This paper presents the design and implementation of a distributed virtual reality (VR) platform that was developed to support the training of multiple users who must perform complex tasks in which situation assessment and critical thinking are the primary components of success. The system is fully immersive and multimodal, and users are represented as tracked, full-body figures. The system supports the manipulation of virtual objects, allowing users to act upon the environment in a natural manner. The underlying intelligent simulation component creates an interactive, responsive world in which the consequences of such actions are presented within a realistic, time-critical scenario. The focus of this work has been on the training of medical emergency-response personnel. BioSimMER, an application of the system to training first responders to an act of bio-terrorism, has been implemented and is presented throughout the paper as a concrete example of how the underlying platform architecture supports complex training tasks. Finally, a preliminary field study was performed at the Texas Engineering Extension Service Fire Protection Training Division. The study focused on individual, rather than team, interaction with the system and was designed to gauge user acceptance of VR as a training tool. The results of this study are presented.

In a virtual environment for small groups of interacting participants, it is important that the physical motion of each participant be replicated by synthetic human forms in real time. Sensors on a user's body are used to drive an inverse kinematics algorithm. Such iterative algorithms for solving the general inverse kinematics problem are too slow for a real-time interactive environment. In this paper we present analytic, constant time methods to solve the inverse kinematics problem and drive an avatar figure. Our sensor configuration has only eight sensors per participant, so the sensor data is augmented with information about natural body postures. The algorithm is fast, and the resulting avatar motion approximates the actions of the participant quite well. This new analytic solution resolves a problem with an earlier iterative algorithm that had a tendency to position knees and elbows of the avatar in awkward and unnatural positions.