New algorithm optimizes spacecraft trajectories under uncertainty with 43-page paper

Researchers Masahiro Fujiwara and Naoya Ozaki have submitted a 43-page paper to the Journal of Guidance, Control, and Dynamics introducing a stochastic differential dynamic programming (SDDP) algorithm for trajectory optimization under partial observability. The method directly addresses the challenge of designing spacecraft trajectories in the presence of stochastic effects like maneuver execution errors and observation uncertainties. Unlike traditional approaches that separate trajectory design, orbit determination, and correction maneuver planning, this algorithm tightly couples these tasks within a belief-space planning framework. It optimizes both the nominal control sequence and feedback gains while complying with general mission constraints, explicitly modeling how covariance propagation depends on the nominal trajectory without relying on the separation principle.

The proposed SDDP algorithm was validated through numerical experiments across various dynamical systems, observation models, and uncertainty levels. Notably, in the circular restricted three-body problem, the method exploited the coupling between trajectory design and orbit determination to produce navigation-aware solutions that achieved substantially lower fuel consumption compared to deterministic local optimization starting from the same initial guess. The paper demonstrates that by accounting for partial observability and uncertainty from the outset, spacecraft can execute more efficient missions. This approach is particularly relevant for deep-space missions where maneuver execution errors and observation uncertainties significantly impact mission success and fuel budgets. The algorithm represents a practical advancement for designing robust control policies and information-aware trajectories in real-world space operations.