AAM-Gym for Artificial Intelligence Practice
Advanced Air Mobility gets a research and development testbed
Practice, it’s been said, make perfect. But what if there’s no place to practice the skills you hope to hone to perfection?
Dr. Marc Brittain, a research scientist at the Massachusetts Institute of Technology (MIT)’s Lincoln Laboratory specializing in artificial intelligence (AI) and machine learning (M/L), recently wondered the same thing. How can industry, academia, or governments predict how eVTOLs and other emerging advanced air mobility vehicles will perform without a means to do so?
He and his colleagues on the Technical Staff at the university – Luis E. Alvarez, Kara Breeden, and Ian Jessen – developed a standardized simulation testbed to address the knowledge gap.
The result is the AAM-Gym, which provides an ecosystem where people can develop and validate their algorithms across a broad spectrum of AAM use cases.
Specifically, the scientists studied two reinforcement learning algorithms that will demonstrate how separation can be assured in AAM corridors. As long as the AI algorithm follows the OpenAI gym protocol, users will be able to simulate potential AAM scenarios.
The Simulations Numbers Game
The researchers spent more than 100 training hours using the AAM-Gym tool to simulate more than 340,000,000 possible aircraft states. The tool provides users a collaborative means to reduce the time needed to develop better airspace metric analyses. These analyses can identify gaps or discover potential barriers that might hinder aircraft certification by regulatory agencies worldwide.