Crowdsensing for Emergency Response in Unknown Environments: A Rapid Strategic Sensing Approach

Integrating Unmanned Aerial Vehicles (UAVs) andautonomous vehicles within the crowdsensing paradigm offers apromising approach to collecting environment-relevant data overlarge spatial areas, particularly in disaster-stricken or high-riskregions. However, deploying crowdsensing systems in emergencyresponse scenarios presents substantial challenges. The lack ofprior environmental knowledge complicates the selection of opti-mal sensing locations and strategy optimization, often relying oncostly trial-and-error methods. Additionally, real-time decision-making is critical in such scenarios, requiring the rapid identifi-cation of optimal deployment strategies. Yet, the absence of priorknowledge further complicates the assessment of the optimalityof these strategies. This gap remains inadequately addressed inexisting research. To address this, we present the first frameworkthat frames these challenges as a rapid online strategy optimizationproblem for mobile agent-based crowdsensing systems operating inunknown environments during emergency response scenarios. Wepropose DGap−UCB, a novel approach within the multi-armedbandit (MAB) framework, which efficiently identifies the optimalsensing strategy with high-confidence guarantees. Leveraging theUpper-Confidence Bound (UCB) technique, DGap−UCB itera-tively refines strategy selection based on reward feedback. To accel-erate learning, we introduce a gap-confidence pair (Δ_t, δ_t)-basedQuick Stopping Criterion, enabling rapid and high-confidence iden-tification of the optimal strategy. Extensive experiments on bothsynthetic and real-world datasets demonstrate the superiority ofDGap−UCB over state-of-the-art techniques.