Udo Frese - Academia.edu (original) (raw)

Papers by Udo Frese

2019 International Conference on Indoor Positioning and Indoor Navigation (IPIN), 2019

Inertial Navigation Systems suffer from unbounded errors on the position and orientation estimate... more Inertial Navigation Systems suffer from unbounded errors on the position and orientation estimate. This drift can be corrected by applying prior knowledge, instead of using exteroceptive sensors. Analysing the state observability induced by prior knowledge motivates us to track bikers in track cycling races. In this paper, we show that the pose of the bikers can be estimated with an IMU as the only sensor by using a heightmap of the track and the knowledge that the biker drives forward. We present a dataset with three 60-round trials and evaluate the state estimate. We show that the influences of the priors match the expectation derived from state observability analysis.

2020 IEEE 23rd International Conference on Information Fusion (FUSION), 2020

Control, tracking, and obstacle detection algorithms for mobile robots, including autonomous cars... more Control, tracking, and obstacle detection algorithms for mobile robots, including autonomous cars, rely on a jump-free estimate of the vehicle's pose. While one cannot completely avoid jumps in global solutions like INS/GNSS and SLAM, relative localization (i.e., odometry) does not suffer from this problem. Methods based on graph optimization are popular in that field, but they do not scale very well with high-frequency measurements. Kalman filters (KFs) are able to cope with those measurements, but they face the issue of a continuously growing covariance. This results in instabilities and eventually jumps in the state estimate. We present an approach to handle this problem by periodically moving the reference state forward in time, which is realized using two filters. The equations for implementing this in both the extended Kalman filter (EKF) and the unscented Kalman filter (UKF) are derived. The algorithm is evaluated using real-world datasets covering different scenarios of ...

Sensors (Basel, Switzerland), 2021

Hybrid systems are subject to multiple dynamic models, or so-called modes. To estimate the state,... more Hybrid systems are subject to multiple dynamic models, or so-called modes. To estimate the state, the sequence of modes has to be estimated, which results in an exponential growth of possible sequences. The most prominent solution to handle this is the interacting multiple model filter, which can be extended to smoothing. In this paper, we derive a novel generalization of the interacting multiple filter and smoother to manifold state spaces, e.g., quaternions, based on the boxplus-method. As part thereof, we propose a linear approximation to the mixing of Gaussians and a Rauch–Tung–Striebel smoother for single models on boxplus-manifolds. The derivation of the smoother equations is based on a generalized definition of Gaussians on boxplus-manifolds. The three, novel algorithms are evaluated in a simulation and perform comparable to specialized solutions for quaternions. So far, the benefit of the more principled approach is the generality towards manifold state spaces. The evaluatio...

In this paper, we present a novel method of incorporating dense (e.g., depth, RGB-D) data in a ge... more In this paper, we present a novel method of incorporating dense (e.g., depth, RGB-D) data in a general purpose least-squares graph optimization framework. Rather than employing a loosely coupled, layered design where dense data is first used to estimate a compact SE(3) transform which then forms a link in the optimization graph as in previous approaches [28, 10, 26], we use a tightly coupled approach that jointly optimizes over each individual (i.e. per-pixel) dense measurement (on the GPU) and all other traditional sparse measurements (on the CPU). Concretely, we use Kinect depth data and KinectFusion-style point-to-plane ICP measurements. In particular, this allows our approach to handle cases where neither dense, nor sparse measurements separately define all degrees of freedom (DoF) while taken together they complement each other and yield the overall maximum likelihood solution. Nowadays it is common practice to flexibly model various sensors, measurements and to be estimated va...

The calibration of cameras is a crucial step in machine vision and usually relies on an accurate ... more The calibration of cameras is a crucial step in machine vision and usually relies on an accurate detection and localization of calibration patterns in images. Therefore, checkerboards are often used, allowing precise subpixel estimation of their corners. However, noise in localization generates a proportional noise in the derived model parameters. Therefore, it is important that the localization has a certain robustness against image noise. This is even more important for deteriorated imaging conditions strongly affecting subpixel detectors. This paper presents a new checkerboard corner detector based on a localized Radon transform implemented by large box filters making it robust to low contrast, image noise, and blur while maintaining high subpixel accuracy.

