Session
Session VIII: Next on the Pad 2
Location
Utah State University, Logan, UT
Abstract
The mission goal of BeaverCube II is to demonstrate autonomous on-orbit image processing and classification using the Xilinx Versal System-On-Chip (SOC). The Versal has 400 dedicated AI engines on an FPGA-based platform, enabling compute-intensive machine learning capabilities. While the Versal can consume over 100 W of power, BeaverCube II will demonstrate its on-orbit function in a low-power, size-constrained system. The system will use machine learning algorithms such as K-means clustering to autonomously assess whether features or changes of interest are present across keyframes.
BeaverCube II has a 1U imaging payload consisting of three Commercial off-the-shelf (COTS) cameras: two identical visible wavelength cameras (112.5m GSD) and one LWIR camera (197.8m GSD). Using this imaging payload, BS II aims to demonstrate autonomous capabilities such as cloud identification, shoreline feature recognition, and change detection. Running a high-power chip such as the Versal has demanding thermal, structural, and power requirements; the solutions to these challenges will be described in this work. BeaverCube II is a 3U CubeSat designed and built by the STAR (Space, Telecommunication, Astronomy and Radiation) Lab at the Massachusetts Institute of Technology and sponsored by the Northrop Grumman Corporation. We are targeting a Q1 2025 launch with deployment from the ISS in Q2 2025.
BeaverCube II: Using AI-Optimized Processors on Earth-Observing CubeSats for Autonomous Image Analysis and Intelligent Data Handling
Utah State University, Logan, UT
The mission goal of BeaverCube II is to demonstrate autonomous on-orbit image processing and classification using the Xilinx Versal System-On-Chip (SOC). The Versal has 400 dedicated AI engines on an FPGA-based platform, enabling compute-intensive machine learning capabilities. While the Versal can consume over 100 W of power, BeaverCube II will demonstrate its on-orbit function in a low-power, size-constrained system. The system will use machine learning algorithms such as K-means clustering to autonomously assess whether features or changes of interest are present across keyframes.
BeaverCube II has a 1U imaging payload consisting of three Commercial off-the-shelf (COTS) cameras: two identical visible wavelength cameras (112.5m GSD) and one LWIR camera (197.8m GSD). Using this imaging payload, BS II aims to demonstrate autonomous capabilities such as cloud identification, shoreline feature recognition, and change detection. Running a high-power chip such as the Versal has demanding thermal, structural, and power requirements; the solutions to these challenges will be described in this work. BeaverCube II is a 3U CubeSat designed and built by the STAR (Space, Telecommunication, Astronomy and Radiation) Lab at the Massachusetts Institute of Technology and sponsored by the Northrop Grumman Corporation. We are targeting a Q1 2025 launch with deployment from the ISS in Q2 2025.
