Abstract
Executing complex time series computations on mobile edge devices in general and augmented reality (AR) devices in particular enables a variety of novel applications. As the complexity of such computations typically exceeds the resources of the mobile device, the task may be offloaded to a remote server. However, the resulting delay of the response substantially impacts the user experience, especially for real-time applications. To overcome this delay, we propose a novel system in which client and server exchange forecasting models instead of simple values. This allows the server to forecast future input values and precalculate the corresponding output. New results can thus be predicted in advance and, considering the communication delay, arrive just in time on the mobile device. The lightweight forecasting models not only reduce the amount of data that needs to be communicated, but also local (mobile device) computation to a minimum. Our results show that we can achieve a medium euclidean distance (MED) error of 0.477 mm for bio-mechanical simulation visualizations, even in the presence of jitter and with a round trip time of 98 ms. Compared to regular offloading, we reduce the MED by 61 % and the number of transmitted messages by 81 %, while providing a better user experience.
Funding Information
  • German Research Foundation (EXC 2075 - 390740016)

This publication has 8 references indexed in Scilit: