Global-Local AI Coordinated Learning over Optical Access Networks for Scalable H2M/R Collaborations

Abstract

While pre-fifth and fifth generation networks have been designed mainly to support machine-to-machine and human-to-human communications via data, audio, and video communications, a major thrust of beyond fifth generation Networks is the abutment of human-to-machine/robot (H2M/R) collaboration. Such collaborations are deeply anchored in the vision of Society 5.0, which promotes a human-centered society that integrates virtual and physical real spaces to resolve social problems, and Industry 5.0, which promotes intelligent manufacturing processes through collaboration between humans and cyber-physical systems. It is expected that real-time H2M/R collaborations will drive new technologies that facilitate the low-latency exchange of control signals and haptic feedback between humans and machines/robots. The low-latency constraints for real-time interaction will, however, place limits on deployable distance and the flexibility of adding new machines/robots to the system. The scalability of H2M/R collaborations across large geographical distances that also allow rapid onboarding of new machines/robots is therefore an open challenge. This article focuses on global-local AI coordinated learning (GLAD), a solution framework that supports the dynamic addition of new machines/robots and their rapid onboarding. The GLAD framework is based on learning the traffic characteristics and correlation of supported robots/machines at the local edge and global cloud. Harnessing data samples from a real H2M/R application for training and validation, GLAD enables human operators to remotely control machines/robots over extended distances through accurately forecasting their haptic feedback. By globally coordinating and sharing this learning across the network, up to 72 percent of the time can be saved when onboarding new machines/robots.