Updated small_maze maze to deepmind maze HEAD
Maze is L-shaped at both ends. Visual results in navigation_visual_results folder
Training results in tensorboard_results folder
Auxiliary tasks
	rp - Reward prediction (skewed sampling)
	vp - Value prediction (unskewed sampling)
	rp_vp
	pc - Pixel control (unskewed sampling)
	rp_vp_pc
	fp - Frame prediction (unskewed sampling)
	rp_vp_fp
	ap - Action prediction (unskewed sampling)
	ftp - Frame threshold prediction (image is thresholded)
	flp - Flow prediction
    
Bigger beta for exploration works best when the episode is short

$ git clone https://github.com/deepmind/lab.git

py_binary(
name = "a3c_train",
srcs = ["A3C/main.py"],
data = [":deepmind_lab.so"],
main = "A3C/main.py",
)

''' Choose task '''
CONFIG = FP

bazel run :a3c_train

tensorboard --logdir=worker_0:'./train_0',worker_1:'./train_1',worker_2:'./train_2',worker_3:'./train_3'

tensorboard_results folder

navigation_visual_results folder