The YAM-9 satellite discovered the desired object for the first time using AI

The breakthrough was made by Loft Orbital, on whose device the Gemma 3 visual and language analysis model (VLM) from Google DeepMind, designed to work on limited hardware, was deployed.

Data processing took place directly in orbit – the satellite analyzed images and classified objects, such as infrastructure in areas of contact between nature and human activity, without transmitting raw data to Earth.

Demonstration is important for two reasons. In the near term, it will reduce the amount of raw data transmitted to analysts and increase the efficiency of their work. In the long term, this is the first step towards creating a large-scale AI computing infrastructure in space.

Loft Orbital head of artificial intelligence Paul Lasserre explained that this opens up the possibility of 24-hour patrols in space, when a virtual space module will be able to monitor borders and report suspicious events. According to him, the experience gained will help in the development of new scientific instruments and autonomous space systems.

Read also:

Almost “American”: a ZIL with a sleeping bag made from the roof of a VAZ-2104 went on sale

1,300 rubles per hour: working conditions at the main Volkswagen plant became known

Residents of Zheleznodorozhny complained about “oil” rain