As of 2021, NASA’s Perseverance rover has achieved several milestones, including returning the first audio recordings from Mars. Now, almost five years after arriving on the Red Planet, it has just achieved something else. This past December, Perseverance successfully completed a route across part of the Jezero crater programmed by Anthropic’s Claude chatbot, marking the first time NASA has used a large-scale language model to drive a car-sized robot.
Between December 8 and 10, Perseverance drove about 400 meters (about 437 yards) into the rocky surface of the Martian surface mapped by Claude. As you can imagine, using an AI model to plan an Endurance course wasn’t as simple as entering a single piece of information.
As NASA explains, piloting Perseverance is no easy task, even for a human. “Every rover drive needs to be carefully planned, lest the rover slip, tip, spin its wheels, or coast,” NASA said. “So since the rover landed, people working on it have been carefully laying out roads – they call it ‘breadcrumb trails’ – for it to follow, using a combination of images taken from space and cameras inside the rover.”
In order for Claude to complete the task, NASA had to give Claude Code, an agent of the Anthropic program, “years” of total data from the rover before the model began to chart Perseverance’s trajectory. Claude then continued the process of mapping the route, including waypoints from the ten meter segments that he would later critique and repeat.
This is NASA we’re talking about, and engineers at the agency’s Jet Propulsion Laboratory (JPL) made sure they double-checked the model’s performance before sending it to Perseverance. The JPL team used Claude’s scores in simulations they use every day to ensure the accuracy of commands sent to the rover. In the end, NASA says it only had to make “minor changes” to Claude’s trajectory, with one tweak coming because the team had access to low-resolution images that Claude didn’t see in its planning process.
“Engineers estimate that using Claude in this way will cut the route planning time in half, and make the trip more consistent,” NASA said. “Less time spent doing tedious manual programming — and less time spent training — allows rover operators to go on more drives, collect more science data, and do more analysis. It means, in short, we’ll learn more about Mars.”
While the productivity benefits provided by AI are often overstated, in the case of NASA, any tool that can allow its scientists to work efficiently will certainly be welcomed. Over the summer, the agency lost about 4,000 workers — about 20 percent of its workforce — as a result of the Trump administration’s cutbacks. Starting in 2026, the president had proposed cutting the agency’s science budget by nearly half before Congress finally rejected that plan in early January. Still, even if its funding is kept below 2025 levels, the agency has a tough road ahead. It is being asked to return to the Moon with less than half the crew it had at the height of the Apollo program.
For Anthropic, on the other hand, this is a big factor. You may remember last spring Claude couldn’t even hit Pokémon Red. In less than a year, the company’s models have gone from struggling to navigate a simple 8-bit Game Boy game to successfully planning a rover’s course on a distant planet. NASA is excited about the possibility of future collaboration, saying that “autonomous AI systems can help explore the most distant parts of the solar system.”
