A platform for the navigation of Unmanned Aerial Vehiclesbased on Reinforcement Learning

Abstract

This article introduces a platform for developing use cases for the automated control of unmanned aerial vehicles (UAVs), utilising the AirSim simulator. This platform allows for the generation of realistic flight scenarios involving multiple UAVs.The proposed platform facilitates the construction of use cases for the development, validation, and verification of reinforcement learning (RL) and neural networks in critical real-time systems. These algorithms can learn to navigate dynamic environments without human intervention. The platform was validated using two UAVs with neural networks that were trained using the Soft Actor-Critical (SAC) algorithm.