TL;DR
Anthropic’s Claude AI generated navigation commands for NASA’s Perseverance Rover on December 8 and 10, 2025, guiding it approximately 400 metres through rocky Martian terrain. This marks the first AI-planned drive on another planet, with engineers noting only minor changes were needed to Claude’s plans.
How It Works
JPL engineers provided Claude with historical rover operation data and context. Claude then used its coding capabilities to write commands in Rover Markup Language—the specialised programming used for Mars rovers.
The process involved analysing overhead images to plot the route in 10-metre segments, iteratively refining waypoints to navigate the terrain safely. The simulation modelled over 500,000 variables to verify safety before transmission.
Minimal Human Intervention Required
When engineers reviewed Claude’s generated plans before transmission to Mars, “only minor changes were needed.” The rover successfully completed the planned path, demonstrating that AI-generated navigation commands can meet the rigorous safety standards required for planetary exploration.
Cutting Planning Time in Half
Engineers estimate this approach will cut route-planning time in half, enabling more frequent rover drives and increased scientific data collection. Traditional manual planning for Mars rovers requires careful consideration of terrain, power constraints and scientific objectives—a time-intensive process that limits how much ground can be covered.
Looking Forward
Anthropic positions this as a prototype for future autonomous AI systems supporting longer space missions, including NASA’s Artemis lunar programme and distant solar system exploration. As missions venture further from Earth, communication delays make autonomous decision-making increasingly valuable.
The collaboration demonstrates a practical application of large language models beyond text generation—using their reasoning capabilities to solve complex spatial and logistical problems in environments where errors carry significant consequences.