NAVI-Orbital reports an in-orbit demonstration of a vision-language model running autonomous multimodal inference onboard a low Earth orbit spacecraft.
The system is designed for Earth observation, where satellites can collect more imagery than they can efficiently downlink or have humans process on the ground. Onboard AI could classify scenes and describe useful content before data ever leaves the spacecraft.
If this pattern scales, space-based sensing may rely more on edge AI to decide what data is worth transmitting, reducing latency and bandwidth pressure.