Attitude Control in Unmanned Aerial Vehicles Using Reinforcement Learning—A Survey

Abstract

Unmanned Aerial Vehicles (UAVs) have great potential in various fields. With ongoing research, the day is not far when they would be directly impacting our lives. The promise of deep learning techniques like reinforcement learning has created a window of opportunity for their use in a plethora of tasks, like the attitude control problem of UAVs. Attitude of a UAV is the angle at which it is flying relative to the ground. Attitude control is the management of the orientation of a UAV with respect to the inertial frame. In this paper, we have surveyed reinforcement learning algorithms to learn attitude control of UAVs and be able to take decisions in unforeseen circumstances. Reinforcement learning is the branch of deep learning where there is no human intervention in training the model. Instead, the system learns over time by trial and error. Since navigation in the air presents scenarios that may be new and unexpected for a UAV, reinforcement learning presents a viable option for their use in attitude control in them.