Space has always been considered the final frontier of human exploration. Now it is becoming the final frontier for artificial intelligence as well.
In December, NASA's Jet Propulsion Laboratory used Anthropic's Claude models to plan two driving routes for the Perseverance rover. Human planners reviewed and adjusted the route before uploading it.
In May, NASA and IBM deployed a compressed AI model on the International Space Station and a satellite. It identified floods and clouds from orbit — the first such test in space.
In July, astronauts on the ISS tested a large language model for answering questions about maintenance procedures.
These experiments reflect a shift in space engineering. For decades, engineers on Earth precisely defined what a spacecraft would do in space. Now researchers are testing whether nondeterministic systems like generative AI can give spacecraft the flexibility to interpret their surroundings and plan tasks.
Robert Ambrose, former head of NASA's software, robotics and simulation division, notes that engineers have always avoided nondeterministic outcomes, because a system's behavior depends on the path by which it arrived at a situation.
However, for missions like diving into the water plume on Europa, a moon of Jupiter, a decision must be made instantly — a signal from Earth would arrive too late. "It's completely impossible without true autonomy," Ambrose says.
The company Icarus Robotics is developing a free-flying robotic system called Joy. After zero-gravity tests in Canada, it is planned to be sent to the ISS to move cargo between modules. The company is starting with teleoperation to gather data for training autonomous operation.
The problem lies in the physics of microgravity: models trained on Earth fail immediately in space. There is almost no data from the real environment, and the company is building its own dataset from simulations and tests.
Ufuk Topcu of the University of Texas at Austin emphasizes that the goal is not to guarantee the absence of errors. One should start with limited applications, study how the systems behave and gradually expand their use.
The commercial boom in space is accelerating innovation: missions are being developed and launched faster than ever before.
