A robot task orchestration system
By adopting a separate architecture design between the cloud and the robot, with the cloud responsible for decision generation and the edge responsible for execution and verification, the rigidity problem of robot task orchestration system is solved, achieving efficient and intelligent task scheduling and response, and improving the system's flexibility and security.
Patent Information
- Application Number
- CN202611262296.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing robot task orchestration systems rely on manual configuration strategies for orchestration decisions when facing complex and ever-changing real-world scenarios. This results in low efficiency, difficulty in dynamic adaptation, rigid task scheduling, and untimely responses, failing to meet the needs of intelligent services across all scenarios.
The system adopts a separate architecture design with cloud orchestration service nodes and robot-side access adapters. The cloud is responsible for decision-making and generating standard tasks, while the edge performs local compliance verification, including intent parsing, unified task pool management, decision adjudication, and task generation, thereby decoupling decision-making from execution.
It improves the intelligence level and response speed of task generation, solves the problem of multi-source task conflicts, enhances the scheduling efficiency and fairness of the system in multi-task concurrent scenarios, reduces the computing load of the robot terminal, and ensures the security of task issuance.
Smart Images

Figure CN123117623A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of software scheduling technology, and in particular to a robot task orchestration system. Background Technology
[0002] In the field of robot task orchestration and scheduling, existing systems typically rely on manual configuration of fixed trigger action rules in the backend management system. This means that when a specific event occurs (such as detecting an elderly person falling), the system can only execute the corresponding action according to preset hard-coded logic. If the backend does not pre-configure the corresponding rules, even if a critical event is detected, no alarm or service action will be triggered, resulting in a lack of dynamic responsiveness. Currently, the industry has a certain technological foundation, including robot point-to-point task execution capabilities (navigation to a point, stopping, broadcasting, interaction, etc.), a unified task expression foundation (capable of describing task objectives, points, parameters, sources, and states), multiple scheduling semantics (timed tasks, sequential tasks, runtime inserted tasks, and default fallback tasks), runtime task access and lifecycle management mechanisms (marking completion, cancellation, failure, timeout, or interruption), and execution diagnostics and logging capabilities.
[0003] However, when faced with complex and ever-changing real-world scenarios, existing solutions still rely on manual configuration of strategy rules for orchestration decisions. When there are many types of scenarios, conflicts between tasks, or a need to comprehensively judge priorities, manual configuration is not only inefficient but also difficult to cover all possibilities. It cannot dynamically adjudicate tasks based on real-time context and automatically generate behavior orchestration schemes that are adapted to the current scenario, resulting in rigid task scheduling, untimely responses, and difficulty in meeting the intelligent service needs under the generalization of all scenarios. Summary of the Invention
[0004] To address the issues of rigid task scheduling and untimely response in existing technologies, which make it difficult to meet the needs of intelligent services across all scenarios, this application provides a robot task orchestration system.
[0005] In a first aspect, this application provides a robot task orchestration system, which adopts the following technical solution: A robot task orchestration system includes a cloud orchestration service node and a robot-end access adapter; The cloud orchestration service node is used to obtain a context snapshot of the current operating environment. The context snapshot includes the current robot state, active task information, external trigger events, current task orchestration status, and scheduled execution plan. Based on the context snapshot and preset scheduling strategy, the node makes decisions on candidate tasks and generates standard tasks. The node then sends the standard tasks to the robot's access adapter. The robot-side access adapter is used to perform local compliance verification on the standard task; when the local compliance verification passes, it controls the underlying execution system to execute the standard task.
[0006] By adopting the above technical solution and using a separate architecture design between the cloud orchestration service node and the robot terminal access adapter, the cloud is responsible for decision-making and generating standard tasks, while the terminal only performs local compliance checks and task submission, thus decoupling decision-making from execution. The cloud centrally processes complex orchestration logic, while the terminal performs only lightweight checks, which reduces the computational load on the robot terminal and ensures the security of task distribution. The cloud can independently upgrade the orchestration strategy without affecting the stability of terminal execution.
[0007] In a preferred embodiment, this application can be further configured such that the cloud orchestration service node includes an intent parsing module, and the intent parsing module is configured as follows: When the external triggering event is a natural language request, the natural language request is semantically parsed through the intent parsing model to identify the intent type; When the external triggering event is a sensor event or triggers the timed execution plan, it is mapped to the corresponding predefined intent type according to the event type and time parameters; Based on the identified intent type, match scene families and task templates to generate candidate tasks, and incorporate the newly generated candidate tasks into a unified task pool.
[0008] By adopting the above technical solution, the intent parsing module performs semantic parsing on natural language requests, types mapping on system timed events or sensor events, and generates candidate tasks by matching scene families and task templates. This transforms unstructured user instructions and structured system events into standardized candidate tasks, enabling the robot to understand diverse inputs and automatically classify them into corresponding business scenarios. This reduces the workload of manual configuration and improves the intelligence level and response speed of task generation.
[0009] In a preferred embodiment, this application can be further configured such that the cloud orchestration service node also includes a unified task pool management module and a decision-making module, wherein the unified task pool management module is configured as follows: Candidate tasks from different task sources are included in a unified task pool, including user immediate request sources, planned configuration task sources, timed policy task sources, and high-priority inserted event sources. Assign task attributes to each candidate task in the unified task pool. The task attributes include task source identifier, time window constraint, priority level and status label. The status label includes pending execution, executing, completed, canceled, failed and timed out. Based on the task attributes and the current robot state, a candidate task to be executed is dynamically selected from the unified task pool, and the selected candidate task is submitted to the decision-making module for decision-making.
[0010] By adopting the above technical solutions, the unified task pool management module incorporates candidate tasks from multiple sources, such as immediate requests, plan configurations, timing strategies, and high-priority events, into unified management. It assigns attributes such as source identifier, time window, priority, and status label to each task and dynamically selects the optimal task for decision-making based on the robot's current state, thus resolving the multi-source task conflict problem. This ensures that all types of tasks are executed in an orderly manner under the same set of decision-making logic, avoiding task chaos and priority inversion, and improving the scheduling efficiency and fairness of the system in multi-task concurrent scenarios.
[0011] In a preferred embodiment, this application can be further configured such that the cloud orchestration service node also includes a combined task processing module, which is configured to perform the following operations after the intent parsing module identifies the intent type and matches the scene family, and before the decision-making module makes a decision: When the external triggering event contains multiple sub-targets or composite instructions, the external triggering event is decomposed into multiple executable sub-tasks; Establish execution relationships for the plurality of executable subtasks, wherein the execution relationships include sequential execution, conditional execution, and failure degradation; The plurality of executable subtasks and the execution relationship are mapped to standard task batches.
