Problem 4

Improving orbit propagation of space objects: learning a propagator error

Brief description

As the number of objects in space continues to increase, managing them becomes crucial, requiring careful monitoring and detection of possible collisions, among other tasks. A fundamental step in this process is the propagation of the current state of the objects (position and velocity) into the future.

An accurate propagation can be obtained by considering complex physical-models of the different forces acting on the object. However, these models can be computationally intensive and impractical for many objects. An alternative approach is to employ a simpler propagator and then rely on a data-driven method to learn and correct the propagation errors.

The proposed challenge is to learn the propagator error from a dataset of past observations. In particular, we consider two interesting settings of different levels of difficulty:

1) Training on historical data of one orbit, predicting for future dates of the same orbit.

2) Training on data from one orbit, predicting for a different orbit.

Mathematical background:

Mathematical background: basic knowledge on python; machine learning.

Coordinator:

Filipa Valdeira, Universidade NOVA de Lisboa