It’s Not Rocket Science: Machine Learning to the Rescue!

Kayla Vecera

The reason for the surge in the application of machine learning techniques to space exploration stems, in part, from the need to manage extremely large datasets. The data sets that are produced by space exploration efforts are enormous and call for the application of machine learning techniques.

The SKA project is an international effort to construct the world’s most enormous radio telescope and is the perfect illustration of data overload. SKA uses thousands of dishes and one million antennas to monitor the entire sky in great detail, transmitting data at higher speeds than any system in existence today. Together, these radio-telescopes generate approximately seven-hundred terabytes of data per second. In perspective, this amount of data is equivalent to the amount of data transmitted through the internet every two days (Kirkovska, 2018).

Screen Shot 2019-04-11 at 4.31.49 PM.png

That being said, transmitting such high volumes of data from deep space to Earth has generated issues. Seeing as host planets differ in their rotational speeds and direction/orbits, these massive data packets must be transmitted to Earth during specific windows of time. This delay, lasting between months and years, is dependent on how far Earth lies from the spacecraft’s host planet (Tian, 2018). This process is critical because a spacecraft’s tracking and communications systems are the only means with which to interact with it after it has left Earth because deep space missions don’t return to Earth after launch (Knosp, 2018). Suppose that data packet transmission is unsuccessful, it is possible for data to be permanently lost if it was overwritten with new data in the onboard memory (Tian, 2018).

Machine learning is instrumental in managing transmission issues. In 2005, the Mars Express AI Tool (MEXAR2), was introduced by the Institute for Science and Technology (ISTC-CNR) (Tian, 2018). The onboard learning algorithm leverages historical data to filter superfluous data and recognize the download schedule, ultimately optimizing data packet transmission (Tian, 2018). This outer data transmission technique is already implored by NASA and other space agencies worldwide in space research programs (Tian, 2018).

Screen Shot 2019-04-11 at 4.50.01 PM.png

MEXAR2 weighs variables that may impact data downloading. More specifically, MEXAR2 weights the overall science observation schedule for all Mars Express instruments in an attempt to predict which on-board data packets might be later lost due to memory conflicts. Based on these predictions, it generates a data download schedule and creates the commands needed to implement the download. Fred Jansen, ESA's mission manager for Mars Express, was quoted saying, ”With MEXAR2, any loss of stored data packets has been largely eliminated” (ESA, 2008).

Screen Shot 2019-04-11 at 4.49.22 PM.png

Thus, machine learning is imperative in the process of storing and processing the exponential amount of incoming data and generating valuable insights.


Chowdhury, Amit Paul. “How Big Data advances are fuelling space exploration.” Analytics India Magazine, January 10, 2017, accessed February 4, 2019.

Tian, Robert. “The New Age of Discovery: Space Exploration and Machine Learning.” Medium, March 31, 2018, accessed February 4, 2019.

Martin, David. “Origin of the Universe (3268): Projects- Square Kilometre Array.” Jet Propulsion Laboratory: California Institute of Technology, accessed February 4, 2019.

Kirkovska, Anita. “Big Data and its impact in the space sector, one bit at a time.” Medium, September 14, 2018, accessed February 4, 2019.

Knosp, Brian. "Deep Space Communications.” NASA: Jet Propulsion Laboratory, accessed February 28, 2019.

ESA. “Artificial Intelligence Boosts Science From Mars.” ESA: Our Activities: Operations, 29 April 2008, accessed April 11, 2019.