Near misses at busy airports have become a growing concern in aviation safety, prompting researchers to explore new ways to predict dangerous situations before they escalate into disasters. A team from Carnegie Mellon University’s Robotics Institute has developed an artificial intelligence system called World2Rules, designed to detect and explain potential airport collision risks using real-world traffic data and historical safety violations, tells Tech Xplore.
The project was inspired by incidents such as a close call at New York’s John F. Kennedy International Airport, where an Air Canada aircraft nearly crossed an active runway in front of a landing EVA Air jet. Although a disaster was avoided, the event highlighted how quickly runway incursions can become catastrophic. The researchers wanted to create a system capable not only of recognizing dangerous aircraft behavior but also of predicting possible collisions early enough to provide pilots and controllers with valuable reaction time.
To train the system, the team developed the Amelia-42 dataset, which contains two years of Federal Aviation Administration airport surface movement data collected from 42 U.S. airports. The dataset includes aircraft and vehicle movements across taxiways and runways, along with information from crash and incident reports. Using the Bridges-2 supercomputer at the Pittsburgh Supercomputing Center, researchers processed the massive volume of data to teach the AI to distinguish between normal operations and unsafe behavior.
World2Rules combines two forms of AI. Neural methods identify patterns hidden within complex airport activity, while symbolic reasoning converts those patterns into readable safety rules. This allows the system not only to flag risks but also to explain which safety rule has been violated and why the situation is dangerous.
Researchers believe the technology could eventually expand beyond aviation into other safety-critical environments where machines and humans must coordinate reliably under pressure.