ASCENT: Transformer-Based Aircraft Trajectory Prediction in Non-Towered Terminal Airspace
Alexander Prutsch, David Schinagl, Horst Possegger
AI summary
Problem
General Aviation flights in non-towered terminal airspace face significantly higher accident rates due to the absence of air traffic control, creating an urgent need for reliable trajectory prediction. Existing deep learning models struggle to capture 3D motion dynamics and generate diverse, multi-modal maneuver hypotheses efficiently.
Approach
ASCENT normalizes historical 3D flight paths using positional and angular alignment to preserve local motion patterns while retaining global context. It then encodes these dynamics with a transformer and decodes future trajectories using learnable mode queries and kinematic parameters (speed, yaw, pitch) for efficient multi-modal forecasting.
Key results
- State-of-the-art accuracy on the TrajAir dataset across multiple evaluation splits
- First trajectory prediction benchmark on the TartanAviation dataset with cross-dataset evaluation
- Superior multi-modal trajectory diversity compared to CVAE and diffusion-based baselines
- Ablation studies confirm the effectiveness of angular normalization, mode queries, and kinematic decoding
Why it matters
Provides a critical safety tool for General Aviation by enabling real-time conflict detection and traffic flow management in non-towered airspace.
Abstract
Accurate trajectory prediction can improve Gen- eral Aviation safety in non-towered terminal airspace, where high traffic density increases accident risk. We present ASCENT, a lightweight transformer-based model for multi- modal 3D aircraft trajectory forecasting, which integrates domain-aware 3D coordinate normalization and parameterized predictions. ASCENT employs a transformer-based motion encoder and a query-based decoder, enabling the generation of diverse maneuver hypotheses with low latency. Experiments on the TrajAir and TartanAviation datasets demonstrate that our model outperforms prior baselines, as the encoder effectively captures motion dynamics and the decoder aligns with struc- tured aircraft traffic patterns. Furthermore, ablation studies confirm the contributions of the decoder design, coordinate- frame modeling, and parameterized outputs. These results es- tablish ASCENT as an effective approach for real-time aircraft trajectory prediction in non-towered terminal airspace.