Autonomous Object Search in the Kibo Module of the International Space Station

Free-flying robots such as SPHERES, Astrobee, and Int-Ball are essential for assisting astronauts and increasing mission efficiency aboard the International Space Station (ISS). As these systems have low processing power, one key challenge for these systems is the ability to locate specific objects in minimal time. This paper represents a complete perception-toaction pipeline designed to solve this problem using Astrobee. The developed approach integrates a robust object detection model, trained on a custom-built dataset, with a time-optimized path avoiding some restricted zones. The system was developed and tested in the NASA Astrobee simulator and validated on the physical Astrobee in the ISS during the 5th KIBO Robot Programming Challenge organized by the Japan Aerospace Exploration Agency (JAXA). In the final on-orbit competition, the developed system demonstrated high efficiency in microgravity conditions, achieving the fastest mission time(191.8s), while the object detection accuracy reached mAP@0.5:0.95 of 0.963.