AI helps find new clues in the 60-year search for Luna 9, the 1st successful moon lander

When you buy through links on our articles, Future and its syndication partners may earn a commission.

Scientists may be closing in on the long-missing landing site of the Soviet Union's Luna 9, the first human-made object to safely touch down on the moon.

On Feb. 3, 1966, Luna 9 descended into Oceanus Procellarum on the western edge of the moon's Earth-facing side and transmitted the first images ever captured from the surface of another celestial body. At the time, scientists were still debating whether the surface of the moon might be too soft to support a lander or future astronauts. The images Luna 9 sent back, revealing solid ground, helped settle that question and paved the way for future missions.

Six decades later, however, the probe's exact resting place remains uncertain. NASA's Lunar Reconnaissance Orbiter (LRO) has mapped nearly the entire moon in remarkable detail since 2009. But Luna 9 is tiny — about the size of a beach ball — making it difficult to distinguish from the rocks and shadows scattered across the lunar surface.

Now, two independent research teams say they may have narrowed down the spacecraft's location.

One team's candidate site lies within about 3 miles (5 kilometers) of the Soviet Union's official landing coordinates for the mission. The other is roughly 15 miles (25 km) away. But those official coordinates were themselves imprecise, based on a method that estimated the spacecraft's position by measuring the direction of its radio signal from antennas on Earth. That technique placed Luna 9 somewhere within a search area roughly 60 miles (100 km) across.

Researchers say that forthcoming images from India's Chandrayaan-2 orbiter — scheduled to pass over the region in March and equipped with a camera capable of capturing sharper views of the lunar surface than LRO — could help determine which, if either, location is correct.

"On their overpasses, I believe the camera will be slightly better and pick up more at 2x to 4x the resolution and actually get a definitive view of the site," Lewis Pinault, a researcher at University College London/Birkbeck's Centre for Planetary Sciences, who led one of the recent efforts, told Space.com.

Revisiting a 60-year-old mystery

Luna 9's arrival on the moon was very different from modern lunar landings. As it approached the surface, a spherical capsule wrapped in inflatable shock absorbers was ejected from the spacecraft and bounced several times before coming to rest. It unfolded four petal-like panels for stability, like a metallic flower, while the heavier descent stage crashed nearby.

That unusual landing sequence may have left multiple pieces of hardware scattered across a small area, which researchers are now attempting to identify from lunar orbit.

One proposed landing site was identified by Vitaly Egorov, a science communicator who runs the blog Zelenyikot and has for years tried to track down Luna 9. He recently relaunched the search as a public, crowdsourced effort, raising funds and livestreaming the process so volunteers could help comb through orbital imagery for small, anomalous pixels that might mark the lander.

Using LROC QuickMap, a Google Street View-like interface for LRO imagery, he compared Luna 9's original 1966 surface panoramas to orbital data. By aligning distant hills, boulders and distinctive ejecta patterns visible in both datasets, he narrowed in on a candidate site about 15 miles (25 km) from the coordinates reported in the Soviet era, according to a Zelenyikot blog post published on Feb. 7.

Around the same time, Pinault's team approached the mystery using artificial intelligence.

In a paper published Jan. 21 in the journal npj Space Exploration, the team reported a cluster of features near the historically reported landing region that may correspond to Luna 9 and hardware jettisoned during its descent.

Rather than manually scanning millions of images, the researchers trained a machine-learning model using LRO images of Apollo landing sites, where spacecraft hardware locations are well documented. To test the system, they provided it with previously unseen images of the Apollo 17 site taken with different lighting and viewing angles. The model successfully detected and localized the Apollo 17 descent stage and also independently identified the Luna 16 landing site in LRO images it had not previously analyzed, the study reported.

The researchers then tasked the system with analyzing imagery near Luna 9's estimated landing region. The algorithm flagged a principal object about 3 miles (5 km) from the reported coordinates, along with several smaller nearby features separated by less than 650 feet (200 meters).

The same features were detected under different lighting conditions — a strong indication that the system was isolating real structures rather than shadows or illumination artifacts, the study noted.

"The machine is tireless, right?" Pinault told Space.com in a recent interview. "It can look at a lot of images and just pause and say, 'This is different.'"

"In the case of a human's attentive focus," he added, "I would say the smaller artifacts would likely have been completely missed."

The model may be detecting subtle combinations of regolith disturbance, deployment geometry and shadow signatures that human observers might not consciously recognize, Pinault said. "The machine is making correlations that, at least currently, we don't fully understand."

An increasingly crowded moon

The team acknowledges that the model — originally designed to identify micrometeoroids in photographs — did produce false positives, including at least one instance in which suspected "hardware" turned out to be a rock. But the researchers deliberately tuned the system to flag promising candidates for human review rather than risk missing subtle evidence, the study noted.

Both teams now hope Chandrayaan-2's camera will be sharp enough to resolve Luna 9's capsule and also distinguish its petal-like panels.

If new images clearly reveal that shape, Pinault said, "I would raise my hand and say, 'Wow, either somebody dropped a cardboard cutout to fool us, or that is this great artifact of humanity's space exploration history.'"

NASA's upcoming Artemis 2 mission, however, is unlikely to help.

Artemis 2 will perform a high-speed slingshot trajectory around the moon before returning to Earth. At closest approach, the four astronauts aboard the mission's Orion spacecraft will pass about 3,700 to 4,300 miles (6,000 to 7,000 km) above the lunar surface, which is far higher than the few hundred kilometers where LRO or Chandrayaan-2 orbit.

Even if Oceanus Procellarum rotates into view — which Pinault called "a stroke of luck" — Artemis 2 will be far too distant to resolve an object as small as Luna 9.

In the long run, the AI-assisted search for Luna 9 offers a glimpse of how artificial intelligence could help monitor an increasingly artifact-filled moon in the decades ahead, Pinault said.

With NASA's Artemis program and China's plans for a long-term lunar presence, and a growing number of commercial missions worldwide heading to the moon, the lunar surface is accumulating more hardware than ever before. AI systems like the one used by Pinault's team could eventually operate directly onboard spacecraft, identifying lunar assets in near real time, helping catalog and track surface equipment, and monitoring how rocket exhaust and impacts disturb the moon’s dust-sized regolith particles.

"There's good science here," said Pinault.