A Robust Approach for LiDAR-Inertial Odometry Without Sensor-Specific Modeling (2509.06593v1)

Abstract: Accurate odometry is a critical component in a robotic navigation stack, and subsequent modules such as planning and control often rely on an estimate of the robot's motion. Sensor-based odometry approaches should be robust across sensor types and deployable in different target domains, from solid-state LiDARs mounted on cars in urban-driving scenarios to spinning LiDARs on handheld packages used in unstructured natural environments. In this paper, we propose a robust LiDAR-inertial odometry system that does not rely on sensor-specific modeling. Sensor fusion techniques for LiDAR and inertial measurement unit (IMU) data typically integrate IMU data iteratively in a Kalman filter or use pre-integration in a factor graph framework, combined with LiDAR scan matching often exploiting some form of feature extraction. We propose an alternative strategy that only requires a simplified motion model for IMU integration and directly registers LiDAR scans in a scan-to-map approach. Our approach allows us to impose a novel regularization on the LiDAR registration, improving the overall odometry performance. We detail extensive experiments on a number of datasets covering a wide array of commonly used robotic sensors and platforms. We show that our approach works with the exact same configuration in all these scenarios, demonstrating its robustness. We have open-sourced our implementation so that the community can build further on our work and use it in their navigation stacks.

• The paper introduces a method that eliminates sensor-specific modeling by leveraging a simplified IMU motion model and adaptive regularization.
• It integrates LiDAR scan-to-map ICP with IMU data averaging to achieve robust real-time pose estimation across varied conditions.
• Extensive experiments demonstrate performance comparable to state-of-the-art LIO systems on diverse datasets and platforms.

Robust LiDAR-Inertial Odometry Without Sensor-Specific Modeling

Introduction

This paper introduces a LiDAR-inertial odometry (LIO) system that eschews sensor-specific modeling, aiming for robust, accurate pose estimation across diverse sensor types, robotic platforms, and operational environments. The approach is motivated by the practical challenges of deploying LIO systems in real-world robotics, where sensor characteristics and mounting configurations vary widely, and calibration procedures for IMU noise and bias are often impractical. The proposed method leverages a simplified motion model for IMU integration and a scan-to-map ICP registration for LiDAR data, augmented by a novel adaptive regularization scheme. The system is designed to operate with a single configuration across all tested scenarios, requiring only the extrinsic calibration between sensors and the robot body.

Methodology

Sensor Data Preprocessing

The system operates in four frames: odometry ($\mathcal{O}$), LiDAR ($\mathcal{L}$), IMU ($\mathcal{I}$), and body base ($\mathcal{B}$). All sensor data are transformed into the body frame using provided extrinsics. IMU data transformation accounts for the transport-rate effect, ensuring correct acceleration and angular velocity representation in $\mathcal{B}$.

Motion Model

The core of the approach is a constant linear acceleration and angular velocity motion model between successive LiDAR frames. For a time interval $\Delta t$, the relative pose is estimated via Euler integration:

$$\Delta \mathbf{p} = \mathbf{v}(t)\, \Delta t + \frac{1}{2}\, \mathbf{a}\, \Delta t^2, \quad \Delta \mathbf{R} = \exp\left([\boldsymbol{\omega} \Delta t]_\times\right)$$

IMU measurements are averaged over the interval to estimate control inputs, reducing the impact of sensor noise. Biases are estimated during initialization, assuming static motion, and gravity alignment is performed using accelerometer data.

LiDAR Scan Registration

The system employs a scan-to-map ICP alignment, using a VDB voxel grid for efficient map representation. Each LiDAR scan is deskewed using the estimated motion model, downsampled, and registered against the local map. The ICP cost is defined as the mean squared residual between corresponding points.

Adaptive ICP Regularization

A key innovation is the addition of an orientation regularization term to the ICP cost, exploiting accelerometer measurements to constrain roll and pitch. The regularization is adaptively weighted based on the standard deviation of acceleration magnitude, $\sigma_a$, allowing the system to prefer LiDAR measurements when IMU data are noisy or the constant acceleration assumption is violated. The total cost minimized during registration is:

$$\chi(\mathbf{T}) = \chi_{\text{icp}}(\mathbf{T}) + \frac{1}{\beta} \chi_{\text{ori}}(\mathbf{T})$$

with $\beta = \beta_0 (1 + \sigma_a^2)$.

Experimental Evaluation

Datasets and Setup

The system is evaluated on multiple public and proprietary datasets, including Oxford Spires (backpack-mounted LiDAR-IMU), Leg-KILO (quadruped robot), HeLiPR (urban driving with sparse LiDARs), DigiForests (forestry, multiple LiDAR-IMU setups), and a long-range car dataset (up to 52 km). All experiments use a single configuration, with only sensor extrinsics provided.

Odometry Performance

Across all datasets, the proposed method achieves performance on par with or better than state-of-the-art LIO systems, including FAST-LIO2, DLIO, and KISS-ICP. Notably, the system maintains strong accuracy without sensor-specific tuning, even in challenging scenarios such as aggressive quadruped motion, long-range driving, and dense forest environments. On the Oxford Spires dataset, the method matches or exceeds the performance of SLAM systems in terms of ATE and RPE. In the car dataset, it demonstrates robustness over long trajectories, outperforming baselines in forest and rural sequences.

Ablation Studies

Ablation experiments confirm the efficacy of the motion model and adaptive regularization. Averaging IMU measurements and enabling regularization consistently improve performance. Disabling double-downsampling for sparse LiDARs yields further accuracy gains, with negligible runtime impact. The system's design allows trivial user configuration for such cases.

Computational Efficiency

The system operates faster than sensor frame rate across all tested datasets, supporting real-time deployment in navigation stacks.

Implications and Future Directions

The presented approach demonstrates that robust LIO can be achieved without detailed sensor-specific modeling or calibration, provided a simplified motion model and adaptive regularization are employed. This has significant practical implications for robotics, enabling rapid deployment across heterogeneous platforms and environments. The method's reliance on only extrinsic calibration and default parameters lowers the barrier for integration into navigation stacks.

Theoretically, the results suggest that the benefits of tightly-coupled filtering or factor graph approaches can be matched by scan-to-map registration with appropriate motion modeling and regularization. The adaptive weighting mechanism provides a principled way to balance exteroceptive and proprioceptive information based on observed sensor noise.

Future work may explore further generalization to multi-sensor fusion (e.g., visual-inertial-LiDAR), online adaptation of regularization parameters, and integration with loop closure modules for full SLAM capability. The open-source release and pip-installable package facilitate community benchmarking and extension.

Conclusion

This paper presents a robust, generalizable LiDAR-inertial odometry system that dispenses with sensor-specific modeling, relying on a simplified motion model and adaptive regularization. Extensive empirical evaluation demonstrates strong performance across diverse sensors, platforms, and environments, with minimal configuration. The approach offers a practical solution for real-world robotics, supporting accurate, real-time odometry without the need for costly calibration or tuning. The open-source implementation further enables adoption and future research in the field.