Scalable Coordinated Learning for H2M/R Applications over Optical Access Networks (Invited)
Sourav Mondal, Elaine Wong
TL;DR
This work addresses scalable human-to-machine/robot (H2M/R) collaboration over converged fiber-optic access networks by introducing the GLAD framework, which coordinates Local AIs at the network edge with a Global AI in the cloud to enable rapid onboarding of new machines while preserving ongoing operations. GLAD employs a binary touch classifier and reinforcement learning-based haptic forecasting to reduce end-to-end latency and leverages a federated-learning–like approach for cross-site knowledge sharing. The authors validate the approach with a VR haptic dataset, showing substantial training-time benefits and the ability to meet sub-millisecond latency targets over extended HO–machine distances, up to about 60 km under moderate network load. The results demonstrate a scalable path for Industry 5.0 H2M/R deployments on converged optical access networks, with potential improvements under lower network loads and larger-scale deployments.
Abstract
One of the primary research interests adhering to next-generation fiber-wireless access networks is human-to-machine/robot (H2M/R) collaborative communications facilitating Industry 5.0. This paper discusses scalable H2M/R communications across large geographical distances that also allow rapid onboarding of new machines/robots as $\sim72\%$ training time is saved through global-local coordinated learning.