/3DCV

3DCV Project Repository

Primary LanguageTeX

3DCV project on critical driving scenarios

Christopher Klammt
Tobias Richstein
Julian Seibel
Karl Thyssen

Link to the report

Scenario generation

To start the scenario generation, make sure to install all requirements using the environment.yml i.e. with conda:

conda env create -f environment.yml

Apart from that the CARLA Python package needs to be installed as described in the CARLA documentation.

(This step is not needed if you only want to use the basic scenarios provided with CARLA v0.9.9) Afterwards, make sure to place all basic scenarios in the ./scenario_generation/basic_scenarios/ folder as xosc OpenSCENARIO configuration files.

Start the process by executing the following command from the root folder:

python ./scenario_generation/openSCENARIO_generator.py [--log] [--number_scenarios] [--save_path]

Explanation of the arguments:

  • logs : Flag, whether logging should be activated. The generator will save one csv file per basic scenario in ./logs/ that includes all new values set for newly generated scenarios
  • number_scenarios : The number of scenarios which should be created from one basic scenario (default is 1)
  • save_path : Folder where to save the generated scenarios (default is ./generated_scenarios/)

Note : To start the Scenario-Generator, make sure you have a running CARLA v0.9.9 instance. The scripts will use the Python-API of the CARLA Server.

Below are some examples of newly generated scenarios compared to the original scenario.

Cyclist scenario

Original cyclist scenario

Above is the original blueprint for the cyclist scenario. It features a hero car and a bike that is about to cross the street forcing the car to brake.

Modified cyclist scenario in same town This is a modified scenario of the cyclist crossing. The bike has been replaced with a car and the situation occurs at a different intersection but within the same map.

Modified cyclist scenario in different town

For this scenario the town has changed but the parameters of the situation remain the same.

Lane Changing Scenario

Original lane changing scenario

This scenario describes the hero car following a car driving in front that abruptly changes its lane to avoid a still standing vehicle.

Modified lane changing scenario in same town

This generated scenario takes place in rougher weather and also features the highway going in the other direction than in the original and also using a different lane.

Modified lane changing scenario in a different town

Above we can see a generated lane changing scenario taking place in a different town from the original scenario.

Data generation

To collect data for some generated scenarios, you have to replace the manual_control.py-file of the Scenario-Runner with the one provided in the data_generation folder. Then (with CARLA running) you just need to execute the run.cmd-file (currently only for Windows). This will:

  • generate the specified amount of scenarios per base scenario (currently defined with 5 in the run.cmd)
  • for each generated scenario, run it via the Scenario-Runner
  • and start the manual_control.py in autopilot mode to gather the data with the expert driving

You can find the collected data in form of image-files and measurements.json (expert control input) inside the _out folder in the Scenario-Runner.