[0012] By adopting the above technical solution, the combined task processing module intervenes after intent parsing and before decision-making, decomposing requests containing multiple sub-goals or composite instructions into multiple executable sub-tasks, and establishing execution relationships of sequential execution, conditional execution, or failure degradation, mapping them to standard task batches; this enables the system to process complex composite instructions, solving the limitation of traditional solutions that can only handle single-step tasks, avoiding logical breaks caused by the isolated execution of each sub-task after composite requests are decomposed, and improving the system's support capability for complex business scenarios.
[0013] In a preferred embodiment, this application can be further configured such that: the decision-making module is configured to make a decision on the candidate tasks and generate a scheduling strategy, including: The precondition verification unit is used to perform precondition verification on the candidate task based on the current robot state and the task parameters of the candidate task; The task interruption judgment unit is used to determine whether to interrupt the current active task and output a task interruption flag when the precondition verification passes, based on the priority of the candidate task and the priority of the current active task in the active task information, and in combination with the interruptible status of the current active task. The scheduling policy generation unit is used to determine a scheduling policy based on the task interruption identifier and update the current task orchestration based on the scheduling policy; wherein, when the task interruption identifier indicates that the currently active task is interrupted, the scheduling policy is a task recovery policy; when the task interruption identifier indicates that the currently active task is not interrupted, the scheduling policy is a task replacement policy or a task waiting policy.
[0014] By adopting the above technical solution, the precondition verification of the decision-making module ensures that the candidate task is feasible to execute under the current robot state. Then, the task interruption judgment unit compares the priority of the candidate task with the priority of the current active task and makes an intelligent decision based on the interruptibility status, and finally generates a task recovery, replacement or waiting strategy. This avoids blind interruption or blind rejection, enables high-priority tasks to reasonably seize resources, and ensures the safe exit and subsequent recovery of the current task, thus realizing intelligent decision-making for task scheduling.
[0015] In a preferred embodiment, this application can be further configured such that the cloud orchestration service node also includes a task generation module, which is configured to: when the currently active task is interrupted to execute the candidate task. Obtain a task template matching the candidate task from the task template library, fill the task template with execution parameters to generate the standard task, and attach the task recovery strategy to the standard task; The task recovery strategy is used to instruct the currently active task to perform recovery, skip, postpone, or end operations after the candidate task has been completed.
[0016] By adopting the above technical solution, after determining that the current task is interrupted, the task generation module matches a template from the task template library and fills in the parameters to generate a standard task. At the same time, it includes a task recovery strategy (recover, skip, postpone or end), which makes the generation process of standard tasks standardized and configurable, avoiding the need to manually write task instructions each time. The included recovery strategy ensures that the interrupted task can be recovered in an orderly manner according to preset rules after the preempted task is completed, improving the predictability of the system and the continuity of user experience.
[0017] In a preferred embodiment, this application can be further configured such that the cloud orchestration service node also includes a task verification module, which is configured as follows: Before the standard task is sent to the robot access adapter, the standard task output by the task generation module is validated for legality. The validation includes data format validation and field integrity validation. When the legality verification passes, the standard task is sent to the robot's access adapter; If the validity check fails, the standard task will not be issued, a check failure log will be recorded, and a retry mechanism will not be triggered.
[0018] By adopting the above technical solution, the cloud performs data format verification and field integrity verification on the generated standard tasks, and directly rejects the delivery without triggering a retry mechanism when the verification fails. This intercepts basic problems such as format errors and missing fields in the cloud, avoiding invalid tasks from occupying terminal resources and communication bandwidth. The absence of a retry mechanism prevents infinite loop delivery caused by structural errors in the task itself, reducing system overhead and improving resource utilization efficiency.
[0019] In a preferred embodiment, this application can be further configured such that the robot-end access adapter includes: The local compliance verification unit is used to perform format verification, physical space and location verification, equipment capability verification and security rule verification on the standard tasks. A local compliance judgment unit is used to determine that the standard task has passed the local compliance verification when all the verification results of the local compliance verification unit have passed. The task submission unit is used to submit the standard task to the underlying execution system through the controlled task access interface when the local compliance verification is passed.
[0020] By adopting the above technical solution, the local compliance verification unit of the robot end access adapter verifies the standard task execution format, physical space and location, equipment capability, and security rules issued by the cloud. Only after ensuring that all the actual conditions on the end side are met can the robot submit the task to the underlying execution system through the controlled interface. This forms the last line of defense on the end side, preventing task execution failures caused by dynamic changes in the end-side environment (such as map updates or equipment failures) due to cloud decisions, and improving the robustness of the system.
[0021] In a preferred embodiment, this application can be further configured such that the robot end access adapter also includes: The failure feedback unit is used to provide feedback on the reason for the failure to the cloud orchestration service node when the local compliance judgment unit determines that the local compliance verification has failed. The reason for the failure includes the failed verification item and the corresponding parameter value. After receiving the verification failure reason, the cloud orchestration service node controls the task generation module to regenerate the standard task according to the verification failure reason and sends it to the robot terminal access adapter. The retry control unit is used to record the number of retries that are regenerated and reissued. When the number of retries exceeds the preset retry threshold, a local fallback strategy is triggered.
[0022] By adopting the above technical solution, the failure feedback unit reports the specific reason for failure to the cloud when the local compliance verification on the edge fails. Based on this, the cloud controls the task generation module to regenerate the standard task and reissue it. At the same time, the retry control unit records the number of retries and triggers a fallback when the threshold is exceeded. Compared with simply giving up or retrying indefinitely, this solution can accurately correct the task parameters according to the specific reason for failure, which improves the final success rate of task execution. Meanwhile, the retry threshold avoids the system wasting resources on unrecoverable errors.
[0023] In a preferred embodiment, this application can be further configured such that the robot-end access adapter also includes a fallback execution unit, used to trigger a local fallback strategy when any of the following conditions are met: Communication between the cloud orchestration service node and the robot terminal access adapter is unavailable or has timed out. The local compliance assessment unit determines that the local compliance verification failed and the number of retries exceeded the preset retry threshold; The standard task returned by the cloud orchestration service node is an illegal task; The local fallback strategy includes at least one of the following: keeping the currently active task running, controlling the robot to return to a preset service point, or controlling the robot to enter a default standby mode.
