Decentralized Critical Area Coverage Using Multi-UAV System with Guided Explorations During Floods

During natural disasters, having access to realtime ground information is paramount for disaster response teams to act swiftly and efficiently. To provide immediate assistance, Unmanned Aerial Vehicles (UAVs) can be quickly deployed to assess the situation on the ground. This paper introduces a decentralized multi-UAV control algorithm based on deep reinforcement learning to perform the task of real-time data collection from the flooded region autonomously. The decentralized policy allows the UAVs to work independently without any central command and control unit. This relieves the rescue teams from deploying any manpower in managing a centralized control unit to handle the UAVs. In this paper, the objective of the multi-UAV system is to independently distribute themselves to collect critical ground information from the flooded areas. The task is time-sensitive as the UAVs have finite energy and human lives might be at risk. To address this, we propose a decentralized training and execution approach with opportunistic communication. This allows each UAV to make decisions based on local information while also intermittently communicating with other UAVs. To develop optimal control policies, we employ the water flow estimation algorithm called D8 which utilizes surface elevation information to generate directed exploration strategies. We conducted experiments on simulated flood environments inspired by real-world scenarios. Our proposed decentralized multi-UAV model called dec-DQNC8 was compared with other prevalent area coverage techniques from the literature. The results signify the impact of our proposed model as it outperformed other techniques during training and demonstrated optimal performance when evaluated over a test environment.