Bayesian active verification of lifeline roads for risk aware emergency logistics with trucks and drones

Post-disaster delivery under uncertain road access is an active decision-making problem under partial observability. An emergency controller must decide not only how to route vehicles, but also whether scarce drone capacity should be used to acquire road information before delivery. This paper treats lifeline road verification as a resource-coupled information-purchase problem in a rolling-horizon truck–drone emergency logistics controller with Bayesian belief updates. The proposed routing-aware value of information (RA-VoI) scores each candidate inspection by the expected downstream delivery-cost reduction it induces. Its conditional value-at-risk (CVaR) extension, RA-CVaR-VoI, moves lower-tail service-loss evaluation upstream into the same verification decision. On synthetic flooded-road networks with a six-stage horizon, route-aware verification reduces the $$n=20$$ catastrophic-loss rate from 2.53% under maximum-entropy probing to 1.25%. At $$n=40$$ under a fixed two-probe budget, broader uncertainty probing regains strength, revealing a capacity–coverage boundary for route-consequence screening. In a focused ablation over four boundary cells, the pure-tail RA-CVaR-VoI endpoint attains the lowest severe-loss rate (0.105) and uses the fewest inspections (2.66). The findings are regime-dependent: route- and tail-aware verification reduces severe service losses where verification capacity can reach route-critical bottlenecks, whereas broad uncertainty probing can be stronger under high sensing pressure and a fixed budget—a capacity–coverage boundary rather than a general superiority of route-aware verification. The evidence comes primarily from synthetic networks and approximate rolling-horizon control, supplemented by a Zhengzhou case grounded in real road geography and documented disruption locations; a global sensitivity analysis over nine factors supports the regime-dependent interpretation.