Pedestrian Trajectory Prediction through LoRA Federated Training

Abstract

In Intelligent Transport Systems, protecting vulnerable road users, specifically pedestrians, is a paramount challenge. The present work contributes to this field, putting forth a distributed architecture in which vehicles connected to the cellular infrastructure collect pedestrians' trajectories and fine-tune a foundation model for pedestrian trajectory forecasting via federated training. A server positioned at the edge of the cellular network is responsible for aggregating the local model's parameters from the vehicles and updating the global model's parameters. To reduce radio resource utilization, Low-Rank Adaptation is proposed; to decrease both the number and the size of updates, an increase in the number of training rounds performed by the vehicles per communication round is put forth. A novel figure of merit is then introduced to quantify the use of radio resources dedicated to Vehicle-to-Infrastructure communications as required by the federated solution. The proposed scheme delivers excellent prediction accuracy while significantly reducing the communication cost compared to alternative federated training methods. This shows that combining fine-tuning, Low-Rank Adaptation, and a reduction of the communication periodicity between the vehicles and the edge server is key to enabling a distributed approach to trajectory prediction in vehicular environments.