Artifacts for the paper "Concretization of Abstract Traffic Scene Specifications Using Metaheuristic Search"

This deposit contains measurement data and additional artifacts pertaining to the Concretization of Abstract Traffic Scene Specifications Using Metaheuristic Search" paper.

Specifically, the artifacts are divided into 6 directories:

0-config/ contains configuration files use to generate input scene specifications (using Scenic) with various number of actors.

1-data/ contains the generated input scene specifications for each map and actor size. It also contains additional scene specifications used for RQ3 (in the `zalaFullcrop/` subfolder).

2-results/ contains measurement results (i.e. (1) visualizations of generated scenes and (2) corresponding scene specifications containing exact positions of each vehicle, as well as (3) measurement data) for each research question.

3-figures/ contains figures derived from the contents of `results/` (including additional figures not included in the publication).

4-TestingSemanticSegmentation/ contains artifacts related to the integration of the CARLA simulator and of three computer-vision components:

  0-sceneConfig/ contains configuration files used to generate scenes.

  1-scenesWithExactCoordinates/ contains (1) visualizations of generated scenes and (2) corresponding Scenic-compatible scene specifications containing exact positions of each vehicle.

  2-staticImages/ contains dashcam images for generated scenes and corresponding ground truth semantic segmentation, obtained through the CARLA integration.

  3-predictions/ contains predicted semantic segmentation results (and their overlay on top of the corresponding dashcam image) obtained using three computer-vision components.

  4-videos/ contains videos of generated scenes running with the default AV stack included in CARLA.

  5-metrics contains initial measurement data for the predicted semantic segmentation.

5-simulationVideos/ contains videos of two scenario-based test cases executed in simulation. The two scenarios are only differenciated by the initial scene, which impacts the outcome of test execution (collision vs. no-collision).

All the code of the proposed MHS-based scene concretization approach is implemented as an extension to the Scenic tool, which is available under the 3-Clause BSD License.