Alphabet’s Loon hands the reins of its internet air balloons to self-learning AI

Alphabet’s Loon, the team responsible for bringing the Internet to Earth from helium stratospheric balloons, has taken the next step: Its navigation system is no longer managed by man-made software.

Alphabet’s Loon
Alphabet’s Loon self-learning AI

Instead, the company's Internet balloons are driven around the world by artificial intelligence - in particular, a set of algorithms written and implemented by the flight control system based on augmented learning. 

Self-learning AI

It is profoundly more efficient and skilled than the old man-made. a. The system now operates the Loon Airship fleet over Kenya, with Loon launching its first commercial internet service in July after testing its fleet as part of a series of disaster relief and disaster relief initiatives. Other test environments for most of the past decade.

Similar to how researchers have made incredible advancements in artificial intelligence by teaching computers to run cutting-edge video games and helping programs learn to manipulate robotic hands realistically, augmented learning is a technology that allows a program to teach itself skills through trial and error. 

Obviously, such redundancy is not possible in the real world when it comes to high altitude balloons which are expensive to operate and even more expensive to repair if they fail.

So Loon, like many other AI labs that turned to augmentative learning to develop cutting-edge AI programs, taught his flight control system how to drive airships to the using computer simulations, with the help of the Montreal Google AI team. In this way, the system could improve over time before being deployed on a real fleet of balloons.

“While the promise of RL (reinforcement learning) for Loon has always been great, when we first started exploring this technology it was not always clear that deep RL was practical or viable for elevated platforms drifting. through the stratosphere independently for long periods of time, ”Candido, technical director of Loon and co-author of a research article on the new flight-control system published this week in the scientific journal Nature, in a blog post. “It turns out that RL is an operation for a fleet of stratospheric balloons. 

Google loon balloon

Nowadays, the most complex tasks of the Loon navigation system are solved through an algorithm that is learned by a computer running a balloon navigation experience. in the simulation. "

Lone says his system is the world's first deployment of this type of AI in a commercial flight system. Not only that, but it surpasses the system that humans have designed. Candido writes: "Frankly, we wanted to confirm that using an RL machine could build a navigation system on par with what we built ourselves." "The acquired deep neural network defining flight controls is encapsulated in an appropriate safety assurance layer to ensure the agent is always driving safely. Through our simulation benchmark, we have not only been able to replicate our navigation system but also dramatically improve it with RL."

Alphabet’s Loon--self-learning AI
Google loon balloon

During the first real-world test in Peru in July 2019, the AI-controlled flight system collided with a traditional flying system, controlled by a man-made algorithm called StationSeeker, which was designed by Loon engineers themselves. `` In a sense, it was the machine - which spent a few weeks building its console - against me - that, along with many others, spent many years carefully modifying the traditional console based on a decade of experience working with Loon Balloons "We were nervous ... We were hoping to lose," Candido says.

The AI-controlled system easily outshone a human by staying close to a device the team used to measure LTE signals in the field, and this test paved the way for further trials to prove the efficiency of the system before it officially replaces the device. The team spent years building by hand. Loon now believes his system can "serve as proof that RL can be useful in controlling complex real-world systems for essentially continuous and dynamic activity."

In his closing remarks, Candido raises the question of whether this type of AI deserves its name, given its specialization and similarity to a traditional robotic system but not self-learning like those that use heavy machinery. or control elements of public transport.

"While there is no chance that the overpressure balloon effectively drifting through the stratosphere will become conscious, we have moved from designing its own navigation system to designing by computers in a data-driven manner.", he said. “Even though this isn't the start of Asimov's novel, it's a good story and maybe something worth calling artificial intelligence.”

Previous Post Next Post