A method and supporting system for dynamic scheduling of social robot swarms

By using a dynamic scheduling method for social robot clusters, combined with device status and task type, precise matching of devices and tasks is achieved, solving the problems of device overload and resource waste, and improving resource utilization and task execution stability.

CN123114748APending Publication Date: 2026-11-17HEFEI YUYI KEXIN TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611296299.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-25
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, the scheduling methods for social robot clusters lack consideration for device performance degradation and load changes, resulting in unreasonable matching of devices and tasks, low resource utilization, and an inability to respond to sudden tasks in advance, which easily leads to task congestion and resource contention.

Method used

A dynamic scheduling method for social robot clusters is adopted, which achieves accurate assessment of equipment status and differentiated scheduling of resources by short-term load prediction, construction of multi-dimensional capability profiles, calculation of matching degree based on attenuation correction terms, and two-level buffer resource allocation.

Benefits of technology

It improves the matching degree between equipment and tasks, avoids equipment overload or resource waste, enhances task completion quality and resource utilization efficiency, and strengthens the stability and adaptability of the cluster.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123114748A_ABST
    Figure CN123114748A_ABST
Patent Text Reader

Abstract

This application relates to the field of intelligent robot cluster task scheduling technology, and discloses a dynamic scheduling method and supporting system for social robot clusters. The scheduling method is executed by a scheduling device and includes the following steps: S1, short-term load prediction and preliminary candidate screening: collect the historical traffic and task complexity distribution of the task queue over the past N minutes, and output the peak task and complexity distribution for the next T seconds through a time series prediction model; pre-traverse all idle robots, screen M robots with an initial matching degree higher than the basic threshold and mark them as preliminary candidate robots, and lock the basic resource quota; allocate tasks by comprehensively considering robot operating losses and real-time working status, changing the previous single judgment method that only looks at instantaneous resources, effectively improving the situation of unreasonable matching between equipment and tasks, avoiding equipment overload or resource idle waste, and steadily improving the overall task completion quality and resource utilization efficiency of the cluster.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of intelligent robot swarm task scheduling technology, and in particular to a dynamic scheduling method and supporting system for social robot swarms. Background Technology

[0002] Social robot swarms need to continuously process a large number of diverse and frequently occurring social tasks. Over long-term operation, robot devices will experience performance degradation and dynamic load changes. Currently, most mainstream robot swarm scheduling methods use fixed resource matching and unified scheduling rules, which are significantly insufficient when dealing with complex and ever-changing task scenarios, resulting in poor overall adaptability.

[0003] On the one hand, traditional scheduling methods allocate tasks based solely on the robot's current remaining resources, without considering the actual state of the equipment, such as performance degradation and fatigue from long-term operation, resulting in an incomplete assessment of the robot's working capacity. This easily leads to a mismatch between tasks and equipment performance, causing some equipment to work under overload and fail to complete tasks, while also leaving high-quality equipment resources idle and wasting, thus reducing the overall task execution efficiency and resource utilization of the cluster.

[0004] On the other hand, traditional scheduling models lack the ability to predict task traffic in advance and cannot reserve available device resources in advance. When faced with a sudden surge in tasks, task congestion and resource contention are likely to occur. At the same time, traditional scheduling uses the same resource allocation and recycling methods for different types of tasks, such as heavy and fast tasks, which cannot adapt to the resource usage needs of different tasks and cannot flexibly respond to various business scenarios, resulting in poor cluster scheduling stability and adaptability. Summary of the Invention

[0005] To address the issues raised above regarding traditional robot scheduling, which relies solely on instantaneous resource assessment, suffers from poor device-task matching, lacks pre-emptive traffic prediction and differentiated resource scheduling, and is prone to task congestion and weak scenario adaptability in high-concurrency scenarios, this application provides a dynamic scheduling method and supporting system for social robot clusters.

