ConnextEdge: A Hierarchical Framework for Resilient Edge Analytics

The US Army aims to integrate multi-modal sensor data streams and advanced AI analytics at the tactical network edge in support of its vision for a future Internet of Battlefield Things. To maintain situational awareness within mission-acceptable levels despite dynamic conditions and infrastructure disruptions, we propose the ConnextEdge framework for resource-aware location-agnostic adaptive AI processing. This Agile AI methodology makes local decisions about where and how to analyze sensor data. It monitors network connectivity and other system resources (e.g. compute, battery power) and determines whether to process data at a local networked compute device or a remote edge server. This enables more powerful AI algorithms on the edge server to extract relevant information with high accuracy, or low-power less-accurate local analytics to maintain continuity of operations during infrastructure disruptions prevalent in battlefield environments. This effort will incorporate years of experience and novel research on mission-critical edge computing systems with a field-tested TRL-9 communications framework currently deployed in many DoD systems. Our team's strong synergy and existing foundations translates to a low risk to the Army and a shortened timeframe to the completion of both Phase I and II of this effort.