NASA has successfully tested generative AI to help plan Perseverance’s first AI-designed drives on Mars, with the rover executing routes created without direct input from human planners on Dec. 8 and Dec. 10. Led by NASA’s Jet Propulsion Laboratory in Southern California, the demonstration used vision-capable AI to generate navigation waypoints — a task normally handled manually by rover operations teams — allowing Perseverance to travel safely across challenging terrain.
The effort relied on vision-language models trained on JPL’s surface mission datasets, using the same imagery and data human drivers typically analyze when mapping rover routes. In collaboration with Anthropic, the team used Claude AI models to interpret high-resolution orbital imagery from the HiRISE camera aboard NASA’s Mars Reconnaissance Orbiter, along with terrain-slope data from digital elevation models. The AI identified hazards such as boulder fields, sand ripples, and rocky outcrops before producing a continuous path marked by waypoints.
Mars’ distance — about 140 million miles from Earth — makes real-time control impossible, forcing rover teams to plan drives in advance and send instructions through NASA’s Deep Space Network. Traditionally, human planners space waypoints carefully, usually within about 330 feet, to reduce risk. In this case, generative AI handled the route-building step for Perseverance’s drives on sols 1,707 and 1,709, demonstrating how autonomy could reduce operator workload over time.