Flight Control of a Multicopter using Reinforcement Learning

Abstract

Machine Learning, and in particular Reinforcement Learning, is a persistent trend in automation and robotics in recent years. Many researchers worldwide are developing intelligent controllers using Reinforcement Learning techniques. This paper aims to present a proof-of-concept Reinforcement Learning flight controller for a multicopter. The agent has been trained in the Airsim simulation environment to achieve stable flight conditions by controlling its roll, pitch, yaw and throttle. After training, the agent has been tested on the same environment to prove its ability to maintain stable flight conditions while following a determined route.