[0006] This application provides a dynamic scheduling method and supporting system for social robot clusters, which adopts the following technical solution:

[0007] A method and supporting system for dynamic scheduling of social robot clusters, wherein the scheduling method is executed by a scheduling device and includes the following steps:

[0008] S1. Short-term load prediction and preliminary candidate screening: Collect the historical traffic and task complexity distribution of the task queue in the past N minutes, and output the task peak and complexity distribution in the next T seconds through the time series prediction model; traverse all idle robots in advance, screen M robots with an initial matching degree higher than the basic threshold and mark them as preliminary candidate robots, and lock the basic resource quota.

[0009] S2. Construct a multi-dimensional robot capability profile incorporating resource attenuation coefficients: After the target task is enqueued, calculate the task complexity score based on the task type, the number of physical servers required, and the content generation method. Real-time collection of remaining CPU quota for robots Available memory Historical success rate Calculate the resource decay coefficient based on the duration of continuous operation. , generate capability portrait Prot = ( , , , );

[0010] S3. Calculate the corrected matching degree based on the attenuation correction term to determine the target social robot: Calculate the matching degree between the candidate robot and the target task using the corrected matching degree formula, which is:

[0011]

[0012] in , The scores are obtained from the preset score database table. W1, W2, and W3 are dynamic resource weights, (1− () is the attenuation correction coefficient; K candidate robots with a matching degree higher than the formal threshold are selected, and the one with the highest matching degree is selected as the target social robot;

[0013] S4. Two-level temporary buffer resource allocation and scheduling: Estimate the resource requirements of the target task and determine whether the idle basic resources of the target robot meet the requirements; if they do, directly allocate the task; if they do not meet the requirements, determine the execution interval between the current task and the previous task: if the interval is less than the preset interval threshold, call the short-term buffer pool to reuse temporary resources; if the interval is greater than the threshold, allocate global long-term temporary resources.

[0014] S5. Tiered recycling of buffer resources and iterative update of capability profile: After the task is completed, resources are recycled in a tiered manner. Short-term buffer resources are retained until the interval threshold expires and then recycled and reused. Long-term temporary resources are recycled immediately. Read the task execution result. If the execution is successful, update the robot's historical success rate and resource decay coefficient, and refresh the capability profile.

[0015] The scheduling device of the supporting system includes:

[0016] The prediction and pre-screening module is used to perform step S1, short-term load prediction and preliminary candidate pre-screening.

[0017] The profile generation module is used to execute step S2 to construct a multi-dimensional attenuation-type capability profile.

[0018] The matching calculation module is used to perform step S3, attenuation correction matching degree calculation, and to screen target robots.

[0019] A two-level buffer processing module is used to perform step S4, resource requirement determination, and two-level buffer resource allocation.

[0020] The profile update module is used to perform step S5, which involves hierarchical resource recycling and iterative updates to the robot profile.

[0021] Preferably, the preset score database is established by: calculating the complexity scores of all historical tasks and grouping them, and estimating the CPU requirement quota for each group based on the content generation method of each group of tasks.

[0022] Preferred, , , Dynamically adjust based on target task type: Improve the performance of reloaded multimedia tasks. , Weighting, increasing the weight of lightweight interactive like-based tasks Weights.

[0023] Preferably, the calculation logic for the resource decay coefficient Dec is as follows: the longer the robot runs continuously, the larger the value of Dec, and the value range of Dec is 0~0.4.

[0024] Preferably, when multiple target tasks compete for the same target social robot, after determining the target robot in step S3 and before determining the resources in step S4, the method further includes: prioritizing each task based on its complexity score and the predicted peak weight, assigning the highest priority task to the target robot, and reallocating the remaining tasks to non-conflicting robots from the pool of candidate robots.

[0025] Preferably, the time series prediction model is a moving average time series prediction model, where N is 3-8 minutes, T is 5-15 seconds, and M is 10-30 units.

[0026] Preferably, the two-level buffer pools are stored in the memory of the scheduling device, the short-term buffer pool stores short-term reused CPU and memory temporary resources, and the global long-term resource pool stores all available idle resources of the entire cluster.

[0027] Preferably, the scheduling device is deployed on an independent physical server and establishes long-term communication with multiple physical servers carrying social robots to synchronize the operating indicators and resource occupancy status of each robot in real time.

