Landing Guidance for Reusable Launch Vehicle Using Sequential Convex Programming With Deep Neural Network Based Initialization

This paper addresses the nonlinear trajectory optimization problem for the landing guidance of vertical takeoff vertical landing reusable launch vehicles, considering translational and rotational constraints. We formulate a 6-degree-of-freedom minimum fuel consumption trajectory optimization problem using dual quaternions and implement onboard sequential convex programming (SCP) with ECOS, a second order cone programming solver. In addition, this paper proposes a deep neural network based method for initial reference trajectory estimation to enhance computation speed of the onboard SCP algorithm. Monte Carlo simulation results show that spline interpolation methods, which more accurately reflect local sharp changes in the initial reference trajectories, outperform polynomial interpolation methods.