Researchers used AI to control a satellite in space for the first time. The agent, developed by Germany-based space scientists, repeatedly pivoted the spacecraft in orbit to a desired position. Satellites’ “attitude” requires careful adjustments, to point their cameras or sensors at specific targets. That usually involves either direct human control or traditional programmed algorithms that can take months to design.
The new AI model, instead, was trained in simulation via reinforcement learning, making it both faster to build and more flexible. The result shows that AI can be safely used in space, the researchers said, which will be critical in deep-space or interplanetary missions where direct control is impossible because signals take hours or days to reach the probe.