example, continuing with the example of coffee "making", a structured execution plan could be as follows: Plan = {path:[current location, next to the coffee machine, refrigerator, desktop], objects:[coffee beans, milk, cup], parameters:{sweetness: sugar-free, milk volume: low}, timing: execute immediately} It is evident that this structured execution plan can clearly answer the four questions of the end-side robot: "Where to go first, what to get, how to do it, and when to do it." It can be directly used as input to the execution module to accurately realize the generation of the structured execution plan.

[0065] Step 205: Execute the structured execution plan and obtain the execution results.

[0066] In this application, after the structured task plan is issued, the end-side robot can directly complete actions such as navigation, object retrieval, production and delivery, and obtain the corresponding execution results after the execution is completed.

[0067] In one possible implementation, the end-side plan execution can adopt a fully local mode, a local priority mode, or a partially synchronous mode.

[0068] Step 206: Update the preference memory, object memory, and habit memory based on the feedback results.

[0069] In this application, feedback can be written back based on the execution results to update preference memory, object memory, and habit memory. The execution feedback can employ binary feedback, score-based feedback, or multi-level feedback; the memory state can be represented using the following simple update method: V_new=(1-ρ)·V_old+ρ·F Where V_new is the updated state value; V_old is the state value before the update; F is the feedback value; and ρ is the update step size (the larger ρ is, the faster the system accepts the new result; the smaller ρ is, the more conservative the system is, and the more it relies on old experience). In this application, if the plan is executed successfully, F can be 1; if the plan is partially executed successfully, F can be a value between 0 and 1; if the plan fails, F can be 0. It can be seen that this formula can be used to gradually update the memory state with the actual execution result, avoiding drastic fluctuations caused by a single feedback.

[0070] For example, continuing with the "making coffee" example, suppose the edge robot believes that "coffee beans @ next to the coffee machine" is credible, V_old = 0.85; after this plan is executed, the edge robot does indeed find coffee beans next to the coffee machine, therefore, it can provide feedback F = 1. If ρ = 0.2, then: V_new = 0.8 × 0.85 + 0.2 × 1 = 0.88; this shows that the system is more certain that coffee beans are indeed often located here. Conversely, if this time the coffee beans are not found next to the coffee machine, the feedback F = 0, then: V_new = 0.8 × 0.85 + 0.2 × 0 = 0.68; this indicates that the system begins to doubt this location, but will not immediately completely deny it.

[0071] In one possible implementation, in order to facilitate edge deployment and avoid over-reliance on complex tree structures, "structured storage and addition, deletion and modification" are also implemented in this application.

[0072] Specifically, in engineering, a combination of "node table + relationship table + index table + event log" can be used to achieve structured storage and CRUD operations, but it is not limited to this. Among them, the node table can be used to store basic information of supporting items, the relationship table can be used to store the association between task templates and preferences, items, and habits, the index table can be used to accelerate retrieval, and the event log can be used to record additions, modifications, and deletions.

[0073] In this application, when adding an event log, the log can be written first, then written to the node table and the index updated; when modifying an event log, history can be preserved through versioning or incremental updates; when deleting an event log, lazy deletion can be prioritized, with background cleanup later. The deletion mechanism can employ immediate deletion, lazy deletion, or delayed cleanup.

[0074] Furthermore, this application not only facilitates edge deployment but also avoids over-reliance on complex tree structures. Of course, this structure can be replaced with trees, forests, graphs, or hybrid indexes, etc.

[0075] In one possible implementation, the memory support structure of this application can be implemented using a tree, forest, graph, DAG, or hybrid indexing method, and is not limited to a single structure.

[0076] In summary, in this application, (1) Since the information related to task execution is divided into three support domains: preference memory, item memory and habit memory, and the task template is the core, a structured execution plan is generated through the "memory-driven slot completion method". Therefore, compared with the prior art, this application can not only solve the problem of the separation between execution planning and memory system and improve the accuracy of execution plan, but also cover the multi-dimensional needs of tasks and facilitate verification and completion.

