Paper: NASA SpaceCube Next-Generation Artificial-Intelligence Computing (NASA 2023)


The paper describes the Space Test Program – Houston 9 – SpaceCube Edge-Node Intelligent Collaboration (STP-H9-SCENIC) experiment that was recently deployed on the International Space Station (ISS) to demonstrate artificial intelligence and machine learning technologies on space-based platforms.

The motivation is that AI/ML concepts can enable autonomous operation, rapid onboard data analysis, and constellation management for future missions but current rad-hard processors lack the computational power. Novel AI accelerators for mobile devices could be leveraged but need space qualification.

The objectives are to evaluate commercial AI chips like Intel Movidius Myriad X and Google Coral TPU in space radiation; collect hyperspectral Earth images to provide public datasets for ML model training; and demonstrate the SpaceCube v3.0 Mini rad-tolerant high-performance space computer.

The hardware architecture consists of the SpaceCube card plus 7 other cards in a 4U form factor payload. These include power, AI accelerator, software-defined radio, GPS receiver, and hyperspectral imager cards. The software architecture uses NASA’s Core Flight System with custom applications.

Vibration, thermal vacuum, and radiation testing was conducted to raise technology readiness level. Launch was in March 2023 on a SpaceX Dragon capsule. The payload is installed on the Japanese Experiment Module Exposed Facility of the ISS.

Early operations have verified nominal functionality. Commissioning is slower than expected but will advance to new ML application deployment soon. Uploaded models will run on the AI accelerator cards using the hyperspectral Earth images collected.

In summary, the experiment demonstrates cutting-edge AI computing for space with both rad-tolerant and commercial technology. The public dataset enables ML research. Successful operations in the ISS radiation environment raise readiness for future science and defense missions needing onboard autonomy and intelligence.

Print Friendly, PDF & Email