Task feature based collaborative scheduling method for edge side cerebellum and cloud side cerebrum

Abstract

This invention discloses a collaborative scheduling method between an edge-side embodied cerebellum and a cloud-based brain based on task characteristics. The method includes: acquiring task characteristic information of a task to be executed, wherein the task characteristic information includes at least the task's real-time requirements, computational complexity, and accuracy requirements; combining the action execution capability and execution latency threshold of the edge-side embodied cerebellum to generate a task scheduling decision, allocating the task to either the edge-side embodied cerebellum or the cloud-based brain for execution; the edge-side embodied cerebellum collecting task execution status information during task execution and sending the task execution status information to the cloud-based brain; and the cloud-based brain updating the scheduling strategy for subsequent tasks based on the task execution status information, thereby achieving closed-loop collaborative scheduling between the edge and the cloud. This invention can dynamically adjust the task execution position according to task characteristics, improve task execution efficiency, reduce overall system latency, and is suitable for collaborative scheduling applications in multiple tasks and scenarios.

Description

Technical Field

[0001] This invention relates to the field of intelligent system task scheduling technology, specifically to a collaborative scheduling method based on task characteristics between an edge-side embodied execution unit (edge ​​cerebellum) and a cloud computing unit (cloud brain), applicable to scenarios such as embodied intelligent devices, robot systems, and edge computing terminals. Background Technology

[0002] With the widespread use of embody smart devices and edge computing devices, systems typically include both edge execution units and cloud computing units. In existing technologies, edge-cloud collaboration mainly suffers from the following problems: The scheduling rules are fixed and mostly use preset rules or static configurations, making it impossible to dynamically adjust the task allocation method according to task characteristics. The lack of a closed-loop feedback mechanism between the cloud and the edge makes it difficult for the cloud to obtain the task execution status and abnormal information on the edge in a timely manner, which makes it difficult to continuously optimize the scheduling strategy. Insufficient hardware adaptation capability at the edge: Existing scheduling methods typically do not consider differences in computing power, I / O latency, and instruction characteristics of edge chips, which affects real-time execution. Multimodal tasks are scheduled independently of each other. Action tasks, vision tasks, and sensing tasks are often processed separately, making it difficult to optimize the overall system performance. Therefore, it is necessary to propose an edge-cloud collaborative scheduling method based on task characteristics to improve the overall operating efficiency and stability of the system. Summary of the Invention

[0003] The purpose of this invention is to provide: Task-feature-based collaborative scheduling method between end-to-end embodied cerebellum and cloud-based brain This addresses the problems of fixed cloud scheduling strategies, insufficient feedback, poor hardware adaptability, and low efficiency of multimodal task collaboration in existing technologies.

[0004] Technical solution To achieve the above objectives, the present invention adopts the following technical solution. A task-feature-based collaborative scheduling method between the end-lateral cerebellum and a cloud-based brain includes the following steps: S1: Task Acquisition and Task Feature Extraction The end-to-end cerebellum acquires the task to be performed and extracts task feature parameters from the task, the task feature parameters including at least: Real-time requirement parameters Accuracy Requirement Parameters Computational complexity parameters Data interaction volume parameters Task type parameters The task type parameter is used to distinguish between action execution tasks, visual recognition tasks, sensor processing tasks, or combinations thereof. S2: Acquisition of End-Side Capability Parameters The end-side embodied cerebellum acquires the capability parameters of the current end-side execution environment, said capability parameters including at least: Edge chip computing power parameters Instruction set supports parameters Storage resource parameters Input / output delay parameters The above capability parameters are used to characterize the real-time processing capability of the task executed on the edge. S3: Initial Scheduling Decision Generation The end-side embodied cerebellum generates an initial scheduling decision based on the task characteristic parameters and the end-side capability parameters. This initial scheduling decision is used to indicate: The task is executed on the device side. Tasks are executed in the cloud. Tasks are split and executed between the edge and the cloud. S4: Cloud-based collaborative scheduling and processing When a task is assigned to the cloud or a collaborative execution between the edge and cloud, the cloud-based system receives task characteristic parameters and edge capability parameters, and generates a cloud scheduling strategy based on preset scheduling rules to determine: Cloud computing resource allocation methods Task processing order The result format returned by the task to the end side S5: End-side task execution and status acquisition The end-sided embodied cerebellum executes tasks according to the scheduling decision and collects task execution status parameters during task execution. The task execution status parameters include at least: Execution delay Execution successful status Abnormal information S6: Edge-to-Cloud Closed-Loop Feedback and Scheduling Update The end-side embodied cerebellum sends the task execution status parameters to the cloud brain. Based on the task execution status parameters, the cloud brain updates the scheduling rules and uses the updated scheduling strategy for subsequent task scheduling. S7: Multimodal Task Cooperative Scheduling When the task is multimodal, the end-to-end embodied cerebellum coordinates the scheduling of action tasks, visual tasks, and sensor tasks according to the correlation between different modalities, so as to optimize the overall task execution performance.