This papers provides two contributions to the problem of Simultaneous Localization and Mapping (S... more This papers provides two contributions to the problem of Simultaneous Localization and Mapping (SLAM): First we discuss properties of the problem itself and of the intended semantics of an uncertain map representation, with the main idea of “representing certainty of relations despite the uncertainty of positions”. We propose some requirements an ideal solution of SLAM should have concerning uncertainty, memory space and computation time and discuss existing approaches in the light of these requirements. The second part proposes a representation based on sparse information matrices together with some properties that motivate this approach. This is shown to comply to the uncertainty and space requirements. To derive an estimated map from the representation a sparse linear equation system has to be solved. However, an update of the representation itself needs only constant time, making it highly attractive for building a SLAM algorithm.

AI Technology for Underwater Robots

This paper addresses visual navigation of autonomous underwater vehicles (AUVs) with and without ... more This paper addresses visual navigation of autonomous underwater vehicles (AUVs) with and without a given map, where the latter is called Simultaneous Localization and Mapping (SLAM). We summarize the challenges and opportunities in underwater environments that make visual navigation different from land navigation and also briefly survey the current state-of-the-art in this area. Then as a position paper we argue why many of these challenges could be met by a proper modeling of uncertainties in the SLAM representation. This would in particular allow the SLAM algorithm to thoroughly handle the ambiguity between “I see the same feature again.”, “I see a different but similar looking feature.” and “The environment has changed and the feature moved.”.

Machine Learning in Medical Imaging, 2016

2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2015

KI - Künstliche Intelligenz, 2010

This paper addresses the problem of simultaneous localisation and mapping (SLAM) by a mobile robo... more This paper addresses the problem of simultaneous localisation and mapping (SLAM) by a mobile robot. An incremental SLAM algorithm is introduced that is derived from so-called multigrid methods used for solving partial di#erential equations. The approach overcomes the relatively slow convergence of previous relaxation methods because it optimizes the map at multiple levels of resolution. The resulting algorithm has an update time that is linear in the number of mapped features, even when closing very large loops, and o#ers advantages in handling non-linearities compared to previous approaches. Experimental comparisons with alternative algorithms using two well-known data sets are also presented.

This paper describes the development of a simula-tion software that advances the idea of a roboti... more This paper describes the development of a simula-tion software that advances the idea of a robotic device inspired by ringed worm's locomotion. Its basic design is made up of an elongated body that is composed of a spring-style skeleton which is coated by a flexible skin. Our principle approach is to choose a shape memory alloy material for the skeleton, resulting in different spring forces exerted by the body under varying temperatures. The overall approach requires the elastic skin to prestress the whole body in its rest position, so that it can spatially extend when thermal energy is induced to the system, and relax when an inbuilt air cooling mechanism dissipates the heat.

This paper addresses the problem of simultaneous localisation and mapping (SLAM) by a mobile robo... more This paper addresses the problem of simultaneous localisation and mapping (SLAM) by a mobile robot. An incremental SLAM algorithm is introduced that is derived from multigrid methods used for solving partial differential equations. The approach improves on the performance of previous relaxation methods for robot mapping because it optimizes the map at multiple levels of resolution. The resulting algorithm has an update time that is linear in the number of estimated features for typical indoor environments, even when closing very large loops, and offers advantages in handling non-linearities compared to other SLAM algorithms. Experimental comparisons with alternative algorithms using two well-known data sets and mapping results on a real robot are also presented.

Lecture Notes in Computer Science, 2009

Sensors

Inertial navigation systems suffer from unbounded errors in the position and orientation estimate... more Inertial navigation systems suffer from unbounded errors in the position and orientation estimates. This drift can be corrected by applying prior knowledge, instead of using exteroceptive sensors. We want to show that the use of prior knowledge can yield full observability of the position and orientation. A previous study showed that track cyclers can be tracked drift-free with an IMU as the only sensor and the knowledge that the bike drives on the track. In this paper, we analyze the observability of the pose in the experiment we conducted. Furthermore, we improve the pose estimation of the previous study. The observability is analyzed by testing the weak observability criterion with a Jacobian rank test. The improved estimator is presented and evaluated on a dataset with three 60-round trials (10 km each). The average RMS is 1.08 m and the estimate is drift-free. The observability analysis reveals that the system can gain complete observability in the curves and observability of t...

Journal of Real-Time Image Processing

2019 International Conference on Indoor Positioning and Indoor Navigation (IPIN), 2019

Inertial Navigation Systems suffer from unbounded errors on the position and orientation estimate... more Inertial Navigation Systems suffer from unbounded errors on the position and orientation estimate. This drift can be corrected by applying prior knowledge, instead of using exteroceptive sensors. Analysing the state observability induced by prior knowledge motivates us to track bikers in track cycling races. In this paper, we show that the pose of the bikers can be estimated with an IMU as the only sensor by using a heightmap of the track and the knowledge that the biker drives forward. We present a dataset with three 60-round trials and evaluate the state estimate. We show that the influences of the priors match the expectation derived from state observability analysis.

2020 IEEE 23rd International Conference on Information Fusion (FUSION), 2020

Control, tracking, and obstacle detection algorithms for mobile robots, including autonomous cars... more Control, tracking, and obstacle detection algorithms for mobile robots, including autonomous cars, rely on a jump-free estimate of the vehicle's pose. While one cannot completely avoid jumps in global solutions like INS/GNSS and SLAM, relative localization (i.e., odometry) does not suffer from this problem. Methods based on graph optimization are popular in that field, but they do not scale very well with high-frequency measurements. Kalman filters (KFs) are able to cope with those measurements, but they face the issue of a continuously growing covariance. This results in instabilities and eventually jumps in the state estimate. We present an approach to handle this problem by periodically moving the reference state forward in time, which is realized using two filters. The equations for implementing this in both the extended Kalman filter (EKF) and the unscented Kalman filter (UKF) are derived. The algorithm is evaluated using real-world datasets covering different scenarios of ...

Sensors (Basel, Switzerland), 2021

Hybrid systems are subject to multiple dynamic models, or so-called modes. To estimate the state,... more Hybrid systems are subject to multiple dynamic models, or so-called modes. To estimate the state, the sequence of modes has to be estimated, which results in an exponential growth of possible sequences. The most prominent solution to handle this is the interacting multiple model filter, which can be extended to smoothing. In this paper, we derive a novel generalization of the interacting multiple filter and smoother to manifold state spaces, e.g., quaternions, based on the boxplus-method. As part thereof, we propose a linear approximation to the mixing of Gaussians and a Rauch–Tung–Striebel smoother for single models on boxplus-manifolds. The derivation of the smoother equations is based on a generalized definition of Gaussians on boxplus-manifolds. The three, novel algorithms are evaluated in a simulation and perform comparable to specialized solutions for quaternions. So far, the benefit of the more principled approach is the generality towards manifold state spaces. The evaluatio...

In this paper, we present a novel method of incorporating dense (e.g., depth, RGB-D) data in a ge... more In this paper, we present a novel method of incorporating dense (e.g., depth, RGB-D) data in a general purpose least-squares graph optimization framework. Rather than employing a loosely coupled, layered design where dense data is first used to estimate a compact SE(3) transform which then forms a link in the optimization graph as in previous approaches [28, 10, 26], we use a tightly coupled approach that jointly optimizes over each individual (i.e. per-pixel) dense measurement (on the GPU) and all other traditional sparse measurements (on the CPU). Concretely, we use Kinect depth data and KinectFusion-style point-to-plane ICP measurements. In particular, this allows our approach to handle cases where neither dense, nor sparse measurements separately define all degrees of freedom (DoF) while taken together they complement each other and yield the overall maximum likelihood solution. Nowadays it is common practice to flexibly model various sensors, measurements and to be estimated va...

The calibration of cameras is a crucial step in machine vision and usually relies on an accurate ... more The calibration of cameras is a crucial step in machine vision and usually relies on an accurate detection and localization of calibration patterns in images. Therefore, checkerboards are often used, allowing precise subpixel estimation of their corners. However, noise in localization generates a proportional noise in the derived model parameters. Therefore, it is important that the localization has a certain robustness against image noise. This is even more important for deteriorated imaging conditions strongly affecting subpixel detectors. This paper presents a new checkerboard corner detector based on a localized Radon transform implemented by large box filters making it robust to low contrast, image noise, and blur while maintaining high subpixel accuracy.

