Skip to main content

Artificial intelligence is learning how to dodge space junk in orbit

In September 2019, ESA's wind-monitoring satellite Aeolus came dangerously close to one of SpaceX's Starlink spacecraft. The space agency had to conduct an avoidance manoeuvre to prevent the collision.
In September 2019, ESA's wind-monitoring satellite Aeolus came dangerously close to one of SpaceX's Starlink spacecraft. The space agency had to conduct an avoidance manoeuvre to prevent the collision. (Image credit: ESA)

An AI-driven space debris-dodging system could soon replace expert teams dealing with growing numbers of orbital collision threats in the increasingly cluttered near-Earth environment.

Every two weeks, spacecraft controllers at the European Space Operations Centre (ESOC) in Darmstadt, Germany, have to conduct avoidance manoeuvres with one of their 20 low Earth orbit satellites, Holger Krag, the Head of Space Safety at the European Space Agency (ESA) said in a  news conference organized by ESA during the 8th European Space Debris Conference held virtually from Darmstadt Germany, April 20 to 23. There are at least five times as many close encounters that the agency's teams monitor and carefully evaluate, each requesting a multi-disciplinary team to be on call 24/7 for several days. 

"Every collision avoidance manoeuvre is a nuisance," Krag said. "Not only because of fuel consumption but also because of the preparation that goes into it. We have to book ground-station passes, which costs money, sometimes we even have to switch off the acquisition of scientific data. We have to have an expert team available round the clock."

The frequency of such situations is only expected to increase. Not all collision alerts are caused by pieces of space debris. Companies such as SpaceX, OneWeb and Amazon are building megaconstellations of thousands of satellites, lofting more spacecraft into orbit in a single month than used to be launched within an entire year only a few years ago. This increased space traffic is causing concerns among space debris experts. In fact, ESA said that nearly half of the conjunction alerts currently monitored by the agency's operators involve small satellites and constellation spacecraft. 

ESA, therefore, asked the global Artificial Intelligence community to help develop a system that would take care of space debris dodging autonomously or at least reduce the burden on the expert teams. 

"We made a large historic data set of past conjunction warnings available to a global expert community and tasked them to use AI [Artificial Intelligence] to predict the evolution of a collision risk of each alert over the three days following the alert," Rolf Densing, Director of ESA Operations said in the news conference. 

"The results are not yet perfect, but in many cases, AI was able to replicate the decision process and correctly identify in which cases we had to conduct the collision avoidance manoeuvre."

 Related: Astronomers ask UN committee to protect night skies from megaconstellations 

The agency will explore newer approaches to AI development, such as deep learning and neural networks, to improve the accuracy of the algorithms, Tim Flohrer, the Head of ESA's Space Debris Office told Space.com.

"The standard AI algorithms are trained on huge data sets," Flohrer said. "But the cases when we had actually conducted manoeuvres are not so many in AI terms. In the next phase we will look more closely into specialised AI approaches that can work with smaller data sets."

For now, the AI algorithms can aid the ground-based teams as they evaluate and monitor each conjunction alert, the warning that one of their satellites might be on a collision course with another orbiting body. According to Flohrer, such AI-assistance will help reduce the number of experts involved and help the agency deal with the increased space traffic expected in the near future. The decision whether to conduct an avoidance manoeuvre or not for now still has to be taken by a human operator. 

"So far, we have automated everything that would require an expert brain to be awake 24/7 to respond to and follow up the collision alerts," said Krag. "Making the ultimate decision whether to conduct the avoidance manoeuvre or not is the most complex part to be automated and we hope to find a solution to this problem within the next few years."

Ultimately, Densing added, the global community should work together to create a collision avoidance system similar to modern air-traffic management, which would work completely autonomously without the humans on the ground having to communicate. 

"In air traffic, they are a step further," Densing said. "Collision avoidance manoeuvres between planes are decentralised and take place automatically. We are not there yet, and it will likely take a bit more international coordination and discussions."

Not only are scientific satellites at risk of orbital collisions, but spacecraft like SpaceX's Crew Dragon could be affected as well. Recently, Crew Dragon Endeavour, with four astronauts on board,  reportedly came dangerously close to a small piece of debris on Saturday, April 24, during its cruise to the International Space Station. The collision alert forced the spacefarers to interrupt their leisure time, climb back into their space suits and buckle up in their seats to brace for a possible impact.

According to ESA, about 11,370 satellites have been launched since 1957, when the Soviet Union successfully orbited a beeping ball called Sputnik. About 6,900 of these satellites remain in orbit, but only 4,000 are still functioning.

Follow Tereza Pultarova on Twitter @TerezaPultarova. Follow us on Twitter @Spacedotcom and on Facebook.  

Join our Space Forums to keep talking space on the latest missions, night sky and more! And if you have a news tip, correction or comment, let us know at: community@space.com.