Back to all articles
•Technology

Earth Observation Satellite Achieves AI Milestone with Independent Data Analysis

View original source

For the first time, an Earth observation satellite identified points of interest autonomously, without human input, marking a significant milestone in space technology.

Key Details:

  • The achievement occurred in April, utilizing a vision-language model, Gemma 3, developed by Google DeepMind.
  • This advancement suggests how AI could revolutionize space-based sensors and their value. Typically, satellites transmit large amounts of data for ground analysts to evaluate, but this innovation conducts initial triage in orbit.
  • The satellite, YAM-9, is by Loft Orbital, aiming to enhance their infrastructure-as-a-service model. Launched in Fall 2025, it includes a Nvidia Jetson Orin AGX GPU.
  • The vision-language model processes limited hardware, identifying where the natural environment meets human development or infrastructure around railway hubs.
  • In the near term, it reduces the volume of raw data, facilitating analyst workload, and offers a long-term potential for large-scale AI operations in space.

Impact:

  • Enables constant monitoring from space with immediate reporting capabilities on suspicious activities.
  • Encourages other companies like Planet Labs and Kepler Communications to explore similar AI applications.
  • The project, led by NASA JPL’s AI team with engineer Juan Delfa Victoria, serves as a base for future AI applications, including assisting astronauts on missions to the moon or Mars.
  • Loft Orbital currently operates 12 spacecraft, aiming for 50-100 for global real-time coverage.

Innovative use cases offer insights into deploying larger compute infrastructures in space, crucial for power and memory management as they develop new scientific tools.