Computation and Sensor Offloading for Cloud-Based Infrastructure-Assisted Autonomous Vehicles

Abstract

Existing autonomous vehicles use a myriad of onboard sensors to detect objects and survey environment. Among these sensors, LiDAR is probably the most critical and a high-end LiDAR can cost several tens of thousands USD. High cost of sensors can be a formidable barrier to the proliferation of autonomous vehicles. This article proposes the idea of installing LiDAR sensors on roadside lamp posts for sharing among passing vehicles. This can avoid installing expensive sensors onboard each vehicle. Due to limited computation resources at lamp posts, the task of processing sensor data into environment maps for broadcast to vehicles are performed at edge-cloud servers. This article develops a scheme to identify which edge-cloud servers to host the virtual machines for which lamp posts, so that the total computation and communication cost can be minimized. The proposed scheme exploits reuse of existing data at servers for lower cost. The scheme is formulated as a mixed integer program, which is NP-hard. Therefore, we further propose an iterative algorithm with LP-relaxation to find good approximate solutions to the optimization. Extensive evaluation results confirm the effectiveness of the proposed algorithm. As an example, for a system with 50 lamp posts, the proposed algorithm can find solutions in less than 5 s while solving the original optimization using exhaustive search takes more than 24 h.