Session
Technical Poster Session 2
Location
Utah State University, Logan, UT
Abstract
With recent exponential advances in AI—particularly with respect to the tremendous power and efficiency accessible for data processing—there is now a countless array of applications for aerospace missions and space exploration, even in experimental CubeSats. However, with the large volume of data acquisition required for satellite missions, downlinking presents an increasingly expensive bottleneck that drastically reduces mission efficiency. At the University of Georgia Small Satellite Research Laboratory, the Multi-view Onboard Computational Imager (MOCI) employs an NVIDIA Jetson TX2i GPU module to process data in situ and downlink only final products. As is the case for many commercial off-the-shelf (COTS) devices carried in CubeSats, the TX2i does not come radiation-hardened, as it houses a vulnerable eMMC disk. MOCI will employ hardware shielding to help counteract this issue, but there is nevertheless a possibility of single event effects (SEEs) reaching the TX2i, calling for software-level mitigation as a final line of defense. Using Yocto, we create a minimized operating system with built-in redundancy to reduce reliance on flash memory, particularly with a custom bootloader utilizing a triple modular redundancy (TMR) partition scheme and a RAM-based file system available upon boot.
Space-operating Linux: An Operating System for High Performance AI Computation on Commercial-Grade Equipment in Low Earth Orbit
Utah State University, Logan, UT
With recent exponential advances in AI—particularly with respect to the tremendous power and efficiency accessible for data processing—there is now a countless array of applications for aerospace missions and space exploration, even in experimental CubeSats. However, with the large volume of data acquisition required for satellite missions, downlinking presents an increasingly expensive bottleneck that drastically reduces mission efficiency. At the University of Georgia Small Satellite Research Laboratory, the Multi-view Onboard Computational Imager (MOCI) employs an NVIDIA Jetson TX2i GPU module to process data in situ and downlink only final products. As is the case for many commercial off-the-shelf (COTS) devices carried in CubeSats, the TX2i does not come radiation-hardened, as it houses a vulnerable eMMC disk. MOCI will employ hardware shielding to help counteract this issue, but there is nevertheless a possibility of single event effects (SEEs) reaching the TX2i, calling for software-level mitigation as a final line of defense. Using Yocto, we create a minimized operating system with built-in redundancy to reduce reliance on flash memory, particularly with a custom bootloader utilizing a triple modular redundancy (TMR) partition scheme and a RAM-based file system available upon boot.
