By Kurtis Archer | 2025-05-30

As its presence continues to grow at impressive rates in many industries, artificial intelligence (AI) is increasingly used in space exploration.

Scientists hope for AI to explore many other worlds someday, with machines knowing exactly what mission information is needed, where the data are located and how to analyze them.

Though human technology is years away from that specific scenario, progress is being made with NASA's Martian rover named Perseverance.

Perseverance makes its own decisions on Mars based on real-time analysis of rocks it finds there. For almost three years, the rover's AI system has been looking for mineral components in the red planet's rocky materials.

Adaptive sampling

One of Perseverance's instruments is the Planetary Instrument for X-ray Lithochemistry (PIXL), which the AI software supports. PIXL is a spectrometer that NASA developed at its Jet Propulsion Laboratory (JPL) in southern California.

PIXL allows scientists to learn whether a rock formed in conditions suitable for microbial life by analyzing the chemical composition of minerals on the surface of the rock.

Without communicating with mission controllers on Earth, the rover uses "adaptive sampling" to find targets. The AI system positions PIXL near a rock and then looks at the scans to see if the minerals present warrant more scrutiny.

Peter Lawson directed the implementation of adaptive sampling while working at JPL and spoke of the AI system.

"The idea behind PIXL's adaptive sampling is to help scientists find the needle within a haystack of data, freeing up time and energy for them to focus on other things," said the now-retired Lawson.

"Ultimately, it helps us gather the best science more quickly," he said.

Abigail Allwood, PIXL principal investigator at JPL, spoke of the autonomous system.

"We use PIXL's AI to home in on key science. Without it, you'd see a hint of something interesting in the data and then need to rescan the rock to study it more," she told The Times of India last July. "This lets PIXL reach a conclusion without humans examining the data."

Microscopic accuracy

Vast temperature swings on Mars cause Perseverance's arm to expand or contract a microscopic amount, which can throw off PIXL's aim, NASA noted in an article published July 16.

The spectrometer sits on six tiny robotic legs called a hexapod at the end of Perseverance's robotic arm. The hexapod automatically adjusts the instrument to bring it exceptionally close without touching the rock.

PIXL then scans an area of the rock about the size of a postage stamp by firing an X-ray beam thousands of times to map a grid of microscopic dots. Each dot reveals information about the chemical composition of that part of the rock.

"We have to make adjustments on the scale of micrometers to get the accuracy we need," Allwood said in the NASA article. "It gets close enough to the rock to raise the hairs on the back of an engineer's neck."

Perseverance is equipped with an autonomous system allowing it to shoot rocks with a laser based on the rock's shape and color, allowing the chemical composition to be revealed by analyzing the gas that burns from the rock after each zap.

The zaps that the instrument makes on rocks can detect certain minerals and then automatically stop to gather additional data -- known as a "long dwell." As machine learning improves the AI system, the list of minerals that encourage long dwells is growing.

This AI system was pioneered by NASA's earlier Curiosity rover, currently thousands of kilometers away from Perseverance on the red planet. Perseverance has a more advanced form of AI that allows it to roam without specific instruction from NASA back on Earth.

AI-driven science

AI allows the Perseverance rover to navigate over rough terrain, observe and avoid obstacles in its path and adjust to environmental conditions without receiving instructions from human controllers.

High-performance processors designed to be able to survive radiation and extreme temperatures, along with resilient, compact and energy-efficient AI hardware for space missions, allow Perseverance to cover more ground more quickly, make decisions instantly and effectively carry out multiple scientific experiments at once.

The increased safety of Mars exploration thanks to Perseverance is undeniable.

Without human drivers, Perseverance uses AI-powered navigation to make real-time decisions, avoiding hazards like sand traps, steep slopes and rock fields. By sending a robot to Mars -- or beyond -- it minimizes risks for astronauts, analysts say.

"Effective, autonomous management of these issues is critical, as missions operate under highly variable Martian conditions with limited real-time communication and constrained resources," Indian Space Research Organization (ISRO) scientist Prajjwal Yash said about Martian landers and their missions in an abstract submission for the 2025 Global Space Exploration Conference.

"This work introduces an AI-driven solution that combines an ensemble of machine learning models to predict dust deposition rates onboard Martian landers."

Testing AI on Mars

Data from PIXL and other instruments allow NASA scientists to decide when a core of rock should be drilled and sealed in a titanium metal container for transport back to Earth for further study under the Mars Sample Return campaign.

Regolith collected will be obtained and returned to Earth with subsequent NASA missions. Minerals in Martian rocks can help scientists learn how water may have formed the rock, or how the rock may have possibly provided nutrients for microbial life.

"PIXL is kind of a Swiss army knife in that it can be configured depending on what the scientists are looking for at a given time," said JPL software developer David Thompson. "Mars is a great place to test out AI since we have regular communications each day, giving us a chance to make tweaks along the way."

Autonomy for missions on other planets is increasingly important because future missions will travel deeper into the solar system and beyond, making communication times much longer than the Earth-Mars time frame.

[James Werner contributed to this report.]