FedADAS: Communication-Efficient Federated Distillation for On-Device Driver Yawn Recognition in Vehicular Networks

2026-05-19

AI Summary

This work addresses the challenges of high communication overhead, strong homogeneity assumptions, and poor personalization in driver fatigue detection models when deployed across heterogeneous vehicular devices. To overcome these limitations, the authors propose a federated distillation framework tailored for vehicular networks, which enables collaborative training of fully heterogeneous models by exchanging soft logits over a shared public dataset. This approach allows each vehicle to customize its model according to its computational capacity, achieving full model heterogeneity—the first such implementation in vehicular federated learning—while significantly reducing communication costs and balancing personalization with generalization. Experimental results demonstrate up to a 9,974-fold reduction in communication overhead across 115 edge clients; the Performance-Efficient model achieves a 98.3% F1-score with 1.99 ms inference latency, while the Memory-Efficient variant requires only 6.12 minutes per training round.

Abstract

Driver fatigue is a critical safety concern in advanced driver assistance systems. Driver monitoring models trained off-site on static datasets adapt poorly to real-world conditions, while standard federated learning imposes high communication overhead, assumes homogeneous architectures, and struggles with personalized driver data. We present FedADAS, a federated distillation framework enabling collaborative on-device learning across heterogeneous vehicular networks. FedADAS enables full model heterogeneity by exchanging only soft logits on a shared public dataset, allowing each vehicle to run a customized model tailored to its computational constraints. Additionally, we introduce a yawn recognition pipeline supporting training and inference on edge devices that provides two robust architectures: Performance-Efficient (99.7 MB) achieving 98.3% F1-score with 1.99ms inference time on a Jetson NANO, and a Memory-Efficient (0.6 MB) that trains an epoch in 6.12 minutes on a Jetson AGX Orin. In experiments with up to 115 edge clients, FedADAS significantly outperforms traditional federated learning approaches at higher client participation, achieving up to 9974x reduction in communication cost while maintaining a superior tradeoff between personalization and generalization under extreme data heterogeneity, demonstrating its suitability for real-world deployment. Code is available at https://opensource.silicon-austria.com/mujtabaa/fedadas

Problem

• driver fatigue
• federated learning
• model heterogeneity
• communication overhead
• personalized data

Innovation

• Federated Distillation
• Model Heterogeneity
• On-Device Learning
• Communication Efficiency
• Driver Yawn Recognition