VoxelMap++: Mergeable Voxel Mapping Method for Online LiDAR(-inertial) Odometry

Yifei Yuan,Chang Wu,Yuan You,Xiaotong Kong,Ying Zhang,Qiyan Li

2023-08-05

Abstract:This paper presents VoxelMap++: a voxel mapping method with plane merging which can effectively improve the accuracy and efficiency of LiDAR(-inertial) based simultaneous localization and mapping (SLAM). This map is a collection of voxels that contains one plane feature with 3DOF representation and corresponding covariance estimation. Considering total map will contain a large number of coplanar features (kid planes), these kid planes' 3DOF estimation can be regarded as the measurements with covariance of a larger plane (father plane). Thus, we design a plane merging module based on union-find which can save resources and further improve the accuracy of plane fitting. This module can distinguish the kid planes in different voxels and merge these kid planes to estimate the father plane. After merging, the father plane 3DOF representation will be more accurate than the kids plane and the uncertainty will decrease significantly which can further improve the performance of LiDAR(-inertial) odometry. Experiments on challenging environments such as corridors and forests demonstrate the high accuracy and efficiency of our method compared to other state-of-the-art methods (see our attached video). By the way, our implementation VoxelMap++ is open-sourced on GitHub which is applicable for both non-repetitive scanning LiDARs and traditional scanning LiDAR.

Robotics

What problem does this paper attempt to address?

The paper primarily focuses on improving the map representation method in the Simultaneous Localization and Mapping (SLAM) system that combines LiDAR (Light Detection and Ranging) and Inertial Measurement Unit (IMU). Specifically, the research team proposed VoxelMap++, a voxel-based map construction method that enhances the accuracy and efficiency of the SLAM system through plane merging techniques.

### Main Issues Addressed

1. **Improving Accuracy and Efficiency**: Existing point cloud-based map construction methods are inefficient during point cloud registration and struggle to effectively maintain point cloud uncertainty information. VoxelMap++ improves map accuracy and overall system efficiency by enhancing plane fitting and uncertainty estimation methods.

2. **Reducing Memory Consumption**: Traditional point cloud map methods consume a large amount of memory resources when storing vast amounts of point cloud data. VoxelMap++ employs a new online voxel merging method that significantly reduces memory usage.

3. **Optimizing Plane Feature Representation**: In existing methods, plane features are typically represented using 6 degrees of freedom (6DOF), which not only wastes memory but also increases system complexity. VoxelMap++ simplifies plane representation to 3 degrees of freedom (3DOF), reducing memory usage and computational complexity.

4. **Handling Plane Relationships Between Adjacent Voxels**: Existing methods overlook the plane relationships between adjacent voxels, leading to redundant representation of features on the same plane. VoxelMap++ effectively addresses this issue by proposing a voxel merging algorithm based on the union-find method, improving the accuracy of plane estimation.

### Experimental Validation

The paper validates the effectiveness of VoxelMap++ through a series of experiments:

- In structured urban environments, VoxelMap++ demonstrates higher localization accuracy in most cases compared to several existing advanced algorithms (such as A-LOAM, LeGO-LOAM, etc.).
- In challenging scenarios (such as unstructured forest and grassland environments and indoor long corridor environments), VoxelMap++ also shows better stability and accuracy.

In summary, VoxelMap++ aims to improve the accuracy and efficiency of LiDAR-Inertial SLAM systems while reducing memory consumption through improved plane feature representation methods and the introduction of voxel merging mechanisms, making it suitable for various complex real-world application scenarios.