PI: Fábio Luciano Verdi
This research strand revolves around a fully distributed computer networking footprint focused on decision making processes for the deployment of network functions driven by the needs of XR applications. Immersive AR/VR concerns interactive applications that compose a myriad of candidate 5G latency-sensitive novel services considered prominent advocates for the establishment of edge computing [Zhao et al., 2019]. Applications related to skill transfer will benefit from XR solutions and will improve the experience when learning and teaching, allowing seamless interaction between remote entities. It is expected that online education and training at different levels will become affordable for remote regions in Brazil with 5G and beyond, being XR applications of paramount importance for social inclusion in a future digitalized world. Experts could be able to instruct a technician performing a repair of a machine at a certain location as if they were physically present [Araújo et al., 2018]. The concept of the network as a computer encompasses the core of this research strand, which in a broader way aims to cope with new mechanisms for inserting computation at the network edge and core in order to enhance XR applications.
Using different virtualization approaches, hardware acceleration, and network offloading techniques, this research strand will investigate which applications could benefit from the in-network computing paradigm. Caching, object detection and classification, frames prefetching and frames adaptation consist of possible tasks related to AR/VR that can be offloaded in the network. Machine learning algorithms play an essential role in the support of such tasks, a fact confirmed by already existing dataplane solutions utilizing deep learning [Ran et al., 2017] and random forests [Busse-Grawitz et al., 2019].
Online gaming contains good examples of applications that can benefit from in-network computing. Recent studies have shown that VR gaming users’ head movement can be predicted with high accuracy for upcoming hundreds of milliseconds [Zhang et al., 2018]. Performance gains of 5G latency-sensitive services by in-network computing establish major improvements in users’ QoE, one of the most challenging barriers for the wide adoption of XR applications.