[0024] By adopting the above technical solution, the fallback execution unit triggers a local fallback strategy in abnormal situations such as communication unavailability / timeout, edge verification failure exceeding the threshold, and illegal task return from the cloud. This strategy allows the current task to continue execution, return to a preset service point, or enter a default standby mode. This ensures that the robot still has basic autonomous behavior capabilities when the cloud is unreachable or data is abnormal. It avoids the robot from falling into a stagnant state with no instructions to execute due to the lack of cloud decision-making, improves the system's fault tolerance and on-site availability, and ensures that basic services are not interrupted.
[0025] In summary, this application includes the following beneficial technical effects: This application employs a decoupled architecture design between the cloud orchestration service node and the robot-side access adapter. The cloud is responsible for decision-making and generating standard tasks, while the edge only performs local compliance checks and task submission, thus decoupling decision-making from execution. The cloud centrally processes complex orchestration logic, while the edge performs only lightweight checks. This reduces the computational load on the robot terminal, ensures the security of task distribution, and allows the cloud to independently upgrade orchestration strategies without affecting the stability of edge execution. Attached Figure Description
[0026] Figure 1 This is a schematic diagram of the structure of a robot task orchestration system provided in an embodiment of this application; Figure 2This is a schematic diagram of the module interaction of a robot task orchestration system provided in an embodiment of this application. Detailed Implementation
[0027] The following is in conjunction with the appendix Figure 1 To be continued Figure 2 This application will be described in further detail.
[0028] This specific embodiment is merely an explanation of this application and is not intended to limit it. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.
[0029] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. 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.
[0030] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.
[0031] It should be noted that all data interaction processes involved in this application have corresponding transmission protocols, including authorized data collection and use, and both parties involved in the data interaction have completed data authorization through the execution of the protocol.
[0032] This application provides a robot task orchestration system, such as... Figure 1 As shown, system 10 includes a cloud orchestration service node 101 and a robot-end access adapter 102. The cloud orchestration service node is used to obtain a context snapshot of the current operating environment, which includes the current robot status, active task information, external trigger events, current task orchestration status, and scheduled execution plan; it makes decisions on candidate tasks based on the context snapshot and preset scheduling strategies to generate standard tasks; it issues the standard tasks to the robot-end access adapter; the robot-end access adapter is used to perform local compliance verification on the standard tasks; when the local compliance verification passes, it controls the underlying execution system to execute the standard tasks.
[0033] The cloud-based orchestration service node is a software service system deployed on a remote cloud server, serving as the decision-making brain for the entire robot task orchestration process. It does not rely on any specific robot hardware, but rather communicates with multiple robots via a network, uniformly receiving various event requests from users, the operations backend, timing strategies, and the robots themselves. It then performs semantic understanding, scene recognition, priority determination, and task generation on these requests. Internally, the cloud-based orchestration service node comprises several functional modules: a context management module maintains a real-time snapshot of the status of all robots; an intent parsing module performs semantic parsing of user natural language requests; a unified task pool management module aggregates candidate tasks from different sources and sorts them by priority; a decision-making module determines interruption and selects scheduling strategies for candidate tasks based on the current context; a task generation module extracts templates from a task template library and fills in parameters to generate standard tasks; and a task verification module verifies the format and completeness of the generated standard tasks. The core advantage of the cloud-based orchestration service node lies in utilizing the abundant computing resources of the cloud (especially large model inference capabilities) to handle semantic understanding and policy determination in complex scenarios, while simultaneously distributing the decision results in the form of standard tasks, avoiding the deployment of heavy computational models on the robot itself.
[0034] The robot-end access adapter is a lightweight software program deployed locally on each robot, acting as a communication gateway and execution front-end processor between the robot and the cloud orchestration service node. It does not directly control the robot's navigation, voice, or chassis movement, but rather interfaces with the robot's existing underlying execution system through standardized interfaces. The main responsibilities of the robot-end access adapter include: real-time acquisition of the robot's current status and reporting it to the cloud orchestration service node; receiving standard tasks from the cloud and performing local compliance checks; submitting the standard tasks to the robot's underlying execution system for execution through the controlled task access interface after successful verification; and simultaneously transmitting status changes and feedback results during execution back to the cloud in real time. The robot-end access adapter also has independent fallback logic. When cloud communication is unavailable, times out, or illegal tasks are issued, it can trigger local fallback strategies to ensure that the robot does not completely lose control due to network problems.
[0035] Specifically, during system operation, the cloud orchestration service node continuously monitors the current robot status, active task information, external trigger events, current task orchestration status, and scheduled execution plans reported by the robot. When any trigger condition is met, the cloud orchestration service node reads the unified context snapshot of the current moment from the context management module. The current robot status includes the robot's real-time position coordinates, current battery level, operating mode (e.g., standby / navigation / interaction), and the health status of each sensor; active task information includes the currently executing task identifier, task type, task source, execution duration, and a flag indicating whether the task can be interrupted; external trigger events include instant requests initiated by the user via voice or touchscreen and high-priority abnormal events detected by sensors (e.g., safety alarms); the current task orchestration status includes at least the orchestration link identifier to which the currently active task belongs, the current execution progress of that orchestration link, and the queue of unexecuted tasks in that orchestration link and their recovery strategies; the scheduled execution plan includes planned configuration tasks issued by the operations backend and periodic tasks triggered by scheduled strategies. The context management module maintains the above information in a structured data format (such as JSON) and marks the version of each snapshot with a timestamp to ensure that the context on which subsequent decisions are based is consistent and traceable.
[0036] The cloud orchestration service node first generates candidate tasks based on the type of externally triggered events: user natural language requests are identified by the intent parsing model to identify the intent type and match the task template, while timed or sensor events are directly mapped to predefined intent types and candidate tasks are retrieved from a unified task pool.
[0037] Subsequently, the decision-making module performs precondition verification based on the context snapshot. After the verification passes, it further determines whether to interrupt the current task to execute the candidate task based on the priority level of the candidate task and the current active task and the interruptibility status of the current task, and determines the scheduling strategy (recovery strategy, replacement strategy or waiting strategy) accordingly.
[0038] The task generation module retrieves a matching task template from the task template library based on the adjudication result, fills the template with execution parameters (target location, broadcast content, interaction method, etc.), and includes the scheduling strategy determined in the decision-making stage to generate a structured standard task. This task adopts a unified data format agreed upon with the robot's end access adapter.
[0039] After the task generation module outputs a standard task, the task verification module first performs a validity check, including data format verification and field integrity verification. If the verification passes, the task is sent to the target robot's access adapter through a secure communication channel. If the verification fails, the task is rejected and a failure log is recorded, without triggering a retry mechanism.
