My ultimate goal is to build agents with a human-like ability to generalize in real and diverse environments. I believe understanding how to continually develop knowledge and acquire new skills from just raw sensory data will play a vital role in achieving this goal. I draw inspiration from psychology to build practical systems at the interface of vision, learning and robotics that can learn using data as its own supervision. My research is supported by fellowships from Facebook, Nvidia, and Snapchat.