ScholarWorks@성균관대학교

초록

This paper proposes a system that sequentially performs autonomous valet parking for multiple vehicles using infrastructure-based multi-camera inputs. Relying solely on onboard sensors mounted on the vehicle is insufficient to safely and autonomously navigate in complex and confined parking environments due to limitations such as obstacles and restricted fields of view. To address these challenges, the proposed system estimates the position and orientation (yaw) of each vehicle externally using a YOLO-based pose estimation model and assigns a discrete ID to each detected vehicle. A central server, which manages the availability of parking spaces in real time, receives this information and allocates a parking spot to each client vehicle. Each vehicle then manages a sequence of waypoints leading to the assigned parking space and identifies its goal position. To generate the shortest feasible path from the vehicle's current location to the goal while accounting for the vehicle's motion constraints, we apply the Hybrid A-star algorithm. The vehicle is then guided along the generated path using a PID controller, which adjusts the steering angle in real time. In particular, to enhance accuracy in the parking system - where precise control is critical - this study replaces the commonly used Pure Pursuit controller with a PID controller to maximize path-following performance. The proposed system was tested and validated in the Gazebo simulation environment. As a result, the system successfully executed sequential autonomous valet parking for multiple vehicles. Compared to the Pure Pursuit method, the PID controller reduced path-tracking error and improved the overall stability and reliability of the system. These results demonstrate the potential of integrating infrastructure sensors and a centralized server to enable autonomous parking in complex environments.

키워드