[0040] After receiving a standard task, the robot's access adapter performs local compliance checks, including format verification, physical space and location verification, device capability verification, and security rule verification. If any one of these checks fails, the verification is considered failed; if all checks pass, the verification is considered successful. After the local compliance checks are successful, the robot's access adapter submits the standard task to the underlying execution system through the controlled task access interface. The underlying execution system executes the task step by step according to the sequence of actions. The robot's access adapter reports the execution feedback to the cloud orchestration service node in real time, allowing the context management module to update the context snapshot, thus forming a complete closed loop.
[0041] Figure 2 This is a schematic diagram of the module interaction of a robot task orchestration system provided in an embodiment of this application.
[0042] In one possible implementation of this application embodiment, the cloud orchestration service node includes an intent parsing module, which is configured as follows: When the external triggering event is a natural language request, the intent parsing model performs semantic parsing on the natural language request to identify the intent type; When an external trigger event is a sensor event or a trigger timed execution plan, it is mapped to the corresponding predefined intent type based on the event type and time parameters. Based on the identified intent type, match scene families and task templates to generate candidate tasks, and incorporate the newly generated candidate tasks into a unified task pool.
[0043] After receiving an external trigger event, the intent parsing module first determines the type of the event, distinguishing between a natural language request and a system-timed / sensor event. Natural language requests refer to voice commands picked up by the robot's microphone array or text commands entered via the touchscreen, such as a user saying "Take me to the activity room" or "Introduce this exhibition area." System-timed events are periodic task events triggered by the operations backend or the timing strategy module according to a preset schedule, such as a scheduled patrol task triggered every morning at 9:00 AM. Sensor events are events automatically generated by various physical sensors on the robot (infrared sensors, depth cameras, drop sensors, collision sensors, etc.) when they detect changes in specific physical conditions, such as a visual recognition event detecting an elderly person falling or an obstacle avoidance event detecting an obstacle.
[0044] When an external trigger event is identified as a natural language request input by the user via voice or text, the intent parsing module submits the request content to the intent parsing model for processing. The intent parsing model is a semantic understanding model built on natural language processing technology. It can employ a finely tuned large language model or a lightweight intent classification model. This model is pre-trained and capable of extracting the user's core needs from unstructured natural language text. The intent parsing model performs semantic parsing on the request content, including word segmentation, entity recognition, and intent classification (mapping the user's intent to predefined intent type labels). Intent types are a predefined set of labels representing the essential category of the user's request, including at least seven categories: navigation, introduction, guidance, cancellation, resumption, reminder, and inquiry.
[0045] When an externally triggered event is identified as a system timed event or a sensor event, since these events already possess clear structured semantics, complex natural language parsing is unnecessary. The intent parsing module directly maps the event type identifier and time parameter to a predefined intent type based on preset mapping rules. These preset mapping rules are a set of mapping relationships maintained using a lookup table. For example, a timed cruise event maps to a guide intent, a fixed-point broadcast event maps to an introduction intent, a security patrol event maps to an inquiry intent, and a fall detection sensor event maps to a high-priority inquiry intent (triggering an emergency response). The time parameter is used to define the effective time window of the event, such as the specific trigger time in a timed task or the timestamp of a sensor event. This parameter is retained and passed to subsequent task generation stages.
[0046] After obtaining the intent type and extracted entity parameters, the intent parsing module matches the corresponding scenario family from the scenario registry based on the intent type and retrieves a task template adapted to the scenario family from the task template library. A scenario family is a high-level classification set that groups scenarios with similar business attributes and execution environments together, such as elderly care scenario families, exhibition hall guidance scenario families, hospital service scenario families, hotel service scenario families, etc. Each scenario family is associated with a specific set of business rules, location semantics, and default strategies. A task template is a predefined task structure blueprint, containing a task type field, an execution action sequence (e.g., navigation → pause → broadcast → wait for feedback → end), a list of required parameters (target location, broadcast text, timeout, etc.), a default timeout time, and a retry strategy. The intent parsing module fills the extracted entity parameters (target location coordinates, broadcast content, reminder time, etc.) into the corresponding parameter fields of the task template, generating a complete and executable candidate task.
[0047] After the intent resolution module generates candidate tasks, it submits the newly generated candidate tasks to the unified task pool management module, which then includes the candidate tasks in the unified task pool. The unified task pool is a global task collection maintained internally by the cloud orchestration service nodes, used to aggregate candidate tasks from different task sources.
[0048] In one possible implementation of this application embodiment, the cloud orchestration service node further includes a unified task pool management module and a decision-making module, wherein the unified task pool management module is configured as follows: Candidate tasks from different task sources are included in a unified task pool. Task sources include user immediate request sources, planned configuration task sources, timed policy task sources, and high-priority inserted event sources. To assign task attributes to each candidate task in the unified task pool, the task attributes include task source identifier, time window constraint, priority level and status label, and the status label includes pending execution, executing, completed, canceled, failed and timed out; Based on the task attributes and the current robot status, the system dynamically selects candidate tasks to be executed from the unified task pool and submits the selected candidate tasks to the decision-making module for decision-making.
[0049] Candidate tasks refer to executable task units generated after processing by the intent parsing module and awaiting scheduling decisions. Task sources include at least the following four categories: User-instantaneous request sources, referring to tasks generated from service requests sent to the robot in real time via voice or text; planned configuration task sources, referring to periodic or one-time tasks pre-configured by operations management personnel through the backend management system, such as scheduled patrols, fixed-point broadcasts, and course activities; timed strategy task sources, referring to tasks automatically triggered by internal system timers according to preset time strategies, such as service point standby rotation and scheduled security patrols; and high-priority insertion event sources, referring to tasks requiring immediate response triggered by anomaly detection, security alarms, or manual emergency designation. When the intent parsing module generates candidate tasks, or when the planned configuration module or timed strategy module generates new tasks, the candidate tasks are submitted to the unified task pool management module. The unified task pool management module incorporates all these candidate tasks into the unified task pool for centralized management and unified scheduling.
[0050] When adding candidate tasks to the unified task pool, the unified task pool management module assigns or refines task attributes for each candidate task, enabling subsequent task selection and adjudication based on these attributes. The task source identifier identifies the source from which the candidate task originated, and its value can be one of a user's immediate request, a planned configuration task, a timed policy task, or a high-priority inserted event. The time window constraint limits the effective execution time range of the candidate task, including the earliest executable time and the latest completion time. The priority level characterizes the execution priority of the candidate task relative to other tasks. Priority levels are set at least three levels based on task source and business importance: Normal level corresponds to regular user requests and planned tasks; High level corresponds to important tasks designated by operations; and Emergency level corresponds to security events and exception handling tasks. The status label marks the current status of the candidate task in its lifecycle, including: pending execution, executing, completed, canceled, and failed. The status label is updated in real-time by the unified task pool management module based on feedback from the decision-making module and the underlying execution system throughout the task's lifecycle.
