An autonomous, multi-agent UAV platform for inspection of civil infrastructure

UAVs (Unmanned Aerial Vehicles) are nowadays being used more and more in Structural Health Monitoring (SHM). Their versatility, speed, and manoeuvrability make them the ideal means to perform inspections autonomously and remotely, instead of relying on visual inspections carried out by human operators. Since commercial drones have limited flight times, the information collected in this short span must be maximised: to tackle the problem of gathering the maximum amount of data in the shortest possible time, we propose a platform where a central controller coordinates multiple UAVs. We address 1) the problem of generating points of interest, i.e., positions from which a sensor reading must be taken, given a 3D model of the structure, 2) the problem of assigning the points to the drones and finding the optimal traversal order of such points, in order to minimise the total flight time and make the best possible use of each drone's battery capacity. We decouple the two problems by first generating points of interest, starting from the structure's virtual model, and then feeding those points to a central mission planner that employs a linear programming formulation to find near-optimal trajectories for each agent, guaranteeing obstacle avoidance. We also address the issue of robustness of the whole system against the failure of an aircraft. We evaluate our method by applying it to the inspection of a virtual model of an existing building. We find that our approach yields good solutions in a reasonably short time, justifying its use as a robust mission planning algorithm.