Robust and accurate visual localization is a fundamental capability for numerous applications, such as autonomous driving, mobile robotics, or augmented reality. It remains, however, challenging task, particularly large-scale environments in presence of significant appearance changes. State-of-the-art methods not only struggle with scenarios, but are often too resource intensive certain real-time applications. In this paper we propose HF-Net, hierarchical approach based on monolithic CNN...

Robust and accurate visual-inertial estimation is crucial to many of today's challenges in robotics. Being able localize against a prior map obtain driftfree pose estimates can push the applicability such systems even further. Most currently available solutions, however, either focus on single session use-case, lack localization capabilities or an end-to-end pipeline. We believe that only complete system, combining state-of-the-art algorithms, scalable multi-session mapping tools, flexible...

Visual robot navigation within large-scale, semistructured environments deals with various challenges such as computation intensive path planning algorithms or insufficient knowledge about traversable spaces. Moreover, many state-of-the-art approaches only operate locally instead of gaining a more conceptual understanding the objective. This limits complexity tasks can accomplish and makes it harder to deal uncertainties that are present in context real-time robotics applications. In this...

Robust, scalable place recognition is a core competency for many robotic applications. However, when revisiting places over and over, state-of-the-art approaches exhibit reduced performance in terms of computation memory complexity accuracy. For successful deployment robots long time scales, we must develop algorithms that get better with repeated visits to the same environment, while still working within fixed computational budget. This paper presents evaluates an algorithm alternates...

Precisely estimating a robot's pose in prior, global map is fundamental capability for mobile robotics, e.g. autonomous driving or exploration disaster zones. This task, however, remains challenging unstructured, dynamic environments, where local features are not discriminative enough and scene descriptors only provide coarse information. We therefore present SegMap: representation solution localization mapping based on the extraction of segments 3D point clouds. Working at level offers...

Robust, scalable localization unlocks path-planning, obstacle avoidance as well manipulation and thus is a core competency for many robotic applications. However, we leave the lab move out in world, models of environment no longer span distances meters but kilometers length. Now, gigabytes instead megabytes memory are required to hold model localization. Discarding data keeping map representation compact essential any meaningful application. This paper presents evaluates compression...

Many robotics applications require precise pose estimates despite operating in large and changing environments. This can be addressed by visual localization, using a pre-computed 3D model of the surroundings. The estimation then amounts to finding correspondences between 2D keypoints query image points local descriptors. However, computational power is often limited on robotic platforms, making this task challenging large-scale Binary feature descriptors significantly speed up 2D-3D...

Many robotics applications require precise pose estimates despite operating in large and changing environments. This can be addressed by visual localization, using a pre-computed 3D model of the surroundings. The estimation then amounts to finding correspondences between 2D keypoints query image points local descriptors. However, computational power is often limited on robotic platforms, making this task challenging large-scale Binary feature descriptors significantly speed up 2D-3D...

Dynamic obstacle avoidance is one crucial component for compliant navigation in crowded environments. In this paper we present a system accurate and reliable detection tracking of dynamic objects using noisy point cloud data generated by stereo cameras. Our solution real-time capable specifically designed the deployment on computationally-constrained unmanned ground vehicles. The proposed approach identifies individual robot's surroundings classifies them as either static or dynamic. are...

Flying and walking robots can use their complementary features in terms of viewpoint payload capability to the best a heterogeneous team. To this end, we present our online collaborative navigation framework for unknown challenging terrain. The method leverages flying robot's onboard monocular camera create both map visual simultaneous localization mapping dense representation environment as an elevation map. This shared knowledge from platform enables robot localize itself against global...

We present a complete map management process for visual localization system designed multi-vehicle long-term operations in resource constrained outdoor environments. Outdoor generates large amounts of data that need to be incorporated into lifelong order allow at all times and under appearance conditions. Processing these quantities is non-trivial, as it subject limited computational storage capabilities both on the vehicle mapping backend. address this problem with two-fold update paradigm...

Large scale, long-term, distributed mapping is a core challenge to modern field robotics. Using the sensory output of multiple robots and fusing it in an efficient way enables creation globally accurate consistent metric maps. To combine data from agents into global map, most existing approaches use central entity that collects manages information all agents. Often, raw sensor one robot needs be made available processing algorithms on other due lack computational resources robot....

