NASA’s Perseverance rover has achieved a groundbreaking milestone by completing the first drives on another world planned entirely by artificial intelligence. The historic demonstration took place on December 8 and 10, 2025, when the six-wheeled explorer traversed the Martian surface using waypoints generated by generative AI rather than human route planners.
The mission, led by NASA’s Jet Propulsion Laboratory in Southern California, marks a significant leap forward in autonomous space exploration. For the past 28 years, rover routes across multiple Mars missions have been carefully plotted by human drivers who analyze terrain and status data to create safe paths. This traditional approach requires waypoint spacing of no more than 330 feet to avoid potential hazards. The recent AI-planned drives demonstrate a new paradigm where machine intelligence can handle complex navigation tasks previously reserved for human operators.
How the AI-Planned Drives Worked
The Perseverance team leveraged vision-language models, a type of generative AI, to analyze high-resolution orbital imagery from the HiRISE camera aboard NASA’s Mars Reconnaissance Orbiter. The system also processed terrain-slope data from digital elevation models to identify critical features including bedrock, outcrops, hazardous boulder fields, and sand ripples. After analyzing this data, the AI generated a continuous path complete with waypoints where the rover could receive new instructions.
Before transmitting commands to Mars, the engineering team subjected the AI-generated instructions to rigorous verification. They processed the drive commands through JPL’s digital twin, a virtual replica of the rover, confirming compatibility with Perseverance’s flight software by checking over 500,000 telemetry variables. This careful validation ensured the AI’s route would work safely with the rover’s existing systems.
On December 8, Perseverance drove 689 feet using AI-generated waypoints stored in its memory. Two days later, on December 10, the rover completed a second AI-planned drive covering 807 feet along the rim of Jezero Crater. The second drive lasted approximately two hours and thirty minutes, with the rover’s navigation cameras capturing the journey for later reconstruction.
Collaboration with Anthropic’s Claude AI
The initiative emerged from JPL’s Rover Operations Center in partnership with Anthropic, the artificial intelligence company behind the Claude AI models. The collaboration allowed NASA to apply advanced generative AI capabilities to the challenges of interplanetary navigation.
NASA Administrator Jared Isaacman praised the demonstration as evidence of advancing capabilities that will reshape how humanity explores other worlds. He noted that autonomous technologies enable missions to operate more efficiently, respond to challenging terrain, and increase scientific returns as spacecraft travel farther from Earth. The administrator described the project as a careful and responsible application of new technology in real operations.
Why Autonomous Navigation Matters
The vast distance between Earth and Mars creates significant communication delays that make real-time remote control impossible. Mars sits approximately 140 million miles away on average, meaning radio signals take many minutes to travel between the planets. This time lag prevents mission controllers from directly steering rovers using joysticks or immediate commands.
Instead, traditional rover operations require human planners to analyze imagery and data, sketch routes, and upload complete instruction sets via NASA’s Deep Space Network. The rover then executes these plans autonomously while Earth awaits confirmation of success. AI-planned routes promise to streamline this process, reducing the workload on human operators while potentially enabling longer, more complex journeys.
Vandi Verma, a space roboticist at JPL and member of the Perseverance engineering team, explained that generative AI shows promise across the fundamental pillars of autonomous navigation. These include perception of terrain features, localization to determine position, and planning and control to select and execute safe paths. She envisions a future where smart tools help rovers handle kilometer-scale drives while automatically flagging scientifically interesting surface features hidden within massive image archives.
Looking Toward Future Missions
Matt Wallace, manager of JPL’s Exploration Systems Office, described the broader implications of integrating intelligent systems directly into space exploration hardware. He foresees AI applications not only in ground-based operations but also in edge computing aboard rovers, helicopters, drones, and other surface assets. These systems could be trained using the collective expertise of NASA engineers, scientists, and astronauts.
Wallace identified this integration as game-changing technology necessary for establishing permanent human presence on the Moon and advancing crewed missions to Mars. As exploration pushes deeper into the solar system, the ability to make real-time navigation decisions without waiting for Earth-based input becomes increasingly critical.
The Perseverance AI demonstration represents a crucial step toward that future, proving that generative AI can safely plan routes across alien terrain. With two successful drives completed, the mission team has opened new possibilities for how humanity’s robotic explorers will navigate the Red Planet and beyond.
