Efficient and Probabilistic Adaptive Voxel Mapping for Accurate Online LiDAR Odometry (2109.07082v5)

Abstract: This paper proposes an efficient and probabilistic adaptive voxel mapping method for LiDAR odometry. The map is a collection of voxels; each contains one plane (or edge) feature that enables the probabilistic representation of the environment and accurate registration of a new LiDAR scan. We further analyze the need for coarse-to-fine voxel mapping and then use a novel voxel map organized by a Hash table and octrees to build and update the map efficiently. We apply the proposed voxel map to an iterated extended Kalman filter and construct a maximum a posteriori probability problem for pose estimation. Experiments on the open KITTI dataset show the high accuracy and efficiency of our method compared to other state-of-the-art methods. Outdoor experiments on unstructured environments with non-repetitive scanning LiDARs further verify the adaptability of our mapping method to different environments and LiDAR scanning patterns. Our codes and dataset are open-sourced on Github

Efficient and Probabilistic Adaptive Voxel Mapping for Accurate Online LiDAR Odometry

The paper by Chongjian Yuan et al. introduces an innovative approach in the domain of LiDAR odometry through proposing an efficient and probabilistic adaptive voxel mapping method. This approach aims at enhancing the accuracy and efficiency of LiDAR-based navigation systems. The method hinges on the adaptive voxel mapping technique, which incorporates probabilistic representations to better capture environmental uncertainties and applies this methodology within an iterated extended Kalman filter framework to achieve robust pose estimation.

Methodological Contributions

The paper presents several distinctive contributions:

1. Adaptive Voxel Construction: The method involves constructing voxels of varying sizes depending on environmental structures and LiDAR scan density. This adaptability addresses the issue of sparse and irregular LiDAR data distribution, a common challenge in LiDAR-based systems. The voxel construction is organized using an octree-hash data structure to ensure efficient voxel creation, update, and inquiry operations.
2. Probabilistic Map Representation: Unlike traditional point cloud maps that overlook inherent measurement noise, the proposed method models plane features within each voxel probabilistically. This captures uncertainties caused not only by point measurement errors but also by pose estimation errors, contributing significantly to the map's reliability in representing the true state of the environment.
3. Implementation and Validation: The paper provides open-source code for real-world application, emphasizing the method's practical viability. Through extensive testing on the KITTI dataset and other real-world environments with various LiDAR configurations, the authors demonstrate their method's superior adaptability and performance across different scenarios.

Experimental Verification

Experiments conducted on well-known datasets such as KITTI showcase the advantages of the proposed adaptive voxel mapping over existing state-of-the-art methods. The results indicated substantial improvements in trajectory accuracy as evidenced by lower Absolute Trajectory Error (ATE). Furthermore, the method demonstrated efficient computational performance relative to other LiDAR odometry methods, largely owing to the streamlined voxel update mechanism and effective outlier rejection powered by probabilistic uncertainty modeling.

Implications and Future Directions

The authors outline meaningful implications for both theoretical exploration and practical application within the field of SLAM (Simultaneous Localization and Mapping) systems. The probabilistic mapping approach enables robust point-to-plane registration, critical for environments characterized by feature sparsity and irregularity. This adaptability is particularly advantageous in dynamic and unstructured environments where conventional mapping techniques may fall short.

Looking ahead, advancements could focus on integrating more diverse feature types within the voxel framework to enhance universality and adaptability in even more complex environments. Furthermore, exploring multi-sensor fusion techniques that incorporate the proposed mapping method could yield breakthroughs in real-time navigation and mapping for autonomous systems operating in challenging and varied settings.

In conclusion, Yuan et al.'s work contributes substantially to the evolution of LiDAR odometry systems by introducing an adaptive and probabilistic mapping technique that finely balances accuracy, efficiency, and robustness. The paper is likely to spur further research and development in probabilistic mapping approaches, driving future innovations in SLAM methodologies and autonomous system navigation.