For the first time in the history of space exploration, a Mars rover has traversed the surface of another planet by following a route entirely designed by artificial intelligence. On December 8 and 10, 2025, the Perseverance rover traversed nearly 400 meters of Jezero Crater on Mars, following a path planned by Claude, the AI model developed by Anthropic. This major technological breakthrough marks a turning point in the way humanity explores distant worlds.
Perseverance, a rover that has already achieved numerous historic firsts
Since landing on Mars in February 2021, Perseverance has achieved a series of firsts: the first audio recordings from the Red Planet, the collection of samples for future return to Earth, and the production of oxygen from the Martian atmosphere. This new milestone, however, marks a significant step forward by entrusting an AI with a critical task of high operational value: autonomous route planning.
The distances covered— 210 meters on December 8 and 246 meters two days later—may seem modest. Yet they are significant in an environment where every movement is risky and costly, both financially and scientifically.
A task that had previously been reserved for humans
Until now, every move Perseverance made required the meticulous work of entire teams of engineers at the Jet Propulsion Laboratory (JPL). They spent hours analyzing orbital images, digital elevation models, and topographic data to plot safe routes. Waypoints were positioned with extreme caution, often spaced less than 100 meters apart, to avoid rocks, steep slopes, boulder fields, or sand ripples.
Although reliable, this method significantly limited the pace of exploration. In practice, Perseverance travels only 100 to 300 meters per sol (Martian day) when it is on the move1. A major obstacle to the mission’s scientific ambitions.
Claude cuts planning time in half
It is precisely this limitation that AI is designed to address. Working closely with JPL engineers, Claude was trained to analyze the same data as human planners: high-resolution images from the HiRISE camera aboard the Mars Reconnaissance Orbiter, digital elevation models, and terrain data.
Thanks to its vision and reasoning capabilities, the model can now identify hazardous areas and generate a continuous route with appropriate waypoints. According to NASA, this approach cuts the time required for route planning in half2. A significant gain that paves the way for more missions, more scientific data, and faster exploration of the planet.
Strict human safeguards
That doesn’t mean, however, that we’re letting the AI operate on its own. Every route generated by Claude has been rigorously verified by the JPL teams. The AI’s proposals were tested in a digital simulator—a true digital twin of Perseverance—which is used daily to validate the commands sent to the rover.
Before transmission to Mars via the Deep Space Network, more than 500,000 telemetry variables were examined3. In most cases, only minor adjustments were necessary. One of these involved sand ripples visible in ground-based images that the AI had not accounted for. The operators then fine-tuned the trajectory locally. Overall, the route remained true to the initial plan, and the rover successfully crossed the area.
Essential range for long-distance missions
For NASA, this demonstration goes far beyond simply saving time. It addresses a fundamental challenge of space exploration: communication latency. Between Earth and Mars, this latency ranges from 4 to 24 minutes, depending on orbital positions. For more distant destinations, such as the moons of Jupiter or Saturn, the latency is measured in hours4.
Under these circumstances, systems capable of making complex decisions autonomously are becoming indispensable. As Vandi Verma, a space robotics engineer at JPL, points out, generative AI shows great potential for streamlining the core elements of off-planet navigation: perception, localization, planning, and control.
Ethical Issues and the Responsibility of Autonomy
Delegating some control to AI, however, raises ethical and operational questions. Who is responsible in the event of an error? How can we ensure that decisions made by AI remain aligned with scientific objectives and safety constraints? NASA emphasizes a key point: humans must remain in the loop. AI is not the final decision-maker, but a decision-support tool, subject to systematic validation.
This incremental approach embodies a philosophy of controlled autonomy, in which AI enhances human capabilities without replacing them. A model that is set to become widespread as missions become more distant and complex5.
A new era for space exploration
Less than a year ago, the same AI model struggled to finish a game of Pokémon Red. Today, it successfully plans routes for a rover worth hundreds of millions of dollars on another planet. This rapid evolution illustrates the potential of generative AI when integrated into critical environments and governed by strict protocols.
By entrusting an AI with the task of planning routes on Mars, NASA isn’t just improving the efficiency of a mission. It’s paving the way for space exploration that is more autonomous, more responsive, and more ambitious. This is a key step in preparing for future missions to even more distant worlds.
Learn more
The growing reliance on artificial intelligence in space missions is part of a broader trend toward integrating AI into the core of both private and public space ambitions. On a related topic, check out our article “SpaceX Acquires xAI: AI at the Heart of Musk’s Space Ambitions”, which analyzes how industry players are now structuring their space strategies around advanced artificial intelligence capabilities.
References
1. NASA JPL. (2023). Mars Rover Mobility and Planning Constraints.
https://mars.nasa.gov
2. NASA. (2025). AI-assisted route planning for Mars rovers.
https://www.nasa.gov
3. Anthropic. (2025). Claude in high-stakes autonomous planning.
https://www.anthropic.com
4. ESA. (2024). Deep space communications and latency.
https://www.esa.int
5. IEEE. (2023). Human-in-the-loop autonomy for space systems.
https://ieeexplore.ieee.org
