NGSIM freeway | openDD roundabout |
---|---|
![]() |
![]() |
We've also trained and benchmarked policies on lane change maneuvers and roundabout navigation tasks.
More details, our article and videos of trained policies are published on our website.
git clone https://github.com/deepsense-ai/carla-real-traffic-scenarios.git && cd carla-real-traffic-scenarios
pip install -r requirements.txt
If working on remote servers with no desktop, use gdown to download from Google Drive links, e.g.
pip install gdown
gdown --id 1FCHL7YJk12AwfxuMPmwXPJj71n3mwSxE
Download and extract CARLA (0.9.9.4 download link). Then, add PythonAPI wheel to your PYTHONPATH
:
cd ~/Downloads
wget https://carla-releases.s3.eu-west-3.amazonaws.com/Linux/CARLA_0.9.9.4.tar.gz
export CARLA_ROOT=/home/$USER/Downloads/CARLA_0.9.9.4
mkdir -p $CARLA_ROOT;
tar zxvf CARLA_0.9.9.4.tar.gz --directory $CARLA_ROOT
export PYTHONPATH=$CARLA_ROOT/PythonAPI/carla/dist/carla-0.9.9-py3.7-linux-x86_64.egg:$PYTHONPATH
Download and import our CARLA package with new maps: 7 roundabout, 2 freeways: Google Drive download link
mv ~/Downloads/carla-package-NGSIM-openDD.tar.gz $CARLA_ROOT/Import/
$CARLA_ROOT/ImportAssets.sh
Download and extract: NGSIM, opendDD
sudo apt install p7zip-full
# no space after -o
7z x ~/Downloads/openDD.7z -oopenDD
mkdir NGSIM && tar zxvf ~/Downloads/NGSIM.tgz --strip-components 1 --directory NGSIM
cd $CARLA_ROOT
./CarlaUE4.sh -benchmark -fps=10
Wait until server boots up. Feel free to play with the code.
# Directory which contains "rdb1to7.sqlite" and "image_georeferenced/"
export OPENDD_DIR=~/Downloads/openDD
# Directory which contains "i80/" and "us101/"
export NGSIM_DIR=~/Downloads/NGSIM
python examples/runnable_template.py --dataset opendd --num-episodes 5
We encourage you send us any kind of feedback on what should be improved, what's not working etc.
Code for interfacing with NGSIM dataset was based on https://github.com/Atcold/pytorch-PPUU
Authors (cannot be disclosed yet):