Copyright Interesting Engineering
A new age of space autonomy just took its first real breath in orbit. A research team from Julius Maximilians Universität Würzburg has completed a world first by running an AI-based attitude controller directly in space. The breakthrough took place aboard the InnoCube nanosatellite during a short morning pass as the AI executed a complete attitude maneuver using only its learned decision-making. The controller repeatedly hit its target orientation during follow-up tests, proving that the system could handle real conditions rather than simulated ones. New intelligence emerging The team behind the LeLaR project says this achievement places autonomous space systems within reach. The project aims to build the next generation of adaptive controllers that stabilize spacecraft and point their instruments without human tuning. What makes the Wurzburg controller different is its use of Deep Reinforcement Learning. Instead of relying on fixed rules, the neural network learned its behavior through long trial-based training sessions in a simulated environment. The approach promises faster development and adaptive performance. Traditional controllers often take months of expert tuning. The DRL system can learn in advance and then adjust itself when faced with unexpected conditions. A frontier in motion Before the flight test, the controller was trained in a high-fidelity simulator and uploaded to InnoCube. One of the largest hurdles was ensuring that behavior learned on Earth would translate to space, an issue often called the Sim2Real gap. “A truly decisive success,” says Kirill Djebko. “We have achieved the world’s first practical proof that a satellite attitude controller trained using Deep Reinforcement Learning can operate successfully in orbit,” he adds. Tom Baumann notes the wider implications. “This successful test marks a major step forward in the development of future satellite control systems. It shows that AI can not only perform in simulation but also execute precise, autonomous maneuvers under real conditions.” The team says the result establishes new trust in artificial intelligence for safety-critical missions. Future spacecraft will need to make autonomous choices during long interplanetary journeys where communication is slow or impossible. Frank Puppe says, “This will significantly increase the acceptance of AI methods in aeronautics and space research,” pointing to the strength of the simulation model. With the success of LeLaR, the group has achieved a central project milestone supported since 2024 by the Federal Ministry for Economic Affairs and Energy and managed by the German Space Agency at DLR. Erik Dilger says, “This success motivates us to extend the technology to new scenarios.” InnoCube itself is a cooperative satellite with TU Berlin that provides an orbital testbed for emerging concepts. One of the experimental platforms on board is the wireless system called SKITH, which replaces traditional cabling with internal wireless communication to reduce mass and failure risk. The team believes the demonstration moves Wurzburg into a leading position in AI-driven space control and lays the foundation for deep space missions that rely on intelligent and self learning systems. “The next goal is to build on this head start,” says Kirill Djebko. “It is a major step towards full autonomy in space,” adds Sergio Montenegro. The significance of the work is underscored by its publication in Acta Astronautica.
