Eventual world domination by self-learned self driving cars.
Installation : Needs: pygame ( Not for training) curses ( Optional, for colorfull terminal debuging ) tensorflow 1.1 or tensorflow 1.2 numpy hdf5 imgaug carla_client
########## RUNNING ############
the toolbox has three different kinds of runs:
drive -> drive a network using one of our amazing interfaces ( carla or GTA) train -> train a network
evaluate -> Run the network a bit more and check the outputs.
########### TRAIN MODE ############
The first step needed is to make a configuration file. There are plenty of configuration files provided at the "config" folder. Just copy a24 file and change to your needs On this configuration file you shoul d ( MOVE THIS TO The chauffeur please....)s
x
To run the train mode, first increase the number of posible open files:
ulimit -Sn 8096
Then run the program for training
python2 chauffer.py train -e <experiment_name>
where the experiment name should be a configuration file.
########## DRIVE MODE ###########
python2 chauffeur.py drive --driver Machine -l 158.109.9.226 -p 2000 -pt /data2/datasets/ -sc -e a24_ca20_boostpedals_loss -cc drive_interfaces/carla_interface/CarlaSettings.ini -cy carla_0
run corl tests:
CUDA_VISIBLE_DEVICES='0' python2 drive_interfaces/carla_interface/run_test_corl.py -e a24_ca20_boostpedals_loss -s a24_test_2_weathers_2_pos.csv -l 158.109.9.226 -p 2000 -t 0,1 -cy carla_0 -w 1,3