A deep hierarchical reinforcement learner for aerial shepherding of ground swarms

Abstract

This paper introduces a deep reinforcement learning method to train an autonomous aerial agent acting as a shepherd to provide guidance for a swarm of ground vehicles. The learner is situated within a high-fidelity robotic-operating-system (ROS)-based simulation environment consisting of an Unmanned Aerial Vehicle (UAV) learning to guide a swarm of Unmanned Ground Vehicles (UGVs) to a target location. Our approach uses a combination of machine education, apprenticeship bootstrapping, and deep-learning-based methodologies to decompose the complex shepherding strategy into sub-problems requiring simpler skills that get fused to form the overall skills required for shepherding. The proposed methodology is effective in training the UAV agent with multiple reward designing schemes.