Reinforcement Learning in CARLA

We release a trained RL agent from the CoRL-2017 paper "CARLA: An Open Urban Driving Simulator". This is only the inference code, the training code is not released yet.

The agent was trained with the asynchronous advantage actor-critic (A3C) algorithm by V. Mnih et al. (2016). We build on this open-source Chainer implementation: https://github.com/muupan/async-rl .

Dependencies

Tested with:

• CARLA 0.8.2
• python 3.6
• chainer 1.24.0
• cached-property 1.4.2
• PIL 5.1.0
• opencv 3.3.1
• h5py 2.7.1

In Anaconda, you can create and activate an environment with installed dependencies (except for CARLA) by running:

conda create -n carla_rl python=3.6 chainer=1.24.0 cached-property=1.4.2 pillow=5.1.0 opencv=3.3.1 h5py=2.7.1
source activate carla_rl

To start evaluation on the CoRL-2017 benchmark:

• Start a CARLA server on town TownXX (Town01 or Town02) and port PORT (this is to be executed in the CARLA server folder):

./CarlaUE4.sh /Game/Maps/TownXX -carla-server -benchmark -fps=10 -windowed -ResX=800 -ResY=600 -carla-world-port=PORT

• Make sure CARLA client is in your python path, e.g. by running:

export PYTHONPATH=/path/to/CARLA/PythonClient:$PYTHONPATH

• Run the evaluation:

python run_RL.py --city-name TownXX --port PORT --corl-2017

The results will be stored in _benchmarks_results.

Paper

If you use this code in your research, please cite our CoRL 2017 paper:

@inproceedings{Dosovitskiy17,
  title = { {CARLA}: {An} Open Urban Driving Simulator},
  author = {Alexey Dosovitskiy and German Ros and Felipe Codevilla and Antonio Lopez and Vladlen Koltun},
  booktitle = {Proceedings of the 1st Annual Conference on Robot Learning},
  pages = {1--16},
  year = {2017}
}