(Subscribe to our Today's Cache newsletter for a quick snapshot of top 5 tech stories. Click here to subscribe for free.)
Google’s moon-shot project Loon’s internet-providing balloons will be controlled and steered by AI, not humans.
Developed by Loon and Google AI, the navigation system leverages deep reinforcement learning (RL) to navigate the stratospheric baloons.
RL is a type of machine learning technique that enables an agent to learn by trial and error in an interactive environment using feedback from its own actions and experiences, Loon said in a statement.
The new technology is different from the conventional automated systems that have fixed procedures developed through traditional methods by engineers.
Instead of engineers building specific navigation machine, the team is making a machine that in turn leverages computational resources to build navigation machines.
Loon claims that it is the world’s first deployment of reinforcement learning in a production aerospace system.
Testing the system
The team deployed the latest RL controllers named perciatelli44, in the stratosphere over the Pacific Ocean. The conventional navigation system had already performed well in that location, it said.
The test was conducted for 39 days, with the RL controller navigating a group of balloons for nearly 3,000 flight hours.
The experiment objective was to remain within 50km of one defined location.
The closer Loon can remain to a defined location, the more stable service it can offer to the people in the area. The results were excellent, it said.
The RL system kept the balloons in range of the desired location more often and also consumed less power. Using less power to steer the balloon leaves more power available to connect people to the internet, information, and other people.
Loon’s system is solar-powered, and the energy collected powers communication equipment while powering the navigation system.