[0077] (2) Because of the introduction of the execution feedback write-back mechanism, the supporting domain can be continuously updated according to the real environment, forming an iterative closed-loop update mechanism. Therefore, compared with the prior art, this application can learn from the actual execution results through the closed-loop process of "instruction → plan generation → execution feedback → memory update" to gradually improve the accuracy of the execution plan.

[0078] (3) Since the “end-side robot” is used to generate and execute the structured execution plan, user data does not need to be transmitted to the cloud compared with existing technologies, thus avoiding the risk of privacy leakage.

[0079] Based on the same inventive concept, embodiments of this application provide a structured execution plan generation device 30, such as... Figure 3 As shown, the structured execution plan generation device 30 is applied in an end-side robot and includes: The intent recognition unit 301 is used to perform task intent recognition on the received user instructions to obtain the target task intent; The task template determination unit 302 is used to determine the target task template associated with the target task intent; wherein, the target task template includes a task intent slot, a parameter slot, an object slot and a time slot; The slot completion unit 303 is used to complete the slots of the target task template based on the target task intent, preference memory, item memory and habit memory, so as to obtain the target completion task; The plan generation unit 304 is used to complete the task based on the objective and generate a structured execution plan; wherein, the structured execution plan includes the execution path, the set of items required for the task, the execution parameters, and the execution timing; The plan execution unit 305 is used to execute the structured execution plan and obtain the execution results.

[0080] Optionally, the intent recognition unit 301 is also used for: The semantic similarity matching method is used to calculate the semantic similarity between the user command and each candidate task intent, and to obtain the task candidate score corresponding to each candidate task intent. The candidate task intent with the highest task candidate score is determined as the target task intent.

[0081] Optionally, the slot completion unit 303 is also used for: Based on the target task intent, complete the task intent slots in the target task template to obtain the first completion task; Based on preference memory, the parameter slots of the first completion task are completed to obtain the second completion task; Based on the item memory, complete the object slots for the second completion task to obtain the third completion task; Based on habitual memory, the time slots of the third completion task are completed to obtain the target completion task.

[0082] Optionally, the slot completion unit 303 is also used for: Based on preference memory, determine the default parameters related to the target task template; Fill the default parameters into the parameter slots of the first completion task to obtain the second completion task.

[0083] Optionally, the slot completion unit 303 is also used for: Based on the object's memory, multiple candidate locations for the object are determined; For any candidate location, the location support score is determined based on the success rate of item retrieval at the candidate location, the historical frequency of the candidate location being confirmed as the item storage location, the confidence level of the candidate location, and the distance the end-side robot travels to the candidate location. Fill the candidate position with the highest positional support score into the object slot of the second completion task to obtain the third completion task.

[0084] Optionally, the slot completion unit 303 is also used for: Based on habitual memory, determine the time frame of the target task and the degree of habit stability; Generate habit items based on the time frame of the target task and the stability of the habit; Fill the habit items into the time slot of the third completion task to obtain the target completion task.

[0085] Optionally, the structured execution plan generation apparatus 30 further includes a feedback update unit 306, used for: Based on the feedback results, the preference memory, object memory, and habit memory are updated.

[0086] The structured execution plan generation device 30 can be used for execution. Figure 2 The method executed by the structured execution plan generation device in the illustrated embodiment can be referenced for understanding the functions that each functional module of the structured execution plan generation device 30 can achieve. Figure 2 The embodiments shown are described in detail below.

[0087] In some possible implementations, various aspects of the methods provided in this application can also be implemented as a program product comprising program code that, when run on a computer device, causes the computer device to perform the steps of the methods according to the various exemplary embodiments of this application described above. For example, the computer device may perform actions such as... Figure 2The method executed by the structured execution plan generation device in the illustrated embodiment.

