TL;DR
NASA's Perseverance rover completed the first Mars drives planned entirely by generative AI in December 2025, cutting human planners out of route generation and signaling a new era of autonomous planetary exploration.
What happened
- On December 8 and 10, 2025 (sols 1,707 and 1,709), Perseverance executed two AI-planned drives covering 210 meters and 246 meters along the rim of Jezero Crater.
- A vision-language model built in collaboration with Anthropic's Claude analyzed HiRISE orbital imagery and digital elevation data to identify hazards (boulder fields, sand ripples, bedrock) and generate waypoints, replacing human route planners for those drives.
- Before transmission, JPL's digital twin of the rover verified over 500,000 telemetry variables to confirm command safety across the 140-million-mile Earth-Mars gap.
- JPL roboticist Vandi Verma, who has worked autonomous systems across Spirit, Opportunity, Curiosity, and Perseverance, led the technical effort and described the AI as advancing all three pillars of off-planet driving: perception, localization, and planning.
- By June 2026, Perseverance had completed a full marathon distance (42.195 km) on Mars, doing it in 5 years and 4 months versus Opportunity's 11 years and 2 months.
Why it matters
- Human planners are no longer required for every route segment: the AI replicates their workflow using the same data sources, freeing mission teams to focus on science rather than navigation logistics.
- The communication delay to Mars (4 to 24 minutes one way) makes real-time control impossible; AI-planned routes directly address the core operational bottleneck of deep-space robotics.
- Commercial AI tools are now embedded in active NASA missions: the Anthropic collaboration marks a precedent for integrating frontier language models into safety-critical planetary operations.
- Longer, more complex traversals become feasible: human planners historically capped waypoint spacing at 330 feet (100 meters) to manage hazard risk, a constraint AI can systematically push beyond.
- NASA Administrator Jared Isaacman framed this as infrastructure for a permanent human presence on the Moon and crewed Mars missions, not just a rover efficiency gain.
What to watch next
- Whether JPL expands AI-planned drives from demonstration to routine operations on Perseverance, and at what waypoint distances, will confirm whether the December tests were a proof of concept or a turning point.
- The Anthropic-JPL collaboration model could attract other commercial AI labs (Google DeepMind, OpenAI) to bid on future rover and lander contracts, reshaping how NASA sources autonomous systems.
- Parallel development of AutoNav plus generative AI planning on future missions (lunar rovers, Mars Sample Return assets) will show whether the two autonomy layers can be unified into a single onboard system rather than a ground-based pre-planning tool.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.