From Voice to Safety: Language AI Powered Pilot-ATC Communication Understanding for Airport Surface Movement Collision Risk Assessment

TL;DR

This work addresses airport surface safety gaps by integrating language AI with surface risk modeling to augment ASSC capabilities. It introduces a hybrid ATC Rule-Enhanced NER to extract callsigns, aircraft states, and destination intents from pilot–ATC transcripts and couples this with a node-link airport graph using log-normal link travel times to compute spatiotemporal collision probabilities. Real-time risk assessment is demonstrated on three case studies (Haneda, KATL, Tenerife), with the Fenton–Wilkinson (FW) approach typically providing earlier warnings than Petri-Net (PN) formulations, highlighting practical gains in lead time for mitigation. Limitations include reliance on domain-specific embeddings, scope limited to pairwise conflicts, and the need for extending to multi-aircraft interactions and dynamic contextual factors such as weather and congestion.

Abstract

This work provides a feasible solution to the existing airport surface safety monitoring capabilities (i.e., Airport Surface Surveillance Capability (ASSC)), namely language AI-based voice communication understanding for collision risk assessment. The proposed framework consists of two major parts, (a) rule-enhanced Named Entity Recognition (NER); (b) surface collision risk modeling. NER module generates information tables by processing voice communication transcripts, which serve as references for producing potential taxi plans and calculating the surface movement collision risk. We first collect and annotate our dataset based on open-sourced video recordings and safety investigation reports. Additionally, we refer to FAA Order JO 7110.65W and FAA Order JO 7340.2N to get the list of heuristic rules and phase contractions of communication between the pilot and the Air Traffic Controller (ATCo). Then, we propose the novel ATC Rule-Enhanced NER method, which integrates the heuristic rules into the model training and inference stages, resulting in a hybrid rule-based NER model. We show the effectiveness of this hybrid approach by comparing different setups with different token-level embedding models. For the risk modeling, we adopt the node-link airport layout graph from NASA FACET and model the aircraft taxi speed at each link as a log-normal distribution and derive the total taxi time distribution. Then, we propose a spatiotemporal formulation of the risk probability of two aircraft moving across potential collision nodes during ground movement. Furthermore, we propose the real-time implementation of such a method to obtain the lead time, with a comparison with a Petri-Net based method.

Figures (19)

• Figure 1: Accidents and Incidents in the continental U.S. from April 2024 to February 2025, documented by the FAA FAA_Accidents_Incidents. In \ref{['fig: accident-map']}, blue dots indicate commercial aviation events, and black dots indicate general aviation events.
• Figure 2: The proposed workflow of ATC communication transcript understanding and surface movement risk assessment.
• Figure 3: The distribution of entity types of the dataset used in the ATC communication transcript information retrieval work.
• Figure 4: Three examples of entity rules of flight callsigns. Based on FAA Order JO 7340.2N, speed bird is the nickname of British Airlines FAA_7340.2N_2024.
• Figure 5: Performance comparison between different embedding models under various setups. The last setup clearly gives better overall performance for any contextual embeddings.

Theorems & Definitions (3)

• Lemma 1
• Lemma 2
• Lemma 3