Reproduce “Synthetic Aircraft Trajectory Generation Using Time-Based VQ-VAE” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Abdulmajid Murad, Massimiliano Ruocco
Researchers from SINTEF Digital and NTNU developed an adaptation of Time-Based VQ-VAE for synthetic aircraft trajectory generation, producing high-fidelity full-flight paths. The model achieved a Fréchet Inception Distance of 0.0029, outperforming a baseline, and demonstrated operational flyability in simulations while accurately replicating real-world flight statistics and structures.
Key insights
- Processing flight data in the time-frequency domain is effective for simultaneously capturing both slow-varying global trends and rapid, fine-grained local dynamics within aircraft trajectories.
- The combination of vector quantization and transformer-based priors allows for the generation of diverse, spatiotemporally coherent full-flight trajectories by modeling intricate long-range dependencies in a discrete latent space.
- A multi-stage training procedure, separating VQ-VAE, prior, and optional fidelity enhancer training, is effective for optimizing different aspects of synthetic trajectory generation, from reconstruction to global coherence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.