Compositional Diffusion Models for Spacecraft Powered-Descent Trajectory Generation
A generative approach to spacecraft trajectory design that models the space of feasible 6-degree-of-freedom powered-descent trajectories as a probability density learned by a diffusion model, then composes separate densities for distinct constraints and design specifications. Composition lets the model adapt to new, out-of-distribution problem variations in a few-shot manner and produce dynamically feasible initial guesses for downstream guidance, including inference-time minimum-fuel landing-site selection.
Diffusion Policies for Generative Modeling of Spacecraft Trajectories Julia Briden∗and Breanna
This paper applies compositional diffusion modeling to 6-degree-of-freedom powered-descent spacecraft trajectory generation. Rather than training a single network to map problem parameters to a solut…