A pytorch implementation to train the conditional imitation learning policy in "End-to-end Driving via Conditional Imitation Learning" and "CARLA: An Open Urban Driving Simulator".
python 3.6
pytorch > 0.4.0
tensorboardX
numpy
torchvision
cuda
h5py
PIL
train-dir and eval-dir should point to where the Carla dataset located. Please put the train dataset and eval dataset in two subfolders.
$ python main.py --batch-size
150
--workers
4
--train-dir
"data/SeqTrain/"
--eval-dir
"data/SeqVal/"
--id
training
--gpu
0
Have the Carla simulator running on server mode. You can run with $CarlaUE4.exe -carla-server -windowed -benchmark-fps=20
$python run_cil.py