Optimal multi-agent path planning for fast inverse modeling in UAV-based flood sensing applications

Abstract —Floods are the most common natural disasters,causing thousands of casualties every year in the world. Inparticular, ash ood events are particularly deadly becauseof the short timescales on which they occur. Unmanned airvehicles equipped with mobile microsensors could be capableof sensing ash oods in real time, saving lives and greatlyimproving the efciency of the emergency response. However,of the main issues arising with sensing oods is the difcultyof planning the path of the sensing agents in advance so asto obtain meaningful data as fast as possible. In this particle,we present a fast numerical scheme to quickly compute thetrajectories of a set of UAVs in order to maximize the accuracyof model parameter estimation over a time horizon. Simulationresults are presented, a preliminary testbed is briey described,and future research directions and problems are discussed. I. I NTRODUCTION Floods are one of the most commonly occurring naturaldisasters, and caused more than 120,000 fatalities in theworld between 1991 and 2005 [1]. They are a major problemin many areas in the world, and are expected to becomeworse due to global warming, which causes more extremeweather events around coastal areas. In 2010, oods were notonly responsible for more than 4000 deaths worldwide butalso caused considerable economic loss. The 2009 Jeddahoods claimed hundreds of lives and caused hundreds ofmillions of dollars of property damage, with for instancemore than 10,000 vehicles lost. While these natural disastersare unavoidable, the loss of lives can be minimized bya proper warning system, which can also give emergencyresponders real time data to organize their operations.Though rain monitoring systems have been used for oodprediction and estimation before [4], ood caused by extremerains cannot be accurately predicted with these systems, asood propagation models require a large number of parame-ters which are difcult to know beforehand. Similarly, xedwater level sensors are only adapted to river monitoring, andare unsuitable to desert environments in which the locationand the extent of a ood cannot be estimated reliably. In alarge scale coastal city such as Jeddah, the surface of thehydrological basin to monitor is in the orders of thousandsof square kilometers, as illustrated in Figure 1, which makesa xed monitoring infrastructure economically infeasible.A recent effort to directly measure water levels in a riverwas investigated in [3], but this technology only applies torivers and cannot monitor large hydrological basins in whichnew water channels that cannot be predicted in advanceare formed during oods. Other such efforts involve theuse of indirect measurements from rain stations or frommeteorological data, combined with ood propagation mod-els that attempt to predict ood parameters such as theextension of ooded areas, water velocities and levels. Themain drawback with this approach is the lack of accuratemodel parameters. The inaccuracy of these model parameterscan drastically change the predicted ood behavior, leadingto a very unreliable ood warning system.A new form of sensing known as Lagrangian sensingmakes the use of mobile (oating) sensors, and has beenrecently investigated in the context of hydrological sensingin [24]. Lagrangian sensing is very promising for largescale sensing, or on demand sensing, as it requires minimalinfrastructure. While operating costs during sensing opera-tions are higher in Lagrangian sensors than in their Eulerian(xed sensors) counterparts, the relatively rare occurrence ofoods makes the Lagrangian sensing very suitable to oodmonitoring. An added economic benet is the possibility toredeploy the mobile sensors on demand.The latest advances in unmanned aerial vehicles (UAVs)signicantly improved their reliability and increased theirapplication range [19]. As a result, an abundance of UAV-assisted projects has been witnessed recently. A Global HawkUAV was deployed in Japan to assess the damage inside a nu-clear plant in the aftermath of an earthquake [6]. A Europeanproject has been launched for the exploration and measure-ment of volcanic activities [18]. Adams et al. [2] presented adamage assessment-based UAV system for hurricane events.All of these recent applications leave no doubt that UAVswill play a major role in future remote sensing technology.This article contributes to the efforts of developing reliablereal-time UAVs-based disaster monitoring systems.This project proposes the use of UAVs as a platform forLagrangian ood sensing using microsensors [5]. In thisframework, a swarm of UAVs would drop small disposablewireless sensors over the areas to monitor. These wirelesssensors would be buoyant, and would be carried away bythe ood. UAVs will receive signals from these sensors,and will map the extent of the ood, transmitting back thisdata in real time to a xed ground station for processing(in particular estimation, forecast or inverse modeling). Thisdirect measurement data will provide the authorities with