Auto-Tuned Primal-Dual Successive Convexification for Powered Descent Guidance

This paper presents discrete-time and continuous-time variants of an auto-tuned primal–dual successive convexification algorithm (AutoSCvx) for powered-descent guidance with signal temporal logic specifications. Virtual controls are commonly introduced to maintain the feasibility of the linearized dynamics, yet their penalty weights typically require manual tuning. We show that both variants (DT-AutoSCvx and CT-AutoSCvx) can automatically tune these penalties through closed-form updates, marking the first application of AutoSCvx to virtual control buffering of the dynamics rather than only to terminal state relaxation. Together, we propose a framework for nonconvex optimal control compatible with the following features: (1) auto-tuning of penalty weights for generalized constraint buffers, (2) a reformulation technique for continuous-time satisfaction of path constraints (3) a smooth and exact parameterization of signal temporal logic specifications used for enforcing high-level mission constraints, and (4) treatment of time-interval length as a free decision variable. Numerical results across representative scenarios demonstrate that the approach produces dynamically consistent trajectories comparable to hand-tuned methods while significantly improving robustness and usability by eliminating manual penalty selection.