New technology achieves both efficiency and low latency through collaboration between SRv6 MUP and AI Agents
SoftBank Corp. ("SoftBank") announced it developed "Autonomous Thinking Distributed Core Routing," a technology that uses AI to analyze communication conditions (traffic characteristics) in real time and autonomously selects optimal network routes depending on the situation.
Autonomous Thinking Distributed Core Routing dynamically switches between a conventional centrally managed core network (UPF: User Plane Function) and SRv6 MUP (Segment Routing v6 Mobile User Plane) through AI agents that utilize the industry-standard QoD API (Quality on Demand Application Programming Interface) defined by the CAMARA Project.
This technology allows for the use of a conventional centralized mobile core for communications where efficiency is important, and for applications requiring low latency. By automatically switching to the shortest network route, optimal communication quality is achieved according to the application.
1. Development Background
In next-generation communication environments, it is anticipated that applications such as AR (Augmented Reality), VR (Virtual Reality), and autonomous driving will require ultra-low latency communications, while also demanding efficient transmission of large volumes of data. Within a single application, these differing requirements are expected to switch dynamically and in a complex manner.
However, in conventional mobile networks, when different latency requirements (SLA: Service Level Agreements) exist within a single application, it has been difficult to flexibly accommodate each requirement, making uniform communication control challenging.
To address this issue, SoftBank has newly developed Autonomous Thinking Distributed Core Routing, a technology in which AI agents autonomously determine the required latency conditions and control communications by switching to the optimal core network route in real time.
2. Features of Autonomous Thinking Distributed Core Routing
- Definition of latency requirements and pre-measurement
Latency requirements for applications are defined as tiered classes ranging from several tens of milliseconds (1 ms = 1/1,000 of a second) down to 10 milliseconds or less.
When launching an application such as cloud gaming that requires a specific latency threshold (e.g., 40 ms or less), the AI observes network latency via measurement servers in real time whether the current network latency satisfies the required condition. - Autonomous optimisation based on pre-measurement
Based on the results of the pre-measurement, the AI determines whether SRv6 MUP should be applied. If it determines that applying SRv6 MUP can satisfy the application's latency requirement, it autonomously switches the network route. - Verification (measurement during service use)
While the application is used in a communication environment optimized by SRv6 MUP, the measurement server continuously measures latency. - Proof (visualisation of results)
After the session ends, the achievement status of the latency requirement is determined. If necessary, a certificate of results is returned via API. Real-time switching within the mobile network
The mobile network switches in real time between two routes with different characteristics: a centrally managed core network (UPF) that provides reliable and efficient communication; and SRv6 MUP-PE (Provider Edge), which provides a shortest-path communication route that suppresses latency.This enables the optimal use of each route according to communication requirements.
3. Field Trial Results
At "JANOG57," an event for network operators held in February 2026, SoftBank conducted a field trial using cloud gaming.
Verification was conducted within SoftBank's commercial 4G mobile network environment. The results showed that while the average latency via the conventional mobile core was 41.9 milliseconds, the average latency when applying the Technology was reduced to 27.4 milliseconds.
These results confirmed that Autonomous Thinking Distributed Core Routing can stably meet the latency requirement (40 milliseconds or less) required for cloud gaming.
In addition, the traffic control accuracy of the AI agent reached 99.7%, confirming the effectiveness of highly precise and stable autonomous control.
4. Future Developments
SoftBank will further enhance Autonomous Thinking Distributed Core Routing by enabling AI agents to learn from a variety of application traffic patterns, advancing toward more generalised control capabilities.
In the future, SoftBank aims to realise a mechanism in which application providers can simply deploy low-latency applications on SoftBank's MEC servers, and optimal network control will be applied autonomously.
SoftBank will promote the commercial deployment and further advancement of Autonomous Thinking Distributed Core Routing as it works toward realizing an autonomous society in which AI and networks are highly integrated.
