Exact and Approximate Relaxation Techniques for Computational Guidance

The focus of this dissertation is in the development and application of relaxation techniques that enable efficient and real-time solution of complex computational guidance problems. Relaxations transform a non-convex constraint into a convex constraint and provides proof that the optimal solutions to the relaxed problem are optimal for the original problem. Unique contributions of this work include: 1) a relaxation technique for solving fixed final time problems between fixed points, 2) a performance analysis on the application of computational guidance for the Mars Ascent Vehicle, and 3) establishment of sufficient conditions for non-singularity of optimal control for problems on a smooth manifold with mixed constraints. The first result states that for annularly constrained linear systems, controllability is a sufficient condition for the problem to be solvable as a sequence of convex programs. The second result applies relaxations to an ascent problem. The third result is the most general result to date for problems with mixed constraints. It uses a minimum principle on manifolds with mixed constraints to analyze the problem in a geometric framework, and shows that strong observability of the dual system is sufficient for non-singularity