This papers provides two contributions to the problem of Simultaneous Localization and Mapping (S... more This papers provides two contributions to the problem of Simultaneous Localization and Mapping (SLAM): First we discuss properties of the problem itself and of the intended semantics of an uncertain map representation, with the main idea of “representing certainty of relations despite the uncertainty of positions”. We propose some requirements an ideal solution of SLAM should have concerning uncertainty, memory space and computation time and discuss existing approaches in the light of these requirements. The second part proposes a representation based on sparse information matrices together with some properties that motivate this approach. This is shown to comply to the uncertainty and space requirements. To derive an estimated map from the representation a sparse linear equation system has to be solved. However, an update of the representation itself needs only constant time, making it highly attractive for building a SLAM algorithm.

AI Technology for Underwater Robots

This paper addresses visual navigation of autonomous underwater vehicles (AUVs) with and without ... more This paper addresses visual navigation of autonomous underwater vehicles (AUVs) with and without a given map, where the latter is called Simultaneous Localization and Mapping (SLAM). We summarize the challenges and opportunities in underwater environments that make visual navigation different from land navigation and also briefly survey the current state-of-the-art in this area. Then as a position paper we argue why many of these challenges could be met by a proper modeling of uncertainties in the SLAM representation. This would in particular allow the SLAM algorithm to thoroughly handle the ambiguity between “I see the same feature again.”, “I see a different but similar looking feature.” and “The environment has changed and the feature moved.”.

Machine Learning in Medical Imaging, 2016

2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2015

KI - Künstliche Intelligenz, 2010

This paper addresses the problem of simultaneous localisation and mapping (SLAM) by a mobile robo... more This paper addresses the problem of simultaneous localisation and mapping (SLAM) by a mobile robot. An incremental SLAM algorithm is introduced that is derived from so-called multigrid methods used for solving partial di#erential equations. The approach overcomes the relatively slow convergence of previous relaxation methods because it optimizes the map at multiple levels of resolution. The resulting algorithm has an update time that is linear in the number of mapped features, even when closing very large loops, and o#ers advantages in handling non-linearities compared to previous approaches. Experimental comparisons with alternative algorithms using two well-known data sets are also presented.

This paper describes the development of a simula-tion software that advances the idea of a roboti... more This paper describes the development of a simula-tion software that advances the idea of a robotic device inspired by ringed worm's locomotion. Its basic design is made up of an elongated body that is composed of a spring-style skeleton which is coated by a flexible skin. Our principle approach is to choose a shape memory alloy material for the skeleton, resulting in different spring forces exerted by the body under varying temperatures. The overall approach requires the elastic skin to prestress the whole body in its rest position, so that it can spatially extend when thermal energy is induced to the system, and relax when an inbuilt air cooling mechanism dissipates the heat.

This paper addresses the problem of simultaneous localisation and mapping (SLAM) by a mobile robo... more This paper addresses the problem of simultaneous localisation and mapping (SLAM) by a mobile robot. An incremental SLAM algorithm is introduced that is derived from multigrid methods used for solving partial differential equations. The approach improves on the performance of previous relaxation methods for robot mapping because it optimizes the map at multiple levels of resolution. The resulting algorithm has an update time that is linear in the number of estimated features for typical indoor environments, even when closing very large loops, and offers advantages in handling non-linearities compared to other SLAM algorithms. Experimental comparisons with alternative algorithms using two well-known data sets and mapping results on a real robot are also presented.

Lecture Notes in Computer Science, 2009

Sensors

Inertial navigation systems suffer from unbounded errors in the position and orientation estimate... more Inertial navigation systems suffer from unbounded errors in the position and orientation estimates. This drift can be corrected by applying prior knowledge, instead of using exteroceptive sensors. We want to show that the use of prior knowledge can yield full observability of the position and orientation. A previous study showed that track cyclers can be tracked drift-free with an IMU as the only sensor and the knowledge that the bike drives on the track. In this paper, we analyze the observability of the pose in the experiment we conducted. Furthermore, we improve the pose estimation of the previous study. The observability is analyzed by testing the weak observability criterion with a Jacobian rank test. The improved estimator is presented and evaluated on a dataset with three 60-round trials (10 km each). The average RMS is 1.08 m and the estimate is drift-free. The observability analysis reveals that the system can gain complete observability in the curves and observability of t...

Journal of Real-Time Image Processing