FAST-LIVO: Fast and Tightly-coupled Sparse-Direct LiDAR-Inertial-Visual Odometry (2203.00893v1)

Abstract: To achieve accurate and robust pose estimation in Simultaneous Localization and Mapping (SLAM) task, multi-sensor fusion is proven to be an effective solution and thus provides great potential in robotic applications. This paper proposes FAST-LIVO, a fast LiDAR-Inertial-Visual Odometry system, which builds on two tightly-coupled and direct odometry subsystems: a VIO subsystem and a LIO subsystem. The LIO subsystem registers raw points (instead of feature points on e.g., edges or planes) of a new scan to an incrementally-built point cloud map. The map points are additionally attached with image patches, which are then used in the VIO subsystem to align a new image by minimizing the direct photometric errors without extracting any visual features (e.g., ORB or FAST corner features). To further improve the VIO robustness and accuracy, a novel outlier rejection method is proposed to reject unstable map points that lie on edges or are occluded in the image view. Experiments on both open data sequences and our customized device data are conducted. The results show our proposed system outperforms other counterparts and can handle challenging environments at reduced computation cost. The system supports both multi-line spinning LiDARs and emerging solid-state LiDARs with completely different scanning patterns, and can run in real-time on both Intel and ARM processors. We open source our code and dataset of this work on Github to benefit the robotics community.

Overview of FAST-LIVO: Advances in Multi-Sensor Fusion for Robust SLAM

The research paper presents FAST-LIVO, a novel LiDAR-Inertial-Visual Odometry system developed to enhance the accuracy and robustness of pose estimation in Simultaneous Localization and Mapping (SLAM) tasks. The system is designed for real-time applications, wherein multi-sensor fusion—involving LiDAR, camera, and IMU—is utilized to surpass the limitations observed in traditional methods that rely on a single sensor. FAST-LIVO is composed of two tightly-coupled subsystems: a visual-inertial odometry (VIO) subsystem and a LiDAR-inertial odometry (LIO) subsystem.

Framework and Methodologies

FAST-LIVO distinguishes itself by adopting a sparse-direct approach for sensor data processing. Unlike traditional feature-point extraction that adds computational overhead, the LIO subsystem registers raw LiDAR points directly to the incrementally-built point cloud map. This map is then used in the VIO subsystem, where image patches attached to the map points align the new image by minimizing photometric errors, eliminating the need for visual feature extraction.

To bolster the robustness and accuracy of the VIO subsystem, the paper introduces an outlier rejection method to remove unstable map points, particularly those positioned on edges or occluded in the image view. Additionally, FAST-LIVO supports both multi-line spinning LiDAR and emerging solid-state LiDAR technologies, indicating versatility across different sensor architectures.

Numerical Results and Performance

Experimental results underscore the superiority of FAST-LIVO in various challenging environments using both open dataset sequences (specifically the NTU VIRAL Dataset) and data collected from a custom device. FAST-LIVO consistently outperforms competing systems, including state-of-the-art LiDAR-inertial and visual-inertial odometry solutions. The system effectively balances accuracy with reduced computational load, and exhibits remarkable real-time performance, operating efficiently on both Intel and ARM processors.

The paper presents detailed comparisons in which FAST-LIVO achieves lower root mean square errors (RMSE) in translational motion estimates compared to other existing solutions, even under sensor-degraded conditions. This underscores the robustness of the integrated multi-sensor fusion in addressing the inadequacies of systems relying on fewer sensing modalities.

Implications and Future Directions

The implications of the FAST-LIVO system extend to various robotics applications, especially those requiring real-time SLAM in dynamic and sensor-degraded environments. This research establishes a strong case for tightly-coupled multi-sensor fusion systems that leverage direct methods for efficient and robust state estimation.

Future exploration could focus on further optimizing the computational efficiency of FAST-LIVO, perhaps integrating learning-based approaches to enhance the system's adaptability to diverse environments. Additionally, applying FAST-LIVO in practical autonomous navigation systems could provide valuable insights into its scalability and performance in real-world scenarios.

In conclusion, FAST-LIVO contributes a significant advancement in the field of SLAM by combining sparse-direct techniques with efficient multi-sensor fusion, setting a new benchmark for high-performance odometry systems in autonomous robotics.