At some point last December, a Mars rover drove somewhere its human operators had not manually planned. Two drives, actually: 210 meters on December 8 and 246 meters on December 10, 2025. Both were designed by an AI.

NASA's Jet Propulsion Laboratory recently published details of a collaboration with Anthropic in which Claude vision-language models were used to autonomously plan drives for the Perseverance rover. The system analyzed orbital imagery from the Mars Reconnaissance Orbiter, assessed terrain for obstacles and hazards, identified a viable path, and generated the Rover Markup Language commands that Perseverance actually executed on the surface.

Human engineers reviewed each plan before it was transmitted to the rover. But the planning itself came from the AI. That distinction matters. Rover navigation is painstaking work that requires expertise in Martian geology, spacecraft dynamics, and the specific constraints of operating hardware across a 20-minute communications delay. The fact that a language model produced plans that passed human review is not a small thing.

For NASA, the appeal is throughput. Perseverance currently takes several days to plan a significant drive, partly because of the specialized human labor involved. If AI can handle the initial planning pass, scientists can redirect that time toward analysis and science decisions. JPL describes the project as an early feasibility demonstration, not a production system, and is careful not to overstate what two successful drives prove.

For Anthropic, the collaboration is a concrete proof point for Claude in a high-stakes operational domain. And for the wider field, it is a real milestone: an AI system handling a genuine operational task on another planet, not a benchmark or a simulation, but actual commands driving an actual vehicle across actual Martian terrain.

The two December drives are a tiny fraction of Perseverance's total distance since landing in 2021. But they are the first where the route was chosen by something other than a human.

Sources

  1. i. www.jpl.nasa.gov
  2. ii. www.nasa.gov
  3. iii. www.space.com

Commentarii · 0

Add · a · Comment