[0028] In summary, this application includes the following beneficial technical effects:

[0029] 1. By comprehensively considering robot operating losses and real-time working status to allocate tasks, the previous single judgment method of only looking at instantaneous resources has been changed. This effectively improves the situation of unreasonable matching between equipment and tasks, avoids equipment overload or resource idle waste, and steadily improves the overall task completion quality and resource utilization efficiency of the cluster.

[0030] 2. By predicting task traffic in advance and selecting backup robot devices, we can cope with sudden large-scale task pressure in advance, effectively improve the problems of task backlog and resource contention in high-concurrency scenarios, make the scheduling process of the entire robot cluster smoother and more stable, and improve the overall operational stability.

[0031] 3. By adopting differentiated resource allocation methods for different types of tasks, it can flexibly adapt to various social business scenarios of varying weight and speed, solving the problem of poor adaptability of traditional scheduling modes and greatly improving the scenario adaptability and dynamic scheduling effect of robot clusters. Attached Figure Description

[0032] Figure 1 This is a flowchart of the overall scheduling method.

[0033] Figure 2 This is a block diagram of the supporting scheduling system architecture. Detailed Implementation

[0034] 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 a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0035] Reference Figure 1-2 This invention provides a dynamic scheduling method and supporting system for social robot clusters. S1, short-term load prediction and pre-selection of candidate robots: collect the historical traffic and task complexity distribution of the task queue in the past N minutes, and output the peak task and complexity distribution in the next T seconds through a time series prediction model; traverse all idle robots in advance, select M robots with an initial matching degree higher than the basic threshold and mark them as candidate robots, and lock the basic resource quota.

[0036] S2. Construct a multi-dimensional robot capability profile incorporating resource attenuation coefficients: After the target task is enqueued, calculate the task complexity score based on the task type, the number of physical servers required, and the content generation method. Real-time collection of remaining CPU quota for robots Available memory Historical success rate Calculate the resource decay coefficient based on the duration of continuous operation. , generate capability portrait Prot = ( , , , );

[0037] S3. Calculate the corrected matching degree based on the attenuation correction term to determine the target social robot: Calculate the matching degree between the candidate robot and the target task using the corrected matching degree formula, which is:

[0038]

[0039] in , The scores are obtained from the preset score database table. W1, W2, and W3 are dynamic resource weights, (1− () is the attenuation correction coefficient; K candidate robots with a matching degree higher than the formal threshold are selected, and the one with the highest matching degree is selected as the target social robot;

[0040] S4. Two-level temporary buffer resource allocation and scheduling: Estimate the resource requirements of the target task and determine whether the idle basic resources of the target robot meet the requirements; if they do, directly allocate the task; if they do not meet the requirements, determine the execution interval between the current task and the previous task: if the interval is less than the preset interval threshold, call the short-term buffer pool to reuse temporary resources; if the interval is greater than the threshold, allocate global long-term temporary resources.

[0041] S5. Tiered recycling of buffer resources and iterative update of capability profile: After the task is completed, resources are recycled in a tiered manner. Short-term buffer resources are retained until the interval threshold expires and then recycled and reused. Long-term temporary resources are recycled immediately. Read the task execution result. If the execution is successful, update the robot's historical success rate and resource decay coefficient, and refresh the capability profile.

[0042] The scheduling device of the supporting system includes:

[0043] The prediction and pre-screening module is used to perform step S1, short-term load prediction and preliminary candidate pre-screening.

[0044] The profile generation module is used to execute step S2 to construct a multi-dimensional attenuation-type capability profile.

[0045] The matching calculation module is used to perform step S3, attenuation correction matching degree calculation, and to screen target robots.

[0046] A two-level buffer processing module is used to perform step S4, resource requirement determination, and two-level buffer resource allocation.

[0047] The profile update module is used to perform step S5, which involves hierarchical resource recycling and iterative updates to the robot profile.

