Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Lidar-only Odometry based on Multiple Scan-to-Scan Alignments over a Moving Window

Aaron Kurda, Simon Steuernagel, Marcus Baum

TL;DR

This paper tackles lidar-only odometry by mitigating drift through a moving-window framework of multiple independent scan-to-scan ICP registrations that establish a dense set of pose constraints. These constraints feed a pose-graph optimization that yields accurate current poses and smooths past estimates, without the computational burden of repeatedly recomputing local maps. The approach combines scan preprocessing with a constant-motion predictor, robust ICP, and a g2o-based Levenberg–Marquardt optimization to maintain a consistent trajectory. Evaluations on KITTI, MulRan, and a custom dataset show competitive or superior performance in most sequences, highlighting strong performance in automotive scenarios and pointing to robustness improvements in feature-sparse environments as future work.

Abstract

Lidar-only odometry considers the pose estimation of a mobile robot based on the accumulation of motion increments extracted from consecutive lidar scans. Many existing approaches to the problem use a scan-to-map registration, which neglects the accumulation of errors within the maintained map due to drift. Other methods use a refinement step that jointly optimizes the local map on a feature basis. We propose a solution that avoids this by using multiple independent scan-to-scan Iterative Closest Points (ICP) registrations to previous scans in order to derive constraints for a pose graph. The optimization of the pose graph then not only yields an accurate estimate for the latest pose, but also enables the refinement of previous scans in the optimization window. By avoiding the need to recompute the scan-to-scan alignments, the computational load is minimized. Extensive evaluation on the public KITTI and MulRan datasets as well as on a custom automotive lidar dataset is carried out. Results show that the proposed approach achieves state-of-the-art estimation accuracy, while alleviating the mentioned issues.

Lidar-only Odometry based on Multiple Scan-to-Scan Alignments over a Moving Window

TL;DR

This paper tackles lidar-only odometry by mitigating drift through a moving-window framework of multiple independent scan-to-scan ICP registrations that establish a dense set of pose constraints. These constraints feed a pose-graph optimization that yields accurate current poses and smooths past estimates, without the computational burden of repeatedly recomputing local maps. The approach combines scan preprocessing with a constant-motion predictor, robust ICP, and a g2o-based Levenberg–Marquardt optimization to maintain a consistent trajectory. Evaluations on KITTI, MulRan, and a custom dataset show competitive or superior performance in most sequences, highlighting strong performance in automotive scenarios and pointing to robustness improvements in feature-sparse environments as future work.

Abstract

Lidar-only odometry considers the pose estimation of a mobile robot based on the accumulation of motion increments extracted from consecutive lidar scans. Many existing approaches to the problem use a scan-to-map registration, which neglects the accumulation of errors within the maintained map due to drift. Other methods use a refinement step that jointly optimizes the local map on a feature basis. We propose a solution that avoids this by using multiple independent scan-to-scan Iterative Closest Points (ICP) registrations to previous scans in order to derive constraints for a pose graph. The optimization of the pose graph then not only yields an accurate estimate for the latest pose, but also enables the refinement of previous scans in the optimization window. By avoiding the need to recompute the scan-to-scan alignments, the computational load is minimized. Extensive evaluation on the public KITTI and MulRan datasets as well as on a custom automotive lidar dataset is carried out. Results show that the proposed approach achieves state-of-the-art estimation accuracy, while alleviating the mentioned issues.

Paper Structure

This paper contains 15 sections, 13 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Contribution
  3. Related Work
  4. Method
  5. Scan Preprocessing
  6. Motion Prediction
  7. Iterative Closest Point
  8. Multiple Scan-to-Scan Registration
  9. Smoothing via Pose Graph Optimization
  10. Keyframe Management
  11. Evaluation
  12. KITTI
  13. MulRan
  14. Custom Dataset
  15. Conclusion

Figures (2)

  • Figure 1: Schematic representation of (a): naive scan-to-scan (b): scan-to-map and (c): our proposed registration method for the estimation of the current pose $\mathbf{x}_5$. Dashed lines indicate motion constraints from previous time steps.
  • Figure 2: Representation of the pose-graph constructed by our method. Blue nodes are the keyframes used for registration. A scan ($\mathbf{x}_7$) is registered against all previous keyframes inside the interval marked with $\gamma$. The constraints of previous time steps are displayed with dashed lines and constraints of the most recent time step are displayed as solid lines. The node $\mathbf{x}_1$ is outside the local window of $\mathbf{x}_7$. It is not used for registration and its state is fixed.