In the high-stakes world of NASA’s space missions, there is zero margin for error. Traditionally, repairing a damaged satellite or landing a craft on the moon has relied on expensive robotics, risky human spacewalks, and on grueling physics-based calculations that can take years to develop and validate. But that is changing.
Building on its deep history of partnership with NASA, student teams from Rice University’s Data to Knowledge (D2K) Lab have recently leveraged artificial intelligence (AI) to conduct a series of data science projects to solve some of the most complex challenges in aerospace engineering—from building autonomously navigating spacecrafts to ensuring safe hypersonic re-entry.
Described below are the four projects the D2K teams have completed in collaboration with NASA in the past two years:
Powering autonomous inspection of spacecrafts
Spacecrafts are often damaged by hazardous conditions and space debris. When they break, they frequently lose communication, becoming "non-cooperative" targets that are difficult to locate and fix. Currently, locating, maintaining, and repairing such an orbiting spacecraft requires risky "spacewalks" by astronauts or use of expensive, complex robotic arms. To save money and keep people safe, NASA has recently developed small inspector spacecrafts. Due to space constraints, these have low-resource onboard computers with limited computational processing power or storage capacity, making AI-driven models an attractive alternative for autonomous and efficient image processing. To develop these models, D2K students, guided by Rice faculty Arko Barman and NASA sponsor James Berck, have created and benchmarked the world’s largest labeled image segmentation library of spacecrafts. This "digital textbook" contains 64,000 high-quality photos of spacecrafts in various shapes and sizes captured under realistic and tricky lighting conditions of space. These were used to train tiny inspector satellites to process images using the algorithms developed by the D2K team at blistering speeds—less than 0.5 seconds per captured image. A paper about this study was recently presented at the 2026 Institute of Electrical and Electronic Engineers (IEEE) Aerospace Conference. Click here to read the paper and access the dataset.
Precision Lunar Landings via Crater Detection
With NASA’s Artemis II mission, human travel to the moon and deep space is making a comeback after a 50-year hiatus. But landing a spacecraft on the lunar surface safely requires more than just luck. Under the guidance of Barman and NASA sponsor Kyle W. Smith, D2K teams are developing AI-based navigation systems that allow spacecraft to land autonomously. By mastering the precise and rapid detection of lunar crater boundaries, these systems allow a spacecraft to determine its precise location by matching the detected craters with a lunar atlas, navigate in an autonomous fashion, and "see" its way down to the surface, ensuring a soft touchdown on the rugged lunar terrain.
Optimizing and Accelerating Mission Planning
Moving a satellite from one orbit to another—a "transfer maneuver"—is a delicate dance of physics where every drop of fuel counts and needs to be performed with utmost precision. Traditionally, these moves are calculated using first principles of physics and require months of meticulous planning. Working with Rice faculty and D2K director Chad Shaw and NASA sponsor Max Widner (with Jacobs Engineering), D2K students are replacing these conventional equations with machine learning methods to speed up mission planning. This shift doesn’t just save fuel; it dramatically increases payload capacity, reduces computational time and improves mission costs and feasibility for missions — from satellite deployment to interplanetary exploration.
Surviving the Heat During Re-entry
The most dangerous part of any space mission is coming home. As the spacecraft hits Earth’s atmosphere at hypersonic speeds, it faces intense heat fluxes that could melt the hull if the Thermal Protection System (TPS) fails during re-entry. Usually, the extreme conditions a spacecraft will face are calculated using traditional physics-based computational fluid dynamics (CFD) models—a process that involves solving massive partial differential equations — which are notoriously time-consuming and computationally expensive. Under the guidance of Shaw and NASA’s Andrew Jay Hyatt, D2K students have developed a hybrid workflow that incorporates machine learning/artificial intelligence models to provide fast, near-real-time predictions. This will allow NASA to screen different flight trajectories and vehicle designs in seconds rather than days and significantly increase the safety of returning astronauts and cargo.
The Future of Space is Data-Driven
“These partnerships demonstrate that the next frontier of space exploration isn’t just about bigger or more expensive rockets—it’s about making space travel smarter and more efficient with improved software and data analysis,” said Shaw. “By integrating machine learning methods with complex physics calculations, NASA is paving the way for long-term and sustained human presence in space with safer, cheaper, and more autonomously controlled missions.”