Scenario-based safety validation is a new approach to handle the increased testing complexity of highly automated vehicles. After categorizing traffic into relevant driving scenarios, every scenario element needs to be modeled in order to be used for testing. While driving maneuvers are typically modeled by experts, data-driven generative models allow to automate the modeling process. We propose an improved unsupervised modeling approach called BézierVAE, which is based on our previously published TraVAE framework. TraVAE is extended by a Bézier-curve output layer, which ensures the trajectories’ smoothness in both the position and speed domain. The results are evaluated on lane changes and acceleration/deceleration trajectories extracted from the highD dataset. In comparison to our previous findings, BézierVAE learns a larger set of intuitive model parameters through unsupervised learning, hence enabling a more detailed modeling process. At the same time, the reconstruction errors are reduced by up to 91.3% and the unsmoothness decreased by 83.4% compared to TraVAE. Partially specifying parameters in a semi-supervised fashion maintains the ability to learn disentangled parameters while allowing the parameter learning process to be guided. Finally, we demonstrate the modeling of multiple maneuvers through a single model by introducing discrete parameters into the latent representation.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    BézierVAE: Improved Trajectory Modeling using Variational Autoencoders for the Safety Validation of Highly Automated Vehicles


    Contributors:


    Publication date :

    2019-10-01


    Size :

    3136719 byte




    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English





    Frugal Incremental Generative Modeling using Variational Autoencoders

    Enescu, Victor / Sahbi, Hichem | ArXiv | 2025

    Free access

    Deep Tracking Portfolios Using Autoencoders and Variational Autoencoders

    Urrego, Daniel Aragón / Nieto, Oscar Eduardo Reyes / Quimbayo, Carlos Andrés Zapata | Springer Verlag | 2024


    Variational Autoencoders

    Ghojogh, Benyamin / Crowley, Mark / Karray, Fakhri et al. | Springer Verlag | 2022