6G-BRICKS

PEDESTRIAN (Protecting vulnErable roaD usErS with federaTed leaRnIng for trAjectory predictioN) proposes a vertical application for automotive that exploits 6G technology to improve road safety, taking advantage of the great amount of data that can be collected from the Internet of Vehicles (IoV) domain.

PEDESTRIAN exploits computing resources in the extreme-edge/edge/cloud continuum to analyse data from Connected Vehicles (CVs) and Road Side Units (RSU). Solutions based on transformer neural networks combined with Federated Learning (FL) are put forth, to predict the future trajectories of Vulnerable Road Users (VRUs) and guarantee their safety. CVs and RSUs transmit sensor data (extracted from public datasets to guarantee realism and reproducibility) to the edge for training, or jointly cooperate to train the trajectory prediction model in a distributed manner while preserving data privacy. Scenarios with central and hierarchical FL aggregators at the edge are compared in terms of uplink data rate and computing resources requirements.