EDHPC 2025 will be take place from 13th to 17th October and at Airbus Crisa & Klepsydra we are proud to showcase a demonstrator running AI for Space Situational Awareness.

In space & defense sector, Space Situational Awareness (SSA) is a strategic capability aimed at detecting, tracking, and characterizing objects and activities in space. Its purpose is to protect national assets, ensure operational security, and maintain space superiority. SSA enables effective threat assessment, provides early warnings of potential hostile actions or collisions, and supports informed decision-making in increasingly contested or congested space environments.

The objective of this demonstrator is to identify and gather information on unidentified objects observed near national space assets or infrastructure. For this demo, Klepsydra has used a lightweight, open-source satellite detection AI algorithm based on the YOLOX architecture. The algorithm was trained using a combination of real and synthetic images from the SpeedPlus dataset.

The hardware platform is Airbus Crisa’s Payload Control Module (PCM), which features a space-grade processor. Optimization through Klepsydra AI enhances the model’s inference and computation performance without modifying the model itself—it is used as-is. Further improvements are anticipated when the companion FPGA on the PCM is leveraged to accelerate and parallelize portions of the AI model. Klepsydra AI orchestrates the data flow between the CPU and FPGA, ensuring optimal performance.

This combination of optimized AI inference and FPGA acceleration opens the door to a range of applications, including shape recognition, edge processing for bandwidth optimization, and on-board autonomous FDIR. It enables customers to run their unaltered models at peak performance on reliable hardware and software, bringing AI in space closer to reality.