AC-LIO: Towards Asymptotic Compensation for Distortion in LiDAR-Inertial Odometry via Selective Intra-Frame Smoothing

TL;DR

AC-LIO addresses residual motion distortion in LiDAR-Inertial Odometry by introducing selective intra-frame smoothing with RTS-inspired backpropagation within a discrete-state EKF framework. The method adds convergence-guided asymptotic distortion compensation, controlled by a point-to-plane residual-based criterion, to improve frame-to-environment consistency without adding extra state variables. Across diverse datasets and sensors, AC-LIO achieves substantial accuracy gains (average RMSE reduction of about $30.4\%$ over the second-best method) while maintaining real-time performance comparable to strong baselines. This approach enhances long-term and large-scale localization and mapping, offering a robust and computationally efficient improvement to LIO systems.

Abstract

Existing LiDAR-Inertial Odometry (LIO) methods typically utilize the prior trajectory derived from the IMU integration to compensate for the motion distortion within LiDAR frames. However, discrepancies between the prior and true trajectory can lead to residual motion distortions that compromise the consistency of LiDAR frame with its corresponding geometric environment. This imbalance may result in pointcloud registration becoming trapped in local optima, thereby exacerbating drift during long-term and large-scale localization. To this end, we propose a novel LIO framework with selective intra-frame smoothing dubbed AC-LIO. Our core idea is to asymptotically backpropagate current update term and compensate for residual motion distortion under the guidance of convergence criteria, aiming to improve the accuracy of discrete-state LIO system with minimal computational increase. Extensive experiments demonstrate that our AC-LIO framework further enhances odometry accuracy compared to prior arts, with about 30.4% reduction in average RMSE over the second best result, leading to marked improvements in the accuracy of long-term and large-scale localization and mapping.

Figures (9)

• Figure 1: (a)A principal comparison of our AC-LIO with traditional approaches. Conventional LIO typically considers an initial distortion compensation via IMU, yet the residual motion distortion may prevent the registration from further convergence. Our AC-LIO employs a selective intra-frame smoothing strategy to asymptotically backpropagate and compensate for residual distortion. (b) Large scale localization and mapping result of BotanicGarden dataset. We achieve better accuracy performance than other benchmark frameworks on its sequence Garden_1008 with VLP16 LiDAR.
• Figure 2: (a)The system overview of our AC-LIO framework. During iterative ESKF, the back propagation and asymptotic compensation for pointcloud distortion are conducted based on the convergence criteria. (b)The on-manifold propagation. The blue shows the forward propagation of IMU integration. The green indicates the backward recursion of the update term according to the propagation matrix and noise relation of error state chains.
• Figure 3: The relationship between original ranging variance of LiDAR point and point-to-plane residual variance, where the residual $z$ is a scalar.
• Figure 4: Our handheld sensor suite for self-collected dataset, including solid-state LiDAR Livox Avia with built-in BMI088 IMU, and three other cameras.
• Figure 5: Accuracy evaluation in campus road scene. The figure shows the result of sequence Outdoor_1. Within the dashed box are, in order, the car, the streetlight, and the pattern on the ground. We keep running after initialization to enhance the motion intensity and complete a closed loop about 300m. The closure trajectory and the map consistency suggest that our AC-LIO achieves better performance.