NajVR: low-latency plug-and-play remote rendering for interactive virtual reality applications

Virtual Reality (VR) setups connected to a computer offer a more realistic experience compared to standalone devices. However, these setups require high-performance hardware, making them expensive and less portable. A more cost-effective and portable alternative involves leveraging shared cloud or edge resources to render VR experiences. This work introduces NajVR, an architecture for remote rendering of VR applications that requires only commodity hardware on the client and includes LHE, a codec designed to minimize latency and resistant to packet loss that uses the user’s gaze data to improve the encoding process. NajVR does not require any modifications to the VR applications or their runtime environment. In NajVR an application on the client captures pose data and events generated by the VR hardware and transmits them to the server. On the server, a driver processes these data, simulating the VR hardware as if it were connected directly to it. This driver also captures the images to be displayed on the Head Mounted Display (HMD), encodes them, and transmits them to the client. NajVR supports an arbitrary number of trackers and takes advantage of the user’s gaze data to improve the perceived quality. The architecture round-trip latency was measured, and the LHE codec was compared with four different configurations of two commercially available codecs. The tests were carried out on both high-bandwidth and resource-constrained networks and in WAN setups. In the best-case scenario, latencies of just over 20 ms were achieved.