NASA Taps AI Model Claude to Chart Course for Mars Rover Perseverance

United States - Ekhbary News Agency

NASA Leverages AI 'Claude' to Navigate Mars Rover Perseverance

In a pioneering stride for space exploration, NASA has enlisted the capabilities of Anthropic's advanced AI model, Claude, to meticulously plan the journey of its Perseverance rover across the Martian surface. This initiative underscores a pivotal shift towards integrating cutting-edge artificial intelligence into the complex demands of interplanetary missions, demonstrating AI's potential to revolutionize how we explore distant worlds. Claude has successfully plotted a course for the rover, marking a significant milestone in autonomous space navigation.

During a recent Martian month, the Perseverance rover traversed approximately 400 meters on the Red Planet, guided by a route generated entirely by AI. This ambitious undertaking received the endorsement of engineers at NASA's Jet Propulsion Laboratory (JPL), who opted to delegate the intricate task of route planning to Anthropic's sophisticated AI. The decision stems from the inherent challenges of navigating Mars, a terrain known for its treacherous landscapes that can easily trap expensive equipment, a fate NASA aims to avoid, recalling the incident with the Spirit rover in 2009.

The process of route planning for Mars rovers is traditionally a time-consuming and labor-intensive endeavor. Perseverance's mission team dedicates substantial effort to analyzing orbital and surface imagery of Mars to establish a series of waypoints, essential for guiding the rover's movements. Once these waypoints are plotted, the navigational data is transmitted across the vast expanse – an average of 140 million miles or 225 million kilometers – to Perseverance, which then executes the plan. Direct, real-time joystick control is rendered impossible by this immense communication delay.

Perseverance is equipped with an AutoNav system designed for real-time decision-making, enabling it to autonomously replan its route around unexpected obstacles like rocks or treacherous terrain en route to its pre-determined destination. However, the effectiveness of AutoNav is significantly enhanced by robust pre-planning. Recognizing the demanding nature of this pre-planning phase, described by Anthropic as "time-consuming" and "laborious," JPL researchers turned to Claude's visual processing capabilities to assist.

Claude analyzed high-resolution orbital imagery captured by the HiRISE camera aboard NASA's Mars Reconnaissance Orbiter, alongside terrain-slope data derived from digital elevation models. Its task was to identify critical geological features – including bedrock, outcrops, hazardous boulder fields, and sand ripples – and subsequently generate a continuous, traversable path complete with the necessary waypoints. The AI model generated commands for the rover in Rover Markup Language (RML), an XML-based format.

Interestingly, when initially queried, the publicly accessible web version of Claude could not emit RML and claimed no knowledge of it. Upon being directed to Anthropic's statement, Claude acknowledged its error and apologized. Although it could not provide an RML example, attributing this to a lack of a publicly documented standard, its demonstrated ability to generate RML when provided with NASA's specific data highlights its functional capability in this context.

Following Claude's generation of the RML path plan, human engineers undertook a critical review process. AI models, while powerful, are not infallible, and human oversight remains crucial, especially when operating multi-billion dollar assets on another planet. JPL engineers utilized a simulator, a virtual replica of the rover and its environment, to meticulously check over 500,000 telemetry variables related to the rover's projected position and potential hazards. They identified and implemented necessary corrections. According to Anthropic, the engineers found only minor adjustments were needed. For instance, ground-level images, unseen by Claude, revealed sand ripples more clearly within a narrow corridor, prompting rover drivers to refine the path's precision. Nevertheless, the AI-generated route proved remarkably robust.

On Martian days (sols) 1,707 and 1,709, corresponding to December 8 and December 10, 2025 (relative to Perseverance's landing date of February 18, 2021), the rover successfully executed routes planned by AI, rather than relying solely on human-generated plans. While the actual path taken on December 10, as depicted in annotated orbital imagery, shows slight deviations from the AI-planned route (likely due to AutoNav's real-time adjustments), the AI's foundational role is undeniable. This successful demonstration signals a future where advanced AI, particularly Vision-Language-Actions models, will play an increasingly integral role in robotic exploration.

NASA Administrator Jared Isaacman stated, "This demonstration shows how far our capabilities have advanced and broadens how we will explore other worlds." He further elaborated that autonomous technologies enhance mission efficiency, improve responses to challenging terrains, and increase scientific returns, especially over vast distances. "It's a strong example of teams applying new technology carefully and responsibly in real operations," he added.

Anthropic reports that JPL engineers estimate Claude can halve the time required for route planning. However, the company did not specify the exact duration being halved, and representatives from both Anthropic and JPL were unavailable for immediate comment to quantify this figure.

Ekhbary News Agency