[0088] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. Alternatively, if the integrated units of this application are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to the prior art, can be embodied in the form of software products. These computer software products are stored in a storage medium and include 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 methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0089] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0090] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for generating structured execution plans, characterized in that, When applied to end-side robots, the method includes: Perform task intent recognition on received user commands to obtain the target task intent; Determine the target task template associated with the target task intent; wherein the target task template includes a task intent slot, a parameter slot, an object slot, and a time slot; Based on the target task intent, preference memory, item memory, and habit memory, the target task template is filled in with slots to obtain the target completion task; Complete the task based on the objective and generate a structured execution plan; wherein, the structured execution plan includes the execution path, the set of items required for the task, execution parameters, and execution timing; The structured execution plan is executed to obtain the execution result.

2. The method as described in claim 1, characterized in that, The step of recognizing the task intent of the received user instructions to obtain the target task intent includes: The semantic similarity matching method is used to calculate the semantic similarity between the user command and each candidate task intent, and to obtain the task candidate score corresponding to each candidate task intent. The candidate task intent with the highest task candidate score is determined as the target task intent.

3. The method as described in claim 1, characterized in that, The step of completing the target task template by filling in the slots based on the target task intent, preference memory, item memory, and habit memory to obtain the target completion task includes: Based on the target task intent, complete the task intent slots in the target task template to obtain the first completion task; Based on preference memory, the parameter slots of the first completion task are completed to obtain the second completion task; Based on the item memory, complete the object slots for the second completion task to obtain the third completion task; Based on habitual memory, the time slots of the third completion task are completed to obtain the target completion task.

4. The method as described in claim 3, characterized in that, The step of completing the parameter slots of the first completion task based on preference memory to obtain the second completion task includes: Based on preference memory, the default parameters related to the target task template are determined; The default parameters are filled into the parameter slots of the first completion task to obtain the second completion task.

5. The method as described in claim 3, characterized in that, The step of completing the object slots for the second completion task based on item memory to obtain the third completion task includes: Based on the object's memory, multiple candidate locations for the object are determined; For any candidate location, the location support score of the candidate location is determined based on the degree of successful retrieval of the item at the candidate location, the historical frequency of the candidate location being confirmed as the item storage location, the confidence level of the candidate location, and the distance from the end-side robot to the candidate location. The candidate position with the highest positional support score is filled into the object slot of the second completion task to obtain the third completion task.

6. The method as described in claim 3, characterized in that, The step of completing the time slot of the third completion task based on habitual memory to obtain the target completion task includes: Based on habitual memory, determine the time frame of the target task and the degree of habit stability; Generate habit items based on the time frame of the target task and the stability of the habit; The habitual items are filled into the time slot of the third completion task to obtain the target completion task.

7. The method as described in claim 1, characterized in that, After executing the structured execution plan and obtaining the execution result, the method further includes: Based on the feedback results, the preference memory, object memory, and habit memory are updated.

8. A structured execution plan generation device, characterized in that, The device, used in end-side robots, includes: The intent recognition unit is used to recognize the task intent of the received user instructions and obtain the target task intent. The task template determination unit is used to determine the target task template associated with the target task intent; wherein, the target task template includes a task intent slot, a parameter slot, an object slot, and a time slot; The slot completion unit is used to complete the slots of the target task template based on the target task intent, preference memory, item memory, and habit memory to obtain the target completed task; The plan generation unit is used to complete the task based on the objective and generate a structured execution plan; wherein, the structured execution plan includes the execution path, the set of items required for the task, execution parameters, and execution timing; The plan execution unit is used to execute the structured execution plan and obtain the execution result.

9. An electronic device, characterized in that, The device includes: Memory, used to store program instructions; A processor is configured to invoke program instructions stored in the memory and execute the method described in any one of claims 1-7 according to the obtained program instructions.

10. A storage medium, characterized in that, The storage medium stores computer-executable instructions for causing a computer to perform the method described in any one of claims 1-7.

Images