[0048] This solution utilizes five functional modules of a dedicated scheduling device to perform a complete dynamic scheduling of a social robot cluster. First, the system collects recent task traffic and complexity data, uses a time-series model to predict short-term task peaks, pre-selects a batch of high-quality idle robots, and locks resources for them as backup devices. When a new task is introduced, the system calculates the task complexity based on the task type, resource requirements, and generation method. Simultaneously, it collects real-time status data from the backup robots, including remaining computing power, memory, task success rate, and runtime, and builds a dedicated robot capability profile based on resource decay patterns. Subsequently, the system introduces a decay correction coefficient to accurately calculate the robot's compatibility with the current task, selecting the optimal target robot to execute the task. If the target robot lacks sufficient resources, the system will allocate short-term buffer resources or global long-term resources to supplement the demand based on the execution interval of previous and subsequent tasks. After the task is completed, the system reclaims various buffer resources in a tiered manner and updates the robot's performance parameters and capability profile based on the task execution results, achieving dynamic iteration. The five modules of the supporting system correspond to the entire process of prediction and selection, profile generation, matching calculation, resource scheduling, and profile updating, forming a closed-loop scheduling system.

[0049] The entire scheduling process anticipates task load and accurately assesses equipment status in advance, solving the problems of blind allocation and poor adaptability in traditional scheduling. By introducing resource decay characteristics, it aligns with the performance degradation of robots during long-term operation, making equipment assessment more consistent with actual working conditions. The two-level buffered resource allocation mode balances resource reuse for high-frequency continuous tasks and resource replenishment for low-frequency heavy-load tasks, effectively improving cluster resource utilization. Simultaneously, the system dynamically updates equipment profiles after each task, continuously optimizing subsequent scheduling accuracy. The modular system architecture ensures stable operation and ease of maintenance, comprehensively improving the task response speed, execution success rate, and resource utilization efficiency of the robot cluster.

[0050] In a preferred embodiment, the preset score database is established by: calculating the complexity scores of all historical tasks and grouping them, and estimating the CPU requirement quota for each group based on the content generation method of each group of tasks.

[0051] This solution is primarily used to establish a standardized task resource reference library. The system aggregates all historical task data from the cluster, categorizes and groups tasks according to their complexity, type, and content generation method, calculates the actual CPU resources consumed by each group of tasks, determines the standard resource requirements for each type of task, and integrates and archives information such as task attributes, generation methods, and resource quotas to form a pre-set score library table that can be accessed at any time. This provides a unified data reference standard for subsequent real-time task resource accounting and matching degree calculation.

[0052] By establishing a resource standard library based on real historical data, traditional experience-based resource estimation methods are replaced, making the determination of resource requirements for various tasks more standardized and accurate. This avoids waste caused by excessive resource allocation and prevents task execution failure due to insufficient resources, effectively reducing scheduling errors, minimizing system computing power consumption, and improving the overall stability and rationality of scheduling.

[0053] In a preferred embodiment , , Dynamically adjust based on target task type: Improve the performance of reloaded multimedia tasks. , Weighting, increasing the weight of lightweight interactive like-based tasks Weights.

[0054] The system can dynamically adjust the weight parameters of the matching calculation based on different task types to achieve differentiated adaptation. When handling heavy multimedia tasks such as video rendering and high-definition content generation, the system focuses on increasing the weights related to robot resource capacity and complex task adaptation, prioritizing the selection of high-performance devices. When handling lightweight, high-frequency tasks such as liking, commenting, and simple interactions, the system increases the weights related to robot response efficiency and high-frequency adaptation, focusing on assessing the device's rapid execution capabilities, ensuring that the matching criteria align with the core requirements of different tasks.

[0055] In a preferred embodiment, the calculation logic of the resource decay coefficient Dec is as follows: the longer the robot runs continuously, the larger the value of Dec, and the value range of Dec is 0~0.4.

