Space Exploration Blasts Off with AI

Shooting rockets into space and peering into faraway galaxies has long hinged on the mathematical and engineering prowess of humans. Spaceships, telescopes, robotic devices, and other tools use complex mechanical systems and sophisticated computer programs to do their jobs.

A new era of space exploration is dawning. Artificial intelligence (AI) is radically reshaping a broad array of systems, tools and applications. Digital twins, machine learning (ML), generative AI, and other tools are now helping scientists unravel the mysteries of the universe, design smarter vehicles and robots, and accomplish myriad other tasks that had previously fallen outside the orbit of what was possible.

“AI and ML are becoming powerful contributors in the overall space exploration ecosystem,” said David Salvagnini, Chief Artificial Intelligence Officer for the U.S. National Aeronautics and Space Administration (NASA). Already, the agency has deployed AI for Martian rovers, developed digital twins to analyze flight telemetries, and used machine learning to discover more than 400 exoplanets from terabytes of satellite data.

AI Takes Flight

From the earliest Sputnik and Mercury missions to the International Space Station and private space ventures, technology has rocketed to the vanguard of space exploration. Yet AI is suddenly changing the trajectory of space exploration. “These technologies can perform tasks that may be repetitive, tedious, or low value per labor hour, thus freeing engineers to focus on higher value functions that require human insight, creativity, and advanced analysis capabilities,” Salvagnini said.

AI also helps engineers venture beyond the limitations of human knowledge and expertise. For example, digital twins can simulate launches, missions and scenarios, including what a colony on the Moon or Mars might look like and how it might respond to different conditions. The technology also can model specific machine components and how they would likely respond to specific conditions, such as a solar storm or asteroid impact.

For example, NASA’s Artificial Intelligence Group has launched a diverse array of projects. It is studying the use of AI for cognitive radio systems that can better adapt to conditions and avoid interruptions, particularly during periods of heavy use or when electromagnetic interference occurs. This system would utilize unused or underused segments of the licensed radio spectrum to automatically adapt to conditions. Once conditions return to normal, it would revert to conventional communications.

At the other end of the technology spectrum, NASA has used AI to handle route planning for its Perseverance Mars Rover, which landed on Mars in 2021. The autonomous system collected soil samples without the need for Earth-to-Mars communication. The European Space Agency (ESA) is currently funding 12 AI projects, including one examining how to establish cognitive cloud computing in outer space. A space network could aid exploratory missions to other planets but also help scientists monitor conditions on earth, including the impacts of climate change.

Meanwhile, the Japanese space agency (JAXA) has developed an Epsilon rocket that is the first to incorporate AI. It performs a self-inspection and monitors performance continuously and autonomously, adjusting to conditions as needed. It also includes a mobile launch control feature that connects to a desktop computer. Japan is using the Epsilon vehicle for satellite launches. “We aim to greatly simplify the launch system by using artificial intelligence,” stated Yasuhiro Morita, Project Manager for the Epsilon Launch Vehicle.

AI also is making its presence felt in mapping the universe. At the European Southern Observatory in Munich, Germany, research fellow Miguel Vioque and colleagues use AI to sift through enormous data sets, find subtle patterns, and identify complex objects—from asteroids to stars—that the human eye cannot see. This includes gravitational fields or electromagnetic influences on celestial bodies that can cause imperceptible image distortions. “It’s impossible to go through hundreds of thousands of images manually. Machine learning and AI algorithms completely change things,” Vioque said.

Into the Stars

AI may be the new star in space exploration, but the technology remains parsecs away from its full potential. Engineers face challenges that, in many cases, are not present on Earth, said Zachary Manchester, an assistant professor at the Carnegie Mellon University Robotics Institute. These include adapting AI algorithms to weightlessness, sending essential data back to Earth, and achieving true autonomy for vehicles and rovers.

“Frequently, humans have to take over when rovers or other systems get stuck,” Manchester said. “Engineers wind up recreating the problem in a lab and then uploading the data to the vehicle.” He is studying how to better adapt robotics for outer space, including low gravity environments that make conventional forms of locomotion, such as walking, difficult. “In many cases, robots do better when they hop. This requires a different form and design,” he explained.

The CMU lab is studying ways to engineer robots that incorporate reaction wheels and predictive algorithms so they can achieve stable footing on uneven surfaces. The systems rely on the same types of actuators that satellites use for orientation; in this case, the actuators provide balance, Manchester said. In addition, the lab is developing more robust motion planning systems that can incorporate more complete data about a planet’s atmosphere, winds, and the vehicle’s position and velocity. This could help NASA drop large payloads on the Moon or Mars.

There also is a need to overcome steep communications obstacles. While a cognitive radio system would help, bandwidth limitations and onboard computing power make it difficult for scientists to obtain the data they desire. At present, only about 10% of the data captured in space makes it back to Earth. The rest is lost. Manchester is studying new satellite control systems, including using onboard AI to determine which data to send to Earth, and what resolution to use. “It would be helpful to instruct a satellite to look for certain types of images, such as a forest fire or glacier melt,” he said.

The impact of AI on space exploration will continue to increase, NASA’s Salvagnini said. This includes using generative AI to handle non-sensitive data and certain design tasks. However, he also noted that all the opportunities also come with challenges, including upskilling teams, managing security and ensuring that AI is used in ethical and responsible ways. “AI represents significant potential. We all must learn to use appropriately,” he said.

Samuel Greengard is an author and journalist based in West Linn, OR, USA.