Our social AI research aims to create advanced AI solutions that are able to perceive and understand humans and their behaviors observed by cameras in relation to the physical space they are interacting with. More specifically, social interaction analysis uses nonverbal behavioral cues e.g., posture, gesture and facial features, to investigate tasks including active speaker detection from body motion analysis, human trajectory and head pose forecasting, intension detection from body pose and 3D action recognition. These problems are addressed in conjunction with object detection and tracking, pose estimation, person/object re-identification, instance and semantic segmentation as well as deep learning methodologies.