We propose a lightweight and ground-optimized lidar odometry mapping method, LeGO-LOAM, for realtime six degree-of-freedom pose estimation with ground vehicles. LeGO-LOAM is lightweight, as it can achieve on low-power embedded system. ground-optimized, leverages the presence of plane in its segmentation optimization steps. first apply point cloud to filter out noise, feature extraction obtain distinctive planar edge features. A two-step Levenberg-Marquardt method then uses features solve...

We propose a framework for tightly-coupled lidar inertial odometry via smoothing and mapping, LIO-SAM, that achieves highly accurate, real-time mobile robot trajectory estimation map-building. LIO-SAM formulates lidar-inertial atop factor graph, allowing multitude of relative absolute measurements, including loop closures, to be incorporated from different sources as factors into the system. The estimated motion measurement unit (IMU) pre-integration de-skews point clouds produces an initial...

We propose a framework for tightly-coupled lidar-visual-inertial odometry via smoothing and mapping, LVI-SAM, that achieves real-time state estimation map-building with high accuracy robustness. LVI-SAM is built atop factor graph composed of two sub-systems: visual-inertial system (VIS) lidar-inertial (LIS). The sub-systems are designed in manner, which the VIS leverages LIS to facilitate initialization. improved by extracting depth information visual features using lidar measurements. In...

Inspection of ship hulls and marine structures using autonomous underwater vehicles has emerged as a unique challenging application robotics. The problem poses rich questions in physical design operation, perception navigation, planning, driven by difficulties arising from the acoustic environment, poor water quality highly complex to be inspected. In this paper, we develop apply algorithms for central navigation planning problems on hulls. These divide into two classes, suitable open,...

In this paper we address the problem of drift-free navigation for underwater vehicles performing harbor surveillance and ship hull inspection. Maintaining accurate localization duration a mission is important variety tasks, such as planning vehicle trajectory ensuring coverage area to be inspected. Our approach only uses onboard sensors in simultaneous mapping setting removes need any external infrastructure like acoustic beacons. We extract dense features from forward-looking imaging sonar...

We discuss the problem of inspecting an underwater structure, such as a submerged ship hull, with autonomous vehicle (AUV). Unlike large body prior work, we focus on planning views AUV to improve quality inspection, rather than maximizing accuracy given data stream. formulate inspection extension Bayesian active learning, and show connections recent theoretical guarantees in this area. rigorously analyze benefit adaptive re-planning for problems, prove that potential adaptivity can be...

We present several new contributions in sampling-based coverage path planning, the task of finding feasible paths that give 100% sensor complex structures obstaclefilled and visually occluded environments. First, we establish a framework for analyzing probabilistic completeness algorithm, derive results on convergence existing algorithms. Second, introduce algorithm iterative improvement path; this relies samplingbased subroutine makes asymptotically optimal local improvements to based...

We consider an autonomous exploration problem in which a mobile robot is guided by information-based controller through priori unknown environment, choosing to collect its next measurement at the location estimated be most informative within current field of view. propose novel approach predict mutual information (MI) using Bayesian optimization. Over several iterations, candidate sensing actions are suggested optimization and added committee that repeatedly trains Gaussian process (GP). The...

We propose a novel framework for distributed,multi-robot SLAM intended use with 3D LiDAR observations. The framework, DiSCo-SLAM, is the first to lightweight Scan Context descriptor multi-robot SLAM, permitting data-efficient exchange of observations among robots. Additionally, our includes two-stage global and local optimization distributed which provides stable localization results that are resilient unknown initial conditions typify search inter-robot loop closures. compare proposed...

To support autonomous, in-water inspection of a ship hull, we propose and implement new techniques for coverage path planning over complex 3D structures. Our main contribution is comprehensive methodology sampling-based design routes, including an algorithm planning, smoothing, analysis probabilistic completeness. The latter two outcomes are the first their kind in area planning. algorithms give high-quality solutions expansive structures, demonstrate this with experiments laboratory on 75 m...

We present a novel algorithm to produce descriptive online 3D occupancy maps using Gaussian processes (GPs). GP regression and classification have met with recent success in their application robot mapping, as GPs are capable of expressing rich correlation among map cells sensor data. However, the cubic computational complexity has limited its large-scale mapping use. In this paper we address issue first by proposing test-data octrees, octrees within blocks that prune away nodes same state,...