Globally localizing in a given map is crucial ability for robots to perform wide range of autonomous navigation tasks. This paper presents OneShot - global localization algorithm that uses only single 3D LiDAR scan at time, while outperforming approaches based on integrating sequence point clouds. Our approach, which does not require the robot move, relies learning-based descriptors cloud segments and computes full 6 degree-of-freedom pose map. The are extracted from current matched against...

Large scale, long-term, distributed mapping is a core challenge to modern field robotics. Using the sensory output of multiple robots and fusing it in an efficient way enables creation globally accurate consistent metric maps. To combine data from agents into global map, most existing approaches use central entity that collects manages information all agents. Often, raw sensor one robot needs be made available processing algorithms on other due lack computational resources robot....

An increasing number of simultaneous localization and mapping (SLAM) systems are using appearance-based to improve the quality pose estimates. However, with growing time-spans size areas we want cover, maps often becoming too large handle consisting features that not always reliable for purposes. This paper presents a method selecting map persistent over time thus suited long-term localization. Our methodology relies on CNN classifier based image patches depth recognizing which suitable...

Precisely estimating the pose of an agent in a global reference frame is crucial goal that unlocks multitude robotic applications, including autonomous navigation and collaboration. In order to achieve this, current state-of-the-art localization approaches collect data provided by one or more agents create single, consistent map, maintained over time. However, with introduction lengthier sorties growing size environments, transfers between backend server where map stored are becoming...

Changes in appearance is one of the main sources failure visual localization systems outdoor environments. To address this challenge, we present VIZARD, a system for urban By combining local algorithm with use multi-session maps, high recall can be achieved across vastly different conditions. The fusion constraints wheel-odometry state estimation framework further guarantees smooth and accurate pose estimates. In an extensive experimental evaluation on several hundreds driving kilometers...

Many robotics and Augmented Reality (AR) systems that use sparse keypoint-based visual maps operate in large highly repetitive environments, where pose tracking localization are challenging tasks. Additionally, these usually face further challenges, such as limited computational power, or insufficient memory for storing of the entire environment. Thus, developing compact map representations improving retrieval is considerable interest enabling large-scale place recognition loop-closure. In...

Flying and walking robots can use their complementary features in terms of viewpoint payload capability to the best a heterogeneous team. To this end, we present our online collaborative navigation framework for unknown challenging terrain. The method leverages flying robot's onboard monocular camera create both map visual simultaneous localization mapping dense representation environment as an elevation map. This shared knowledge from platform enables robot localize itself against global...

An accurate estimate of the 3D-structure in environment is key to robotic applications such as autonomous inspection, obstacle avoidance and manipulation. Recent years have seen substantial algorithmic advances towards creating highly models small objects well large scale architectural structures. Most commonly a rich set images covering static scene are used jointly pose cameras observed 3D-structure. For many practical application however assumption scenes sufficient coverage by does not...

A variety of end-user devices involving keypoint-based mapping systems are about to hit the market e.g. as part smartphones, cars, robotic platforms, or virtual and augmented reality applications. Thus, generated map data requires automated evaluation procedures that do not require experienced personnel ground truth knowledge underlying environment. particularly important question enabling commercial applications is whether a given sufficient quality for localization. This paper proposes...

Industrial facilities often require periodic visual inspections of key installations. Examining these points interest is time consuming, potentially hazardous or special equipment to reach. MAVs are ideal platforms automate this expensive and tedious task. In work we present a novel system that enables human operator teach inspection task an autonomous aerial vehicle by simply demonstrating the using handheld device. To enable robust operation in confined, GPS-denied environments, employs...

Accurate and robust real-time map generation onboard of a fixed-wing UAV is essential for obstacle avoidance, path planning, critical maneuvers such as autonomous take-off landing. Due to the computational constraints, required robustness reliability, it remains challenge deploy with an online-capable, accurate framework. While photogrammetric approaches have underlying assumptions on structure view camera, generic simultaneous localization mapping (SLAM) are computationally demanding. This...