Gaia Maselli

Thesis opportunities

Networks of Aerial Drones

in the context of the NATO project "SeaSec: DroNets for Maritime and Border Port Security"

Routing

During a mission, drones can collect a significant amount of data. When should drones offload data? During the mission or at the end? We will investigate different approaches to route data.

Task assignment

Assign a task to a drone is a big challenge especially when drones a flying over a wide areas and the network may be jeopardized

In collaboration with Francesco Restuccia @ the Roux Institute in Portland, Maine

The execution of state-of-the-art neural networks on unmanned autonomous vehicles (UAVs) is quintessential for mission-critical applications such as real-time obstacle detection and navigation. On the other hand, today's neural network models present computational requirements that go far beyond what the vast majority of UAVs can offer. While offloading to edge servers can drastically decrease the UAV's computational burden, erratic patterns in channel quality and edge server load can lead to severe disruption of the system's key operations. This critical aspect calls for strategies where inference tasks are distributed to multiple edge servers according to the current priority. At the same time, achieving maximum spectrum efficiency is fundamental to avoid network congestion. In this context, multi-user, multiple-input multiple-output (MU-MIMO) technologies are ideal to support the continuous and parallel execution of different streams of tasks without increasing spectrum congestion. The target of this project will be the investigation of data-driven algorithms that will adaptively choose among different MU-MIMO offloading and beamforming strategies to achieve the desired trade-off between reliability and efficiency for a given set of inference tasks. We will prototype and investigate the algorithms using state-of-the-art WiFi-6 chips and embedded boards, by also referring to modern neural networks such as DenseNet and ResNet.

Report abuse