Localizing a camera in a given map is essential for vision-based navigation. In contrast to common methods for visual localization that use maps acquired with cameras, we propose a novel approach, which tracks the pose of monocular camera with respect to a given 3D LiDAR map. We employ a visual odometry system based on local bundle adjustment to reconstruct a sparse set of 3D points from image features. These points are continuously matched against the map to track the camera pose in an online fashion. Our approach to visual localization has several advantages. Since it only relies on matching geometry, it is robust to changes in the photometric appearance of the environment. Utilizing panoramic LiDAR maps additionally provides viewpoint invariance. Yet low-cost and lightweight camera sensors are used for tracking. We present real-world experiments demonstrating that our method accurately estimates the 6-DoF camera pose over long trajectories and under varying conditions.

Observing human activities can reveal a lot about the structure of the environment, the objects contained therein and also their functionality. This knowledge, in turn, can be useful for robots interacting with humans or for robots performing mobile manipulation tasks. In this paper, we present an approach to infer the geometric and functional structure of the environment and the position of certain relevant objects in it from human activity. We observe this activity using a full-body motion capture system consisting of a set of inertial measurement units. This is a hard problem since our data suit provides odometry estimates only, which severely drift over time. Therefore, we regard the objects inferred from the activities as landmarks in a graph-based simultaneous localization and mapping problem, which we optimize to obtain accurate estimates about the poses of the objects and the trajectory of the human. In extensive experiments, we demonstrate the effectiveness of the proposed method for the reconstruction of 3D representations. The resulting models not only contain a geometric but also a functional description of the environment and naturally provide a segmentation into individual objects.

Networks of laser range finders are a popular tool for monitoring large cluttered areas and to track people. Whenever multiple scanners are used for this purpose, one major problem is how to determine the relative positions of all the scanners. In this paper, we present a novel approach to calibrate a network of multiple planar laser range finders scanning horizontally. To robustly deal with the potentially restricted overlap between the fields of view, our approach only requires a dynamic object, e.g., a person, moving through the observed area. We employ a RANSAC-like algorithm to find the correspondences between the measurements of the different laser range finders. Based on these correspondences we formulate a graph-based optimization problem to determine the maximum likelihood extrinsic parameters of the sensor network. Furthermore, we present a method to evaluate the consistency of the resulting calibration based on visibility constraints. Experiments on real and simulated data show that the proposed approach yields better results than techniques that only perform pairwise calibration.

To determine the pose of a vehicle is a fundamental problem in mobile robotics. Most approaches relate the current sensor observations to a map generated with previously acquired data of the same system or by another system with a similar sensor setup. Unfortunately, previously acquired data is not always available. In outdoor settings, GPS is a very useful tool to determine a global estimate of the vehicles pose. Unfortunately, GPS tends to be unreliable in situations in which a clear view to the sky is restricted. Yet, one can make use of publicly available map material as prior information. In this paper, we describe an approach to localize a robot equipped with a 3D range scanner with respect to a road network created from OpenStreetMap data. To successfully localize a mobile robot we propose a road classification scheme for 3D range data together with a novel sensor model, which relates the classification results to a road network. Compared to other approaches, our system does not require the robot to actually travel on the road network. We evaluate our approach in extensive experiments on simulated and real data and compare favorably to two state-of-the-art methods on those data.

Laser range scanners are commonly used in mobile robotics to enable a robot to sense the spatial configuration of its environment. In addition to the range measurements, most scanners provide remission values, representing the intensity of the returned light pulse. These values add a visual component to the measurement and can be used to improve reasoning on the data. Unfortunately, a remission value does not directly tell us how bright a measured surface is in the infrared spectrum. Rather, it varies with respect to the incidence angle and the range at which it was measured. In addition, multiple scanners typically do not agree upon the values of a certain surface. In this paper, we present a calibration method for remission values of multiple laser scanners considering dependencies in range, incidence angle of the measured surface, and the respective scanner unit. Our system learns the calibration parameters based on a set of registered point clouds. It uses a graph optimization scheme to minimize the error between different measurements, so that all involved scanners yield consistent reflection values, independent of the perspective from which the corresponding surface is observed.

In this paper, we present an appearance-based visual SLAM approach that focuses on detecting loop closures across seasons. Given two image sequences, our method first extracts one descriptor per image for both sequences using a deep convolutional neural network. Then, we compute a similarity matrix by comparing each image of a query sequence with a database. Finally, based on the similarity matrix, we formulate a flow network problem and compute matching hypotheses between sequences. In this way, our approach can handle partially matching routes, loops in the trajectory and different speeds of the robot. With a matching hypothesis as loop closure information and the odometry information of the robot, we formulate a graph based SLAM problem and compute a joint maximum likelihood trajectory.

In this paper, we describe a novel approach to construct textured 3D environment models in a hierarchical fashion based on local surface patches. Compared to previous approaches, the hierarchy enables our method to represent the environment with differently sized surface patches. The reconstruction scheme starts at a coarse resolution with large patches and in an iterative fashion uses the reconstruction error to guide the decision as to whether the resolution should be refined. This leads to variable resolution models that represent areas with few variations at low resolution and areas with large variations at high resolution. In addition, we compactly describe local surface attributes via sparse coding based on an overcomplete dictionary. In this way, we additionally exploit similarities in structure and texture, which leads to compact models. We learn the dictionary directly from the input data and independently for every level in the hierarchy in an unsupervised fashion. Practical experiments with large-scale datasets demonstrate that our method compares favorably with two state-of-the-art techniques while being comparable in accuracy.

