Generative AI Gives Spacecraft New Autonomy

Generative AI is beginning to give spacecraft a new level of autonomy, a capability that engineers once viewed with caution because of reliability and control concerns. IEEE Spectrum reports that NASA's Jet Propulsion Laboratory used Anthropic's Claude models to help plan two drives for the Perseverance rover on Mars, with human planners remaining in the loop. The work is an early sign of how AI could assist future missions that face long communication delays and limited opportunities to send commands from Earth. Spacecraft autonomy has traditionally been narrow and rules-based. A rover might navigate around obstacles or select a safe path, but mission planning still depends heavily on teams on the ground. Generative AI could broaden that autonomy by helping interpret science targets, draft drive plans, identify anomalies, and propose responses. For Mars, a signal can take several minutes to travel one way, so real-time joystick control is impossible. A more capable onboard or ground-assisted AI could let a rover make better decisions between check-ins, or help operators quickly evaluate complex terrain and instrument data. The JPL experiment did not hand over control to AI. Human planners reviewed and approved the plans, which is important for safety and accountability. That human-in-the-loop approach is likely to remain standard for high-stakes space operations, at least in the near term. The appeal of generative AI is not replacing experts but augmenting them: reducing planning time, exploring more options, and helping teams manage increasing mission complexity. Still, challenges remain. AI models can hallucinate, misunderstand constraints, or behave unpredictably. Space hardware must also operate with limited power and computing resources, and radiation can affect electronics. Engineers will need rigorous validation, transparent reasoning, and fail-safe mechanisms. If those hurdles are managed, generative AI could become a useful partner for navigation, science targeting, and anomaly response, helping spacecraft operate more independently when Earth is far away.