Predicting Like A Pilot: Dataset and Method to Predict Socially-Aware Aircraft Trajectories in Non-Towered Terminal Airspace (2109.15158v2)

Abstract: Pilots operating aircraft in un-towered airspace rely on their situational awareness and prior knowledge to predict the future trajectories of other agents. These predictions are conditioned on the past trajectories of other agents, agent-agent social interactions and environmental context such as airport location and weather. This paper provides a dataset, $\textit{TrajAir}$, that captures this behaviour in a non-towered terminal airspace around a regional airport. We also present a baseline socially-aware trajectory prediction algorithm, $\textit{TrajAirNet}$, that uses the dataset to predict the trajectories of all agents. The dataset is collected for 111 days over 8 months and contains ADS-B transponder data along with the corresponding METAR weather data. The data is processed to be used as a benchmark with other publicly available social navigation datasets. To the best of authors' knowledge, this is the first 3D social aerial navigation dataset thus introducing social navigation for autonomous aviation. $\textit{TrajAirNet}$ combines state-of-the-art modules in social navigation to provide predictions in a static environment with a dynamic context. Both the $\textit{TrajAir}$ dataset and $\textit{TrajAirNet}$ prediction algorithm are open-source. The dataset, codebase, and video are available at https://theairlab.org/trajair/, https://github.com/castacks/trajairnet, and https://youtu.be/elAQXrxB2gw respectively.

• The paper introduces a novel open-source TrajAir dataset and TrajAirNet algorithm to enhance trajectory prediction in non-towered terminal airspace.
• The methodology integrates Temporal Convolutional Networks, Graph Attention Networks, and Conditional Variational Autoencoders to model spatial, dynamic, and social interactions.
• Experimental results show significant improvements in forecasting accuracy using ADE and FDE metrics compared to traditional constant velocity models.

Overview of "Predicting Like A Pilot" Study

The paper "Predicting Like A Pilot: Dataset and Method to Predict Socially-Aware Aircraft Trajectories in Non-Towered Terminal Airspace" addresses the complex challenge of predicting aircraft trajectories in non-towered terminal airspace and introduces both a novel dataset, TrajAir, and a predictive algorithm, TrajAirNet. The authors offer both as open-source resources to advance research in autonomous aviation, particularly in less controlled environments where pilot judgment significantly impacts flight safety and efficiency.

Contributions and Methodology

The primary contributions of this research include the TrajAir dataset, which provides comprehensive records of aircraft trajectories at a non-towered general aviation airport over a period of 111 days, and the TrajAirNet trajectory prediction algorithm tailored for this dataset. The dataset captures the intricacies of aircraft movement, interaction patterns, and integrates corresponding METAR weather data, marking one of the first attempts to address socially-aware navigation in a three-dimensional aerial context. TrajAirNet, the proposed predictive model, leverages recent advancements in deep learning such as Temporal Convolutional Networks, Graph Attention Networks, and Conditional Variational Autoencoders. It aims to predict future trajectories by modeling spatial positions, dynamic context like weather, and social interactions among various airspace agents.

Experimental Results

TrajAirNet's performance is gauged using the Average Displacement Error (ADE) and Final Displacement Error (FDE) metrics. Comparative evaluations against existing benchmarks—such as constant velocity models and trajectory prediction algorithms from the autonomous vehicle domain—demonstrate TrajAirNet's superior accuracy and reliability in predicting aircraft pathways over long horizons. These results underscore the effectiveness of utilizing explicit spatial contextualization and modeling dynamic environmental factors.

Theoretical and Practical Implications

The introduction of TrajAir and TrajAirNet represents a significant advancement in unsupervised aerial navigation, particularly in the context of growing UAV operations which require seamless integration with manned aircraft. The availability of such a dataset is crucial for developing predictive models that can be generalized across different environments, including the anticipated increase in traffic at non-towered airfields.

Theoretically, this paper enhances the understanding of multi-agent interactions in unregulated airspace and sets a foundation for more sophisticated social navigation models. The dynamic nature of such environments demands agile prediction algorithms fine-tuned to the behavioral nuances of human pilots as they adapt to ever-changing contexts and adhere to loosely defined procedural norms in terminal areas.

Future Developments

The authors acknowledge areas for future research, including the need to generalize this model for varied airport configurations and incorporate additional contextual inputs like real-time communication and visual data. The exploration of more advanced architectures, such as transformers, for sequence modeling might yield even greater prediction accuracy and adaptability.

In summary, this research lays a foundational step towards safe, efficient navigation of aerial vehicles in non-towered terminal airspaces, offering a valuable dataset and a pioneering algorithm for trajectory prediction. It opens avenues for future innovations in autonomous aviation while acknowledging the complexity and necessity of effectively integrating dynamic environmental and social interaction factors.