In this paper, we describe a novel approach to construct compact colored 3D environment models representing local surface attributes via sparse coding. Our method decomposes a set of colored point clouds into local surface patches and encodes them based on an overcomplete dictionary. Instead of storing the entire point cloud, we store a dictionary, surface patch positions, and a sparse code description of the depth and RGB attributes for every patch. The dictionary is learned in an unsupervised way from surface patches sampled from indoor maps. We show that better dictionaries can be learned by extending the K-SVD method with a binary weighting scheme that ignores undefined surface cells. Through experimental evaluation on real world laser and RGBD datasets we demonstrate that our method produces compact and accurate models. Furthermore, we clearly outperform an existing state of the art method in terms of compactness, accuracy, and computation time. Additionally, we demonstrate that our sparse code descriptions can be utilized for other important tasks such as object detection.

Over the past years, there has been a tremendous progress in the area of robot navigation. Most of the systems developed thus far, however, are restricted to indoor scenarios, non-urban outdoor environments, or road usage with cars. Urban areas introduce numerous challenges to autonomous mobile robots as they are highly complex and in addition to that dynamic. In this paper, we present a navigation system for pedestrian-like autonomous navigation with mobile robots in city environments. We describe different components including a SLAM system for dealing with huge maps of city centers, a planning approach for inferring feasible paths taking also into account the traversability and type of terrain, and a method for accurate localization in dynamic environments. The navigation system has been implemented and tested in several large-scale field tests in which the robot Obelix managed to autonomously navigate from our university campus over a 3.3 km long route to the city center of Freiburg.

In this paper, we propose an approach to obtain highly accurate 3D models from range data. The key idea of our method is to jointly optimize the poses of the sensor and the positions of the surface points measured with a range scanning device. Our approach applies a physical model of the underlying range sensor. To solve the optimization task it employs a state-of-the-art graph-based optimizer and iteratively refines the structure of the error function by recomputing the data associations after each optimization. We present our approach and evaluate it on data recorded in different real world environments with a RGBD camera and a laser range scanner. The experimental results demonstrate that our method is able to substantially improve the accuracy of SLAM results and that it compares favorable over the moving least squares method.

In this paper we describe an algorithm for learning highly accurate laser-based maps that treats the overall mapping problem as a joint optimization problem over robot poses and laser points. We assume that a laser range finder senses points sampled from a regular surface and we utilize an improved likelihood function that accounts for two phenomena affecting the laser measurements that are often neglected: the conic shape of the laser beam and the incidence angle. To solve the entire problem we apply an optimization procedure that jointly adjusts the position of all the robot poses and all points in the scans. As a result, we obtain highly accurate maps. We evaluated our approach using simulated and real-world data and we show that utilizing the estimated maps greatly improves the localization accuracy of robots. The results furthermore suggest that the accuracy of the resulting map can exceed the resolution of the laser sensors used.

Place recognition, i.e., the ability to recognize previously seen parts of the environment, is one of the fundamental tasks in mobile robotics. The wide range of applications of place recognition includes localization (determine the initial pose), SLAM (detect loop closures), and change detection in dynamic environments. In the past, only relatively little work has been carried out to attack this problem using 3D range data and the majority of approaches focuses on detecting similar structures without estimating relative poses. In this paper, we present an algorithm based on 3D range data that is able to reliably detect previously seen parts of the environment and at the same time calculates an accurate transformation between the corresponding scan-pairs. Our system uses the estimated transformation to evaluate a candidate and in this way to more robustly reject false positives for place recognition. We present an extensive set of experiments using publicly available datasets in which we compare our system to other state-of-the-art approaches.

In this paper we describe a novel algorithm for constructing a compact representation of 3D laser range data. Our approach extracts an alphabet of local scans from the scene. The words of this alphabet are used to replace recurrent local 3D structures, which leads to a substantial compression of the entire point cloud. We optimize our model in terms of complexity and accuracy by minimizing the Bayesian information criterion (BIC). Experimental evaluations on large real-world data show that our method allows robots to accurately reconstruct environments with as few as 70 words.

We present an algorithm for learning 3D object models from partial object observations. The input to our algorithm is a sequence of 3D laser range scans. Models learned from the objects are represented as point clouds. Our approach can deal with partial views and it can robustly learn accurate models from complex scenes. It is based on an iterative matching procedure which attempts to recursively merge similar models. The alignment between models is determined using a novel scan registration procedure based on range images. The decision about which models to merge is performed by spectral clustering of a similarity matrix whose entries represent the consistency between different models.

In this paper, we address the problem of creating an objective benchmark for comparing SLAM approaches. We propose a framework for analyzing the results of SLAM approaches based on a metric for measuring the error of the corrected trajectory. The metric uses only relative relations between poses and does not rely on a global reference frame. The idea is related to graph-based SLAM approaches in the sense that it considers the energy needed to deform the trajectory estimated by a SLAM approach to the ground truth trajectory. Our method enables us to compare SLAM approaches that use different estimation techniques or different sensor modalities since all computations are made based on the corrected trajectory of the robot. We provide sets of relative relations needed to compute our metric for an extensive set of datasets frequently used in the SLAM community. The relations have been obtained by manually matching laser-range observations. We believe that our benchmarking framework allows the user an easy analysis and objective comparisons between different SLAM approaches.