Project Loon gets new AI autopilot system

Google started its Project Loon several years back with the goal of putting a balloon into the stratosphere that carried with it tech needed to beam internet access to people on the ground. The goal was to use the balloons to connect parts of the world where internet was unavailable or very slow. Over the summer, a balloon of this sort was launched in Peru and it stayed there for 98 days.

The challenge with putting these balloons in to the stratosphere and having them provide a stable and persistent connection is that balloons are at the mercy of the winds and often float away. This tendency to float away is why keeping the balloon in Peruvian airspace for 98 days is such a big deal.

Project Loon used artificial intelligence to keep the balloon where it could perform at its best. In the beginning of Project Loon, the balloons were guided with algorithms made by hand that were able to respond to certain predetermined variables such as altitude, location, wind speed, and time. The new algorithms used in Peru make use of machine learning.

They algorithm is able to analyze data and learn based on what happened in the past. By using these machine learning algorithms the team is able to keep the balloons in place where they want them for longer periods. This means that the team can provide internet access to a location with fewer balloons required. The form of machine learning that the Project Loon team is using is called "Gaussian processes" that allows the balloon to predict the course it should take an when it should move up or down controlled by pumping air into or out of the balloon inside the balloon.

SOURCE: Wired and Google