A flexible framework for accurate LiDAR odometry, map manipulation, and localization (2407.20465v2)

Abstract: LiDAR-based SLAM is a core technology for autonomous vehicles and robots. One key contribution of this work to 3D LiDAR SLAM and localization is a fierce defense of view-based maps (pose graphs with time-stamped sensor readings) as the fundamental representation of maps. As will be shown, they allow for the greatest flexibility, enabling the posterior generation of arbitrary metric maps optimized for particular tasks, e.g. obstacle avoidance, real-time localization. Moreover, this work introduces a new framework in which mapping pipelines can be defined without coding, defining the connections of a network of reusable blocks much like deep-learning networks are designed by connecting layers of standardized elements. We also introduce tightly-coupled estimation of linear and angular velocity vectors within the Iterative Closest Point (ICP)-like optimizer, leading to superior robustness against aggressive motion profiles without the need for an IMU. Extensive experimental validation reveals that the proposal compares well to, or improves, former state-of-the-art (SOTA) LiDAR odometry systems, while also successfully mapping some hard sequences where others diverge. A proposed self-adaptive configuration has been used, without parameter changes, for all 3D LiDAR datasets with sensors between 16 and 128 rings, and has been extensively tested on 83 sequences over more than 250~km of automotive, hand-held, airborne, and quadruped LiDAR datasets, both indoors and outdoors. The system flexibility is demonstrated with additional configurations for 2D LiDARs and for building 3D NDT-like maps. The framework is open-sourced online: https://github.com/MOLAorg/mola

• The paper introduces a novel view-based mapping approach that creates diverse metric maps for real-time localization and obstacle avoidance.
• It presents a configurable pipeline inspired by deep learning networks, enabling users to define mapping processes without coding.
• The framework integrates dynamic variable optimization within the ICP loop, enhancing robustness and accuracy in aggressive motion scenarios without an IMU.

A Flexible Framework for Accurate LiDAR Odometry, Map Manipulation, and Localization

The paper "A flexible framework for accurate LiDAR odometry, map manipulation, and localization" by Jose Luis Blanco-Claraco presents a methodology and implementation for 3D LiDAR-based simultaneous localization and mapping (SLAM) tailored for autonomous vehicles and mobile robots. This work addresses prominent challenges in mapping using 3D LiDAR by introducing innovative concepts and robust solutions while maintaining flexibility and modularity.

The key contributions can be summarized into three primary aspects: a novel map representation, a flexible pipeline framework akin to deep learning network designs, and incorporating dynamic variable optimization within the iterative closest point (ICP) algorithm.

Core Contributions

1. View-based Maps: The proposed system utilizes view-based maps, also termed "simple maps," as the principal representation for environmental mapping. Such maps store sparse key-frames encapsulating raw sensor data and the vehicle's kinematic state at specific timestamps. This abstraction can generate diverse metric maps optimized for varied tasks, such as obstacle avoidance or real-time localization, by applying different pipeline configurations.
2. Flexible Mapping Pipelines: Inspired by the configurability of deep learning networks, this framework introduces a paradigm where users can define mapping pipelines without the need to write code. These pipelines consist of reusable blocks, enabling various data transformation processes such as down-sampling, metric map generation or update, dynamic object removal, temporal alignment adjustments, and more. This approach not only enhances reusability but also significantly reduces the development time for mapping applications.
3. Dynamic Variable Optimization in ICP: The inclusion of dynamic variable optimization within the ICP loop is another notable innovation. By incorporating vehicle kinematics as additional variables to be optimized, the system attains higher robustness against aggressive motion profiles. This improvement is achieved without relying on an inertial measurement unit (IMU), showcasing its efficiency in maintaining accurate pose estimates under challenging conditions.

Experimental Validation and Performance

The presented system, encapsulated under the name MOLA-LO, has undergone rigorous testing across various datasets and scenarios, encompassing a wide array of motion models, environments, and sensors ranging from high-resolution LiDARs with up to 128 rings to simpler 2D LiDARs. This extensive validation demonstrates its robustness and adaptability.

Quantitative Performance

• Relative Translational Error (RTE) and Relative Rotational Error (RRE):

Metrics like RTE and RRE were assessed on datasets such as KITTI and MulRan. MOLA-LO consistently performs on par with, or outperforms, existing state-of-the-art methods such as KISS-ICP and SIMPLE.

• Absolute Trajectory Error (ATE):

Evaluations on metrics such as ATE indicate the high precision of the proposed system. For example, in the KITTI dataset, MOLA-LO shows a competitive edge in some sequences when compared to other modern algorithms like SIMPLE.

This balance of accuracy and computational efficiency highlights the framework's potential for real-time applications, with most computations being faster than sensor rates, enabling real-time deployment on robotic platforms.

Practical Implications and Future Directions

This framework's modularity and flexible configuration make it particularly valuable for research and development in robotics. The ability to easily switch between different metric map representations and to debug ICP iterations in detail can accelerate development cycles and facilitate the testing of new hypotheses in SLAM research.

Potential Future Developments:

• Integration of Learning-Based Methods:

While the current system does not utilize learning-based methods, its highly modular design allows seamless integration of such techniques. For instance, machine learning algorithms could be incorporated for feature detection or to improve down-sampling strategies.

• Advanced Sensor Fusion:

Extending the system to incorporate additional sensor modalities, such as cameras or advanced IMUs, can further improve robustness and accuracy, especially in complex environments.

• Enhanced Loop Closure Detection:

Incorporating topological loop closure techniques based on visual or LiDAR descriptors, as mentioned in the paper, could significantly enhance the SLAM system's ability to detect global loop closures efficiently.

In conclusion, the paper presents a comprehensive solution to the existing challenges in 3D LiDAR-based SLAM by leveraging view-based maps, configurable mapping pipelines, and adaptive optimization strategies within the ICP framework. The versatility and performance of this system position it as a significant contribution to the field of robotics, opening avenues for future advancements and practical deployments in various autonomous applications.