Effective Learning Algorithms for Search and Rescue Missions in Unknown Environments

Abstract

During the last decade, there has been a huge attention on deployment of unmanned aerial vehicles (UAVs) for both civilian and military missions. Alike swarming phenomena in animals and insects, these herds of smart robots could perform complicated task when working together. However, doing so needs smart coordination between individual subsystems. In this paper, aerial coordination by a UAV (also known as aerial shepherd) for a group of unmanned ground vehicles (UGVs) is investigated. The problem is separated in two sections. In the first stage, the aerial leader learns how to find a target using both Q-Learning and DQN techniques. In the next step, the optimal path is sent through a communication link to a group of mobile robots for ground coordination. Through simulations, we study the problem in a model-free environment with existence of stochasticity in control input and the environment. Once the target has been localized, the optimal path is sent to ground mobile robots to provide rescue. The obtained results for a primitive environment indicate successful coordination of mobile robots on the ground.