Deep Reinforcement Learning Multi-UAV Trajectory Control for Target Tracking

In this article, we propose a novel deep reinforcement learning (DRL) approach for controlling multiple unmanned aerial vehicles (UAVs) with the ultimate purpose of tracking first responders (FRs) in challenging 3-D environments presence obstacles and occlusions. We assume that UAVs receive noisy distance measurements from FRs which are two types, i.e., Line Sight (LoS) non-LoS (NLoS) used by UAV agents order to estimate state (i.e., position) FRs. Subsequently, proposed DRL-based controller selects optimal joint control actions according Cramér–Rao lower bound (CRLB) measurement likelihood function achieve high performance. Specifically, quantified reward function, considers both CRLB entire system each UAV’s individual contribution system, called global difference reward, respectively. Since take reduce accuracy is improved ensuring reception quality LoS probability. Our simulation results show provides highly accurate target solution very low runtime cost.