AI Successfully Controls Satellite Attitude in Orbit for the First Time (scienceclock.com)

On 30 October 2025 researchers at Julius‑Maximilians‑Universität Würzburg demonstrated the world’s first in‑orbit, fully autonomous satellite attitude controller: a Deep Reinforcement Learning (DRL) policy running onboard the 3U nanosatellite InnoCube that, during a nine‑minute pass (11:40–11:49 CET), steered the craft from its initial orientation to a precise target attitude using reaction wheels. The LeLaR in‑orbit demonstrator trained the controller entirely in simulation and then deployed it to space; subsequent trials confirmed the DRL agent could reliably command the satellite without any human intervention. The result is significant because it closes a critical Sim2Real gap—showing a controller learned in simulated physics can operate robustly in the unpredictable conditions of orbit—and promises to slash the months of manual tuning required by traditional attitude control systems. Technical implications include faster mission development, onboard adaptability to changing dynamics or faults, and practical autonomy for missions with long communication delays (e.g., deep‑space or interplanetary probes). Funded by the German Federal Ministry for Economic Affairs and Energy and managed by DLR, LeLaR marks a milestone for AI‑driven space systems and paves the way for smarter, more cost‑effective satellite operations.