Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
184 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

IMLS-SLAM: scan-to-model matching based on 3D data (1802.08633v1)

Published 23 Feb 2018 in cs.RO

Abstract: The Simultaneous Localization And Mapping (SLAM) problem has been well studied in the robotics community, especially using mono, stereo cameras or depth sensors. 3D depth sensors, such as Velodyne LiDAR, have proved in the last 10 years to be very useful to perceive the environment in autonomous driving, but few methods exist that directly use these 3D data for odometry. We present a new low-drift SLAM algorithm based only on 3D LiDAR data. Our method relies on a scan-to-model matching framework. We first have a specific sampling strategy based on the LiDAR scans. We then define our model as the previous localized LiDAR sweeps and use the Implicit Moving Least Squares (IMLS) surface representation. We show experiments with the Velodyne HDL32 with only 0.40% drift over a 4 km acquisition without any loop closure (i.e., 16 m drift after 4 km). We tested our solution on the KITTI benchmark with a Velodyne HDL64 and ranked among the best methods (against mono, stereo and LiDAR methods) with a global drift of only 0.69%.

Citations (293)

Summary

  • The paper introduces IMLS-SLAM, a novel 3D LiDAR SLAM algorithm utilizing an Implicit Moving Least Squares surface representation for robust scan-to-model matching.
  • The algorithm achieved a low global drift of 0.40% over a 4 km route and 0.69% on the KITTI benchmark, competitive with state-of-the-art methods like LOAM.
  • IMLS-SLAM demonstrates robustness in varied environments and operates without GPS or IMU, enhancing autonomous navigation deployment.

IMLS-SLAM: Advancements and Evaluations in 3D LiDAR-Based SLAM

The paper presents a novel SLAM (Simultaneous Localization and Mapping) algorithm specifically designed to utilize 3D LiDAR data for odometry, named IMLS-SLAM. This methodology introduces a scan-to-model matching framework that effectively addresses challenges related to data sparsity and processing volumes in LiDAR-based autonomous navigation systems. These challenges have historically limited the adoption of LiDAR-focused approaches in SLAM solutions.

Significance and Methodology

The core contribution of this research lies in its use of an Implicit Moving Least Squares (IMLS) surface representation in the scan-to-model matching framework. This representation allows for more effective handling of noise and sparsity in LiDAR data compared to traditional methods like Iterative Closest Point (ICP). By computing the normal from the denser portions of the data using the vehicle frame axes, the authors enhance the robustness of the pose estimation and the subsequent data alignment process.

Remarkably, the algorithm achieves a global drift of just 0.40% over a 4 km acquisition without employing loop closure strategies. Such precision highlights the algorithm's efficacy in minimizing drift, which is a significant issue in LiDAR-based systems due to their inherent data sparsity and environmental noise.

Experimental Evaluations

The authors present a comprehensive experimental evaluation of their system on various datasets, including their Velodyne HDL32 dataset and the public KITTI benchmark. Testing on their dataset in urban environments resulted in minimal drift, as highlighted by the accurate lane alignment in subsequent trajectory loops. Furthermore, they adapted their method for execution on the KITTI benchmark using Velodyne HDL64 data, yielding a drift of 0.69%, which is highly competitive with, and in some cases superior to, state-of-the-art methods such as LOAM, which reported a 0.64% drift in their updated evaluations.

Robustness and Parameter Sensitivity

The robustness of the IMLS-SLAM method was further demonstrated through sensitivity analysis of its numerous parameters, including the number of sampled points and the extent of overcome spans within the model. Findings indicated a significant advantage when incorporating dynamic object removal prior to matching, corroborating the method's strength in enhancing localization precision by reducing ambiguity in dynamic environments.

Likewise, the research underscores the flexibility of the IMLS framework in performing well across varied environments, from urban landscapes with distinct planar features to more unpredictable rural settings comprising significant vegetation and other potential noise-inducing elements.

Implications and Future Directions

The practical implications of this work are manifold, particularly in enhancing the deployment of autonomous vehicles in dynamically changing environments without the need for extensive pre-mapping. As the algorithm functions without reliance on integrated GPS or IMU data, it represents a critical step towards independent, real-time SLAM solutions capable of operating over large-scale environments.

Future research directions may include refinement towards real-time processing capabilities. Given the non-real-time implementation on the KITTI dataset, the paper suggests improvements in normal computation and optimized k-d trees as pathways forward. Additionally, integrating machine learning techniques for adaptive sampling strategies might further reduce computational overhead and improve matching reliability.

In summary, the IMLS-SLAM leverages an innovative approach to 3D LiDAR odometry, exhibiting substantial potential in both academic circles and practical applications related to autonomous vehicle navigation. The methodology not only signifies advancements in minimizing global drift in LiDAR-based SLAM but also provides a framework ripe for subsequent refinement and wider applicability.