February marks the five year anniversary of ENDURANCETime of Mars. But when it comes to testing new things, NASA’s rover is as active as the day it first reached the Martian regolith.

According to NASA JPL, Persistence completed two driving stints using an itinerary created by Claude, a generative AI model developed by California-based firm Anthropic. To create the rover’s travel plan, the AI ​​program refers to the same data that human engineers use to create waypoints—a “breadcrumb trail” for the rover to follow, Anthropic explained in a statement.

These drives occurred on the 1,707th and 1,709th Martian days of the mission, during which Endurance traveled 689 feet (210 meters) and 807 feet (246 meters), respectively.

“This demonstration shows how far our capabilities are advancing and expanding how we explore other worlds,” NASA Administrator Jared Isaacman said in a JPL statement. “This is a strong example of teams applying new technology carefully and responsibly in real operations.”

Drive a rover from Earth

Suffice it to say, driving a rover on Mars is very different from driving a car on Earth. On average, Mars is located about 140 million miles (225 million kilometers) from Earth. The distance between the Earthbound drivers and the actual rovers created a huge communication lag, leaving the rover unprepared for unexpected environmental obstacles on the Martian surface.

“This is a high-stakes job,” Anthropic said in its statement. For example, Endurance’s predecessor, Spirit, fell into a sand trap and “didn’t move again,” it added.

To prevent such travesties, researchers carefully study the images of the Martian terrain taken from space and the rover. Based on the images, the engineers developed a travel plan that would hopefully avoid natural obstacles such as bedrock, outcrops, bumpy boulder fields, and sand ripples.

The AI ​​took the wheel, sort of

This is a process that requires a lot of time and effort. And so, NASA and collaborators are wondering if AI can help ease the intensity of manually planning rover routes.

Before the demonstration, JPL engineers fed Claude data on how the rover was operating, which the AI ​​used to make its commands. The commands are then checked for accuracy through simulations that verify more than 500,000 variables, NASA said.

Engineers found Claude’s commands to be surprisingly trouble-free, requiring only minor fixes to address some images that Claude didn’t see, Anthropic explained. Once the team was satisfied with the results, the plan written by AI was sent to Endurance and implemented in December.

“The fundamental elements of generative AI show great promise in streamlining the pillars of autonomous navigation for off-planet driving,” said Vandi Verma, an engineer on JPL’s Perseverance team, in a NASA release.

“This means, in short, that we can learn a lot more about Mars,” Anthropic added, while engineers expect the program to help cut route planning time in half.

Although Endurance has only driven a few hundred feet so far, the AI ​​could start planning miles of routes across the Martian terrain, Verma said.