this is an annotated PDF. you can read the reader’s remarks in your browser.

We present the first deep learning model to successfully learn control policies directly from high-dimensional sensory input using reinforcement learning.