Multi-agent reinforcement learning with information sharing for optimal drone mapping

Abstract

Multi-drone systems have been widely used for various applications which will otherwise be difficult to be done us ing only single drone operation. One important challenge in such multi-drone system operation is how to optimize their performance when required to operate in unknown environment. In this paper, a multi-agent reinforcement learning (MARL) scheme is proposed to initiate a cooperative operation of a multidrone system that is tasked to perform a 3D space mapping or a region. The proposed MARL method is designed to optimize the multi-drone system's energy consumption by introducing a sparse cooperative interaction scheme. In this regard, each drone either communicates with the other when needed or perform its individual learning otherwise. Simulation results are shown to illustrate the convergence/optimality of the used MARL scheme.