Stanford University researchers recently demonstrated that a robot aboard the International Space Station (ISS) can autonomously navigate its complex interior using machine-learning control for the first time, tells Stanford Report.
The robot in question is Astrobee, a compact, cube-shaped, fan-powered device designed to float freely through the ISS. The challenge is real: the station’s corridors are narrow, cluttered with equipment, wiring, and storage modules. Traditional path-planning algorithms commonly used on Earth are too computationally demanding for space-qualified hardware, and executing them safely within the ISS environment is especially challenging due to strict safety constraints and unpredictable disturbances.
To overcome these limitations, the research team led by Somrita Banerjee trained a machine-learning model on thousands of previously solved navigation paths. When given a new navigation task, the model generates an initial “warm start,” a trajectory estimate based on learned patterns, such as the locations of corridors and the areas where obstacles tend to be. That warm start then feeds into a conventional optimization routine (sequential convex programming), which refines the path while strictly respecting safety constraints.
When tested aboard the ISS, Astrobee followed this AI-guided process and completed its motion planning 50–60% faster than with the conventional method, especially in tight, complicated spaces. The experiments covered 18 distinct trajectories, each executed with and without the warm-start approach to compare performance.
This milestone brings us closer to an era where robots aboard space stations, and on future lunar or Martian bases, perform routine tasks independently. With proven reliability, autonomy like this could free astronauts for more critical work, reduce crew workload, and pave the way for more frequent, efficient missions beyond Earth orbit.