Project Loon uses machine learning to make internet balloons ‘dance on the wind’

The aptly named Project Loon started as a “moonshot” experiment at Google’s X division in 2011. The goal was to provide wireless internet access to remote areas with high-altitude balloons, which does indeed sound loony. It has become one of the most well-known endeavors by X, now a division of Google’s parent company Alphabet. It’s not just a wacky idea anymore — it’s starting to work. X has announced improvements to its navigation algorithm that could vastly reduce the number of balloons needed to cover an area.

Early experiments in Project Loon were basically designed to prove it wouldn’t work. Was it even possible for these balloons to cover large distances? Could they navigate on air currents? Could they be steered to specific locations? The answer to all those questions turned out to be an emphatic and surprising “yes.” X leader Astro Teller says that one of Project Loon’s early successes came in 2014, when balloons released in New Zealand were able to circumnavigate the globe and arrive within a few tens of kilometers of a target in Australia.

The Project Loon balloons rely on air currents to traverse the planet. They can alter their altitude in order to catch different gusts, but that’s it. Originally, Loon was intended to be giant rings of balloons sailing around the globe. As one balloon drifted out of range of a certain area, another would move in to takes its place. X has applied Google’s machine learning technology to navigation, which has improved the capabilities of the balloons. It may no longer be necessary to deploy those giant rings of floating hotspots, because the balloons can stay put.

Teller explains that in early 2016, the team noticed that some balloons in the test fleet were behaving strangely. Rather than sail for long distances, they were lingering around the same area. X researchers realized the algorithms could be used to do this intentionally. The balloons could essentially “dance on the wind” for a long time, providing more reliable connectivity with fewer balloons.

Throughout 2016, the team tested this refined algorithm by launching balloons from Puerto Rico , then flying them to Peru. They were able to keep some balloons there for as long as 90 days. According to Teller, an area that might have required a rotating network of hundreds of balloons before might only require dozens now.

This improvement to Loon navigation could mean an accelerated timeline for actual deployment of the balloons to provide connectivity to people.