Students take on Earth-focused NASA challenge

NASA’s Space to Soil Challenge wasn’t about asteroids, heat shields or a mission to Mars. This one came from its Earth Science Technology Office, inviting teams to develop or describe innovative technologies for small satellites that provide us with observations about down-to-Earth problems such as regenerative agriculture and sustainable forestry.  

Students in The Grainger College of Engineering, University of Illinois Urbana-Champaign responded to the unique request and were among the 10 finalists selected to present their design solution at NASA’s Washington, D.C. headquarters.

The team developed LOOP: Lightweight Onboard Observation Pipeline for Small Satellites.

“Their design uses on-board processing to determine what targets to capture on a given pass based on cloud and weather forecasts, what images and data to keep based on their quality and if certain criteria/metrics need to be transmitted to the ground or if the target needs to be observed again,” said Matt Hausman, the team’s faculty adviser.

Hausman said he is particularly proud of these students because they weren’t just competing against other student teams. The other finalists included established start-ups.

The team is part of the Laboratory for Advanced Space Systems at Illinois. It was led by AE Ph.D. student, Phoenix Alpine and included aerospace undergrads Ayaan Ahmad, Dibyansh Choudury, and Denver Haslett as well as materials science undergrad Hailey Pohl.

Choudury and Pohl traveled to D.C. to present at NASA’s headquarters.

According to Choudury, the most difficult aspect of the challenge was balancing the satellite's limited onboard resources. The team wanted their system to intelligently analyze Earth observation data, but it had to consider constraints such as power, storage, computing capability, and communication bandwidth.

“Instead of transmitting every image back to Earth, LOOP uses onboard AI models to evaluate the value of the collected data,” Choudury said. “The system determines what should be captured, stored, and downlinked, allowing the satellite to prioritize the most useful information while reducing unnecessary data transmission. This made the mission more efficient and better suited for the constraints of a CubeSat.”

The team used AI to analyze multispectral satellite imagery and identify winter cover crops directly onboard the satellite. Beyond image classification, Choudury said AI also helped prioritize which observations were most valuable, enabling the satellite to make autonomous decisions about data management instead of relying entirely on ground stations.

“One of the biggest challenges with AI was making sure it was accurate and lightweight enough to run on the limited computing hardware available onboard a CubeSat,"  he said. “There is always a trade-off between model complexity, accuracy, power consumption, and processing speed. We also had to consider the risk of incorrect classifications, because mistakes could lead to valuable data being discarded or less important data being prioritized. We had to carefully balance performance with reliability and efficiency.”