My research interests are broadly in computer vision, machine learning, and AI with a focus on multi-task transfer/selfsupervised/unsupervised learning and perception-for-robotics. The goal of my research is going beyond narrow and offline perception methods toward a general multi-task visual perception that operates as part of an active body in the real world. Here are some of my works on this topic: Taskonomy, Cross-Task Consistency, Mid-Level Vision, Gibson. I'm also interested in structured prediction, video understanding, and 3D vision. I'm a fan of Slow Science.