A major goal of the Department of Science, Innovation and Technology, the project will introduce a radically new type of wireless network.
Professor Yansha Deng, from the Department of Engineering, has been awarded €3 million by Horizon Europe to build a new kind of goal-oriented multi-agent semantic networking (GoAgentNet) for 6G.
Alongside Professor Sir Bashir Al-Hashimi and a consortium from industry and academia (including Nokia Bell Labs and the University of Padova), Professor Deng hopes to introduce a new kind of network architecture, GoAgentNet, that will deliver on the historical promises of 6G through goal-oriented networking at scale.
Today's wireless networks rely on a separation between the network controlling the data and the applications and devices, such as mobile phones and autonomous vehicles, which use it. This results in networks acting as opaque data pipes carrying messages and data packets with no understanding of an application's goals, while applications make decisions without awareness of real-time network conditions.
By combining a goal-oriented approach with edge computing, this network architecture could just as easily connect a smart city in Milton Keynes as run a production line of internet-enabled robotic pickers in the rainforests of Brazil. People have been trying to create a dynamic, high-performance network with universal application for some time - and this work is potentially pioneering."
Professor Yansha Deng
This could look like a device attempting to play video with poor connectivity to the central network, causing it to fail. Such isolation prevents optimisation for speed and connectivity and limits the network's ability to prioritise its resources to support goal-oriented services. As the number of users and applications joining networks balloon, this precarious balancing act can result in bottlenecks which slow speeds.
By introducing two new layers between the application and network layer, and a multi-AI agent approach into traditional network architectures, Yansha hopes to reimagine 6G as an ecosystem where applications and the network work together to achieve a user's goals.
Using 6G-enabled remote robotic control in a warehouse, a high-pressure use case currently pushing the boundaries of network connectivity, Professor Deng explains. "To remote control robots using today's network requires streaming massive volumes of video and sensor data, then processing them at the central network or on the cloud. This approach is bandwidth-hungry and imposes a limit on the number of robot devices you can connect.
"With our platform, a user could say an explicit goal 'Move package A to pallet B, as fast as possible' and a series of AI agents takes this request and breaks it down into several constituent parts. Using methods like knowledge graphs, the robots then talk with each other with only essential goal-oriented information to work together to achieve this goal. This is much less bandwidth hungry than relaying raw sensor feeds."
By having a knowledge layer that essentialises semantic information and an AI agent layer throughout the network infrastructure, the network can now dynamically prioritise resources based on how urgent an action is to a user's goal.
To help achieve that goal, this new approach reduces data traffic and latency, optimises resource utilisation, and makes AI algorithms more interpretable, providing transparency into their decision-making processes.
Drawing on Yansha's specialism in edge computing, GoAgentNet also dynamically pushes the computationally intense work of network resource management to devices when connection to the central servers/networks are poor, and vice versa when the connection is strong. This ensures devices can operate even when network connectivity is limited.
Our GoAgentNet takes a first step by providing a universal framework that puts the user first."
Professor Yansha Deng
Professor Deng explains, "When 6G was first mentioned several years ago, it promised enormous connectivity, lower latency, greater energy efficiency and universal use across a range of applications, from smart cities to smart manufacturing. But until now, that has yet to come to pass.
"Our GoAgentNet takes a first step by providing a universal framework that puts the user first. By combining a goal-oriented approach with edge computing, this network architecture could just as easily connect a smart city in Milton Keynes as run a production line of internet-enabled robotic pickers in the rainforests of Brazil. People have been trying to create a dynamic, high-performance network with universal application for some time - and this work is potentially pioneering."
In the future, the team hope to operationalise the network for use in smart cities and for information recovery in smart manufacturing hubs, using the goal-oriented agent layer to optimise production and robotic picking.
Much of this work on goal-oriented communication was first demonstrated at the Mobile World Congress (2025), including work monitoring warehouse inventory with Nokia and 3D weather forecasting with VIAVI Marconi Lab and Nokia Bell Labs.
