Crosscheck Request: A3C Continuous CNN

Hey guys,
I know that this is probably not the right place but I was wondering whether someone would be keen to have look at my A3C continuous end-to-end reimplementation (shoutouts at jing-weiz and andrew liao at this point btw)?
Github: Agent
Github: model

I first tried to follow the proposed model setup with two convolutional layers without pooling non non-linearity, but changed to a more common setup, as I initially thought that the model was simply too unsophisticated to generalise over the seen states for a broad range of setups.