FLUDITY Unleashed: Dynamic Blockchain-Enabled Federated Learning Architecture for Tactile IoT Applications

Abstract

The Tactile Internet of Things (TIoT) demands ultra-reliable, low-latency communication for real-time haptic applications in domains such as remote surgery and human-robot collaboration. Conventional Federated Learning (FL) architectures-centralized, hierarchical, or purely decentralized-each suffer from bottlenecks in scalability, latency, or trust. We propose Federated Learning Using Distributed Infrastructure for TIoT (FLUDITY), a hybrid FL framework leveraging Blockchain, Multi-Edge Computing (MEC), and a Dynamic Aggregation Decision Algorithm (DADA). Evaluated via the TACTO (a fast and flexible tactile simulator) tactile simulator in a remote grasping task, FLUDITY achieves up to 30% reduction in convergence time (880 s vs. 1,258 s), 15% fewer Floating Point Operations (FLOPs) (74.8 × 10¹² vs. 88 × 10¹²), and 25% lower communication overhead (1,575 MB vs. 2,099.9 MB) compared to static blockchain FL.