[0051] At the start of each decision cycle (triggered by an external event or a timed polling mechanism), the unified task pool management module evaluates and filters all candidate tasks in the unified task pool to select the most suitable candidate task for execution at the current moment and submits it to the decision-making module. The specific selection logic includes the following steps: First, the unified task pool management module iterates through all candidate tasks in the unified task pool with the status tag "to be executed," performing a first round of filtering based on the time window constraints of the task attributes, eliminating candidate tasks that do not meet the time window constraints at the current moment (e.g., not yet reaching the earliest execution time or exceeding the latest completion time). Then, for the candidate tasks filtered through the time window, a second round of filtering is performed based on the current robot status: if the target point of the candidate task is unreachable from the current robot location (e.g., the path is blocked or the target point is outside the map range), the task is marked as temporarily unexecutable; if the required equipment capabilities of the candidate task are not supported by the current robot model or the relevant modules are offline, the task is marked as temporarily unexecutable; if the power required for the execution of the candidate task exceeds the robot's current remaining power, the task is marked as temporarily unexecutable. Finally, among the remaining executable candidate tasks, the unified task pool management module sorts them from highest to lowest priority. When there are multiple candidate tasks with the same priority, they are further sorted according to the order of task generation time, and the candidate task with the highest priority is selected as the current candidate task to be executed.
[0052] After selecting candidate tasks to be executed, the unified task pool management module submits the selected candidate tasks, along with their complete task attributes, to the decision-making module. Simultaneously, the unified task pool management module updates the status label of the selected candidate tasks to "in execution" and retains other unselected tasks in the unified task pool, awaiting re-evaluation and selection in the next decision cycle.
[0053] In one possible implementation of this application embodiment, the cloud orchestration service node further includes a combined task processing module, which is configured to perform the following operations after the intent parsing module identifies the intent type and matches the scene family, and before the decision-making module makes a decision: When an external triggering event contains multiple sub-targets or composite instructions, the external triggering event is broken down into multiple executable sub-tasks; Establish execution relationships for multiple executable subtasks, including sequential execution, conditional execution, and failure degradation; Map multiple executable subtasks and their execution relationships to a standard task batch.
[0054] The combined task processing module performs a composite test on the parsing results output by the intent parsing module. A composite instruction refers to a natural language instruction that expresses multiple execution goals or contains conditional logic within a single request. For example, "Take me to the rehabilitation area and introduce it to me" contains both the goals of guiding and introducing; "Return to the service point and broadcast a notification after inspecting the activity room" contains a sequential execution relationship; and "Go and greet the receptionist if someone is there, otherwise continue patrolling" contains conditional branching logic. The combined task processing module determines whether the current triggering event contains multiple sub-goals or composite instructions by detecting the number of intent types output by the intent parsing module, whether the extracted entity parameters contain multiple target points or action sequences, and whether there are sequential or conditional connectors in the request text. If the detection result is negative, meaning the external triggering event is only a simple instruction with a single intent, the combined task processing module does not perform any additional processing and directly passes it to the decision-making module.
[0055] If the current triggering event contains multiple sub-goals or composite instructions, the entire external triggering event is decomposed into multiple executable subtasks based on the intent type sequence and entity parameter set output by the intent parsing module. An executable subtask is the smallest task unit that can be executed independently at the underlying execution system. Each subtask has a single, clear execution goal and complete execution parameters. The decomposition rules are as follows: For composite instructions containing multiple independent goals (such as "Take me to the rehabilitation area and introduce it to me"), the combined task processing module decomposes the "Navigate to the rehabilitation area" corresponding to the navigation intent into one subtask, and the "Provide a voice introduction in the rehabilitation area" corresponding to the introduction intent into another subtask. Each subtask is bound to independent execution parameters (target point coordinates, broadcast text content, etc.). For compound instructions containing conditional logic (such as "go and greet the receptionist if someone is there, continue patrolling if no one is there"), the combined task processing module breaks down the condition judgment into a condition detection subtask, "greet the receptionist" into an execution subtask, and "continue patrolling" into another execution subtask, and binds judgment parameters (detection target location, detection object type, etc.) to the condition detection subtask. For compound instructions containing sequential actions (such as "return to the service point and broadcast a notification after patrolling the activity room"), the combined task processing module breaks down "patrol the activity room" into a subtask, "return to the service point" into a second subtask, and "broadcast a notification" into a third subtask, and binds corresponding execution parameters to each subtask.
[0056] After decomposing the subtasks, the composite task processing module establishes clear execution relationships between the executable subtasks based on the semantic logic in the original composite instructions. Sequential execution means that subtasks are executed in a fixed order; the next subtask can only be started after the previous one has successfully completed. Conditional execution means that the execution of a subtask depends on the judgment result of a precondition check subtask, and different execution branches are selected based on the judgment result. Failure degradation means that when a subtask fails, the system does not directly cause the overall composite task to fail, but instead executes a degradation substitution action to maintain system availability.
[0057] A standard task batch refers to a collection of tasks containing multiple subtasks and their orchestration logic. This batch is treated as a single combined task and submitted to subsequent processes.
[0058] In one possible implementation of this application embodiment, the decision-making module is configured to make decisions on candidate tasks and generate a scheduling strategy, including: The precondition verification unit is used to perform precondition verification on the candidate task based on the current robot state and the task parameters of the candidate task; The task interruption judgment unit is used to determine whether to interrupt the current active task and output the task interruption flag when the precondition verification passes, based on the priority of the candidate task and the priority of the current active task in the active task information, and in combination with the interruptible status of the current active task. The scheduling policy generation unit is used to determine the scheduling policy based on the task interruption flag and update the current task orchestration based on the scheduling policy; wherein, when the task interruption flag indicates that the currently active task is interrupted, the scheduling policy is the task recovery policy; when the task interruption flag indicates that the currently active task is not interrupted, the scheduling policy is the task replacement policy or the task waiting policy.
