LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping (2007.00258v3)

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.

Insightful Overview of "LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping"

In the development of mobile robotic systems, accurate state estimation, localization, and mapping are crucial. The paper "LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping" by Shan et al. tackles this by presenting a framework that integrates lidar-based and inertial measurement systems within a factor graph paradigm for real-time state estimation and mapping. This combination aims to enhance performance and reliability, particularly in complex, dynamic environments.

Core Contributions

The authors introduce several key innovations in the field of lidar inertial odometry, explicitly designed to outperform existing methods like LOAM and LIOM. The main contributions are as follows:

1. Tightly-coupled multi-sensor fusion framework: The proposed methodology integrates IMU measurements, lidar scan-matching, GPS, and loop closure data within a factor graph, enabling robust and accurate state estimation.
2. Local scan-matching with sliding window approach: Instead of global map-based scan-matching, the framework employs a local sliding window technique that confines optimizations to a subset of keyframes, thereby enhancing computational efficiency and real-time performance.
3. Extensive validation: The authors validate their approach across various platforms and environments, demonstrating the method's versatility and robustness.

Technical Details

IMU Preintegration

The preintegration of IMU measurements significantly improves the computational efficiency of lidar-inertial odometry by reducing the frequency of integration steps. Using Equations 3-6, the framework integrates angular velocities and accelerations to derive velocity, position, and rotation changes over time, even in high-dynamic scenarios.

Lidar Odometry and Scan-Matching

The extraction and utilization of edge and planar features differ from point cloud matching processes like ICP and GICP due to the computational benefits. The authors introduce a keyframe-based approach to retain efficiency, selectively adding frames by evaluating transformation thresholds for position and rotation (1m and 10° respectively). Furthermore, by combining IMU preintegration with initial pose estimates, the system can match the scans more accurately, as detailed in Section 3.3.

Factor Graph Optimization

The factor graph framework incorporates four types of factors:

• IMU preintegration factors
• Lidar odometry factors
• GPS factors (when available)
• Loop closure factors

These factors collectively construct a graph that allows for optimal inference and pose estimation by leveraging iSAM2 for incremental smoothing and mapping.

Experimental Validation

The framework underwent rigorous testing using datasets collected on various robotic platforms, including a handheld device, an unmanned ground vehicle (UGV), and a boat. The evaluation spanned different environments such as urban spaces, parks, and waterways. Key findings include:

• Accuracy: LIO-SAM consistently matched or outperformed other methods like LOAM and LIOM in terms of trajectory accuracy, maintaining low-drift and precise mapping.
• Real-time Performance: The proposed system demonstrated the ability to operate in real-time, processing data up to 13 times faster than the data acquisition rate in stress tests.
• Robustness: The system exhibited robust performance even in GPS-denied environments by effectively utilizing loop closure mechanisms to correct drift.

Numerical Results and Efficiency

The research highlighted significant numerical results; for example, the end-to-end translation error in the Campus dataset was just 0.12m for LIO-SAM compared to 192.43m for LOAM. Similarly, the Park dataset showed an RMSE of 0.96m for LIO-SAM compared to 47.31m for LOAM.

Implications and Future Work

The implications of this research are two-fold:

• Practical Applications: The advancements in real-time and accurate state estimation equip mobile robots to function effectively in urban planning, autonomous vehicle navigation, and unmanned site surveying, among other applications.
• Theoretical Impact: The adoption of a factor graph framework for lidar-inertial odometry can inspire further research in multi-sensor fusion, optimization techniques in real-time mapping, and scalable methods for long-term robotic deployment.

Future explorations may involve deploying the LIO-SAM system on unmanned aerial vehicles to test its scalability and adaptability in aerial environments, where rapid dynamics and altitude changes pose additional challenges.

Conclusion

Overall, "LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping" introduces a well-rounded approach to addressing the challenges of real-time state estimation and mapping in dynamic environments. The combination of tightly-coupled multi-sensor fusion and efficient local scan-matching sets a solid foundation for the next generation of autonomous mobile systems.