Existing LiDAR-Inertial Odometry (LIO) frameworks typically utilize prior state trajectories derived from IMU integration to compensate for the motion distortion within LiDAR frames, and demonstrate outstanding accuracy and stability in regular low-speed and smooth scenes. However, in high-speed or intense motion scenarios, the residual distortion may increase due to the limitation of IMU's accuracy and frequency, which will degrade the consistency between the LiDAR frame with its represented geometric environment, leading pointcloud registration to fall into local optima and consequently increasing the drift in long-time and large-scale localization. To address the issue, we propose a novel asymptotically and consistently converging LIO framework called AC-LIO. First, during the iterative state estimation, we backwards propagate the update term based on the prior state chain, and asymptotically compensate the residual distortion before next iteration. Second, considering the weak correlation between the initial error and motion distortion of current frame, we propose a convergence criteria based on pointcloud constraints to control the back propagation. The approach of guiding the asymptotic distortion compensation based on convergence criteria can promote the consistent convergence of pointcloud registration and increase the accuracy and robustness of LIO. Experiments show that our AC-LIO framework, compared to other state-of-the-art frameworks, effectively promotes consistent convergence in state estimation and further improves the accuracy of long-time and large-scale localization and mapping.

翻译：现有的激光雷达惯性里程计（LIO）框架通常利用从IMU积分推导出的先验状态轨迹来补偿激光雷达帧内的运动畸变，并在常规低速平滑场景中展现出卓越的精度与稳定性。然而，在高速或剧烈运动场景下，由于IMU精度与频率的限制，残余畸变可能增大，这将降低激光雷达帧与其所表征的几何环境之间的一致性，导致点云配准陷入局部最优，从而增加长时间、大规模定位中的漂移。为解决此问题，我们提出了一种新颖的、具有渐近且一致收敛特性的LIO框架，称为AC-LIO。首先，在迭代状态估计过程中，我们基于先验状态链反向传播更新项，并在下一次迭代前渐近地补偿残余畸变。其次，考虑到当前帧的初始误差与运动畸变之间的弱相关性，我们提出了一种基于点云约束的收敛判据来控制反向传播。这种基于收敛判据引导渐近畸变补偿的方法，能够促进点云配准的一致收敛，并提升LIO的精度与鲁棒性。实验表明，与其他先进框架相比，我们的AC-LIO框架能有效促进状态估计中的一致收敛，并进一步提升长时间、大规模定位与建图的精度。