Integrating PINN Gravity Field with Nyx for LRO Orbit Determination

[ChristopherRabotin]

Hi John,

I've been following your work on the PINN gravity field since your impressive 2021 paper. As you may be aware, I've been developing Nyx for several years, a high fidelity mission design and orbit determination toolkit. It's been used on several high profile missions.

Recently, I successfully demonstrated Nyx's capability to solve the orbit determination of the Lunar Reconnaissance Orbiter with precision that meets mission requirements. Nyx uses Pines' algorithm for gravity field computation. In the third-generation of your work, you presented a trained neural network model of the Moon's gravity field.

I am interested in demonstrating this neural network for estimating the orbit of LRO in Nyx. As a first step, I imagine that the network will need to be serialized and that Nyx would offload the gravitational field accelerations to a large tensor multiplication. If time allows, could you guide me on this integration? Alternatively, could you direct me to the relevant scripts in your repo where acceleration computations are performed, and specify which neural network outputs should be serialized?

Thanks for your help