🌟 We updated the repo with training code! 🌟

Conditional Affordance Learning

In CoRL 2018 [Paper] [Video] [Talk]

Axel Sauer ^{1, 2}, Nikolay Savinov ¹, Andreas Geiger ^{1, 3
1} ETH Zurich, ² TU Munich, ³ MPI for Intelligent Systems and University of Tubingen

If you use this implementation, please cite our CoRL 2018 paper.

@inproceedings{Sauer2018CORL,
  author={Sauer, Axel and Savinov, Nikolay and Geiger, Andreas},
  title={Conditional Affordance Learning for Driving in Urban Environments},
  booktitle={Conference on Robot Learning (CoRL)},
  year={2018}
}

Installation

# install anaconda if you don't have it yet
wget https://repo.continuum.io/archive/Anaconda3-5.3.0-Linux-x86_64.sh
bash Anaconda3-5.3.0-Linux-x86_64.sh
source ~/.profile
# or use source ~/.bashrc - depending on where anaconda was added to PATH as the result of the installation

Now we will:

1. create a conda environment named CAL and install all dependencies
2. download the binaries for CARLA version 0.8.2 [CARLA releases]
3. download the model weights

git clone https://github.com/xl-sr/CAL.git
cd CAL

# create the training environment
conda env create -f requirements.yml
source activate CAL

# run download script
./download_binaries_and_models.sh

Run the Agent

In CARLA_0.8.2/ start the server with for example: (see the CARLA documentation for details)

./CarlaUE4.sh Town01 -carla-server -windowed -benchmark -fps=20 -ResX=800 - ResY=600

Open a second terminal, and run:

python python_client/driving_benchmark.py -c Town01 -v -n test --corl-2017

If you want to run the CORL 2017 benchmark add '--corl-2017'

python python_client/driving_benchmark.py -c Town01 -v -n test --corl-2017

If you want to continue an experiment, you can add the 'continue-experiment' flag.

Training

cd training/

# download and untar the dataset
wget https://s3.eu-central-1.amazonaws.com/avg-projects/conditional_affordance_learning/dataset.tar.gz
tar -xzvf dataset.tar.gz

Now, open training_CAL.ipynb. The notebook walks you through the steps to train a network on the dataset.