Low-Level Control of a Quadrotor using Twin Delayed Deep Deterministic Policy Gradient (TD3)
Mazen Shehab, Ahmed Zaghloul, Ayman El-Badawy
Abstract
Unmanned Aerial Vehicles (UAVs) like Quadrotors are inherently under-actuated and unstable systems. The mechanical complexity and non-linearity of such systems make it a difficult task to control the flight of the mentioned systems. However, due to recent advancements in the fields of data science and machine learning, new algorithms for flight stabilization and trajectory control were developed using Deep Reinforcement Learning. This paper presents two low-level Quadrotor controllers based on the same algorithm. The first designed controller aims to stabilize the Quadrotor at a certain preset point given any random initial position. The second is to track any target position given in the 3D space. Twin Delayed Deep Deterministic Policy Gradient (TD3) is used to train the agents to achieve the required tasks. This method is an off-policy Actor-Critic based method. It was used as it does not require a system model and works on environments with continuous action and state spaces. The superb performance of the trained policies is demonstrated in a simulation to illustrate the effectiveness of the proposed controllers.