Workshop on Collision Avoidance Challenge Results

Europe/Amsterdam
Francesca Letizia (European Space Agency), Klaus Merz (ESA)
Description

Collision avoidance have become common practice for spacecraft operators in the last decade. While every operator follows its own approach with own procedures and tools a common input are standardised conjunction data messages as issued by maintainers of object catalogues, such as the US Space Surveillance Network.

The typical current approach of experts analysing critical events is reaching limits due to the increase of the known number of such close approaches, which is driven by both a rapidly expanding spaceflight activity, i.e. new spacecraft being launched, but also by the increase of catalogue sizes due to enhanced surveillance systems.

In order to explore whether enhanced techniques such as machine learning can help assessing criticality of a close approach and ultimately in the decision whether or not perform an avoidance manoeuvre, ESA prepared a public challenge on its Kelvins platform (https://kelvins.esa.int/). This challenge made for the first time a larger set of Collision Data Messages (CDM) accessible to the wider public and asked for predicting the collision probability of the CDM obtained closest to the time of closest approach (thus representing the best knowledge of the collision risk of the event) however using only CDMs at least 2 days before the conjunction.

In this workshop, some of the leading teams of the challenge will present their approach, lessons learnt and potential ways ahead. It will take place alongside the 8th European Conference on Space Debris (https://space-debris-conference.sdo.esoc.esa.int/) which covers collision avoidance including machine-learning techniques during the same day.

The challenge details can be found at https://kelvins.esa.int/collision-avoidance-challenge/

Results have been published at https://link.springer.com/article/10.1007/s42064-021-0101-5 and http://iafastro.directory/iac/paper/id/57288/summary/

The dataset can be downloaded at https://zenodo.org/record/4463683#.YA8BVBZG02w 

Registration
Connection details
    • 1
      Introduction (ESA)
    • 2
      Engineered Decision Tree for Judging Spacecraft Collision Risks, Sven Rehban (Honda Research Institute Europe)
    • 3
      Learning from an imbalanced dataset of conjunction data messages, Rasit Abay (NeuraSpace)
    • 4
      Machine learning-powered algorithms for predicting the risk of satellite collisions, Michal Myller (KP Labs)
    • 5
      Outlook (ESA)