My research interests primarily focus on robotics and computer vision. I am interested in studying how can an intelligent system or a robot understand its 3D environment from sensing and accomplish tasks in the real world, which is a very challenging and unsolved problem. Perception serves as an interface between an intelligent system and the 3D world, which provides useful information for planning and control of the system in conducting different tasks. I am interested in integrating perception, planning and control in a systematic way and deploying robots in the real world which are capable of accomplishing tasks for humans. I apply machine learning, especially deep learning, to tackle the challenges in robot perception. I explore how to introduce domain knowledge such as geometric constraints into a deep neural network architecture to learn a useful representation of the 3D environment for perception. I am also interested in how to learn a joint representation for perception, planning and control with deep neural networks, and how to enable robots to learn skills in a self-supervision way by interacting with the 3D environment.