Voxel-SLAM: A Complete, Accurate, and Versatile LiDAR-Inertial SLAM System

Zheng Liu,Haotian Li,Chongjian Yuan,Xiyuan Liu,Jiarong Lin,Rundong Li,Chunran Zheng,Bingyang Zhou,Wenyi Liu,Fu Zhang

2024-10-12

Abstract:In this work, we present Voxel-SLAM: a complete, accurate, and versatile LiDAR-inertial SLAM system that fully utilizes short-term, mid-term, long-term, and multi-map data associations to achieve real-time estimation and high precision mapping. The system consists of five modules: initialization, odometry, local mapping, loop closure, and global mapping, all employing the same map representation, an adaptive voxel map. The initialization provides an accurate initial state estimation and a consistent local map for subsequent modules, enabling the system to start with a highly dynamic initial state. The odometry, exploiting the short-term data association, rapidly estimates current states and detects potential system divergence. The local mapping, exploiting the mid-term data association, employs a local LiDAR-inertial bundle adjustment (BA) to refine the states (and the local map) within a sliding window of recent LiDAR scans. The loop closure detects previously visited places in the current and all previous sessions. The global mapping refines the global map with an efficient hierarchical global BA. The loop closure and global mapping both exploit long-term and multi-map data associations. We conducted a comprehensive benchmark comparison with other state-of-the-art methods across 30 sequences from three representative scenes, including narrow indoor environments using hand-held equipment, large-scale wilderness environments with aerial robots, and urban environments on vehicle platforms. Other experiments demonstrate the robustness and efficiency of the initialization, the capacity to work in multiple sessions, and relocalization in degenerated environments.

Robotics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to build a complete, accurate and multi - purpose LiDAR - Inertial SLAM (Laser Radar - Inertial Navigation System) in the field of autonomous mobile robots to achieve real - time state estimation and high - precision map construction. Specifically, the paper proposes a system named Voxel - SLAM, aiming to improve the real - time performance and map accuracy of the system by making full use of short - term, medium - term, long - term and multi - map data associations.

### Main Problems
1. **Real - time State Estimation and High - precision Map Construction**:
- Existing LiDAR - Inertial SLAM systems are prone to cumulative drift during long - term operation, resulting in a decline in map accuracy.
- The method proposed in the paper reduces cumulative drift by using multiple data associations (short - term, medium - term, long - term and multi - map data associations) to achieve high - precision map construction and real - time state estimation.

2. **Initialization under Dynamic Initial State**:
- Most existing filter - baseline LiDAR - Inertial SLAM systems assume that the initial state of the sensor is stationary, which is often not true in practical applications.
- The paper proposes a robust and efficient initialization method that can provide accurate state estimation and a consistent initial local map under a dynamic initial state.

3. **Efficient and Accurate Local Mapping**:
- Existing systems usually use simple point - cloud matching methods in local mapping, resulting in insufficient accuracy and robustness.
- Voxel - SLAM uses efficient LiDAR - Inertial Bundle Adjustment (BA) to optimize the local map and state, improving the accuracy and robustness of the system.

4. **Closed - loop Detection in Multiple Sessions**:
- Existing systems have weak capabilities in detecting closed - loops in multiple sessions, resulting in poor map consistency.
- Voxel - SLAM can detect closed - loops in current and historical sessions and perform global optimization to improve map consistency.

5. **Efficient Global Mapping**:
- Global mapping requires processing a large amount of data, and existing methods have deficiencies in efficiency and accuracy.
- Voxel - SLAM uses a hierarchical global BA method to achieve efficient global optimization and ensure high - precision and consistency of the map.

### Solutions
- **Unified Adaptive Voxel Map**: All modules (initialization, odometry, local mapping, closed - loop detection and global mapping) use the same adaptive voxel map, which provides sufficient features for different tasks and adapts to various scenarios.
- **Robust and Fast Initialization**: Initialization or re - initialization can be completed with only short - term data (1 second), supporting dynamic initial states.
- **Efficient Local Mapping**: Use efficient LiDAR - Inertial BA to optimize the local voxel map and state, enhancing the accuracy and robustness of the system.
- **Closed - loop Detection in Multiple Sessions**: Be able to detect closed - loops in current and historical sessions and perform global optimization.
- **Efficient Global Mapping**: Use a hierarchical global BA method to achieve efficient global optimization and ensure high - precision and consistency of the map.
- **Make Full Use of Four Kinds of Data Associations**: Realize real - time operation and global map consistency through short - term, medium - term, long - term and multi - map data associations.

Through these innovations, Voxel - SLAM has demonstrated excellent performance in multiple scenarios (indoor, outdoor, urban environment), providing new solutions for the navigation and map construction of autonomous mobile robots.