Learning-based Localizability Estimation for Robust LiDAR Localization

Abstract

LiDAR-based localization and mapping is one of the core components in many modern robotic systems due to the direct integration of range and geometry, allowing for precise motion estimation and generation of high quality maps in real-time. Yet, as a consequence of insufficient environmental constraints present in the scene, this dependence on geometry can result in localization failure, happening in self-symmetric surroundings such as tunnels. This work addresses precisely this issue by proposing a neural network-based estimation approach for detecting (non-)localizability during robot operation. Special attention is given to the localizability of scan-to-scan registration, as it is a crucial component in many LiDAR odometry estimation pipelines. In contrast to previous, mostly traditional detection approaches, the proposed method enables early detection of failure by estimating the localizability on raw sensor measurements without evaluating the underlying registration optimization. Moreover, previous approaches remain limited in their ability to generalize across environments and sensor types, as heuristic-tuning of degeneracy detection thresholds is required. The proposed approach avoids this problem by learning from a collection of different environments, allowing the network to function over various scenarios. Furthermore, the network is trained exclusively on simulated data, avoiding arduous data collection in challenging and degenerate, often hardto-access, environments. The presented method is tested during field experiments conducted across challenging environments and on two different sensor types without any modifications. The observed detection performance is on par with state-ofthe-art methods after environment-specific threshold tuning 1 .

X-ICP: Localizability-Aware LiDAR Registration for Robust Localization in Extreme Environments

“…This work has been adopted by multiple LiDAR-based SLAM frameworks [28], [48]- [51] and is considered state-ofthe-art; however, certain aspects can limit its efficacy. i) Being binary in nature, the method depends on the heuristic tuning of thresholds for operation in different environments [14]. ii) As eigenvalues represent the scale of their respective eigenvectors, thresholds for translation and rotation cannot be represented by a singular value.…”

“…Although successful, these methods rely on extensive ground truth data for learning and are unsuitable for real-time operation. To alleviate reliance on data, [14] proposes to leverage simulation for training and only consider the current LiDAR scan to predict a 6-DoF localizability metric. The authors use sparse 3D convolutions to show generalization across different sensors and environments through real-world experiments.…”

“…1. As discussed in recent research [14]- [16], to enable robots to operate in these challenging scenarios, and to handle the solution ill-conditioning caused by degenerate planar or tunnel-like environments, localizability-awareness is required. c) Proposed Approach: As a response to the LiDAR degeneracy problem, this work proposes a robust localizabilityaware point cloud registration (ICP) framework that enables LiDAR-based SLAM systems to operate in featureless eXtreme environments, called X-ICP.…”

“…The Loc.-Module utilizes the point and surface-normal correspondences between the scan and the map to analyze the contribution strength along the optimization's principal directions. In contrast to prior work [14], the localizability detection is applied to the scan-to-map registration, and the localizability detection is done in the optimization eigenspace, which allows the detection to be independent of the sensor orientation in the map. The resulting proposed solution enables reliable localizability detection in various environmental configurations, such as underground environments and large open spaces, without parameter tuning.…”

Dense 3D Reconstruction Through Lidar: A New Perspective on Computer-Integrated Surgery

“…In particular, lidar sensors provide long-range, highaccuracy measurements of the 3D environment. Research and commercial applications for lidar in mobile robotics span from autonomous cars [8], aerial vehicles [9], [10], and construction machines [11] to more recent frameworks for deployment on legged robots [12]. While sensor size and weight matter less in the aforementioned applications, the introduction of such technology in surgery has mostly been limited by device size, heat production, and minimum imaging distance.…”