Abstract
We propose a framework for tightly-coupled lidar inertial odometry via smoothing and mapping, LIO-SAM, that achieves highly accurate, real-time mobile robot trajectory estimation and map-building. LIO-SAM formulates lidar-inertial odometry atop a factor graph, allowing a multitude of relative and absolute measurements, including loop closures, to be incorporated from different sources as factors into the system. The estimated motion from inertial measurement unit (IMU) pre-integration de-skews point clouds and produces an initial guess for lidar odometry optimization. The obtained lidar odometry solution is used to estimate the bias of the IMU. To ensure high performance in real-time, we marginalize old lidar scans for pose optimization, rather than matching lidar scans to a global map. Scan-matching at a local scale instead of a global scale significantly improves the real-time performance of the system, as does the selective introduction of keyframes, and an efficient sliding window approach that registers a new keyframe to a fixed-size set of prior "sub-keyframes." The proposed method is extensively evaluated on datasets gathered from three platforms over various scales and environments.
Introduction
Simultaneous Localization and Mapping (SLAM) is a foundational capability for mobile robotics. However, existing Lidar-based methods often face a trade-off between speed and accuracy. Popular methods like LOAM (Lidar Odometry and Mapping) are computationally efficient but "loosely-coupled," using the Inertial Measurement Unit (IMU) primarily for de-skewing Lidar data rather than as a core optimization constraint. This often leads to drift during long-duration missions or aggressive maneuvers. Conversely, "tightly-coupled" methods like LIOM offer superior accuracy but are computationally intensive, frequently failing to maintain real-time performance on embedded systems.
LIO-SAM bridges this gap by providing a tightly-coupled fusion framework that maintains the speed required for real-time deployment while achieving state-of-the-art accuracy.
System Architecture and Methods
The core of LIO-SAM is a Factor Graph formulation that treats robot positions as nodes and sensor measurements as constraining factors.
Tightly-Coupled Fusion
Unlike previous loosely-coupled approaches, LIO-SAM incorporates IMU pre-integration, Lidar odometry, GPS, and loop closures as factors in a single optimization problem. The IMU is not just a preprocessing tool; its biases are estimated and corrected alongside the robot's trajectory. This allows the IMU to act as a high-quality "prior," enabling the system to handle aggressive motions that would cause Lidar-only systems to diverge.
Local Sliding Window Optimization
To maintain real-time performance, LIO-SAM avoids the computational bottleneck of matching new Lidar scans against a global map. Instead, it employs a local sliding window approach. A new keyframe is matched against a set of recent sub-keyframes. This local scan-matching, combined with the incremental smoothing and mapping (iSAM2) algorithm, allows the system to perform fast updates even as the overall map grows.
Experimental Evaluation
The authors validated LIO-SAM across five diverse datasets using three distinct platforms to test the system's versatility and robustness.
- Handheld (MIT Campus): Tested aggressive rotations and long-distance walking in complex urban environments.
- Clearpath Jackal UGV (Park): Evaluated performance on forested hiking trails with high-frequency vibrations and intermittent GPS.
- Boat (Amsterdam Canals): Represented a "degenerate" environment with few Lidar features and numerous bridges, a scenario that typically breaks standard SLAM algorithms.
Key Results and Insights
- Superior Accuracy: In the "Campus" dataset, LIO-SAM achieved an end-to-end translation error of just 0.12 meters. In contrast, LOAM diverged due to aggressive motion, resulting in an error of 192.43 meters.
- Exceptional Speed: LIO-SAM is highly efficient, processing data up to 13x faster than real-time in stress tests. It consistently uses less CPU time per scan than competing tightly-coupled frameworks.
- Robustness in Degenerate Environments: In the Amsterdam canal tests, LIO-SAM produced consistent maps well-aligned with satellite imagery by successfully utilizing intermittent GPS and loop closures, whereas LOAM and LIOM failed completely.
- Modular Factor Design: The factor graph architecture allows LIO-SAM to be "plug-and-play." Additional sensors like altimeters or compasses can be added as new factors to further reduce drift.
Conclusion and Future Work
LIO-SAM demonstrates that tightly-coupled Lidar-inertial fusion can be both highly accurate and computationally efficient. By shifting from global to local scan-matching and utilizing a factor graph for multi-sensor fusion, it provides a robust solution for diverse robotic platforms. Future work includes implementing more sophisticated place recognition for loop closure in repetitive environments and extending testing to Unmanned Aerial Vehicles (UAVs) with extreme 6-DOF motion.
Summary updated