Application of modern deep learning to satellite attitude control

卫星自主和机器学习是日益增长的研究兴趣领域。卫星姿态控制器通常是为具有特定质量分布的某些卫星开发的，这些卫星利用特定的执行器来尝试实现特定的姿态对准目标。虽然有一些通用的控制器可以用于各种卫星，如PID控制器，但这些控制器往往不能达到最佳性能。一个新的方向是考虑将深度强化学习用于卫星姿态控制。

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.