Federated Learning for UAV Perception Task : A Survey
Yanis Bardes, Hassan Soubra, Zineb Noumir, Amar Ramdane-Chérif · 2025
Unmanned Aerial Vehicles (UAVs) are increasingly deployed for perception tasks such as surveillance, object detection, and traffic monitoring, which play a crucial role in intelligent vehicle systems. Federated Learning (FL) offers a decentralized framework that not only enhances data privacy but also facilitates collaborative intelligence between UAVs and ground-based vehicles, addressing the challenges of distributed data environments. This survey investigates FL applications in UAV perception, with a particular focus on advancing vehicle environment perception through cooperative learning. Key challenges addressed include the need to manage heterogeneous and often unlabeled data, optimize limited computational resources, and navigate communication constraints inherent to mobile, UAV and vehicular networks. Through an in-depth analysis, we assess the impact of data heterogeneity on model performance and explore state-of-the-art learning methods, including semi-supervised, unsupervised, and transfer learning techniques adapted for FL in autonomous systems. We further examine robust algorithms designed to support FL in complex, resourceconstrained settings, laying a foundation for future research in vehicle-centric IoT applications, where UAVs and vehicles interact in shared environments to enhance safety, efficiency, and perception.