Adaptive Multiscale Optimization: Concept and Case Study on Simulated UAV Surveillance Operations
Irwin O. Reyes, Peter A. Beling, Barry Horowitz · IEEE Systems Journal · 2015
This paper presents a resource-usage management scheme called adaptive multiscale optimization (AMO). AMO mitigates dynamically emerging bottlenecks in shared resources by enabling system components to individually select the appropriate operational mode to collectively improve the whole system performance. Components periodically collect and broadcast relevant local measurements to other components, in aggregate, producing an overall view of the performance of the system. These measurements, combined with the preset objective function that encodes the operator's priorities and desired tradeoffs, drive the logic that selects component modes as system state, and user demand changes over time. As a representative use case, we have analyzed its behavior as applied in a three-dimensional object-recognition application involving multiple unmanned aerial vehicles utilizing shared communications and data processing resources. We demonstrate AMO's benefits and tradeoffs through a series of simulator runs, covering such use cases as increasing contention, sudden reduction in system capacity, and varying AMO coordination overhead.