We consider an autonomous exploration problem in which a range-sensing mobile robot is tasked with accurately mapping the landmarks priori unknown environment efficiently real-time; it must choose sensing actions that both curb localization uncertainty and achieve information gain. For this problem, belief space planning methods forward- simulate estimation may often fail real-time implementation, scaling poorly increasing size of state, action spaces. propose novel approach uses graph...

We propose a methodology for robust, real-time place recognition using an imaging lidar, which yields image-quality high-resolution 3D point clouds. Utilizing the intensity readings of we project cloud and obtain image. ORB feature descriptors are extracted from image encoded into bag-of-words vector. The vector, used to identify cloud, is inserted database that maintained by DBoW fast queries. returned candidate further validated matching visual descriptors. To reject outliers, apply PnP,...

We discuss the problem of inspecting an underwater structure, such as a submerged ship hull, with autonomous vehicle (AUV). In scenarios, goal is to construct accurate 3D model structure and detect any anomalies (e.g., foreign objects or deformations). propose method for constructing meshes from sonar-derived point clouds that provides watertight surfaces, we introduce uncertainty modeling through non-parametric Bayesian regression. Uncertainty novel cost functions planning path AUV minimize...

We consider an autonomous mapping and exploration problem in which a range-sensing mobile robot is guided by information-based controller through priori unknown environment, choosing to collect its next measurement at the location estimated yield maximum information gain within current field of view. propose novel time-efficient approach predict most informative sensing action using deep neural network. After training network on series thousands randomly-generated "dungeon maps", predicted...

We consider the problem of autonomous mobile robot exploration in an unknown environment, taking into account a robot's coverage rate, map uncertainty and state estimation uncertainty. In this article, we present novel framework for underwater robots operating cluttered environments, built upon simultaneous localization mapping with imaging sonar. The proposed system comprises path generation, place recognition forecasting, belief propagation utility evaluation using <italic...

Simultaneous localization and mapping (SLAM) is a critical capability for any autonomous underwater vehicle (AUV). However, robust, accurate state estimation still work in progress when using low-cost sensors. We propose enhancing typical sensor package widely available often free prior information; overhead imagery. Given an AUV's sonar image partially overlapping, globally-referenced image, we convolutional neural network (CNN) to generate synthetic predicting the above-surface appearance...

We present a novel path planning algorithm that, starting from probabilistic roadmap, efficiently constructs product graph used to search for near optimal solution of multiobjective optimization problem. The goal is find paths that minimize primary cost, such as the length start goal, subject bound on secondary cost state estimation error covariance. proposed efficient it relies scalar metric, related largest eigenvalue covariance, and adaptively quantizes yielding whose number vertices...

In this paper, we consider the problem of building descriptive three-dimensional (3-D) maps from sparse and noisy range sensor data. We expand our previously proposed method leveraging Bayesian kernel inference for prediction occupancy in locations not directly observed by a sensor. show that approach generalizes previous "counting model" approaches discrete grids to continuous maps. Our enables about regions unobserved based on local measurements, smoothly transitions prior lacking...

We introduce an algorithm to achieve complete sensor coverage of complex, three-dimensional structures surveyed by autonomous agent with multiple degrees freedom. Motivated the application ocean vehicle performing ship hull inspection, we consider a planning problem for fully-actuated, six degree-of-freedom hovering AUV using bathymetry sonar inspect complex underneath hull. discrete model structure be inspected, requiring only that provided in form closed triangular mesh. A dense graph...

We consider the problem of building accurate and descriptive 3D occupancy maps an environment from sparse noisy range sensor data. seek to accomplish this task by constructing a predictive model online inferring probability regions we have not directly observed. propose novel algorithm leveraging recent advances in data structures for mapping, kernels, Bayesian nonparametric inference. The resulting inference has several desirable properties comparison existing methods, including speed...

We present a novel formulation of Hilbert mapping in which we construct global occupancy map by incrementally fusing local overlapping maps. Rather than maintain single supervised learning model for the entire map, new is trained with each robot's range scans, and queried at all points within perceptual field. treat probabilistic output classifier as sensor, employing sensor fusion to merge This allows be used real-world scenarios overlap between observations. The methodology applied...