[0059] The precondition verification includes target point reachability verification, equipment capability support verification, and battery adequacy verification. Target point reachability verification uses the robot's real-time positioning coordinates and map data, combined with the target point coordinates in the candidate task, to calculate whether a feasible path exists from the current location to the target point using a path planning algorithm. It also checks whether the target point is located within a traversable area on the map and is not blocked by obstacles. If no feasible path can be planned or the target point itself does not exist on the map, the target point reachability verification fails. Equipment capability support verification checks whether the execution capability type required by the candidate task is in the robot's current equipment capability list. Battery adequacy verification compares the estimated execution time and minimum required battery threshold of the candidate task with the robot's remaining battery percentage to determine whether the current battery can support the entire execution process of the candidate task.
[0060] When all three checks pass, the precondition verification unit determines that the precondition verification has passed and outputs the candidate task to the task interruption judgment unit for further processing. When any one check fails, the precondition verification unit determines that the precondition verification has failed, rejects the candidate task and records the reason for rejection (specifically marked as location unreachable, equipment capability not supported, or insufficient power). The candidate task is returned to the unified task pool or terminated directly and will not enter the subsequent adjudication process.
[0061] The task interruption judgment unit first obtains the currently active tasks from the active task information in the context management module. Currently active tasks refer to tasks that are currently executing in the underlying execution system and have not yet been completed. Each active task is assigned a task source identifier, priority level, and interruptibility status during scheduling. The interruptibility status is a flag indicating whether the active task is allowed to be interrupted and suspended during execution by a higher-priority candidate task.
[0062] If the priority level of the candidate task is lower than or equal to the priority level of the currently active task, the task interruption judgment unit directly determines that the currently active task will not be interrupted, and the output task interruption flag is "no interruption".
[0063] If the priority level of the candidate task is higher than that of the currently active task, the task interruption determination unit further reads the interruptibility status flag of the currently active task. If the flag indicates that the currently active task is uninterruptible, the task interruption determination unit will still determine not to interrupt the currently active task, even if the candidate task has a higher priority, and the output task interruption flag will be "No Interruption" to avoid violating business security constraints. If the flag indicates that the currently active task is interruptible, the task interruption determination unit will determine to interrupt the currently active task, and the output task interruption flag will be "Interrupted," indicating that the candidate task is allowed to preempt the execution resources of the currently active task.
[0064] A task recovery strategy is a strategy that records complete information about interrupted currently active tasks so that interrupted tasks can be handled according to a specified recovery method after candidate tasks have been completed.
[0065] The task replacement strategy involves canceling or discarding candidate tasks while keeping the currently active task running without any changes. The task waiting strategy involves keeping candidate tasks in the unified task pool and maintaining their status as pending execution, waiting for the next decision cycle after the currently active task has completed before they are reselected and submitted for adjudication by the unified task pool management module. The scheduling strategy generation unit judges based on the time window constraints of candidate tasks: if the time window constraints are relatively lenient, the task waiting strategy is selected; if the time window constraints are close to the deadline, the task replacement strategy is selected and the candidate task is marked as canceled.
[0066] In one possible implementation of this application embodiment, the cloud orchestration service node further includes a task generation module. When the currently active task is interrupted to execute a candidate task, the task generation module is configured as follows: Retrieve task templates that match candidate tasks from the task template library, populate the task templates with execution parameters to generate standard tasks, and attach task recovery strategies to the standard tasks; The task recovery strategy is used to instruct the currently active task to be recovered, skipped, postponed, or terminated after the candidate task has been completed.
[0067] The task template library is a pre-built structured repository within the cloud orchestration service nodes. Internally, it categorizes and indexes execution blueprints for various tasks according to intent type and scenario family. The general structure of a task template includes at least the following fields: task type, execution action sequence (a set of atomic actions listed in execution order, such as [navigation → pause → broadcast → wait for feedback → end]), required parameter list, default timeout, and retry strategy. The task generation module uses the intent type tag carried by the candidate task as the retrieval key to perform a matching query in the task template library, locate the task template record corresponding to that intent type, and extract all structural information of that template.
[0068] After obtaining a matching task template, the task generation module fills the corresponding fields in the required parameter list of the task template with the entity parameters and context information carried in the candidate task.
[0069] When an active task is interrupted to execute a candidate task, the task generation module, while generating a standard task, appends the task recovery policy output by the scheduling policy generation unit to the scheduling policy field of the standard task. The task recovery policy includes the following information fields: the unique identifier of the interrupted task, the reason for the interruption, and the recovery method. Recovery methods include: resume execution (continue execution from the pause point after interruption); skip (do not execute the original task after interruption); postpone (postpone execution to a later time slot after interruption); and end (directly end the original task and mark it as completed after interruption). The specific choice of recovery method is determined by the scheduling policy generation unit based on the task type, task source, and preset rules of the interrupted task. For example, a user-generated task created on an immediate request is usually resumed after being interrupted, while a scheduled cruise task may be postponed.
[0070] In one possible implementation of this application embodiment, the cloud orchestration service node further includes a task verification module, which is configured as follows: Before the standard task is sent to the robot access adapter, the standard task output by the task generation module is validated for legality. The validation includes data format validation and field integrity validation. Once the validity check passes, the standard task will be sent to the robot's access adapter; If the validity check fails, the standard task will not be issued, the verification failure log will be recorded, and the retry mechanism will not be triggered.
[0071] Data format validation checks whether the data types of each field in the standard task conform to the communication protocol specifications agreed upon between the cloud and the robot's access adapter. Field integrity validation checks whether all fields marked as required in the standard task are filled and not empty; required fields are pre-defined by the communication protocol between the cloud and the robot's access adapter.
[0072] If either the data format validation or the field integrity validation fails, the task validation module determines that the standard task has failed the validity validation.
[0073] The verification failure log contains the following information: a unique identifier for the failed standard task, the specific type of verification failure, a detailed description of the reason for the failure, the timestamp of the verification occurrence, and the identifier of the verification module that performed the verification. This log is written to the decision log service of the cloud orchestration service node for subsequent querying and analysis of the root cause of the problem by operations and maintenance personnel and development and debugging personnel.
[0074] One possible implementation of this application embodiment includes a robot-side access adapter comprising: The local compliance verification unit is used to verify the standard task execution format, physical space and location, equipment capability, and security rules. The local compliance judgment unit is used to determine whether the standard task has passed the local compliance verification when all the verification results of the local compliance verification unit pass. The task submission unit is used to submit standard tasks to the underlying execution system through the controlled task access interface when local compliance verification is passed.
