Infrastructure Sensor-Based Modular Autonomous Valet Parking System Using ROS 2

This paper proposes an infrastructure sensor-based Autonomous Valet Parking (AVP) system. The AVP system is a technology in which, upon the vehicle's arrival at the parking lot entrance, it cooperates with infrastructure sensors to search for an available parking space and performs autonomous parking. The system is composed of three main components: the AVP server, the edge device server, and the vehicle control system. The AVP server centrally manages parking space information, while the edge device server estimates the vehicle's position using a deep learning-based pose estimation model applied to images collected from camera sensors. The vehicle control system uses the received information to perform autonomous driving and precise parking. The system was developed based on ROS 2 and was designed to enable multiple vehicles and sensors to operate simultaneously through the use of namespaces and lifecycle management. Experiments were conducted in a parking lot environment constructed using a real children's electric car, a Jetson AGX Orin, and a total of 13 cameras. Through a modular structure based on ROS 2, the system's scalability and stability were secured, confirming the feasibility and efficiency of the proposed infrastructure-based AVP system.