A significant breakthrough has been achieved by the research team at Julius-Maximilinas-Universitat Wurzburg (JMU) in the use of Artificial Intelligence (AI) to control the attitude controller for satellites directly in orbit for the first time.
This marks the first time that the test was carried out aboard the 3U nanosatellite InnoCube.
The satellite passes between 11:40 and 11:49 a.m. CET on 30 October 2025, when the AI agent developed at JMU performed a complete maneuver in orbit.
The maneuver in orbit was entirely controlled by artificial intelligence. It has been observed that the AI brought the satellite from its original attitude to a specified target attitude.
This project was special because the Wurzburg controller was not built using traditional and fixed algorithms; rather, researchers applied a deep reinforcement learning (DRL) approach which involves a process of machine learning in which an artificial neural network independently learns the effective control strategy in a stimulated environment.
The prime advantages of the DRL approach incorporate its speed and flexibility as compared to the transfer function approach.
The AI controller before deployment was trained on earth in a highly realistic simulation and then uploaded to the satellite’s flight model in orbit.
One of the biggest challenges was crossing over the Sim2Real gap and ensuring that a controller trained in simulation is also functioning on the real satellite in space.
This successful step marks a breakthrough for future development and demonstrates that AI retains the ability to outperform in simulation and ensures execution in reality.
It is one of the pivotal steps forward towards future autonomous missions and it presents a central building block for future deep-space exploration projects.
Nonetheless, it’s a significant step forward towards faster and more cost-effective development of new, complex AI-based controllers and the achievement of more efficient and resilient missions.
