Edge-Opportunistic Loop-Closure Verification for Resource-Constrained UAV Swarm SLAM

In swarm SLAM the most expensive and the most dangerouscomputation is one and the same: the geometric verification ofan inter-UAV loop closure. A wrong closure corrupts the mapof every vehicle, while a light UAV that verifies its own candidatesmisses its deadlines. We propose Edge-OpportunisticLoop-Closure Verification (EO-LCV), which ships each candidateunder a time-bounded lease to whichever reachable peercurrently promises the shortest expected verification. Peer speedsare learned online by shrinkage-Beta Thompson sampling, andacceptance is a Beta-binomial posterior test whose prior is reanchoredby a CUSUM detector when the swarm enters a perceptuallyaliased region. We prove that speed-weighted dispatchis stable throughout the opportunistic capacity region, with anexplicit backlog bound. We verify by exhaustive model checking(988,429 and 1,950,358 reachable states) that no closure is evercommitted without an accepting verdict under a live lease, andwe show that two seeded protocol bugs are found with counterexamplesof 7 and 4 steps. On a six-UAV swarm at 80% load,EO-LCV loses 8.5% of candidates to the deadline, against 14.5%for join-shortest-queue and 44.2% for on-board verification.