Airport surface operations are a safety-critical component of the transportation industry. Although airports are nominally centrally deconflicted by an air traffic controller, stochastic pilot behavior and planning errors still result in dangerous outcomes such as runway incursions and loss of separation. Intention prediction and motion forecasting for taxiing aircraft is thus essential for robust and safe airport operations. In this work we propose a ground monitoring and trajectory forecasting system built with state-of-the-art normalizing flows. Normalizing flows are powerful generative models that provide both density estimation and sampling capabilities, allowing a runway monitoring agent to perform trajectory likelihood evaluation and future trajectory rollouts. Furthermore, we use lightweight liquid neural networks to learn context embeddings for observed trajectory data and parameterize the flows. We train our models on weeks of historical ADS-B data and evaluate our approach on real-world ground movements at the Seattle-Tacoma International Airport.
Runway Ground Movement Monitoring with Normalizing Flows
24.09.2024
2194532 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch