Designing AI-Support VR by Self-supervised and Initiative Selective Supports

To provide flexible support ways and intelligent support contents for users in VR contexts, compared with the existing support ways of either single or combination of sensing functions, e.g., support of gesture, head or body movement. In our proposal, to provide flexible support functions conditioned on VR contexts or user’s feedbacks, we propose to use a semi-automatic selection of interactive supports. In modeling of semi-selection by user’s feedbacks and VR contexts, we propose to evaluate the performance by consideration of both intelligent AI evaluation, based on data of users’ performance in VR, and user’s initiative feedbacks. Furthermore, to provide customizable or personalized estimation in the VR support, we propose to apply the machine learning of self-supervised learning. Therefore, we are able to train or retrain estimation models with efficiency of low-cost of data works, including reduction of data-labeling cost or reuse of existing models. We require to evaluate the timing of applying selection or modification of support ways, the balance of ratios of automatics or user-initiative due to user preference or experiences or smoothness of VR contexts, and even user awareness or understanding, etc. Further, we require to evaluate the scale, numbers, size, and limitation of data or training that are needed for stable, accurate, and useful estimations of VR support.