Conflict-aware AoI Minimization in Multi-UAV Data Collection Networks

Abstract

Unmanned aerial vehicles (UAVs) provide a flexible platform for collecting time-sensitive status updates from distributed sensors, where age of information (AoI) is widely used to measure data freshness. In multi-UAV data collection networks, decentralized execution requires UAVs to make simultaneous scheduling decisions, which may cause sensor-selection conflicts and degraded AoI performance. To address these limitations, we propose CP-MAPPO, a conformal prediction-augmented multi-agent proximal policy optimization framework for conflict-aware AoI minimization in multi-UAV networks. Rather than predicting a single action, CP-MAPPO constructs adaptive prediction sets for other UAVs’ possible scheduling actions. The calibrated prediction sets are then encoded as auxiliary inputs to MAPPO, providing each UAV with action-level coordination information about the possible scheduling decisions of other UAVs. By jointly exploiting global network states and conformal action-set information, CP-MAPPO learns an optimized action-selection policy for each UAV, balancing AoI minimization with conflict mitigation under decentralized execution. Simulation results show that CP-MAPPO outperforms representative baselines in terms of AoI reduction and conflict mitigation.