Shared World Models for Decentralised AI Agents
Abstract
Modern AI agents increasingly rely on world models: learned internal simulators of their environment that allow them to plan, predict and imagine the consequences of their actions. These models are trained centrally on vast datasets; yet many agents (robots, vehicles, personal assistants) each observe only a fragment of the world, and often cannot share raw data. This project asks how such agents can build a shared, reliable world model by combining partial models, without a central server, and while some contributors may be faulty or malicious. Applicants from machine learning, robotics or applied mathematics are welcome.
How to Apply
Please contact Dr Pietro Ferraro if you are interested in applying for this project.
For more information on submitting a PhD application, please consult our application pages here. When you submit your application, please cite this project title in the 'how you are planning to fund your PhD' section.
The deadline for submitting your application is 11 January 2027.