Scalable, Partially Observable MARL for UAV Aerial Base Stations

Context

Terrestrial base stations cannot always guarantee adequate coverage in temporary events with high user density, disaster scenarios where infrastructure is damaged, or sparsely populated areas where dense deployment is not cost-effective. Unmanned aerial vehicles (UAVs) acting as aerial base stations (ABSs) offer a flexible complement. Their mobility and altitude allow coverage to be deployed on demand and adapted to user distribution, making them a key component of the three-dimensional network architectures envisioned for 6G.

When several ABSs operate together, the placement of each UAV affects the coverage and interference experienced by users served by the others. Efficient collective coverage therefore requires coordination, while centralized control becomes impractical as the fleet grows and the user population evolves. Each UAV also has only partial knowledge of its environment, so cooperation must rely on limited local observations and on information exchanged between UAVs.

Internship objective

Building on a cooperative multi-agent reinforcement learning framework developed within the EU Horizon Europe 6G-TWIN project by Université Bourgogne Europe, the internship will address the scalability of cooperative multi-UAV ABS deployment along three axes:

1. Number of users: Develop policies that remain effective as the user population grows and evolves.
2. Number of UAVs: Develop policies that remain effective as the fleet size varies.
3. Inter-UAV information exchange: Study how the nature and extent of the information shared between UAVs affect coordination and coverage performance.

Other approaches are not excluded.

Required skills and candidate profile

• Knowledge of machine learning, ideally multi-agent reinforcement learning.
• Solid Python programming skills and experience with PyTorch or TensorFlow.
• Basic knowledge of mobile networks, including 5G/6G, wireless coverage and interference.
• Autonomy, rigor, and the ability to run and analyze simulation experiments.
• Experience with UAV networks or network simulation is appreciated.
• Target profile: Master's student (M2 or final-year engineering school) in computer science, data science/AI, or telecommunications.

Duration and start

• Duration: 4 to 6 months.
• Start: Around January–March 2027.

Location

DRIVE Laboratory, Université Bourgogne Europe, Nevers, France.

Application file

• Detailed CV
• Transcripts