L1 Penalized Sequential Convex Programming for Fast Trajectory Optimization: With Application to Optimal Missile Guidance

Abstract

In this paper, an L1 penalized sequential convex programming (LPSCP) method for trajectory optimization is proposed. Sequential convex methods dramatically reduce computation time for nonlinear trajectory optimizations, based on the desirable properties of convex optimization algorithms. Due to its fast convergence speed, the sequential convex method is considered as a near-future candidate for real-time optimal online guidance. However, the generic sequential convex method seldom suffers from poor robustness to crude the initial trajectory estimates and constraint perturbations, which limits its application to real-world systems. Suggested LPSCP method resolves the robustness issue of generic sequential convex methods, by addressing the convex subproblem infeasibility. While improving its robustness, proposed LPSCP method maintains its fast convergence property. Consequently, LPSCP method enhances overall versatility of sequential convex method for trajectory optimization, which includes ameliorated robustness toward initial trajectory estimates, and improved stability to state and constraint perturbations. Throughout the paper, details of LPSCP method are outlined along with an application example to missile trajectory optimization problems. Desirable properties of LPSCP method are demonstrated through several simulated examples. Optimized trajectories are compared with results from the pseudospectral method. Numerical simulation and Monte Carlo analysis show that LPSCP method accelerates optimization process dramatically (-98.2% on average) while producing the same optimal trajectory.