A Cloud-Connected Autonomous Driving System

Current autonomous driving systems are designed in terms of ego-vehicle, specifically a self-driving vehicle locates itself, perceives the environment, plans plausible paths, and executes actions to safely reach a given goal location all by itself. The systems are complex, expensive, power-consuming and moreover information isolated. In this paper, we propose a Cloud-Connected Autonomous Driving System (CCADS) in terms of traffic network. Self-driving vehicles share data with all other running self-driving cars via the traffic network cloud, and traffic network plays a central role in the CCADS. It analyzes traffic data and manages all the self-driving vehicles to achieve high efficiency and safety traffic environment. The self-driving vehicles are remaining locally autonomous but globally controlled. The proposed CCADS consists of two parts: cloud side and vehicle side. Cloud side keeps collecting and monitoring real-time vehicles and traffic infrastructure data, and the traffic scheduling module dynamically plans safe trajectory task for many vehicles at the same time given traffic status. Vehicle side sends position, velocity and other valuable data to cloudside and executes trajectory task to get to goal location. The proposed autonomous driving system considers network delay, fault tolerance, and network security. We implement the system in a park scene, and verify its system functionality, fault tolerance, and traffic schedule. Experiments show that our system could work robustly and safely.