[0056] The resource attenuation coefficient Dec is used to quantify the performance loss of a robot during long-term operation, with a fixed value range of 0 to 0.4. The longer the robot runs continuously, the higher the hardware fatigue and the more significant the resource performance loss, resulting in a larger Dec value. Conversely, when the equipment operates intermittently or for short periods, the performance loss is small, and the Dec value is lower. The system uses this coefficient to quantify the implicit performance degradation of the equipment, thereby correcting task matching results and ensuring that scheduling closely matches the actual operating state of the equipment.

[0057] In a preferred embodiment, when multiple target tasks compete for the same target social robot, after determining the target robot in step S3 and before determining the resources in step S4, the method further includes: prioritizing each task based on its complexity score and predicted peak weight, assigning the highest priority task to the target robot, and reallocating the remaining tasks to non-conflicting robots from the pool of candidate robots.

[0058] When multiple tasks compete for the same optimal robot, the system activates a priority ranking mechanism. Combining the complexity score and predicted peak weight of each task, the system calculates task priorities and prioritizes allocating the device to the highest-priority core task. The remaining unassigned low-priority tasks are then re-selected from the pool of candidate robots, undergoing a secondary allocation process to resolve resource contention conflicts among multiple tasks.

[0059] In a preferred embodiment, the time series prediction model is a moving average time series prediction model, where N is 3-8 minutes, T is 5-15 seconds, and M is 10-30 units.

[0060] This solution defines the core parameter standards for the prediction and screening process, and uses a moving average time series prediction model for short-term load prediction. The historical data collection duration N is set to 3-8 minutes, which can be adaptively adjusted according to business load; the future task situation prediction duration T is set to 5-15 seconds to adapt to the characteristics of short-term, high-frequency tasks; the number of candidate robots M for pre-screening is set to 10-30 units, with an appropriate increase in the number of candidate devices for high-load scenarios and a moderate decrease for low-load scenarios to ensure reasonable resource reserves.

[0061] In a preferred embodiment, the two-level buffer pools are stored in the scheduling device memory, the short-term buffer pool stores short-term reused CPU and memory temporary resources, and the global long-term resource pool stores all available idle resources for the entire cluster.

[0062] The scheduling device incorporates two independent memory buffer pools with clearly defined functions and mutual isolation. The short-term buffer pool specifically stores quickly reusable temporary CPU and memory resources, serving high-frequency tasks with short intervals and continuous execution. It eliminates the need for immediate resource reclamation, facilitating rapid reuse. The global long-term resource pool integrates all idle long-term resources within the cluster, providing stable global resource replenishment for long-interval, high-load tasks and adapting to the resource allocation needs of different tasks.

[0063] In a preferred embodiment, the scheduling device is deployed on an independent physical server and establishes long-term communication with multiple physical servers that host social robots to synchronize the operating indicators and resource occupancy status of each robot in real time.

[0064] The scheduling device is independently deployed on a dedicated physical server, hardware isolated from the business servers that host the robots, to prevent task execution from interfering with scheduling. Simultaneously, the scheduling server establishes a long-term, stable communication connection with all robot servers, synchronizing core data such as CPU usage, memory usage, operating status, and task success rate of each robot in real time. This allows for comprehensive monitoring of the cluster's equipment status, providing real-time and accurate data support for various scheduling decisions.

[0065] The foregoing description, with reference to preferred embodiments, illustrates an exemplary implementation of a dynamic scheduling method and supporting system for a social robot cluster provided by this disclosure. However, those skilled in the art will understand that various modifications and alterations can be made to the above specific embodiments without departing from the spirit of this disclosure, and various combinations can be made to the various technical features and structures proposed in this disclosure without exceeding the protection scope of this disclosure, the protection scope of which is determined by the appended claims.

