Distributed robot federated learning network optimization method and system
By optimizing network uploads for robot federated learning through real-time monitoring and predictive scheduling, the communication storm problem in federated learning is solved, communication efficiency and training success rate are improved, and training time is reduced.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-09
- Publication Date
- 2026-04-14
AI Technical Summary
Existing federated learning suffers from communication storms in industrial robot clusters, leading to network congestion and training round failures. Existing solutions cannot adapt to traffic bursts and lack cross-layer collaborative optimization.
The robot client uploads network status and resource usage in real time. The training center agent monitors server resources. The controller predicts and allocates upload time slots based on the overall situation. ICMP and ipef are used to measure network parameters to achieve predictive scheduling.
It effectively avoids network congestion, improves communication efficiency, increases model training success rate, and reduces overall training time.
Smart Images
Figure CN121864697A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, specifically providing a method and system for optimizing distributed robot federated learning networks. Background Technology
[0002] Federated learning, as a distributed machine learning paradigm, has been widely applied in industrial robot swarms. However, its inherent synchronous round-robin mechanism leads to severe communication storm problems, characterized by the following: Cycle synchronization: Federated learning is carried out in training rounds, and at the end of each round, a large number of clients simultaneously upload model updates.
[0003] Characteristics of sudden traffic: Model update files are usually large (from a few MB to several hundred MB), and simultaneous uploads by N clients generate N times the traffic peak.
[0004] The chain reaction of network congestion: link congestion → packet loss → transmission timeout and retransmission → increased congestion → training round failure.
[0005] Existing solutions have shortcomings: Static bandwidth allocation: cannot adapt to the bursty characteristics of federated learning traffic.
[0006] Passive congestion control: Mitigation measures are taken only after congestion has occurred, resulting in a delayed response.
[0007] Single-point optimization: Optimization is done only from the application layer or network layer, lacking cross-layer collaboration. Summary of the Invention
[0008] This invention addresses the shortcomings of the prior art by providing a highly practical method for optimizing distributed robot federated learning networks.
[0009] A further technical objective of this invention is to provide a reasonably designed, safe, and applicable distributed robot federated learning network optimization system.
[0010] The technical solution adopted by this invention to solve its technical problem is: A method for optimizing distributed robot federated learning networks includes the following steps: S1. Each robot client uploads its own network bandwidth of each network interface to the server in real time, as well as jitter and packet loss rate information. S2, The Training Center Agent uploads its own resource usage information in real time; S3. After the training task is completed, the robot uploads its own training result package size, training rounds, and data importance to the controller. S4. The controller receives the request for the robot to upload data, calculates the predicted upload time of the packet based on the overall situation of the robot's network exits and the current server resource usage, and issues instructions to the robot. S5. The robot upload time begins, and data is transmitted to the training center according to the upload time window.
[0011] Furthermore, the available bandwidth, latency jitter, and packet loss rate of each link of the robot client are tested. ICMP is used to measure latency, ipef is used to test bandwidth, probe packets are sent to the server periodically to collect network status, measurement information is reported to the controller periodically, and the training model size, training rounds, and data importance are uploaded after the robot training is completed.
[0012] Furthermore, the training center Agent tests the CPU utilization, memory utilization, and available network bandwidth of the test server, monitors the inbound and outbound bandwidth utilization and packet loss rate of the network interface, measures latency through ICMP, and calculates the overall CPU utilization and kernel utilization.
[0013] Furthermore, the controller allocates upload time slots to each robot by comprehensively considering the current network status of the client and the current resource utilization of the server, and monitors the network status in real time.
[0014] A distributed robot federated learning network optimization system, firstly, each robot client uploads its own network bandwidth of each network interface to the server in real time, as well as jitter and packet loss rate information, and the training center agent uploads its own resource usage in real time; Then, after the training task is completed, the robot uploads its own training result packet size, training rounds and data importance to the controller. The controller receives the robot's request to upload data, calculates the predicted packet upload time based on the overall situation of the robot's network exits and the current server resource usage, and issues instructions to the robot. Finally, the robot upload time begins, and data is transmitted to the training center according to the upload time window.
[0015] Furthermore, the available bandwidth, latency jitter, and packet loss rate of each link of the robot client are tested. ICMP is used to measure latency, ipef is used to test bandwidth, probe packets are sent to the server periodically to collect network status, measurement information is reported to the controller periodically, and the training model size, training rounds, and data importance are uploaded after the robot training is completed.
[0016] Furthermore, the training center Agent tests the CPU utilization, memory utilization, and available network bandwidth of the test server, monitors the inbound and outbound bandwidth utilization and packet loss rate of the network interface, measures latency through ICMP, and calculates the overall CPU utilization and kernel utilization.
[0017] Furthermore, the controller allocates upload time slots to each robot by comprehensively considering the current network status of the client and the current resource utilization of the server, and monitors the network status in real time.
[0018] Compared with existing technologies, the distributed robot federated learning network optimization method and system of the present invention have the following outstanding advantages: This invention deeply integrates software-defined networking with robot federated learning, avoids network congestion through predictive scheduling mechanisms, significantly improves the communication efficiency of concurrent traffic, increases the success rate of model training, and reduces the overall training time. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 This is a schematic diagram of a distributed robot federated learning network optimization method. Detailed Implementation
[0021] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] The following is a preferred embodiment: like Figure 1 As shown in this embodiment, a distributed robot federated learning network optimization method has the following steps: S1. Each robot client uploads its own network bandwidth across various network interfaces to the server in real time, along with information on jitter and packet loss rate.
[0023] S2, the training center agent uploads its own resource usage information in real time, including: CPU usage, memory usage, network bandwidth, and latency information.
