Publications

  1. PlanTRansformer: Unified Prediction and Planning with Goal-conditioned Transformer
    PlanTRansformer: Unified Prediction and Planning with Goal-conditioned Transformer
    Selzer, C., & Flohr, F. B. (2026, June). . IEEE Intelligent Vehicles Symposium (IV).
    IEEE Intelligent Vehicles Symposium (IV), 2026
    arXiv DOI 
    @inproceedings{selzer2026ptr,
  author = {Selzer, C. and Flohr, F. B.},
  title = {{PlanTRansformer}: Unified Prediction and Planning with Goal-conditioned Transformer},
  booktitle = {IEEE Intelligent Vehicles Symposium (IV)},
  year = {2026},
  month = jun,
  doi = {10.48550/arXiv.2602.03376},
  arxiv = {2602.03376},
  thumb = {ptr.svg}
}

    Trajectory prediction and planning in autonomous driving operate as fundamentally disconnected systems. Prediction models must forecast the movements of all surrounding agents with multimodal distributions, despite unknown intentions. Planning, in contrast, requires known objectives to generate deterministic paths. We introduce PTR, a unified Gaussian Mixture Transformer framework integrating goal-conditioned prediction, dynamic feasibility, interaction awareness, and lane-level topology reasoning. Our approach uses progressive masking of agent commands during training to match real-world inference conditions. PTR achieves 4.3%/3.5% improvement in marginal/joint mAP compared to the baseline Motion Transformer (MTR) and 15.5% planning error reduction at 5s horizon compared to GameFormer.
  2. DeepUrban: Interaction-aware Trajectory Prediction and Planning for Automated Driving by Aerial Imagery
    DeepUrban: Interaction-aware Trajectory Prediction and Planning for Automated Driving by Aerial Imagery
    Selzer, C., & Flohr, F. B. (2024). . 27th IEEE International Conference on Intelligent Transportation Systems (ITSC), 221–227.
    27th IEEE International Conference on Intelligent Transportation Systems (ITSC), 2024
    arXiv DOI 
    @inproceedings{selzer2024deepurban,
  author = {Selzer, C. and Flohr, F. B.},
  title = {{DeepUrban}: Interaction-aware Trajectory Prediction and Planning for Automated Driving by Aerial Imagery},
  booktitle = {27th IEEE International Conference on Intelligent Transportation Systems (ITSC)},
  address = {Edmonton, AB, Canada},
  year = {2024},
  month = sep,
  pages = {221--227},
  doi = {10.1109/ITSC55196.2024.10919855},
  arxiv = {2601.10554},
  thumb = {deepurban.svg}
}

    We address the gap in autonomous driving benchmarks by introducing DeepUrban, a drone-based dataset focusing on dense urban traffic scenarios. The dataset comprises 3D traffic objects extracted from high-resolution images captured over urban intersections at approximately 100 meters altitude, along with map and scene data. Evaluation of state-of-the-art methods demonstrates significant improvements when combining DeepUrban with nuScenes, yielding gains up to 44.1% / 44.3% on the ADE / FDE metrics.