Abstract:

Classical powered descent guidance (PDG) algorithms focus on low-level trajectory op- timization, such as minimizing propellant consumption subject to soft-landing constraints. However, in future missions to unexplored bodies, the dominant challenge is high-level decision making: during descent, the lander must continuously trade off gathering information about terrain hazards, preserving divert options, and committing to a final landing site. In this paper we formalize such a powered descent decision-making (PDDM) problem as a belief Markov decision process whose objective is to maximize the probability of a safe landing. Directly solving the resulting belief- and set-valued optimal control problem is intractable, so we propose a reachability-steering guidance algorithm with a one-step utility function with precomputed constrained controllable sets. The utility function balances exploitation of the current safest candidate site against the preservation of backup divert options, while the containment constraint in the visibility-safe controllable set guarantees recursive feasibility along the descent. Offline, the controllable sets are computed efficiently using constrained zonotopes combined with a lossless convexification of the PDG problem. Numerical simulations with stochastic safety belief evolution demonstrate that the proposed approach improves the probability of a safe landing relative to a greedy retargeting baseline. Moderate weighting of divert options yields trajectories that remain resilient to the stochastic safety belief process without incurring excessive conservatism in fuel usage.