Milestone: Claude AI Successfully Plans Driving Route for NASA’s Perseverance Rover
NASA recently announced that its Jet Propulsion Laboratory (JPL) has successfully used the Claude AI model, developed by Anthropic, to plan and generate driving commands for the Perseverance rover currently on a mission on Mars. This event marks the first time an AI has independently completed a path-planning task in an extraterrestrial environment, representing a significant milestone in the automation of human deep space exploration.
Autonomous Navigation Across 360 Million Kilometers
This historic driving task took place on December 8, 2025 (Martian Sol 1707), in the Jezero Crater on Mars. Due to the vast distance of approximately 360 million kilometers between Earth and Mars, the one-way communication delay is as long as 20 minutes, making real-time remote control impossible. In this test, Perseverance, guided by the commands generated by Claude, successfully drove 400 meters autonomously, precisely avoiding complex terrain of rocks and sand pits. Although the distance was short, it validated the feasibility of AI making decisions in a complex physical world with minimal real-time human intervention.
How AI Became a “Martian Driver”
JPL engineers did not use Claude as a simple chatbot. Instead, they integrated it into a specialized programming agent environment called “Claude Code” and subjected it to a series of rigorous training and adaptation steps:
- Data Injection: Engineers fed Claude massive amounts of “experience” data, including years of NASA’s rover driving logs, terrain analysis data, and high-resolution images from orbiters.
- Language Learning: The model was trained to master the rover-specific “Rover Markup Language” (RML). This is an XML-based command set, the only language the rover’s onboard computer can understand and execute.
- Path Generation: In the mission, Claude was responsible for analyzing terrain data from satellites and the rover’s own cameras. It decomposed the 400-meter target path into a series of 10-meter micro-segments and calculated the optimal waypoints and driving commands for each segment.
Human-AI Collaboration and Efficiency Gains
To ensure the safety of the multi-billion dollar rover, this mission employed a strict human-AI collaborative workflow. The route commands generated by Claude were not sent directly. They first underwent a comprehensive review by an advanced validation system that simulated over 500,000 physical variables. Subsequently, human engineers made minor adjustments to the AI-planned route based on information from the rover’s ground-level perspective blind spots (e.g., the texture of certain specific sand dunes). The results showed that the AI-generated plan was highly consistent with that of human experts and demonstrated stable and reliable performance overall.
JPL’s engineering team estimates that introducing AI-assisted planning could reduce the time required for route planning by up to 50%. This efficiency boost means scientists can spend more time on data analysis and scientific discovery rather than on repetitive path planning, thereby accelerating humanity’s understanding of Mars.
Towards Autonomous Deep Space Exploration in the Future
The significance of this successful demonstration extends far beyond Mars. As humanity’s exploration targets expand to the subsurface oceans of Europa or the methane lakes of Titan, communication delays will be measured in hours or even days. In these more distant and unknown environments, probes must possess a high degree of autonomous decision-making capability to handle unexpected situations. An “onboard brain” like Claude will become an indispensable core component of future deep space missions, ensuring that missions can continue to execute scientific objectives continuously and safely during long periods without human intervention. Against the backdrop of NASA’s budget and personnel challenges, AI, as an “efficiency multiplier,” is becoming a key driving force for humanity’s exploration of the cosmos.
