Search papers, labs, and topics across Lattice.
This paper introduces ASCENT, a transformer-based model for predicting aircraft trajectories in non-towered terminal airspace, addressing safety concerns in high-traffic areas. ASCENT uses a transformer encoder to capture motion dynamics and a query-based decoder to generate diverse maneuver hypotheses. Experiments on TrajAir and TartanAviation datasets show ASCENT outperforms existing baselines, demonstrating the effectiveness of its decoder design, coordinate-frame modeling, and parameterized outputs.
A lightweight transformer can accurately forecast diverse aircraft trajectories in complex airspace, outperforming prior methods and enabling real-time safety applications.
Accurate trajectory prediction can improve General 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 structured aircraft traffic patterns. Furthermore, ablation studies confirm the contributions of the decoder design, coordinate-frame modeling, and parameterized outputs. These results establish ASCENT as an effective approach for real-time aircraft trajectory prediction in non-towered terminal airspace.
