Traj-LIO: A Resilient Multi-LiDAR Multi-IMU State Estimator Through Sparse Gaussian Process (2402.09189v1)

Abstract: Nowadays, sensor suits have been equipped with redundant LiDARs and IMUs to mitigate the risks associated with sensor failure. It is challenging for the previous discrete-time and IMU-driven kinematic systems to incorporate multiple asynchronized sensors, which are susceptible to abnormal IMU data. To address these limitations, we introduce a multi-LiDAR multi-IMU state estimator by taking advantage of Gaussian Process (GP) that predicts a non-parametric continuous-time trajectory to capture sensors' spatial-temporal movement with limited control states. Since the kinematic model driven by three types of linear time-invariant stochastic differential equations are independent of external sensor measurements, our proposed approach is capable of handling different sensor configurations and resilient to sensor failures. Moreover, we replace the conventional $\mathrm{SE}(3)$ state representation with the combination of $\mathrm{SO}(3)$ and vector space, which enables GP-based LiDAR-inertial system to fulfill the real-time requirement. Extensive experiments on the public datasets demonstrate the versatility and resilience of our proposed multi-LiDAR multi-IMU state estimator. To contribute to the community, we will make our source code publicly available.

• The paper presents a novel multi-LiDAR, multi-IMU state estimator that employs sparse Gaussian processes for continuous-time interpolation of sensor data.
• It innovatively replaces traditional SE(3) representation with SO(3) for rotation and vector space for translation, boosting resilience against sensor failures.
• The estimator is validated across handheld and aggressive UAV scenarios, demonstrating robust accuracy even when faced with sensor limitations.

Advancing LiDAR-Inertial Odometry through Sparse Gaussian Process in "Traj-LIO"

Introduction to "Traj-LIO"

The research presented in this paper introduces "Traj-LIO," a novel multi-LiDAR multi-IMU state estimator that significantly enhances the resiliency and versatility of sensor suites equipped with redundant LiDARs and IMUs. This advancement tackles common issues associated with sensor failure, leveraging Sparse Gaussian Process (GP) to predict continuous-time limited control states. Unlike conventional discrete-asynchronized LiDAR-inertial odometry (LIO) systems, Traj-LIO's kinematic model does not rely on external sensor measurements, thus offering robustness against sensor failures.

Key Contributions and Methodology

The paper articulates three primary contributions:

1. A versatile continuous-time multi-LiDAR multi-IMU state estimator that adapts well to a variety of sensor configurations. This estimator employs sparse GPs to accommodate asynchronous sensor data.
2. A self-driven kinematic model based on GP priors, which operates independently of IMU inputs, therefore enhancing system resilience.
3. A real-time capable GP-based LiDAR-inertial odometry method that simplifies the state space into rotation in SO(3) and translation in vector space, facilitating direct integration with IMU measurements.

To achieve these contributions, the researchers replaced the conventional SE(3) state representation with a combination of SO(3) for rotation and vector space for translation. This approach allowed for efficient GP-based interpolation and facilitated the handling of multiple asynchronized sensors.

Experiments and Results

The evaluation conducted on public datasets showcased Traj-LIO's effectiveness and generalizability from handheld scenarios to aggressive UAV situations. Particularly in environments where IMU data might exceed the sensor's measurement range, Traj-LIO demonstrated impressive resilience and maintained competitive accuracy. Furthermore, to contribute to the community, the authors committed to making their implementation publicly available.

Theoretical and Practical Implications

From a theoretical standpoint, Traj-LIO's introduction of a sparse GP for multi-LiDAR multi-IMU systems represents a significant step forward in continuous-time state estimation. Practically, this means robots equipped with Traj-LIO can better navigate and understand their environment, even in the face of sensor failures or limitations.

Looking Forward

The promising results of Traj-LIO open the door to further advancements in robotic state estimation and navigation. Future developments could explore the integration of additional sensor types, further enhance the system's resilience, and minimize computational requirements for real-time applications.

Conclusion

"Traj-LIO" represents a compelling advancement in the field of robotic state estimation, offering a resilient and versatile solution suited for complex navigation tasks. By effectively leveraging sparse GPs, this research showcases significant improvements over traditional LIO systems, ensuring robots can remain operational even in challenging environments.