Super intelligent(?) ViZdoom agent SIST CS181 project

https://openreview.net/pdf?id=Hk3mPK5gg ICLR2017 FAIR paper 2016 A3C+Curriculum
https://arxiv.org/abs/1609.05521 Arnold CMU - DQRN + split network (Skill Learning?)

We need curriculum learning because the reward (e.g. kill an enemy) is too sparse at the training begining on complex tasks (deathmatch)
https://arxiv.org/pdf/1705.06366.pdf BAIR paper for generating learning goals using GAN
http://bair.berkeley.edu/blog/2017/12/20/reverse-curriculum/ Good blogs from BAIR
https://www.ijcai.org/proceedings/2017/0757.pdf IJCAI short paper

https://nlp.stanford.edu/pubs/liu2018reinforcement.pdf

https://github.com/glample/Arnold Pytorch + Arnold
https://github.com/mwydmuch/ViZDoom/blob/master/doc/Types.md ViZdoom APIdoc - Python
https://github.com/flyyufelix/VizDoom-Keras-RL Tensorflow + Keras

tf-gpu==1.4.0, keras==2.1.6, scikit-image vizdoom suggest to use anaconda, create new env with pip conda create -n 'vizdoom' pip python=3.5 and use pip install to install needed packages

in experiments/drqn folder
usage:
python drqn_curriculum.py -h