ROS 2 기반 자율주행 로봇의 비정형 환경에서 주행성능 향상에 관한 연구

다국어 초록 (Multilingual Abstract)

This study aims to quantitatively evaluate the driving performance of an autonomous mobile robot in both simple and obstacle-dense complex environments. Furthermore, it seeks to analyze performance metrics under each environmental condition to examine the potential for performance improvement given the constraints.
The driving experiments were categorized into three main types. The first performance evaluation focused on comparing and analyzing the autonomous driving system's environmental adaptability and path following performance in two distinct settings: open spaces with low obstacle density, which allow for relatively free movement, and high-density environments characterized by narrow passages and a concentration of numerous static obstacles, which necessitate complex path planning. This experiment was conducted using the Robot Operating System 2 (ROS 2) Foxy distribution, applying the default configuration file (nav2_params.yaml) of the Navigation 2 (Nav2) package. The experimental results indicated that while an average localization error of approximately 0.84 meters occurred in the open space, a significantly larger average error of about 2.6 meters was observed in the high-density, obstacle-cluttered environment. This outcome empirically confirms that the obstacle density along the driving path has an impact on the robot's localization accuracy that exceeds common expectations.
By applying multi-sensor fusion during the SLAM process, the influence of sensor noise was reduced and the feature-based registration process was smoothed, leading to a notable decrease in the overall travel time and map generation time.
These results demonstrate that sensor fusion is effective not only in improving the accuracy of spatial information but also in enhancing real-time responsiveness within unstructured environments. This suggests its potential applicability in various real-world scenarios, such as outdoor mobile robotics and industrial automation.
Furthermore, the study quantitatively analyzed the robot's ability to stably generate and execute paths under unstructured conditions, which involved a combination of complex structural layouts and moving obstacles. While the robot exhibited high driving performance in simple environments, the limitations of the existing algorithms became apparent in the complex static environment: the path length increased by approximately 80%, the travel time increased eightfold, and both the number of collisions and path replanning incidents also rose.
Crucially, normal navigation was impossible in situations that included dynamic obstacles, which clearly highlights the limitations of static map-based planning methods such as Navfn and Dijkstra's algorithm.
These results demonstrate that, while the advancement of sensors like LiDAR is a core requirement for autonomous driving robots, performance in complex environments can also be enhanced solely through appropriate software-based configuration depending on the environment. This suggests a practical approach to addressing diverse environmental conditions even when hardware modification is constrained.
In summary, this study presented key technical directions—including sensor fusion, path planning improvement, and SLAM parameter optimization—to enable autonomous mobile robots to maintain high reliability and efficiency even in unstructured environments.