Synthetic Collision Dataset for Spacecraft Collision Avoidance

This dataset is intended to be used as a banchmark for testing collision avoidance strategies.

It is made of 21000000 relative geometries between LEO space objects, 1000 of which are true collision (miss-distance smaller than combined hard body radius).These relative geometries are expressed as target and chaser 6-dimensional state vectors (cartesian coordinates) at time of closest approach.

The relative geometry of the encounters are statistically matched to the ESA's Kelvins dataset for the collision avoidance challenge through statistical fitting methods.

The collision proportion is tuned to reflect a 1year mission in LEO orbit with an a-priori collision probability of 1e-3 (yearly) and a 21 collision warnings per year.

* Implementation Description *

The dataset is made of a series of .mat files storing the following variables:

'rv_t', target's cartesian state at TCA (km, km/s) 6xN vector

'rv_c', chaser's cartesian state at TCA (km, km/s) 6xN vector 

'Ct', target's covariance state matrix at TCA (km, km/s) 3x3xN 

'Cc', chaser's covariance state matrix at TCA (km, km/s) 3x3xN 

'Rc', combined hard body radius (m) 1xN

'CollFlag', logic value of collision 1xN (0: no-collision, 1: collision)

'missDistance', miss distance at TCA (km) Nx1

The name of the .mat file is formatted as:

batch_.mat

Each batch file has a maximum dimension of N = 1e5.