[0005] Beneficial effects Compared with the prior art, the present invention has at least the following beneficial effects: Implement task-feature-driven dynamic scheduling, which schedules tasks based on real-time performance, accuracy, and computational complexity, avoiding resource waste caused by fixed rules. A closed-loop optimization mechanism is formed between the cloud and the edge to continuously optimize the scheduling strategy based on feedback from the edge side, thereby improving system stability. To improve the adaptability of edge chips, the scheduling process takes into account the characteristics of edge hardware to ensure low-latency execution.

[0006] It supports multimodal task collaborative processing and unified scheduling of different types of tasks, thereby improving the overall system efficiency.

[0007] Example Example 1 In a embodied robotic system: Motion control tasks are marked as high real-time tasks; Visual recognition tasks are labeled as high computational complexity tasks; The end-lateral embodied cerebellum allocates motor control tasks to the end side for execution based on task characteristics; Distribute visual recognition tasks to the cloud for execution; The cloud will return the recognition results to the device to assist in the execution of actions.

[0008] Through the edge-cloud closed-loop feedback mechanism, the system continuously adjusts its scheduling strategy to adapt to different working environments.

[0009] Industrial applicability This invention can be widely applied to: Embossed intelligent robot Edge computing terminal Automated equipment Intelligent manufacturing system It has promising prospects for industrial applications. Attached Figure Description To more clearly illustrate the technical solution of the present invention, the drawings used in the specification will be briefly described below. Obviously, the drawings described below are merely illustrative of the present invention; those skilled in the art can obtain other drawings based on these drawings without any creative effort. Figure 1 is a schematic diagram of the overall architecture of the end-side embodied cerebellum and cloud brain collaborative scheduling system of the present invention. Figure 2 is a schematic diagram of the functional module structure of the end-sided cerebellum of the present invention; Figure 3 is a schematic diagram of the functional module structure of the cloud brain of the present invention; Figure 4 is a flowchart of the task-feature-based end-side embodied cerebellum and cloud brain collaborative scheduling method of the present invention. Figure 5 is a schematic diagram of the closed-loop feedback scheduling mechanism between the end-side embodied cerebellum and the cloud brain of the present invention. Figure 6 is a schematic diagram of the multimodal task collaborative scheduling of the present invention.

Claims

1. A collaborative scheduling method between the end-to-end cerebellum and a cloud-based brain based on task characteristics, characterized in that, Includes the following steps: 1) Obtain the task feature information of the task to be executed, wherein the task feature information includes at least the task's real-time requirements, computational complexity, and accuracy requirements; 2) Based on the task feature information and combined with the action execution delay threshold of the edge-side embodied cerebellum, a task scheduling decision is generated to allocate the task to the edge-side embodied cerebellum or the cloud brain for execution. 3) The end-side embodied cerebellum collects task execution status information during task execution, and the task execution status information includes at least execution delay, resource usage, or abnormal status information; 4) Send the task execution status information to the cloud-based brain; 5) The cloud-based brain updates the scheduling strategy for subsequent tasks based on the task execution status information.

2. The method as described in claim 1, characterized in that, The task scheduling decision dynamically determines whether the task is executed on the edge embodied cerebellum or in the cloud brain based on the task's real-time requirements, computational complexity, and accuracy requirements.

3. The method as described in claim 1, characterized in that, When the cloud-based brain updates the scheduling strategy for subsequent tasks based on the task execution status information, it includes at least adjusting the allocation position of subsequent tasks based on task execution delay or abnormal status information.

4. The method as described in claim 1, characterized in that, The task scheduling decision is further adjusted based on the computing power, instruction set characteristics, or input / output latency of the chip used in the edge-side embodied cerebellum.