Machine Learning Application for Lambert's Problem

This study presents a neural network Lambert's approximator, a novel approach to address Lambert's problem using deep learning. The main contribution lies in providing a means to approximate transfer trajectories for preliminary mission design, utilizing machine learning in conjunction with prior knowledge in astrodynamics. By normalizing the gravitational parameter of a central body and initial distance, this approach is applicable to any two-body system, encompassing not only interplanetary trajectories but also moon tour problems. The proposed neural network model, trained on a dataset of transfer trajectories, demonstrates efficiency in approximating these trajectories. The study delves into the neural network architecture, training methodology, and performance evaluation, highlighting the potential of this approach to enhance the efficiency of preliminary mission design processes. The performance evaluation of the presented approximation approach is conducted with a Porkchop plot validation, speed test, and mission design applications such as the Cassini trajectory and Europa Clipper trajectory.