By Yunzhu Li, Jiaming Song, Stefano Ermon
Modified codebase of TORCS, with the ability to extract dashboard views.
InfoGAIL implementation, attached with two examples: pass & turn.
If you find this codebase useful in your research, please consider citing:
@article{li2017inferring,
title={InfoGAIL: Interpretable Imitation Learning from Visual Demonstrations},
author={Li, Yunzhu and Song, Jiaming and Ermon, Stefano},
journal={arXiv preprint arXiv:1703.08840},
year={2017}
}
In the following section, we only show the commands for running example 0 (indicating pass). Users can replace 0 with 1 to run example 1 (indicating turn).
sudo apt-get install libplib-dev
sudo apt-get install libopenal-dev
sudo apt-get install libalut-dev
See http://torcs.sourceforge.net/ for more information
torcs-1.3.4/
folder, type the following commands to install TORCS:./configure
make
sudo make install
sudo make datainstall
Default installation directories are:
/usr/local/bin
/usr/local/lib/torcs
/usr/local/share/games/torcs
Run the torcs
command to play TORCS.
modified_tracks/
folder to the torcs folder /usr/local/share/games/torcs/tracks/road
rm -rf ~/.torcs
cp -r torcs_config_ai_0 ~/.torcs
Training data
Pretrained weights
cd to wgail_info_0/
open drive.py
and edit line 20-21 in concert with downloaded data and weights
change variable code
in line 14 into different values (0
or 1
) to observe different behaviors
type the python drive.py
to run pass with pretrained weights
Run pass with different latent codes (0
or 1
):
Run turn with different latent codes (0
or 1
):
wgail_info_0/
wgail_info.py
and edit line 17-19 in concert with downloaded data and weightsmodels.py
and edit line 508 to specify the place to store the weights, edit line 422 to specify the place to store the log filepython wgail_info.py
to train on pass, note that the weights trained via behavior cloning are used to initialize the policy networkTrack selection:
torcs-1.3.4 in this repository is a modified version of the original codebase released by Bernhard Wymann. This version holds the ability of extracting and transmitting visual information.
Using Keras and Deep Deterministic Policy Gradient to play TORCS
TORCS - The Open Race Car Simulator
Simulated Car Racing Championship Competition Software Manual