Bài báo khoa học

Autonomous Mobile Robot (AMR) navigation heavily relies on Simultaneous Localization and Mapping (SLAM) as a cornerstone technology. While recent SLAM approaches have achieved high performance, there remains a challenge in ensuring robustness and adaptability to diverse real-world conditions. Many existing works have explored recursive Bayesian filtering to integrate LiDAR and cameras, employing either loosely or tightly coupled methods. Loosely coupled techniques are simple but need more precision, while tightly coupled methods face difficulties in handling 2D LiDAR-visual coupling and are susceptible to sensor degradation. In response to these challenges, this work introduces a pioneering SLAM framework centered around a LiDAR-centric binary pose fusion technique specifically designed for industrial AMRs. The framework incorporates the synergy of multiple 2D LiDARs, cameras, and an Inertial Measurement Unit (IMU). Notably, our proposed SLAM structure adopts a hybrid coupled technique, combining LiDAR-centric front-end processing with visual inertial odometry and a back-end structure featuring factor graph optimization and LiDAR-cameras for loop closure detection. To validate the efficacy of our approach, we implemented the SLAM framework in practical scenarios, mainly showcasing its efficiency when utilizing multiple 2D LiDAR-visualinertial sensors. The results demonstrate exceptional accuracy and out-performance compared to previous LiDAR and visual SLAM systems. Moreover, a low-cost maintenance sensor system offers a cost-effective alternative for integrating industrial AMRs, significantly reducing expenses compared to conventional designs. This research contributes a comprehensive and innovative solution to the challenges of SLAM integration in AGV navigation, emphasizing the importance of fusion consistency and practical applicability in industrial settings.