An Engineering Solution for Multi-sensor Fusion SLAM in Indoor and Outdoor Scenes

Abstract

In this contribution, an engineering solution for multi-sensor fusion simultaneous localization and mapping (SLAM) is proposed for both indoor and outdoor scenarios, targeting at enhanced robustness, accuracy, and scene adaptability. It consists of three powerful schemes. A scheme-switching mechanism is designed based on a thorough performance evaluation and selects the most suitable multi-sensor fusion SLAM method flexibly based on the characteristics of the scenes and the setting of the robot products. Among the three schemes, LVI-SAM-Stereo is a novel multi-sensor fusion SLAM approach that tightly couples a stereo camera with a 3D light detection and ranging (LiDAR) sensor and an inertial measurement unit (IMU). Its stereo-inertial odometry provides a more robust initial guess for the LiDAR registration compared to its monocular counterpart in critical mapping scenarios. Moreover, a visual verification mechanism for the LiDAR loop closure detection is proposed to effectively avoid incorrect LiDAR loop closures. A thorough evaluation with both datasets and real-world experiments in various indoor and outdoor scenarios verify that our proposed engineering solution achieves a satisfactory performance and meets the engineering requirements for autonomous navigation of robot products.