[0075] Format validation checks the data structure of the standard task to ensure its integrity and the correctness of field types. This ensures the standard task can be correctly read and parsed by the robot's local parsing module, preventing parsing errors due to data transmission corruption or format incompatibility. Physical space and location validation verifies the existence of target points in the standard task on the map, their location within the robot's accessible physical area, and their absence of obstacles, based on the robot's current coordinates and locally stored map data. This ensures the robot's navigation module can effectively drive the robot to the target location. Equipment capability validation checks whether the execution capabilities required for the standard task are supported by the current robot model and whether the corresponding hardware modules (such as voice broadcast speakers and facial recognition cameras) are online and ready, ensuring the robot has the physical hardware to execute the task. Safety rule validation checks whether the standard task violates locally preset safety policies, such as whether the target point is located within a system-marked restricted area, whether the execution path passes through a dangerous area, and whether the task actions conflict with currently active emergency safety instructions. This ensures the task execution will not pose a safety hazard to the robot itself, surrounding personnel, or the environment.
[0076] When format verification, physical space and location verification, equipment capability verification, and security rule verification all return "pass," the local compliance judgment unit determines that the standard task has passed the local compliance verification and outputs the verification result to the task submission unit for further processing. When any of the above four verifications returns "fail," the local compliance judgment unit determines that the standard task has failed the local compliance verification, records the specific verification item and reason for failure, prevents the standard task from entering the underlying execution system, and feeds back the reason for verification failure to the cloud orchestration service node.
[0077] The controlled task access interface is a standardized interaction channel between the robot's end-user access adapter and the robot's underlying execution system. This interface follows the established task submission protocol of the underlying execution system, uses data formats and calling methods recognized by the underlying execution system for task injection, and ensures that new tasks enter the execution queue according to the system's standardized process, without intruding on or bypassing the existing execution process. After receiving a standard task, the underlying execution system drives the corresponding hardware modules to execute item by item according to the sequence of execution actions in the task.
[0078] In one possible implementation of this application embodiment, the robot-end access adapter further includes: The failure feedback unit is used to report the reason for the failure to the cloud orchestration service node when the local compliance judgment unit determines that the local compliance verification has failed. The reason for the failure includes the failed verification item and the corresponding parameter value. After receiving the verification failure reason, the cloud orchestration service node controls the task generation module to regenerate the standard task based on the verification failure reason and sends it to the robot's access adapter. The retry control unit records the number of retries that are regenerated and reissued. When the number of retries exceeds the preset retry threshold, a local fallback strategy is triggered.
[0079] The reason for verification failure includes at least two fields: the failed verification item (indicating which item or items failed in format verification, physical space and location verification, device capability verification, or safety rule verification) and the corresponding parameter value (recording the specific data that caused the verification failure, such as the target point coordinates exceeding the map boundary, the requested voice broadcast capability not supported by the current robot model, or the target point being located within a restricted area). The failure feedback unit encapsulates the above verification failure reasons into a failure feedback message according to the data format agreed upon between the cloud and the client, and reports it to the cloud orchestration service node through a secure communication channel.
[0080] The cloud orchestration service node takes appropriate adjustments based on the type of failure, controlling the task generation module to regenerate the standard task: if the physical space and location verification fails (e.g., the target location is unreachable), the task parameters are refilled by selecting the nearest reachable alternative location from the map data; if the device capability verification fails (e.g., the required capability is not currently supported by the robot), the task is reassigned to another robot with that capability or the unsupported actions in the task are replaced with supported action sequences; if the safety rule verification fails (e.g., the target location is located in a restricted area), the task is rejected and the user is notified that it cannot be executed. After the task generation module completes the regeneration, the corrected standard task is reissued to the same robot's access adapter.
[0081] The retry control unit continuously tracks the cumulative number of retries for the same original task from its initial deployment to the current moment. The preset retry threshold is the maximum number of retries pre-configured by the system to prevent unlimited consumption of system resources due to repeated failures of the same task. When the verification passes without exceeding the preset retry threshold, the re-deployed standard task normally enters the local compliance verification process.
[0082] In one possible implementation of this application embodiment, the robot-end access adapter further includes a fallback execution unit, used to trigger a local fallback strategy when any of the following conditions are met: Communication between the cloud orchestration service node and the robot's access adapter is unavailable or has timed out. The local compliance assessment unit determines that the local compliance verification failed and the number of retries exceeded the preset retry threshold; The standard task returned by the cloud orchestration service node is an illegal task; The local fallback strategy includes at least one of the following: keeping the currently active task running, controlling the robot to return to the preset service point, or controlling the robot to enter the default standby mode.
[0083] Communication unavailability means that the robot's access adapter detects through network connection detection mechanisms (such as heartbeat packets) that the network connection between it and the cloud orchestration service node has been lost, making it unable to send uplink data or receive downlink data. Communication timeout means that after the robot's access adapter sends a request to the cloud orchestration service node, it does not receive a response from the cloud within a preset timeout period, indicating that the cloud service may be overloaded, down, or experiencing high network latency. The robot's access adapter continuously maintains the heartbeat connection with the cloud through a built-in communication status monitoring thread and records the response time for each request. Once the number of consecutive heartbeat failures reaches a preset number or a single request times out, the trigger condition is determined to be met.
[0084] An illegal task refers to data returned by the cloud orchestration service node that, after being parsed by the robot's access adapter, is found to be inconsistent with the basic data specifications agreed upon by the cloud and the client. For example, the data structure may be incomplete, leading to parsing failure; the task type may be an enumeration value not defined by the system; or key execution parameters may be missing, making execution impossible.
[0085] The local fallback strategy is an emergency response plan that the robot's on-device adapter executes autonomously locally when it cannot rely on cloud orchestration service nodes for normal task decision-making. Its purpose is to ensure the robot does not become uncontrollable due to cloud unavailability or task anomalies. The fallback execution unit selects at least one of the following three fallback strategies to execute based on the robot's current operational status: Keep the currently active task running: When the robot has an active task that is not dependent on subsequent instructions from the cloud, the fallback execution unit controls the underlying execution system to keep the currently active task running without interruption or pause until the task is completed naturally or reaches its own timeout condition. This strategy is suitable for scenarios where the current task itself has a complete execution sequence and does not require cloud intervention.
[0086] Controlling the robot to return to the preset service point: When the robot has no active task or the current task can be safely interrupted, the fallback execution unit obtains the coordinates of the preset service point configured by the system (such as the location of the robot's charging pile, standby docking point or designated duty area), and controls the navigation module of the underlying execution system to drive the robot to automatically return to the preset service point and dock, ensuring that the robot returns to a safe fixed position to wait for cloud recovery or manual intervention.
