Net-driving: An Alternative to Autonomous Driving

Autonomous driving will become pervasive in the coming decades. Net-driving is a different point in the design space for autonomous driving, motivated by increasing deployments of 5G and edge computing, and the drop in prices of advanced sensors like LiDAR. In Net-driving, the network remotely “drives” vehicles by offloading perception and planning from the vehicle to roadside infrastructure. In this paper, we show that Net-driving can meet the performance needs for autonomous driving by processing voluminous sensor data at full frame rate with a tail latency of less than 100 ms, without sacrificing accuracy. It achieves this using an accurate and lightweight perception component that reasons on composite views derived from overlapping sensors, and a planner that jointly plans trajectories for multiple vehicles. Net-driving is safer than autonomous driving, and can enable higher throughput traffic management at intersections.