[0024] S3. After the training task is completed, the robot uploads its own training result package size, training rounds, and data importance to the controller.
[0025] S4. Upon receiving the robot's request to upload data (training rounds, packet size, data importance), the controller calculates the predicted upload time (including possible delays to ensure data upload is completed within a short period) based on the overall status of the robot's network exits (5G, Wi-Fi, wired) and the current server resource usage, and then issues instructions to the robot. These instructions include (network exit point and upload time window).
[0026] S5. The robot upload time begins, and data is transmitted to the training center according to the upload time window.
[0027] This includes testing the available bandwidth, latency jitter, packet loss rate, and other information for each link of the test robot client.
[0028] Use ICMP to measure latency and ipef to test bandwidth. Periodically send probe packets to the server to collect network status data. Periodically report measurement information to the controller.
[0029] After the robot training is completed, the training model size, training rounds, and data importance need to be uploaded.
[0030] The training center agent tests the server's CPU usage, memory usage, available network bandwidth, latency jitter, and other information.
[0031] Implementation method: Monitor the inbound and outbound bandwidth utilization and packet loss rate of the network interface, and measure latency via ICMP. Calculate overall CPU utilization, kernel utilization, etc.
[0032] The controller allocates upload time slots to each robot by combining the current network status of the client and the current resource utilization of the server, and monitors the network status in real time.
[0033] Based on the above method, a distributed robot federated learning network optimization system is proposed. First, each robot client uploads its own network bandwidth of each network interface to the server in real time, as well as jitter and packet loss rate information. The training center agent uploads its own resource usage in real time. Then, after the training task is completed, the robot uploads its own training result packet size, training rounds and data importance to the controller. The controller receives the robot's request to upload data, calculates the predicted packet upload time based on the overall situation of the robot's network exits and the current server resource usage, and issues instructions to the robot. Finally, the robot upload time begins, and data is transmitted to the training center according to the upload time window.
[0034] The test includes measuring the available bandwidth, latency jitter, and packet loss rate of each link in the robot client, using ICMP to measure latency, using ipef to test bandwidth, periodically sending probe packets to the server to collect network status, periodically reporting measurement information to the controller, and uploading the training model size, training rounds, and data importance after the robot training is completed.
[0035] The training center agent tests the server's CPU usage, memory usage, and available network bandwidth, monitors the network interface's inbound and outbound bandwidth usage and packet loss rate, measures latency via ICMP, and calculates overall CPU usage and kernel usage.
[0036] The controller allocates upload time slots to each robot by combining the current network status of the client and the current resource utilization of the server, and monitors the network status in real time.
[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for optimizing distributed robot federated learning networks, characterized in that, It has the following steps: S1. Each robot client uploads its own network bandwidth of each network interface to the server in real time, as well as jitter and packet loss rate information. S2, The Training Center Agent uploads its own resource usage information in real time; S3. After the training task is completed, the robot uploads its own training result package size, training rounds, and data importance to the controller. S4. When the controller receives the request for the robot to upload data, it calculates the predicted upload time of the packet based on the overall situation of the robot's network exits and the current server resource usage, and then issues instructions to the robot. S5. The robot upload time begins, and data is transmitted to the training center according to the upload time window.
2. The method for optimizing a distributed robot federated learning network according to claim 1, characterized in that, Test the available bandwidth, latency jitter, and packet loss rate of each link on the robot client. Use ICMP to measure latency and ipef to test bandwidth. Periodically send probe packets to the server to collect network status. Periodically report measurement information to the controller. Upload the training data after the robot training is completed, including the size of the training model, the number of training rounds, and the importance of the data.
3. The method for optimizing a distributed robot federated learning network according to claim 2, characterized in that, The training center Agent tests the CPU utilization, memory utilization, and available network bandwidth of the server, monitors the inbound and outbound bandwidth utilization and packet loss rate of the network interface, measures latency through ICMP, and calculates the overall CPU utilization and kernel utilization.
4. The method for optimizing a distributed robot federated learning network according to claim 3, characterized in that, The controller allocates upload time slots to each robot by combining the current network status of the client and the current resource utilization of the server, and monitors the network status in real time.
5. A distributed robot federated learning network optimization system, characterized in that, First, each robot client uploads its own network bandwidth of each network interface to the server in real time, as well as information on jitter and packet loss rate. The training center Agent uploads its own resource usage in real time. Then, after the training task is completed, the robot uploads its own training result packet size, training rounds and data importance to the controller. The controller receives the robot's request to upload data, calculates the predicted packet upload time based on the overall situation of the robot's network exits and the current server resource usage, and issues instructions to the robot. Finally, the robot upload time begins, and data is transmitted to the training center according to the upload time window.
6. A distributed robot federated learning network optimization system according to claim 5, characterized in that, Test the available bandwidth, latency jitter, and packet loss rate of each link on the robot client. Use ICMP to measure latency and ipef to test bandwidth. Periodically send probe packets to the server to collect network status. Periodically report measurement information to the controller. Upload the training data after the robot training is completed, including the size of the training model, the number of training rounds, and the importance of the data.
7. A distributed robot federated learning network optimization system according to claim 6, characterized in that, The training center Agent tests the CPU utilization, memory utilization, and available network bandwidth of the server, monitors the inbound and outbound bandwidth utilization and packet loss rate of the network interface, measures latency through ICMP, and calculates the overall CPU utilization and kernel utilization.
8. A distributed robot federated learning network optimization system according to claim 7, characterized in that, The controller allocates upload time slots to each robot by combining the current network status of the client and the current resource utilization of the server, and monitors the network status in real time.