Alphabet’s Loon team, which is responsible for transporting the Internet from stratospheric helium balloons to the ground, has reached a new milestone when the balloon’s navigation system is no longer run by human designed software. < / P > < p > this artificial intelligence flight control system based on deep reinforcement learning is more efficient and skilled than the old, man-made system. The system is now managing the loon balloon formation over Kenya, which launched its first commercial Internet service in Kenya in July after testing it in a series of disaster relief operations and other test environments for most of the past decade. Similar to teaching computers to play complex video games and helping software learn how to operate manipulators, reinforcement learning is a technology that allows software to improve its skills through trial and error. Obviously, in the real world, such repetition is impossible when facing high-altitude balloons, because the operating costs of high-altitude balloons are very high, and once they crash, the maintenance costs are higher. As a result, Loon, with the help of Google’s artificial intelligence team in Montreal, taught his flight control system how to use computer simulation to fly balloons. In this way, the system can be improved over time before being deployed to a real-world balloon formation. It has been proved that reinforcement learning is practical for stratospheric balloon formation. Nowadays, the most complex task of Loon navigation system is solved by the algorithm learned by computer in simulated balloon navigation experiment. < / P > < p > in the first practical test over Peru in July 2019, the researchers found that the artificial intelligence controlled system felt better than the human developed system because it brought itself closer to the team’s field equipment for measuring LTE signals, and this test paved the way for more experiments to prove the effectiveness of the system before formally replacing the team A system that took years to build by hand. Global Tech

By ibmwl