This post is also available in:
עברית (Hebrew)
Airports operate in highly complex environments where even small communication or coordination errors can escalate into serious safety risks. Runway incursions, misinterpreted instructions, and unexpected aircraft movements remain difficult to manage in real time, especially as air traffic volumes continue to grow. Existing systems can detect conflicts as they occur, but predicting dangerous situations early enough to prevent them remains a major challenge.
Researchers have now developed an AI-based system, called World2Rules, designed to identify potential collision scenarios before they unfold. The platform analyzes airport surface movement data together with historical incident records to forecast risky interactions between aircraft and ground vehicles. Rather than simply reacting to events already in progress, the system is intended to provide controllers and crews with additional time to intervene.
According to TechXplore, the approach combines two different types of artificial intelligence. Traditional neural AI models are effective at recognizing patterns in large and complex datasets, but they often operate as “black boxes”, making their decisions difficult to interpret. Symbolic AI systems, by contrast, generate human-readable logic but struggle with noisy or incomplete data. The new system merges both approaches into a neuro-symbolic framework that can both detect patterns and explain why a potential risk has been identified.
To train the system, researchers used a large aviation dataset containing two years of airport surface movement information collected across dozens of airports. The AI learned from both routine operations and historical accident or incident data, allowing it to distinguish between normal traffic flow and potentially dangerous behavior. According to the researchers, the system outperformed purely neural and purely symbolic AI models in identifying collision risks.
One of the more notable features is its ability to filter unreliable information and discard misleading outputs, reducing false assessments that can overwhelm operators. The system also works alongside trajectory-forecasting tools that predict aircraft movement patterns in advance.
From a defense and security perspective, predictive traffic analysis has applications beyond civilian aviation. Similar approaches could support military air operations, autonomous vehicle coordination, or any environment where multiple moving assets must operate safely in shared spaces under time pressure.
The researchers say future versions may incorporate more dynamic time-based modeling, allowing the system to better account for uncertainty and continuously evolving traffic conditions.
The research was published here.
