For more than two decades, driving on Mars has been a meticulous, high-stakes game of follow the leader played across millions of miles of void.
Human operators at NASA’s Jet Propulsion Laboratory (JPL) would spend hours squinting at satellite imagery, plotting safe waypoints to ensure billion-dollar rovers didn’t end up beached in a sand trap or tumbled down a crater wall.
But in December 2025, the leader changed. For the first time, NASA handed the map to artificial intelligence (AI).
Using Anthropic’s Claude AI, mission controllers successfully plotted a course for the Perseverance rover that spanned nearly 1,500 feet of treacherous Martian terrain. The demonstration, which took place over two Martian days (sols), represents a profound shift in how humanity explores the cosmos: turning AI from a passive data analyst into a strategic navigator.
Claude is better known on Earth for drafting emails or debugging software, but JPL engineers saw potential in its ability to process complex constraints. To prepare for the AI chatbot, the team fed it years of rover telemetry, safety protocols, and high-resolution digital elevation models provided by the Mars Reconnaissance Orbiter.
The AI didn’t just suggest a direction. It spoke the language of the stars. Claude wrote its navigation commands directly in Rover Markup Language, an XML-based code used specifically for Mars missions. It analyzed slopes, ridges, and boulder fields to string together 10-meter segments into a coherent “breadcrumb trail.”
“This demonstration shows how far our capabilities have advanced,” said NASA Administrator Jared Isaacman. “Autonomous technologies like this can help missions operate more efficiently and increase science return as distance from Earth grows.”
The need for autonomy is born of physics. Because signals take roughly 20 minutes to travel between Earth and Mars, real-time joystick driving is impossible. While Perseverance has an onboard AutoNav system to dodge immediate rocks, the strategic long-game of route planning has always remained a human burden.
During the December test, Claude’s proposed route was subjected to a grueling simulation, modeling over 500,000 variables to predict any potential hazards. When the JPL team reviewed the AI’s work, they found it remarkably accurate. Humans made only minor tweaks, such as adjusting for small sand ripples visible only in ground-level photos, before shipping the code to Mars.
The timing of the test is symbolic. Perseverance recently surpassed its so-called marathon milestone, having driven over 25 miles since its 2021 landing. As the rover begins its ambitious climb out of the Jezero Crater — a journey that already saw it scale 1,640 vertical feet — the workload for human planners is peaking.
By delegating the laborious work of waypoint-setting to AI, NASA hopes to free up human scientists to focus on the big questions: The search for ancient microbial life and the characterization of Martian geology. Space roboticist Vandi Verma noted that AI could eventually flag interesting rocks for study, eliminating the need for humans to manually sift through “huge volumes of rover images.”
As humanity looks toward missions further into the solar system, where communication delays stretch into hours rather than minutes, the success of Perseverance’s AI-driven stroll highlights that the explorers of the future won’t just be robots, they’ll be robots that can think for themselves.
