We present a deep learning model for playing games with high level input (image/raw pixel) using reinforcement learning. The games are action limited (like snakes, catcher, air-raider etc.). The model consists of convolution ...