Spacecraft Control Design Methodology Based on Reinforcement Learning Methods

The paper formulates a methodology for reducing the general problem of optimal spacecraft control to a problem of reinforcement machine learning. The methodology includes a method for assessing the quality of a control algorithm based on probability theory inequalities. An author’s software toolbox is presented for reducing optimal control problems to reinforcement learning. Two examples of application of the methodology are considered. The proposed technique may also be of interest for general mechanical systems control design.