Drone patrolling with reinforcement learning

When a camera-equipped drone is assigned a patrolling task, it typically follows a pre-defined path that evenly covers the whole environment. In this paper instead we consider the problem of finding an ideal path under the assumption that not all the areas have the same coverage requirements. We thus propose a reinforcement learning approach that, given a relevance map representing coverage requirements, autonomously chooses the best drone actions to optimize the coverage.