From hurricane-affected citizens of Puerto Rico to rural parts of Kenya, Google Loon’s AI-controlled balloons are providing wireless Internet connectivity to people around the world while aiming to keep costs low. As the world is increasingly remote and online due to the ongoing COVID-19 pandemic, a constant connection to the Internet—and thus information—is starting to become an essential service, or at least a necessity in the current age.
Google’s Loon team tested fixed-volume, super-pressure balloons, which are typically used to carry out experiments in the upper atmosphere, and trained an AI control system to hold the balloons within a particular range (imagine your LTE connection cutting out because a balloon was blown off course—definitely not ideal). Maintaining a position within a certain horizontal distance of a ground location is commonly known as station-keeping. Typically, this concept is applied to orbital station-keeping when applied to spacecraft like satellites.
Loon’s AI-controlled balloons maintain their range by changing their height to move between regions in which winds blow in various directions—when a balloon is pushed away from the station at one altitude, the AI changes the altitude of the balloon such that winds blowing from a different direction will be able to push it back again to the range where it is supposed to stay. Maintaining the position is key to Loon’s success, as no one will want to work with a spotty coverage, especially since there may be other, more reliable (although not necessarily cheaper) options available.
In order to stay within range of their stations, the balloons perform one of two tasks: when a balloon is outside its range, the AI seeks winds that point within a small angle of their stations’ locations; when a balloon is inside its range, it seeks out lighter winds to stay within range and minimize its horizontal movements.
One important constraint to worth noting is that although balloons which are more active at exploring the winds around them are more likely to maintain their station-keeping goals, these altitude changes are made with the use of the internal battery power, which must be shared with other functions of the balloon setup, such as relaying telecommunications and monitoring the environment around the balloons.
The main issue holding back automated machines from being more ubiquitous in our everyday lives is the fact that the environments in which they are deployed greatly differ from the idealized environments that they are conceived in, whether they be theoretical or in a lab. The real world is filled with noise, chaos and various unknowns—it just isn’t possible for any computational system to track and have stored knowledge of all the features of the world. How, then, can this seemingly insurmountable problem be overcome in order to train an artificial agent to make optimal decisions?
Loon’s answer is a type of machine learning known as reinforcement learning. In the case of these balloons, their altitudinal decisions are based on a combination of local forecast and observed winds, the historical records of global winds, and the balloons’ projected future flight paths. The wind data is, of course, not complete, and the engineers filled in the informational gaps with procedural “noise” simulations, which bolstered the AI’s decision-making skills.
The engineers used reinforcement learning by giving balloons a reward depending on how well they kept their positions, while incentivizing exploration. Although initially the reward setting punished the AI harshly for having navigated outside of its station range, Loon’s engineers found that softening the transition of the penalty incurred at the boundaries of the zone actually improved the controller’s score.
Autonomous balloons have several possible commercial applications, but they are currently favored for use in telecommunications. Loon’s balloons have previously been used in Puerto Rico when Hurricane Maria wiped out cell towers, and made a commercial debut providing a 4G LTE network to an area of over 80,000 square kilometres in the east African nation of Kenya. The balloons act as cell towers that transmit information to ground stations and personal devices. They can last over 100 days before needing to descend to Earth for maintenance.
The relatively inexpensive infrastructure seeks to bridge the last-mile connectivity issues faced by those in rural areas—there’s no need for expensive excavation, laying of lines, or building cell towers. The cost of accessing the Internet is the main thing barring more people within the African continent from accessing it, as only 28 percent of Africa’s 1.3 billion people were using the Internet in 2019. This cheaper infrastructure allows providers to offer services at lower price points, thus enabling more customers to connect to the Internet, and more importantly, the information it provides.
Loon has also recently signed a deal with Vodacom to bring its Internet services to Mozambique, and a partnership with Telefónica will be bringing connectivity to remote areas of the Amazon rainforest in Peru. Executives at Loon have withheld details of the contract or any financial arrangements, but SpaceX’s project Starlink, which also seeks to bring worldwide Internet connectivity, especially to remote areas, was able to raise over $100 million. It seems that the lower costs and closer distance to Earth of Loon’s autonomous balloons could address these issues far more effectively and with less latency than satellite-based worldwide coverage due to their closer proximity to the devices using their signal.
Loon’s hopes of connecting the world to the Internet are reminiscent of Facebook’s attempt to use solar-powered unmanned aerial vehicles (UAVs) to provide Internet connectivity in the African continent. The project has since been largely shuttered, although the company maintains that it merely allowed the project to continue with its partners, and that the company is simply moving away from designing and building its own drones in-house; Facebook has since closed its Aquila aircraft facility in Bridgewater, England, and there appear to be no updates on the project since 2018.
Facebook’s attempt at providing Internet via drones was criticized as an attempt to build the customer base further of its namesake social network, as many industry experts begin to see the social network’s growth and influence stagnating. Loon’s autonomous balloons seem like a more sustainable solution as they require less expensive materials to build, their development has allowed for advances within artificial intelligence training, and they do not require long runways to be built for the launch, as the inflated balloons are roughly the size of a tennis court.
Loon’s ability to fix the balloon’s geographical position to a particular geographical area allows for applications beyond telecommunications—keeping a balloon operational for months at a time would allow for long-term environmental monitoring of things like air quality, animal migration patterns, as well as the tracking and monitoring of various borders. Not only do these autonomous balloons cost less and require less work than most traditional infrastructure, they also have the added benefit of being relatively environmentally friendly compared to many other space and atmospheric technologies as they do not burn fossil fuels and require fewer materials that are less expensive to manufacture.