모형 차량에서의 영상센서와 초음파센서를 활용한 자율 발렛 주차 궤적 수정

Abstract

This paper outlines a proposed method of modifying parking trajectory during the autonomous valet parking process. Most autonomous valet parking systems calculate trajectories by first finding the location of a parking space and then tracking the path between the current position and the space while driving or parking. However, these systems do not have a process of modifying parking trajectories in the middle of the action. This increases the risk of collisions due to a system failure to properly detect new obstacles or incorrectly generated parking trajectories arising from hardware errors. To address these problems, it is necessary to conduct research on how to modify parking trajectories while driving. The methods outlined in this paper utilize visual simultaneous localization and mapping (Visual SLAM), ultrasonic sensors, and models with Ackerman geometry to adjust parking trajectories. Visual SLAM determines the location of the model vehicle and the chosen parking space. Once these locations are established, ultrasonic sensors determine whether the situation requires trajectory modification. From this information, the transformed Ackerman geometric model is applied to compute any required trajectory changes. This new method is applied to experimental parking scenarios to determine whether the proposed trajectory modification method was viable. The experimental results of this study indicated that autonomous parking was successfully carried out in 81.66% and 78.33% of autonomous parking events where the parking space was located on the left and right sides of the model vehicle, respectively. These results suggest that applying this model to autonomous parking and driving scenarios makes trajectory generation a viable solution for collision avoidance.