Application of modern deep learning to satellite attitude control

Satellite autonomy and machine learning are areas of growing research interest. Satellite attitude controllers are often developed for certain satellites which have specific mass distributions, which make use of a specific actuator, and attempt to achieve a specific attitude alignment objective. While there are general controllers which can be used for various satellites, such as PID controllers, these often do not achieve optimal performance. A new direction is to consider deep reinforcement learning for satellite attitude control.