Signal delays of about twenty minutes between Earth and Mars make real-time control of rovers impossible, so routes must be preplanned before each drive. In December 2025, NASA’s Jet Propulsion Laboratory used Anthropic’s Claude model to generate commands in Rover Markup Language for the Perseverance rover to follow a roughly four-hundred-meter “breadcrumb trail” through rocky terrain in Jezero crater, marking the first AI-planned rover drive on another planet. Perseverance, active on Mars since 2021, investigates geology, climate, and potential ancient microbial life, and its drives are high-risk operations because hazards like sand traps can permanently immobilize a rover. Traditionally, human experts design waypoints using orbital imagery and rover cameras, then transmit plans across hundreds of millions of kilometers via the Deep Space Network.
To support route planning, engineers supplied Claude Code with years of operational data and domain knowledge so the model could interpret overhead images, propose ten-meter route segments, and iteratively refine waypoints by critiquing its own output. The AI-generated plans for sols 1707 and 1709 were checked in Perseverance’s existing simulation environment, which modeled over “500,000 variables” to validate rover positions and identify potential hazards. Human drivers made only small adjustments—such as fine-tuning around sand ripples visible in ground-level images that Claude had not seen—before transmitting the commands that Perseverance successfully executed. Engineers estimate that this approach can cut planning time in half, increase consistency, and free operators to schedule more drives and collect more scientific data.
The experiment serves as a proof of concept for wider use of autonomous AI in space missions, where models will need to understand unfamiliar situations, control complex instruments, and make rapid decisions with limited human oversight. These capabilities are especially relevant for NASA’s Artemis campaign to establish a base at the Moon’s south pole, where efficient resource use and robust automation will be vital. Looking further ahead, increasingly independent AI systems could operate probes in harsh, distant environments with long communication delays, potentially enabling missions to icy moons like Europa and Titan and allowing robots to “chart their own course” through subsurface oceans. Claude’s four-hundred-meter Mars drive is presented as an early glimpse of a future in which autonomous machines extend human exploration deeper into the solar system.
Reference
Anthropic. (2026). Claude AI powers NASA’s first AI-planned Mars rover drive. Anthropic. https://www.anthropic.com/features/claude-on-mars