Training in Virtual Environments via a Hybrid Dynamic Trainer Model
Hasan Esen · mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) · 2007
This thesis presents a novel virtual reality (VR) training concept that integrates the trainer or the trainer model into training sessions. As an extension to conventional VR training systems that rely only on realistic interaction, the students are given the chance to be corrected by the trainer in a multi-user schema. The trainer is connected to the same virtual environment as the student via an individual haptic display. As an alternative, task performing skill of the trainer is captured with hybrid identification methods and the trainer is replaced with the identified model allowing for a single-user training schema. Two different identification approaches are successfully applied and presented in this thesis: The weighted K-means clustering-based method and the stochastically switching dynamics method. Observations and corrections of the trainer or trainer model are multi-modal, i.e. can be represented in the form of visual, acoustic and/or haptic signals. The combination of these possible signals allows for the definition of different training strategies. Enhancing the training systems with extra features that are not available in a real task is investigated as well. Two different VR scenarios are developed as test-beds: A bone drilling medical training system and a push button system. The efficiency of the different training strategies is checked through a series of user tests. To assess the training results objectively, a metric depending on the distance between the trainer and student in n dimensional Euclidean space is introduced and applied. The results validate the efficiency and usability of the training strategies and hybrid identification methods.