The recent leaps in virtual reality (VR) technology have unleashed revolutionary potential for applications in a wide variety of areas, including education, training, psychological-therapy, etc. As part of the effort on understanding how users interact with VR, I focused on studying full-body joint action using a road crossing task which involves perception, decision-making, action, and joint action. I have been heavily involved in the design, implementation, and construction of two large-screen, room-like stereoscopic virtual environment (VE) simulators. Using this system, I developed a three-part research plan with a series of studies to examine how people engage in full-body joint-action with a partner under three scenarios: 1. two people who are physically present in a co-occupied virtual environment; 2. one person who shares a virtual environment with a computer-generated agent (CG agent); 3. two people who share the same virtual environment remotely in physically separate places where each person is motion tracked and presented in the environment as a graphic avatar. The behaviors of participants were recorded and processed through a customized pipeline that captures important performance metrics, such as how participants pick crossable gaps and time their movements. The VE system, user study designs, and findings are introduced in this dissertation.