Efficient UAV path planning using coverage map‐based value iteration

Abstract

AbstractThis letter presents an efficient coverage map‐based unmanned aerial vehicle (UAV) navigation framework in cellular communication systems. Unlike previous research that focused on viewing UAV navigation as a Markov decision process in unknown continuous state space and leveraged various model‐free and deep neural network‐based reinforcement learning algorithms, a more straightforward and efficient model‐based value iteration algorithm is proposed. The algorithm leverages prior knowledge obtained through empirical channel models to develop a sampled coverage map that can be used in value iteration. A deep neural network is subsequently trained with supervised learning to approximate the optimal Q function in continuous state space. Finally, the trained neural network is applied to obtain a UAV trajectory that optimizes the objective function.