Modeling Edge-Assisted Virtual Reality Systems

Current virtual reality (VR) systems remain far from providing high-fidelity life-like VR experiences. A viable solution to enabling high-fidelity VR is the edge-assisted approach. To understand the characteristics of VR user behavior and improve the VR experiences accordingly, we are currently exploring the performance analysis of VR systems by modeling the 6DoF pose of users’ head movement. We then apply the statistical pose model to characterize the inter-frame redundancy that can be eliminated to save communication and computation resource of edge-assisted VR systems.

Adaptive Edge-Assisted SLAM for Augmented Reality

Seamless augmented reality (AR) experience requires accurate camera pose estimation and 3D reconstruction of the environment in real-time, which is the problem that simultaneous localization and mapping (SLAM) aims to solve. To save computational and power resources on mobile AR devices, computationally expensive parts of SLAM pipelines are offloaded to the edge server. We are building an analytical model to characterize how each keyframe contributes to the overall SLAM performance. By the adaptive uplink keyframe offloading and downlink map update, we aim to maximize the localization and mapping performance for mobile AR under the constraints of limited and unstable wireless transmission.