Precise Rendezvous Guidance in Low Earth Orbit via Machine Learning

This paper presents a precise rendezvous guidance technique in low Earth orbit (LEO) via machine learning. The guidance law must be tolerant of such disturbances as variations of initial conditions due to navigation and/or control errors. Precise relative trajectory control is a particularly key technology in the proximity operation in terms of safety requirements such as collision avoidance. The conventional guidance law in LEO utilizes the Clohessy-Wiltshire solution that is the analytical solution of Hill's equation restricted in circular orbits. This solution is characterized by being able to control the position of a spacecraft at a specific timing, whereas the spacecraft's velocity at the same timing is dependent on estimated orbital states at the timing of impulsive maneuver. Excessive guidance error in terms of velocity results in a wider dispersion of relative trajectories, possibly resulting in trajectories with less safety. It may lead to strict accuracy requirements on navigation sensors or control devices, thereby requiring the recognition of guidance accuracy as a potential cost driving factor in a space mission project. The proposed method aims at improving guidance velocity errors by controlling maneuver impulse timing. It utilizes a linear regression modeling approach to predict optimum impulse timing. The linear regression model that estimates guidance velocity error is formulated by utilizing all elements of disturbances and maneuver impulse timing. An optimum maneuver impulse timing is predicted by utilizing all elements of disturbances as derived from the onboard navigation solution, given the predefined target value of the guidance velocity error. Conventional rendezvous guidance trajectory planning assumes a fixed maneuver impulse timing to simplify the planning, which results in less precise guidance accuracy. The proposed methodology is an innovative way to overcome this concern by implementing a simple algorithm with a light computation load.