Pipeline

Features

  • Generalised scenario generation from DeepScenario drone data
  • Map encoded information
  • Comparable/Combinable with existing prediction and planning algorithms
  • Comparable with other SOTA algorithms

How to

Scenario Generator

If Preprocessed Data is available for the version + location you can directly go to TrajData Implementation

  1. Download the DeepUrban ScenarioPreprocessor
    • Construct the file system as suggested in the repository
  2. Download the scenario split of the location
    • Splits should be placed int DeepUrban/split to be used for ScenarioCreator and TrajData
  3. Choose one of the supported locations datasets
    • This step will redirect you to corresponding Location data from DeepScenario
    • Sign Up or Sign In on the DeepScenario Website
      • Currently supported Versions: Data V1, V2 and Interface V0.8, V0.9)
  4. Set source and output folder in DeepUrban/deepurban_scenariocreator/config/default.yaml
  5. DeepUrban/deepurban_scenariocreator/src/scenario_preprocessor.py to be used to build Scenarios
  6. Scenarios will be saved to DeepUrban/deepurban_scenarios/<location>

TrajData Implementation

  1. Download of corresponding lanelet2 map of the location
    • Maps should be placed in DeepUrban/maps to be loaded into TrajData
  2. Download the modified TrajData Dataloader
    • An example for the usage of DeepUrban Scenarios has been added deepurban_trajdata/examples/deepurban_example.py
    • Source directory is the DeepUrban/deepurban_scenarios folder as mentioned in Scenario Generator
  3. Further details can be found in our extension of the trajdata dataloader DATSETS.md.

Supported Locations

Version – Data V1 – split

Version – Data V1 – Preprocessed Data

coming soon

Version – Data V2 – split

Further Locations Coming Soon

Updates

2024-09-24

Possible version miss allignments will be tackled soon

Citation

If you use part of this pipeline, please cite it as follows:

@Inproceedings{selzer2024deepurban,
  author = {Selzer, Constantin and Flohr, Fabian},
  title = {{DeepUrban}: Interaction-aware Trajectory Prediction and Planning for Automated Driving by Aerial Imagery },
  booktitle = {{IEEE International Conference on Intelligent Transportation Systems (ITSC)}},
  month = sept,
  year = {2024},
  address = {Edmonton, Canada},
}

Commercial Use

The DeepUrban dataset is free for non-commercial use only.