My research focuses on robust feature representation learning for visual perception and generation. I have been developing novel representation learning methods including: (1) Contrastive/Siamese Learning for video and 3D scene understanding; (2) Masked Visual Modeling for human action analysis from videos and 3D point clouds/meshes; and (3) Generatively Pretrained Vision-Language Models for open-vocabulary 3D scene understanding and generation. I have published first-authored papers in top computer vision conferences such as CVPR, ICCV, and ECCV. My research has been featured by CMU News and Science Daily, and deployed by the U.S. Army Research Lab and Federal Emergency Management Agency.