dqn code adapted from Deepmind's original here https://sites.google.com/a/deepmind.com/dqn. Has been adapted to deal with different domains as well as for currculum learning on the trashCollecter domain. Agent can focus on a subset of the image to transfer policies