By Hope Hodge Seck,
Published by Space News, 4 November 2026
Researchers directed an ESA satellite to launch the game ‘Doom.’
Operators on the ground successfully directed a satellite on orbit with a voice command for the first time. Researchers told SpaceNews that their experimental success could be a step toward a future in which responsive satellites can listen for distress calls and emergency signals from ships at sea.
On July 27, Oslo-based amateur radio operators sent a voice command to the European Space Agency’s three-year-old CubeSat OPS-SAT PRETTY: a directive to launch the 1993 first-person shooter video game “Doom.” While the team chose to incorporate that particular game the experiment for comedic value (The OPS-SAT mission has booted up “Doom” in the past), satellites that are able to respond to voice calls could have far more practical uses in the long run, said Georges Labrèche, the principal investigator on the project and a systems engineer and founder of Estonian-based AI and space consultancy Tanagra Space.
While Labrèche told SpaceNews he didn’t think voice commanding satellites would become standard “anytime soon,” he said he saw potential for training satellites to recognize and respond to the “unstructured and unpredictable” patterns of the human voice versus just predefined patterns and formats in radio signals.
Being able to alert a satellite with a distress call from Earth may mean a more rapid rescue.
“Maybe there is an application there in terms of capturing [distress calls] on a global scale and acting on them immediately, instead of waiting until a Coast Guard or another ship,” he said. “Maybe this could be turned into a service where suddenly you can point to the air, call out for mayday and then a network of satellites is able to relay that immediately.”
There may be space observation applications too, he said.
“When you do astronomical observations, like observing gamma-ray bursts or black holes colliding, you’re looking for signals with very specific patterns,” he said. “So maybe there is a near future where instead of training a model to recognize a human voice, we train a model to recognize these astronomical events directly on board the satellite, so we can detect them more immediately.”
The experiment was possible because the satellite, having completed its original mission of GNSS reflectometry and radiation monitoring, was transferred to ESA’s OPS-SAT mission for open-ended experimentation in 2025 by its builder, Austria’s Graz University of Technology. Tangara Space was then able to upload voice-listening software to the satellite and lay the groundwork for the voice command experiment. With the increase in compute power onboard satellites in recent years, Labrèche said, the team was able to select for the software a 27-megabyte neural network model: “small enough to fit, but big enough to accomplish the mission.” Labrèche enlisted the radio amateurs, Eskil Hadland and Jon Bjergli, to provide a ground station for the test. From there, it took 59 days to achieve successful voice command recognition.
Labrèche explained that the team spent time testing different models and distorting voice signals with artificial noise “because we knew there would be when the voice capture was done it would be with a lot of noise, but still [have to] recognize the key words. So operating with those constraints on board was a technical challenge,” he said.
Credit: Truls Johansen (LA7IJ), LA4O / Tanagra Space
Hadland told SpaceNews that they tested some 30 different voices to ensure the satellite would not struggle to understand accents or differences in inflection. He added that he and Bjergli had to develop a makeshift way to contact OPS-SAT PRETTY during its sun-synchronous orbit, which had it passing overhead about once every 10 minutes.
“Normally, people who do communication with satellites like this, they will use automated tracking for estimates on altitude … we don’t have funding in our budget for this,” he said. “I went to a store and bought some street chalk … and we put up some markets on the roof of where we expected the satellite to come to and where it would end. And we just tried to time it.”
Following the voice command, according to Labrèche, the satellite sent back screenshots and a frame-by-frame animation of the Doom gameplay and details specific to each radio capture, along with a recording of the live voice command itself. According to the Oslo team, the whole voice command experiment cost about $780 to execute, “pizza included.”
“It’s really nice to push the envelope in terms of what complex artificial intelligence-related processing can happen on board a satellite,” Labrèche said. “At the end of the day, this is a 27-megabyte model, but still a complex neural network. So it’s cool to see another instance of how these neural networks can be used on board a satellite, and use that experience to approach future mission planning and design with that capability in mind.”
See: Original Article