[0087] Controlling the robot to enter default standby mode: When the robot's current position is safe and no movement is required, the fallback execution unit controls the underlying execution system to switch the robot's operating mode to default standby mode. Default standby mode means that the robot stops all active actions (stops navigation movement, stops voice broadcasting, stops interactive operations), only maintaining basic system operation and network monitoring, waiting for the cloud to restore the connection or for manual wake-up, in order to minimize energy consumption and equipment wear.
[0088] While executing the fallback strategy, the fallback execution unit writes the reason for triggering the fallback, the trigger time, and the type of fallback strategy executed to the local log. It also notifies on-site personnel that the robot has entered fallback mode via a local alarm mechanism (such as flashing indicator lights or a warning sound from the local speaker), allowing them to promptly understand the robot's status and intervene as needed. Once the fallback execution unit confirms that cloud communication has been restored or a valid new task has been received, the robot can exit fallback mode and resume normal scheduling and execution.
[0089] The above are only some embodiments of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.
Claims
1. A robot task orchestration system, characterized in that, This includes cloud orchestration service nodes and robot-end access adapters; The cloud orchestration service node is used to obtain a context snapshot of the current operating environment. The context snapshot includes the current robot status, active task information, external triggering events, current task orchestration status, and scheduled execution plan. Based on the context snapshot and preset scheduling strategy, decision-making is made on candidate tasks to generate standard tasks; The standard task is sent to the robot's terminal access adapter; The robot-side access adapter is used to perform local compliance verification on the standard task; when the local compliance verification passes, it controls the underlying execution system to execute the standard task.
2. The robot task orchestration system according to claim 1, characterized in that, The cloud orchestration service node includes an intent parsing module, which is configured as follows: When the external triggering event is a natural language request, the natural language request is semantically parsed through the intent parsing model to identify the intent type; When the external triggering event is a sensor event or triggers the timed execution plan, it is mapped to the corresponding predefined intent type according to the event type and time parameters; Based on the identified intent type, match scene families and task templates to generate candidate tasks, and incorporate the newly generated candidate tasks into a unified task pool.
3. The robot task orchestration system according to claim 1, characterized in that, The cloud orchestration service node also includes a unified task pool management module and a decision-making module. The unified task pool management module is configured as follows: Candidate tasks from different task sources are included in a unified task pool, including user immediate request sources, planned configuration task sources, timed policy task sources, and high-priority inserted event sources. Assign task attributes to each candidate task in the unified task pool. The task attributes include task source identifier, time window constraint, priority level and status label. The status label includes pending execution, executing, completed, canceled, failed and timed out. Based on the task attributes and the current robot state, a candidate task to be executed is dynamically selected from the unified task pool, and the selected candidate task is submitted to the decision-making module for decision-making.
4. The robot task orchestration system according to claim 1, characterized in that, The cloud orchestration service node also includes a combined task processing module, which is configured to perform the following operations after the intent parsing module identifies the intent type and matches the scene family, and before the decision-making module makes a decision: When the external triggering event contains multiple sub-targets or composite instructions, the external triggering event is decomposed into multiple executable sub-tasks; Establish execution relationships for the plurality of executable subtasks, wherein the execution relationships include sequential execution, conditional execution, and failure degradation; The plurality of executable subtasks and the execution relationship are mapped to standard task batches.
5. The robot task orchestration system according to claim 1, characterized in that, The decision-making module is configured to make decisions on the candidate tasks and generate scheduling strategies, including: The precondition verification unit is used to perform precondition verification on the candidate task based on the current robot state and the task parameters of the candidate task; The task interruption judgment unit is used to determine whether to interrupt the current active task and output a task interruption flag when the precondition verification passes, based on the priority of the candidate task and the priority of the current active task in the active task information, and in combination with the interruptible status of the current active task. The scheduling policy generation unit is used to determine a scheduling policy based on the task interruption identifier and update the current task orchestration based on the scheduling policy; wherein, when the task interruption identifier indicates that the currently active task is interrupted, the scheduling policy is a task recovery policy; when the task interruption identifier indicates that the currently active task is not interrupted, the scheduling policy is a task replacement policy or a task waiting policy.
6. The robot task orchestration system according to claim 5, characterized in that, The cloud orchestration service node also includes a task generation module, which is configured to: When the currently active task is interrupted to execute the candidate task, the task generation module is configured to: Obtain a task template matching the candidate task from the task template library, fill the task template with execution parameters to generate the standard task, and attach the task recovery strategy to the standard task; The task recovery strategy is used to instruct the currently active task to perform recovery, skip, postpone, or end operations after the candidate task has been completed.
7. The robot task orchestration system according to claim 1, characterized in that, The cloud orchestration service node also includes a task verification module, which is configured as follows: Before the standard task is sent to the robot access adapter, the standard task output by the task generation module is validated for legality. The validation includes data format validation and field integrity validation. When the legality verification passes, the standard task is sent to the robot's access adapter; If the validity check fails, the standard task will not be issued, a check failure log will be recorded, and a retry mechanism will not be triggered.
8. The robot task orchestration system according to claim 1, characterized in that, The robot-end access adapter includes: The local compliance verification unit is used to perform format verification, physical space and location verification, equipment capability verification and security rule verification on the standard tasks. A local compliance judgment unit is used to determine that the standard task has passed the local compliance verification when all the verification results of the local compliance verification unit have passed. The task submission unit is used to submit the standard task to the underlying execution system through the controlled task access interface when the local compliance verification is passed.
9. The robot task orchestration system according to claim 1, characterized in that, The robot-end access adapter also includes: The failure feedback unit is used to provide feedback on the reason for the failure to the cloud orchestration service node when the local compliance judgment unit determines that the local compliance verification has failed. The reason for the failure includes the failed verification item and the corresponding parameter value. After receiving the verification failure reason, the cloud orchestration service node controls the task generation module to regenerate the standard task according to the verification failure reason and sends it to the robot terminal access adapter. The retry control unit is used to record the number of retries that are regenerated and reissued. When the number of retries exceeds the preset retry threshold, a local fallback strategy is triggered.
10. The robot task orchestration system according to claim 1, characterized in that, The robot-end access adapter also includes a fallback execution unit, used to trigger a local fallback strategy when any of the following conditions are met: Communication between the cloud orchestration service node and the robot terminal access adapter is unavailable or has timed out. The local compliance assessment unit determines that the local compliance verification failed and the number of retries exceeded the preset retry threshold; The standard task returned by the cloud orchestration service node is an illegal task; The local fallback strategy includes at least one of the following: keeping the currently active task running, controlling the robot to return to a preset service point, or controlling the robot to enter a default standby mode.