Claims

1. A method and supporting system for dynamic scheduling of social robot swarms, characterized in that: The scheduling method is executed by a scheduling device and includes the following steps: S1. Short-term load prediction and preliminary candidate screening: Collect the historical traffic and task complexity distribution of the task queue in the past N minutes, and output the task peak and complexity distribution in the next T seconds through the time series prediction model; traverse all idle robots in advance, screen M robots with an initial matching degree higher than the basic threshold and mark them as preliminary candidate robots, and lock the basic resource quota. S2, construct multi-dimensional robot capability image of fusion resource attenuation coefficient: after the target task is queued, calculate the task complexity score according to the task type, the required number of physical servers and the content generation mode ; real-time collection of remaining CPU quota of the robot , available memory , historical success rate and continuous running time, calculate the resource attenuation coefficient , generate the capability image Prot= ( , , , ) S3. Calculate the corrected matching degree based on the attenuation correction term to determine the target social robot: Calculate the matching degree between the candidate robot and the target task using the corrected matching degree formula, which is: wherein , is obtained by querying the preset score database table, W1, W2, and W3 are dynamic resource weights, and (1− ) is a decay correction coefficient; K candidate robots with a matching degree higher than a formal threshold value are screened, and one with the highest matching degree is selected as the target social robot; S4. Two-level temporary buffer resource allocation and scheduling: Estimate the resource requirements of the target task and determine whether the idle basic resources of the target robot meet the requirements; if they do, directly allocate the task; if they do not meet the requirements, determine the execution interval between the current task and the previous task: if the interval is less than the preset interval threshold, call the short-term buffer pool to reuse temporary resources; if the interval is greater than the threshold, allocate global long-term temporary resources. S5. Tiered recycling of buffer resources and iterative update of capability profile: After the task is completed, resources are recycled in a tiered manner. Short-term buffer resources are retained until the interval threshold expires and then recycled and reused. Long-term temporary resources are recycled immediately. Read the task execution result. If the execution is successful, update the robot's historical success rate and resource decay coefficient, and refresh the capability profile. The scheduling device of the supporting system includes: The prediction and pre-screening module is used to perform step S1, short-term load prediction and preliminary candidate pre-screening. The profile generation module is used to execute step S2 to construct a multi-dimensional attenuation-type capability profile. The matching calculation module is used to perform step S3, attenuation correction matching degree calculation, and to screen target robots. A two-level buffer processing module is used to perform step S4, resource requirement determination, and two-level buffer resource allocation. The profile update module is used to perform step S5, which involves hierarchical resource recycling and iterative updates to the robot profile.

2. The social robot cluster dynamic scheduling method and the matching system according to claim 1, characterized in that: The preset score database table is established by: calculating the complexity scores of all historical tasks and grouping them, and estimating the CPU requirement quota for each group based on the content generation method of each group of tasks.

3. The method and supporting system for dynamic scheduling of social robot clusters according to claim 1, characterized in that: 、 、 Dynamic adjustment based on target task type: overload multimedia type task to improve 、 weight, and reduce the weight of light interaction like tasks. weight.

4. The social robot cluster dynamic scheduling method and the matching system according to claim 1, characterized in that: The calculation logic for the resource decay coefficient Dec is as follows: the longer the robot runs continuously, the larger the value of Dec. The value range of Dec is 0~0.

4.

5. The method and supporting system for dynamic scheduling of social robot clusters according to claim 1, characterized in that: When multiple target tasks compete for the same target social robot, after determining the target robot in step S3 and before determining the resources in step S4, the process also includes: prioritizing tasks based on their complexity scores and predicted peak weights, assigning the highest priority task to the target robot, and reallocating the remaining tasks to non-conflicting robots from the pool of candidate robots.

6. The method and supporting system for dynamic scheduling of social robot clusters according to claim 1, characterized in that: The time series prediction model is a moving average time series prediction model, where N is 3-8 minutes, T is 5-15 seconds, and M is 10-30 units.

7. The method and supporting system for dynamic scheduling of social robot clusters according to claim 1, characterized in that: The two-level buffer pools are stored in the scheduling device memory. The short-term buffer pool stores short-term reused CPU and memory temporary resources, while the global long-term resource pool stores all available idle resources for the entire cluster.

8. The method and supporting system for dynamic scheduling of social robot clusters according to claim 1, characterized in that: The scheduling device is deployed on an independent physical server and establishes long-term communication with multiple physical servers that host social robots to synchronize the operating indicators and resource usage status of each robot in real time.