Despite noticeable research progresses during the past several decades, how to efficiently acquire, synthesize, and edit highly realistic, personality-embodied virtual humans is still significantly away from being resolved in computer graphics and animation community. In this talk I will present selected research efforts on virtual human modeling and animation that have been done in my research group during the past several years, with its focus on facial animation and skinning deformation. Specifically, I will first describe a fully automated approach to generating realistic and synchronized head-and-eye motion on talking avatars based on live speech input. Then, I will present a robust approach to accurately extract the linear blend skinning model (the most widely employed skinning technique to date) with rigid bones and/or skeleton from a set of example poses, and a two-layer sparse compression technique to substantially reduce the computational cost of a dense-weight skinning model with negligible loss of its visual quality. Finally, to demonstrate how our virtual human systems could be applied to tackle challenging, interdisciplinary research, I will briefly present two selected projects: real-time MRI guide, robot-assisted, minimally invasive cardiac surgery, and promoting physical activities for youth using self-embodied avatars.