Alphabet subsidiary Loon released details of an AI technology being deployed to improve flight management for its fleet of connectivity balloons, crediting the approach with improving efficiency and its ability to maintain coverage in specific areas.

In a blog, Loon CTO Salvatore Candido said the system uses a type of machine learning which allows an AI agent to “learn by trial and error in an interactive environment using feedback from its own actions and experiences”.

He added extensive testing showed the deep reinforcement-learning AI controller was able to steer balloons better than the human-designed algorithms used previously, noting it was able to keep the fleet over a targeted area for a longer while using less power during a 39-day trial conducted over the Pacific Ocean.

“Obviously, the more closely Loon can remain to a defined location the better as it allows more stable service to people there.”

Candido stressed energy efficiency gains were also key since the balloons’ on-board navigation and steering systems use the same solar power source: “Using less power to steer the balloon means more power is available to connect people to the internet, information and other people.”