Time in Flight Based MPQUIC Scheduler for Remote Monitoring of Autonomous Driving

Driverless autonomous driving on public roads has finally been approved in Japan under certain conditions such as real-time remote monitoring. However, real-time monitoring is difficult to stably provide with a single mobile network due to the lack of uplink bandwidth and fluctuations in radio quality. While using multiple networks with Multipath QUIC (MPQUIC) could improve the stability of remote monitoring, it may worsen video latency and noise because of packet delay and loss caused by reordering packet arrivals between paths. Conventional packet scheduling algorithms select transmission paths on the basis of the smoothed round trip time (SRTT) and congestion window. These metrics cannot immediately detect the sudden quality degradation in mobile networks. In this paper, we propose a MPQUIC forwarding system for remote monitoring and a novel packet scheduling algorithm that can avoid sudden quality degradations by determining the transmission paths on the basis of time in flight, which is the elapsed time since the oldest unacked packet was sent. We also implement the proposed method and evaluate it in virtual networks and real mobile networks.