Robust Satellite Formation Flying Through Online Trajectory Optimization using LQR and Neural Networks

Utilizing the well-established linear quadratic regulator (LQR) theory and augmenting it with online trained neural networks, an innovative nonlinear online trajectory optimization technique is presented in this paper. Neural networks are trained online using close form expression for weight update rule and do not require any iterative process. The overall structure leads to robust optimal control synthesis and works well despite the presence of un-modeled dynamics. This technique is subsequently applied to the challenging problem of satellite formation ying. Simulation studies show that the presented control synthesis approach is able to ensure close formation ying catering for large initial separation, high eccentricity orbits, uncertain semi-major axis of chief satellite and J2 gravitational e_ects, which is usually considered as an exogenous perturbation.