Automated design of efficient swarming behaviours: a Q-learning hyper-heuristic approach

While the sector of unmanned aerial vehicles (UAVs) is experiencing an exponential growth since several years, the majority of applications consider single devices which come with limitations such as flight duration or payload capacity. A promising way to overcome these is the usage of multiple autonomous UAVs synergistically, also referred to as swarms. Many metaheuristics have been manually designed to optimise the performance of swarms of unmanned vehicles. However developing and fine tuning efficient collective behaviours can be a challenging and time-consuming task. This article proposes to automate the generation of UAV swarming behaviours which optimise the Coverage of a Connected UAV Swarm (CCUS) problem where both the coverage time and the network connectivity are considered. To this end, we introduce a novel generative hyper-heuristic based on Q-Learning (QLHH) and evaluate the performance of the heuristics it generates to manually designed heuristics using state-of-the-art coverage and connectivity metrics. The obtained results demonstrate the capacity of QLHH to generate efficient distributed heuristics for the CCUS optimisation problem.