Research and Implementation of Networked Cloud-Controlled Autonomous Driving System

This paper proposes a framework for a networked cloud-controlled autonomous driving system. It involves transmitting the vehicle’s location, structured perception data, and real-time images to a cloud server. The cloud server autonomously makes decisions based on the received data and sends the decision results to the vehicle. The vehicle, in turn, plans trajectories and controls based on these decision results. Additionally, a remote operator can engage in manual teleoperation using visual real-time images and structured abstract data, assisting the vehicle in making better decisions or manually setting more complex tasks. This paper designs and builds a complete autonomous driving real-word platform and networked cloud control platform. The neural network, trained through reinforcement learning and reinforcement learning, is deployed on the cloud for behavior decision-making. It controls the vehicle to drive over 27 kilometers at approximately $60 \mathrm{~km} / \mathrm{h}$ in real-world open scenarios, successfully completing multiple interactions with other vehicles. This validates the feasibility of the system.