6U+ CubeSat SONATE-2: Operation of an Optical AI Payload in Low Earth Orbit
- Paper ID
87877
- DOI
- author
- company
Julius Maximilians Universität Würzburg; University Wuerzburg
- country
Germany
- year
2024
- abstract
Optical payloads and other instruments can generate large amounts of data. Data acquisition is often limited by downlink capabilities, especially in the context of space exploration, as the NASA deep space network is said to be at capacity, for example. Intelligent onboard data processing can decrease the amount of data that has to be transferred to Earth. The Nanosatellite SONATE-2, developed and operated at the University of Würzburg, tackles this problem by demonstrating artificial intelligence technologies in a low Earth orbit. SONATE-2's AI payload has specialized hardware, featuring an NVIDIA Jetson NX System on Module (SoM) with a 6-Core ARM CPU and an NVIDIA Volta architecture GPU with Tensor cores that can infer and train neural networks. This SoM enables data processing using pre-trained models on the ground and training neural networks onboard. The Linux operating system of the AI Payload allows for the use of standard libraries, like CUDA, CudNN, or TensorFlow. The 6U+ CubeSat SONATE-2 was launched with the Transporter 10 Mission by Space X into a sun-synchronous orbit in March 2024. This work presents the results of the first months of operation of the SONATE-2 Nanosatellite. The primary mission objective is to demonstrate AI technologies in space, including anomaly detection, image segmentation, and object detection using four cameras, two in the visible spectrum and two in the near-infrared. This work focuses on the technical challenges and presents the methods used to operate the AI Payload in space. In addition, we give an overview of our experiences with managing the payload, including the thermal and power considerations. A brief outlook on the planned experiments within the mission duration of one year is presented.