Scalable, Anytime Constraint Optimization through Iterated, Peer-to-Peer Interaction in Sparsely-Connected Networks

Approach: Scan Scheduling • Each radar’s actions are scheduled over a reasonable period – targets are reasonably predictable for ~15 seconds – rescheduling allows radars to adapt to changes (e.g., a target turning) • Metrics quantify schedule quality w.r.t. target behavior – high scores when targets are scanned simultaneously by about three radars – also take into account cost of scanning and constraint violations • Objective is to determine scan schedules that optimize overall metric – an overall metric can be defined in terms of expected values – in practice, need simplifying assumptions to reduce computations Distributed Computation of Scan Schedules • Define a local, per-radar version of the quality metric – assume that each radar knows about targets in its vicinity – assume that each radar knows the scan schedules of nearby radars – then each radar can compute the quality of its scan schedule based on local information • Each radar optimizes its own schedule w.r.t. its local quality metric – assume that a stable set of locally-good schedules is computed • given stable target behavior – then overall quality is expected to be good for practical sensor applications • can not in general make claims about true optimality – there may be pathological metrics for which achieving everywhere locallygood quality results in overall poor quality • probably will not occur in sensor domain – more practical concern is rate of convergence and stability • how to validate assumptions ...