Raspberry Pi Top Gun Style: AI Triumphs Over US Air Force Pilot In Dogfight Simulation


Artificial Intelligence (AI) is taking over the world, and the dawn of the man-versus-machine era could be happening soon across the skies.


Though maybe not as much as world domination, just for manned and unmanned cooperative air flight situations.

Previous AI accomplishments include Giraffe, a chess AI with skills equivalent to that of an international chess grandmaster AlphaGO, a deep learning AI that beat the former South Korean Go champion Lee Sedol and IBM's Watson, previously a Jeopardy champ, now an analytics powerhouse.

Now, AI technology is aiming for broader horizons as a recent AI accomplishment is expected to conquer the skies soon.

ALPHA, a fully AI-controlled aircraft pilot program, has been defeating aircraft experts as well as a retired but well-seasoned Air Force pilot Col. Gene Lee in multiple dogfight simulations since October. Even more so, the program did not only beat these pilots every single time but also won in scenarios wherein programmers gave it a major disadvantage in terms of "speed, turning, missile capability and sensors."

"I was surprised at how aware and reactive it was. It seemed to be aware of my intentions and [reacted] instantly to my changes in flight and my missile deployment," Lee describes, adding that ALPHA "knew how to defeat the shot [he] was taking" and could move instantaneously, depending on the situation, "between defensive and offensive actions."

And as if not all of that are incredible feats, developers behind the ALPHA project claim that the AI runs on consumer-friendly products, specifically, the low-end computer systems found on most store shelves that retail for as low as $35, such as the Raspberry Pi. These affordable systems can only provide so much processing power that it pales in comparison to supercomputers used by large computing companies, i.e. Google's D-Wave quantum computer.

ALPHA's ability to run on such limited systems is worthy of note because flight scenarios involve a handful of parameters and conditions to consider, which one would normally assume needs a more powerful and capable processing unit. ALPHA effectively disproves this assumption as evident in its successful runs.

Despite its seemingly capable current build, developers will still continually work on improving and enhancing ALPHA's algorithms and flight responses in the hopes of making it more suitable for real-world flight scenarios. Its so-called Genetic Fuzzy Tree programming language helps the system instinctively learn from past simulations, "genetic," and perform complex computations using only the needed variables, "tree," and thus its growing and highly efficient learning process.

In the long run, the goal is to make teams of unmanned aircraft operate seamlessly beside manned aircraft, to make the AI respond effectively, according to the instructions of its overseeing human pilot. After all, "air combat ... is a highly dynamic application of aerospace physics, skill, art, and intuition" where microseconds really do matter as "the cost for a mistake is very high."

Detailed information can be accessed through an article by the Journal of Defense Management.

ⓒ 2018 All rights reserved. Do not reproduce without permission.
Real Time Analytics