While space has been hailed as “the final frontier,” it holds no fear for Anthropic’s Claude, which has broken new ground by plotting a route for a NASA vehicle on Mars.
The 400-meter excursion across rugged Martian terrain, which took place in December, constituted the first time NASA has used an AI model to determine a path for its Perseverance Rover on the Red Planet.
The exercise was led by the agency’s Jet Propulsion Laboratory (JPL) in Southern California, with Claude creating waypoints for Perseverance — a decision-making task described by NASA as “complex” and one that is generally performed by human planners.
The Perseverance Rover is a car-sized robot bedecked with cameras and scientific equipment that has been used for research on Mars since 2021.
But as the planet is some 225 million kilometers from Earth, and its surface rocky and tricky to negotiate, maneuvering the vehicle is a challenge, with the prospect of it tipping over or getting stuck an ever-present risk.
As such, NASA typically deploys human operators to laboriously lay out waypoints — known as a “breadcrumb trail” — for the rover to follow, using a combination of images from space and onboard cams.
On this occasion, the waypoints were provided by Claude after JPL engineers decided to see if AI could plan a route as effectively as a human. Doing so involved providing Claude Code, Anthropic’s programming agent, with masses of data collected over 28 years of missions to Mars, to allow it to write commands for Perseverance.
“[By] using its vision capabilities to analyze the overhead images, Claude planned Perseverance’s breadcrumb trail point by point,” according to Anthropic’s account. In doing so: “It strung together ten-meter segments into a path, then iterated to refine the waypoints, critiquing its own work and suggesting revisions.”
Given the importance of the task, JPL engineers reviewed Claude’s work before approving it, running it through a digital twin simulation, and verifying more than 500,000 telemetry variables. They found only minor changes were required.
In early December, the commands were sent to Perseverance via NASA’s Deep Space Network, with each taking around 20 minutes to be transmitted due to the distance between Earth and Mars. The result was encouraging: successful drives of 210 meters and 246 meters respectively.
Now the hope is that Claude’s contribution to the successful demo will pave the way for even more input from AI.
“The fundamental elements of generative AI are showing a lot of promise in streamlining the pillars of autonomous navigation for off-planet driving: perception (seeing the rocks and ripples), localization (knowing where we are), and planning and control (deciding and executing the safest path),” Vandi Verma, a JPL space roboticist and a member of the Perseverance engineering team, explained. “We are moving towards a day where generative AI and other smart tools will help our surface rovers handle kilometer-scale drives while minimizing operator workload.”
This kind of AI-driven functionality could be crucial on future missions, including NASA’s Artemis campaign to take humans back to the moon.
