Visual place recognition is a challenging problem due to the vast range of ways in which appearance real-world places can vary. In recent years, improvements visual sensing capabilities, an ever-increasing focus on long-term mobile robot autonomy, and ability draw state-of-the-art research other disciplines-particularly computer vision animal navigation neuroscience-have all contributed significant advances systems. This paper presents survey landscape. We start by introducing concepts...

A robot that can carry out a natural-language instruction has been dream since before the Jetsons cartoon series imagined life of leisure mediated by fleet attentive helpers. It is remains stubbornly distant. However, recent advances in vision and language methods have made incredible progress closely related areas. This significant because interpreting navigation on basis what it sees carrying process similar to Visual Question Answering. Both tasks be interpreted as visually grounded...

Accurately ranking the vast number of candidate detections is crucial for dense object detectors to achieve high performance. Prior work uses classification score or a combination and predicted localization scores rank candidates. However, neither option results in reliable ranking, thus degrading detection In this paper, we propose learn an Iou-Aware Classification Score (IACS) as joint representation presence confidence accuracy. We show that can more accurate based on IACS. design new...

The application of deep learning in robotics leads to very specific problems and research questions that are typically not addressed by the computer vision machine communities. In this paper we discuss a number robotics-specific learning, reasoning, embodiment challenges for learning. We explain need better evaluation metrics, highlight importance unique robotic simulation, explore spectrum between purely data-driven model-driven approaches. hope provides motivating overview important...

After the incredible success of deep learning in computer vision domain, there has been much interest applying Convolutional Network (ConvNet) features robotic fields such as visual navigation and SLAM. Unfortunately, are fundamental differences challenges involved. Computer datasets very different character to camera data, real-time performance is essential, priorities can be different. This paper comprehensively evaluates compares utility three state-of-the-art ConvNets on problems...

The success of deep learning techniques in the computer vision domain has triggered a range initial investigations into their utility for visual place recognition, all using generic features from networks that were trained other types recognition tasks. In this paper, we train, at large scale, two CNN architectures specific task and employ multi-scale feature encoding method to generate condition- viewpoint-invariant features. To enable training occur, have developed massive Specific PlacEs...

Current SLAM back-ends are based on least squares optimization and thus not robust against outliers like data association errors false positive loop closure detections. Our paper presents evaluates a back-end formulation for using switchable constraints. Instead of proposing yet another appearance-based technique, our system is able to recognize reject during the optimization. This achieved by making topology underlying factor graph representation subject instead keeping it fixed. The...

In this letter, we use two-dimensional (2-D) object detections from multiple views to simultaneously estimate a 3-D quadric surface for each and localize the camera position. We derive simultaneous localization mapping (SLAM) formulation that uses dual quadrics as landmark representations, exploiting their ability compactly represent size, position orientation of an object, show how 2-D can directly constrain parameters via novel geometric error formulation. develop sensor model detectors...

For intelligent robots to interact in meaningful ways with their environment, they must understand both the geometric and semantic properties of scene surrounding them. The majority research date has addressed these mapping challenges separately, focusing on either or mapping. In this paper we address problem building environmental maps that include semantically meaningful, object-level entities point- mesh-based geometrical representations. We simultaneously build point cloud models...

Dropout Variational Inference, or Sampling, has been recently proposed as an approximation technique for Bayesian Deep Learning and evaluated image classification regression tasks. This paper investigates the utility of Sampling object detection first time. We demonstrate how label uncertainty can be extracted from a state-of-the-art system via Sampling. evaluate this approach on large synthetic dataset 30,000 images, real-world captured by mobile robot in versatile campus environment. show...

Current state of the art solutions SLAM problem are based on efficient sparse optimization techniques and represent as probabilistic constraint graphs. For example in pose graphs nodes poses edges between them express spatial information (e.g. obtained from odometry) loop closures. The task constructing graph is delegated to a front-end that has access available sensor information. optimizer, so called back-end system, relies heavily topological correctness structure not robust against...

The ability to recognize known places is an essential competence of any intelligent system that operates autonomously over longer periods time. Approaches rely on the visual appearance distinct scenes have recently been developed and applied large scale SLAM scenarios. FAB-Map maybe most successful these systems. Our paper proposes BRIEF-Gist, a very simplistic appearance-based place recognition based BRIEF descriptor. BRIEF-Gist much more easy implement efficient compared recent approaches...

In this paper we focus on the challenging problem of place categorization and semantic mapping a robot without environment-specific training. Motivated by their ongoing success in various visual recognition tasks, build our system upon state-of-the-art convolutional network. We overcome its closed-set limitations complementing network with series one-vs-all classifiers that can learn to recognize new classes online. Prior domain knowledge is incorporated embedding classification into...

Changing environments pose a serious problem to current robotic systems aiming at long term operation. While place recognition perform reasonably well in static or low-dynamic environments, severe appearance changes that occur between day and night, different seasons local weather conditions remain challenge. In this paper we propose learn predict the an environment. Our key insight is occurring are part systematic, repeatable therefore predictable. The goal of our work support existing...

Fine-grained leaf classification has concentrated on the use of traditional shape and statistical features to classify ideal images. In this paper we evaluate effectiveness hand-crafted propose deep convolutional neural network (Conv Net) features. We introduce a range condition variations explore robustness these features, including: translation, scaling, rotation, shading occlusion. Evaluations Flavia dataset demonstrate that in imaging conditions, combining Conv Net yields state-of-the...

We introduce Probabilistic Object Detection, the task of detecting objects in images and accurately quantifying spatial semantic uncertainties detections. Given lack methods capable assessing such probabilistic object detections, we present new Probability-based Detection Quality measure (PDQ). Unlike AP-based measures, PDQ has no arbitrary thresholds rewards label quality, foreground/background separation quality while explicitly penalising false positive negative contrast with existing mAP...

There has been a recent emergence of sampling-based techniques for estimating epistemic uncertainty in deep neural networks. While these methods can be applied to classification or semantic segmentation tasks by simply averaging samples, this is not the case object detection, where detection sample bounding boxes must accurately associated and merged. A weak merging strategy significantly degrade performance detector yield an unreliable measure. This paper provides first in-depth...

In open set recognition, deep neural networks encounter object classes that were unknown during training. Existing classifiers distinguish between known and by measuring distance in a network's logit space, assuming cluster closer to the training data than classes. However, this approach is applied post-hoc trained with cross-entropy loss, which does not guarantee clustering behaviour. To overcome limitation, we introduce Class Anchor Clustering (CAC) loss. CAC distance-based loss explicitly...

For robots to navigate and interact more richly with the world around them, they will likely require a deeper understanding of in which operate. In robotics related research fields, study is often referred as semantics, dictates what does "mean" robot, strongly tied question how represent that meaning. With humans increasingly operating same world, prospects human-robot interaction also bring semantics ontology natural language into picture. Driven by need, well enablers like increasing...

We present the design and implementation of a taskable reactive mobile manipulation system. In contrary to related work, we treat arm base degrees freedom as holistic structure which greatly improves speed fluidity resulting motion. At core this approach is robust motion controller can achieve desired end-effector pose, while avoiding joint position velocity limits, ensuring manipulator manoeuvrable throughout trajectory. This support sensor-based behaviours such closed-loop visual grasping....

We show that ensembling effectively quantifies model uncertainty in Neural Radiance Fields (NeRFs) if a density-aware epistemic term is considered. The naive ensembles investigated prior work simply average rendered RGB images to quantify the caused by conflicting explanations of observed scene. In contrast, we additionally consider termination probabilities along individual rays identify due lack knowledge about parts scene unobserved during training. achieve new state-of-the-art...