adversarial_carla_env

Adversarial CARLA Gym Environment

Based on ScenarioRunner, modified for adaptive stress testing (AST).

⚠ WORK IN PROGRESS ⚠

Example Experiment

using AVExperiments

config = ExperimentConfig(
    seed=0xC0FFEE,
    agent=WorldOnRails,
    N=100,
    dir="results_wor",
    use_tree_is=true,
    leaf_noise=true,
    resume=false,
)

results = run_carla_experiment(config)

Install Instructions

CARLA

• Download CARLA 0.9.13: https://github.com/carla-simulator/carla/releases/tag/0.9.13
  • On Linux:

  wget https://carla-releases.s3.eu-west-3.amazonaws.com/Linux/CARLA_0.9.13.tar.gz
  mkdir CARLA_0.9.13
  tar -xvzf CARLA_0.9.13.tar.gz -C CARLA_0.9.13
  

Scenario Runner

• Download scenario_runner 0.9.13

  NOTE: Make sure scenario_runner is placed in the <CARLA_ROOT>/PythonAPI/ directory.

  git clone https://github.com/carla-simulator/scenario_runner
  cd scenario_runner
  git checkout v0.9.13
  

  • Install scenario_runner Python packages:

    pip install -r requirements.txt
    

    If you get errors, try running this beforehand: pip install --upgrade pip

  • Apply patch to scenario_runner:

    git apply ../../../adversarial_carla_env/adv_scenario_runner.patch
    

Adversarial CARLA Gym Environment

1. Install adversarial_carla_env Python packages:

   cd <PATH_TO>/adversarial_carla_env
   pip install -r requirements.txt
   

   • Install adv-carla-v0 gym environment (from the same directory):

   pip install -e .
   

2. Set up environment variables

   • For Windows, either do this every time:

   :: %CARLA_ROOT% is the CARLA installation directory
   :: %SCENARIO_RUNNER% is the ScenarioRunner installation directory
   :: In a build from source, the .egg files may be in: ${CARLA_ROOT}/PythonAPI/dist/ instead of ${CARLA_ROOT}/PythonAPI
   
   :: <<<<CHANGE THIS TO YOUR CARLA_ROOT, EVERYTHING ELSE IS UNCHANGED!>>>>
   set CARLA_ROOT=<YOUR_PATH_TO>\CARLA_0.9.13
   
   set SCENARIO_RUNNER_ROOT=%CARLA_ROOT%\PythonAPI\scenario_runner
   set PYTHONPATH=%PYTHONPATH%;%CARLA_ROOT%\PythonAPI\carla\dist\carla-0.9.13-py3.7-win-amd64.egg
   set PYTHONPATH=%PYTHONPATH%;%CARLA_ROOT%\PythonAPI\carla\agents
   set PYTHONPATH=%PYTHONPATH%;%CARLA_ROOT%\PythonAPI\carla
   set PYTHONPATH=%PYTHONPATH%;%CARLA_ROOT%\PythonAPI
   

   • (Windows) Or, add these to your permanent environment variables:

     • CARLA_ROOT

       • <YOUR_PATH_TO>\CARLA_0.9.13

     • SCENARIO_RUNNER_ROOT

       • %CARLA_ROOT%\PythonAPI\scenario_runner

     • PYTHONPATH (append this):

       ;%CARLA_ROOT%\PythonAPI\carla\dist\carla-0.9.13-py3.7-win-amd64.egg;%CARLA_ROOT%\PythonAPI\carla\agents;%CARLA_ROOT%\PythonAPI\carla;%CARLA_ROOT%\PythonAPI

   • For Linux:

   # <<<<CHANGE THIS TO YOUR CARLA_ROOT, EVERYTHING ELSE IS UNCHANGED!>>>>
   export CARLA_ROOT=/path/to/CARLA_0.9.13
   
   export SCENARIO_RUNNER_ROOT=$CARLA_ROOT/PythonAPI/scenario_runner
   export PYTHONPATH=$PYTHONPATH:$CARLA_ROOT/PythonAPI/carla/dist/carla-0.9.13-py3.7-linux-x86_64.egg
   export PYTHONPATH=$PYTHONPATH:$CARLA_ROOT/PythonAPI/carla/agents
   export PYTHONPATH=$PYTHONPATH:$CARLA_ROOT/PythonAPI/carla
   export PYTHONPATH=$PYTHONPATH:$CARLA_ROOT/PythonAPI

NEAT (Image-based Agent)

• Install fork of the NEAT agent:
  • (Clone it into the CARLAIntegration directory):
    • git clone https://github.com/sisl/neat
    • Download NEAT pretrained models (from their README):

    mkdir model_ckpt
    wget https://s3.eu-central-1.amazonaws.com/avg-projects/neat/models.zip -P model_ckpt
    unzip model_ckpt/models.zip -d model_ckpt/
    rm model_ckpt/models.zip
    

    • (DEBUG): If you get an error Torch not compiled with CUDA enabled then run the following:
      • pip install torch===1.5.0 torchvision===0.6.0 -f https://download.pytorch.org/whl/torch_stable.html

WorldOnRails (Image-based Agent)

• Install fork of the WorldOnRails agent:
  • (Clone it into the CARLAIntegration directory):
    • git clone https://github.com/sisl/WorldOnRails
    • Download the WorldOnRails pretrained model main_model_10.th and place it in the WorldOnRails/ directory.

Running Instructions

1. Start CARLA using the carla_start.sh (Unix) or carla_start.bat (Windows) scripts.
2. Execute stress testing and record the final trajectory (requires setting environment variables, see above).

   python test/run_episode.py
   

Note: you may need to specify the correct python verion:

sudo update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.7 1