Generative AI Gives Spacecraft the Autonomy Engineers Once Feared
IEEE Spectrum Jackie Snow
NASA is testing generative AI on Mars, the ISS, and orbiting satellites. It could make spacecraft less helpless when Earth is too far away to call the shots.
Based on reporting by IEEE Spectrum, Jackie Snow — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Space agencies used to build machines they could predict down to the last move. Now they’re testing whether AI that can improvise might be the thing that finally makes deep-space missions practical.
NASA’s Jet Propulsion Laboratory used Anthropic’s Claude models last December to help plan two drives for the Perseverance rover on Mars, with humans still checking and changing the route before it went up. In May, NASA and IBM put a compressed AI model on the International Space Station and a satellite to spot floods and clouds from orbit, which the source calls the first model of its kind shown in space. Then in July, astronauts on the ISS tried a large language model for help with maintenance questions.
These are small steps, but they point in the same direction. For decades, spacecraft autonomy mostly meant carefully scripted behavior. Generative AI is different. It can be useful precisely because it does not behave the same way every time, and that makes engineers nervous. It also makes them wonder whether the old model of Earth-bound control is starting to break down as missions grow more complex, more distant, and more numerous.
Robert Ambrose, who led NASA’s Software, Robotics and Simulation Division, has lived through that tension. He worked on autonomy for Orion, on NASA’s deep space and lunar orbiter spacecraft, and on Robonaut 2, the humanoid robot that went to space in 2011. He says the testing burden exploded as autonomy increased, because engineers had to account not just for what a spacecraft might do, but how it got there. Still, he argues the problem can be managed, even with more automation in the testing itself.
Farther from Earth, the case gets stronger. Ambrose points to a possible mission to Europa, where a spacecraft might need to dive into a plume of water with almost no warning. By the time humans on Earth saw it, the moment would be gone. That is the basic argument running through the field now: not that AI is ready to fly everything, but that some missions may be impossible without giving machines more freedom.
Companies are already trying to adapt robotics to microgravity. Icarus Robotics is building what it calls a robotic labor force for space, starting with Joy, a free-flying system that recently finished zero-gravity testing in Canada before a planned trip to the ISS. The first use case is simple: move cargo bags between modules. The company wants to begin with teleoperation, collect data, and only later move toward partial autonomy and then more.
And that data problem matters. A robot trained on Earth learns physics that do not apply in orbit, where objects keep moving instead of falling. Icarus says there is no meaningful off-the-shelf dataset for that environment, so it has to build one itself from microgravity demonstrations, simulations, and Earth tests. The broader shift is clear enough: space is no longer just a place where machines endure. It’s becoming a place where machines may have to decide.
My take — AI-written commentary, not fact-checked reporting
The real story is not that AI is boldly going where no model has gone before; it’s that space engineers are running out of patience with rigid systems that only work when nothing unexpected happens. The cautious route is still the right one, but pretending autonomy is the danger instead of the tool is classic old-guard thinking. Space has always punished hesitation, and now it’s punishing dumb predictability too.
Read more about this at: IEEE Spectrum