Deterministic Video Streaming with Deep Learning Enabled Base Station Intervention for Stable Remote Driving System

Abstract

A remote driving system (RDS) via wireless net- works is a promising solution to guarantee the safety of autonomous vehicles. In the system, a remote operator controls vehicles while watching video frames transmitted from the controlled vehicles. The conventional video streaming method decides the video resolution using the statistical quality of services (QoS) to guarantee the delay constraints; however, it may yield a long delay in best-effort wireless networks if the quality of the wireless channel suddenly changes. To cope with the issue, we propose a deterministic networking approach. A base station (BS) predicts a future QoS using a radio map and driving route of vehicles to decides the adequate video resolution that satisfies the delay constraints of RDS. BS also has a super-resolution (SR) function to enhance the quality of experience (QoE) in video streaming. Thanks to the proposed QoS prediction and video frame resolution decision, BS can use the adequate SR model for each video frame, further enhancing QoE. Using the measurement datasets of radio access networks, we confirm that the proposed RDS can provide high-quality video streaming while satisfying the delay constraints in any time-series wireless channel situations.