Japan's SoftBank unveiled new tech for autonomously optimizing network routes, as tested in a cloud gaming-based field trial.

Dubbed Autonomous Thinking Distributed Core Routing, the technology uses AI to monitor network traffic in real time and automatically choose the best route for data transmission depending on network conditions. Specifically, it uses AI agents to dynamically switch between a conventional centrally managed core network, defined by user plane function (UPF), and segment routing v6 mobile user plane (SRv6 MUP), an IPv6-native plane which works across commercial SRv6 networks.

The AI agents in question utilize quality on demand (QoD) API, an industry standard defined by the Camara Project.

SoftBank deployed a field trail using cloud gaming, with verification conducted within SoftBank's commercial 4G mobile network environment. It claimed average latency when applying the technology was reduced to 27.4 milliseconds, compared to the average 41.9 milliseconds when using the conventional mobile core, and meeting the 40 milliseconds or less latency requirement required for cloud-based gaming.

The field trial also revealed agentic traffic control accuracy reached 99.7%.

SoftBank plans to enhance its service by enabling AI agents to learn from diverse application traffic patterns, moving toward more generalized and adaptive network control. The firm also aims to establish a framework where application providers can deploy low-latency services on SoftBank’s mobile edge compute (MEC) servers, allowing the network to autonomously apply optimal control.