FAST-LIO: A Fast, Robust LiDAR-inertial Odometry Package by Tightly-Coupled Iterated Kalman Filter (2010.08196v3)

Abstract: This paper presents a computationally efficient and robust LiDAR-inertial odometry framework. We fuse LiDAR feature points with IMU data using a tightly-coupled iterated extended Kalman filter to allow robust navigation in fast-motion, noisy or cluttered environments where degeneration occurs. To lower the computation load in the presence of large number of measurements, we present a new formula to compute the Kalman gain. The new formula has computation load depending on the state dimension instead of the measurement dimension. The proposed method and its implementation are tested in various indoor and outdoor environments. In all tests, our method produces reliable navigation results in real-time: running on a quadrotor onboard computer, it fuses more than 1,200 effective feature points in a scan and completes all iterations of an iEKF step within 25 ms. Our codes are open-sourced on Github.

• The paper introduces a tightly-coupled iterated Kalman filter that fuses LiDAR and IMU data to achieve robust, real-time state estimation.
• The new Kalman gain formula shifts computational dependency from measurement to state dimension, processing over 1,200 feature points within sub-millisecond times.
• The framework is validated through indoor and outdoor UAV tests, demonstrating reliable navigation and open-source implementation for further research.

Review of FAST-LIO: A LiDAR-Inertial Odometry Framework

This paper presents FAST-LIO, a rapidly executed and robust framework for LiDAR-inertial odometry. The paper focuses on efficiently integrating LiDAR feature points with Inertial Measurement Unit (IMU) data using a tightly-coupled iterated Kalman filter (iEKF), enabling significant improvements in environments subject to fast motion, noise, or clutter where potential degeneration can occur. The key innovation is a re-calculated Kalman gain that substantially reduces the computational burden by having the dependence on state dimension instead of the measurement dimension, making the integration process feasible in real-time applications.

Core Contributions

The paper highlights several core contributions:

1. Tightly-Coupled Iterated Kalman Filter: The framework utilizes an iEKF that couples LiDAR feature points with IMU measurements to deliver robust and real-time state estimation. A formal back-propagation process compensates for the motion distortion inherent in LiDAR data capture.
2. Computational Efficiency: A new formula for Kalman gain is proposed, having computation complexity depending on the state dimension instead of the measurement dimension. This improvement is critical as it allows the framework to handle LiDAR scans consisting of over 1,200 feature points while maintaining a real-time processing rate.
3. Practical Implementation: The implementation realizes robust and fast LiDAR-inertial odometry on a UAV platform. The authors provide comprehensive experiments across indoor and outdoor settings validating the system's real-time operation and robust navigation capabilities.
4. Open-source Accessibility: The codebase is openly available on GitHub, facilitating community engagement and further research advancements.

Results and Implications

The authors conducted experiments in both challenging indoor and outdoor environments to evaluate the performance of FAST-LIO. The proposed system consistently delivered reliable navigation outputs, completing iEKF steps within 25 ms while running on limited onboard computational resources indicative of many mobile robots, such as UAVs.

In the practical UAV flight test, FAST-LIO successfully provided accurate navigation paths as confirmed by direct trajectory comparisons and the built 3D environment maps. These results are particularly noteworthy given the challenging conditions, such as fast and diverse movement patterns, which typically challenge odometry solutions.

The computational efficiency tested against other existing formulations showed significant improvement. For example, the paper's proposed method drastically reduced the Kalman gain calculation time from milliseconds to sub-milliseconds, showcasing the potential of the new Kalman gain formula to significantly reduce load, allowing for hundreds of additional feature points to be processed without impacting real-time capability.

Theoretical and Practical Implications

The main theoretical implication of this research is the validation and formalization of the reduced-complexity Kalman gain calculation for iEKF in LiDAR-inertial applications. This efficiently bridges the gap between extensive LiDAR data and the traditionally computationally intensive filter-based integration methods.

Practically, the development of this framework provides a critical advance in autonomous navigation systems where resources are constrained, such as in small-scale UAVs. The ability to navigate accurately in fast-moving and dynamic environments, capture accurate 3D maps, and adapt to mixed lighting conditions opens new possibilities for deployment in industrial and exploratory missions.

Future Developments in AI and Robotics

Future research following this framework may explore pushing the boundaries of LiDAR-inertial fusion across more complex and extensive datasets. Additionally, adaptation to variable data rates and dynamic environmental changes could leverage machine learning models to enhance prediction and estimation accuracy. Furthermore, the framework's impact on multi-agent systems and cooperative navigation scenarios could prove transformational, leveraging the increased efficiency to facilitate synchronous operations with multiple UAVs or robots. The open-source release encourages modifications and iterations upon this promising baseline, inviting innovative improvements and applications in the field.