CPBA-LIWO: Continuous-Time LiDAR-Inertial-Wheel Odometry Based on Probabilistic Bundle Adjustment
Song Wu, Yunzhou Zhang, Yuezhang Lv, Wu Li, Sizhan Wang, Su Yan, Hengwang Ding
AI summary
Problem
Existing LiDAR odometry methods suffer from accumulated errors, motion distortion, and reduced accuracy in structurally degraded or long-distance GPS-denied environments due to unmodeled LiDAR observational uncertainties and rigid sensor fusion.
Approach
The method models robot motion with a continuous-time B-spline trajectory and fuses LiDAR, IMU, and wheel encoder data within a sliding-window probabilistic bundle adjustment backend that weights voxel plane constraints by observational uncertainty.
Key results
- Continuous-time LIW odometry framework with B-spline trajectory and adaptive wheel scale constraint
- Probabilistic voxel plane model enabling uncertainty-aware sliding-window bundle adjustment
- Wheel-assisted dynamic-static IMU initialization for reliable system startup
- State-of-the-art accuracy and robustness on M2DGR-plus and KAIST datasets
Why it matters
Enables reliable long-distance navigation and mapping for ground robots in GPS-denied, structurally challenging environments where traditional LiDAR odometry fails.
Abstract
LiDAR-based odometry is widely used in ground robot localization. However, current methods encounter chal- lenges in accuracy and robustness due to structural degra- dation, system observational error, and accumulated error. To address the above issues, we propose CPBA-LIWO, a continuous-time LiDAR-Inertial-Wheel (LIW) odometry based on probabilistic bundle adjustment (PBA) within a sliding window. This method constructs a general wheel model, which is used for the complementary fusion of LiDAR, IMU and wheel data through a continuous-time trajectory using a B-spline curve, thereby improving the robustness of the system in struc- turally degraded environments. Furthermore, to improve the accuracy of long-distance odometry, we propose a probabilistic model for the voxel plane and implement a sliding-window voxel PBA backend based on this model. The experimental results on the M2DGR-plus and KAIST datasets demonstrate that our method outperforms state-of-the-art LiDAR-based odometry in terms of accuracy and robustness.