BRAVE: Benefit-aware data offloading in UAV edge computing using multi-agent reinforcement learning

Abstract

Edge computing has emerged as a transformative technology in public safety and has the potential to support the rapid data processing and real-time decision-making during critical events. This paper introduces the BRAVE framework, a cutting-edge solution where the UAVs act as Mobile Edge Computing (MEC) servers, addressing users’ computational demands across disaster-stricken areas. An accurate UAV energy consumption model is introduced, including the UAV’s travel, processing, and hover energy. BRAVE accounts for both the users’ Quality of Service (QoS) requirements, such as latency and energy constraints, and UAV energy limitations in order to determine the UAVs’ optimal path planning. The BRAVE framework consists of a two-level decision-making mechanism: a submodular game-based model ensuring the users’ optimal data offloading strategies, with provable Pure Nash Equilibrium properties, and a reinforcement learning-driven UAV path planning mechanism maximizing the data collection efficiency. Furthermore, the framework extends to collaborative multi-agent reinforcement learning (BRAVE-MARL), enabling the UAVs’ coordination for enhanced service delivery. Extensive experiments validate the BRAVE framework’s adaptability and effectiveness and provide tailored solutions for diverse public safety scenarios.