Abstract—We describe a new formulation of appearance-only SLAM suitable for very large scale navigation. The system navigates in the space of appearance, assigning each new observation to either a new or previously visited location, without reference to metric position. The system is demonstrated performing reliable online appearance mapping and loop closure detection over a 1,000 km trajectory, with mean filter update times of 14 ms. The 1,000 km experiment is more than an order of magnitude larger than any previously reported result. The scalability of the system is achieved by defining a sparse approximation to the FAB-MAP model suitable for implementation using an inverted index. Our formulation of the problem is fully probabilistic and naturally incorporates robustness against perceptual aliasing. The 1,000 km data set comprising almost a terabyte of omni-directional and stereo imagery is available for use, and we hope that it will serve as a benchmark for future systems. I.

For many centuries, philosophers and scientists have pondered the origins and nature of human intuitions about the properties of points, lines, and figures on the Euclidean plane, with most hypothesizing that a system of Euclidean concepts either is innate or is assembled by general learning processes. Recent research from cognitive and developmental psychology, cognitive anthropology, animal cognition, and cognitive neuroscience suggests a different view. Knowledge of geometry may be founded on at least two distinct, evolutionarily ancient, core cognitive systems for representing the shapes of large-scale, navigable surface layouts and of small-scale, movable forms and objects. Each of these systems applies to some but not all perceptible arrays and captures some but not all of the three fundamental Euclidean relationships of distance (or length), angle, and direction (or sense). Like natural number (Carey, 2009), Euclidean geometry may be constructed through the productive combination of representations from these core systems, through the use of uniquely human symbolic systems.

Abstract — Spatial knowledge constitutes a fundamental component of the knowledge base of a cognitive, mobile agent. This paper introduces a rigorously defined framework for building a cognitive spatial map that permits high level reasoning about space along with robust navigation and localization. Our framework builds on the concepts of places and scenes expressed in terms of arbitrary, possibly complex features as well as local spatial relations. The resulting map is topological and discrete, robocentric and specific to the agent’s perception. We analyze spatial mapping design mechanics in order to obtain rules for how to define the map components and attempt to prove that if certain design rules are obeyed then certain map properties are guaranteed to be realized. The idea of this paper is to take a step back from existing algorithms and literature and see how a rigorous formal treatment can lead the way towards a powerful spatial representation for localization and navigation. We illustrate the power of our analysis and motivate our cognitive mapping characteristics with some illustrative examples. I.

Abstract — In this paper we present a topological map building algorithm based on a Vocabulary Tree that is robust to features present in dynamic or similar environments. The algorithm recognises incorrect loop closures, not supported by the odometry, and uses this information to update the feature weights in the tree to suppress further associations from these features. Two methods of adjusting these feature entropies are proposed, one decreasing entropy related to incorrect features in a uniform manner and the other proportional to the contribution of the said feature. Preliminary results showing the performance of the proposed method are presented where it is found that by adjusting the feature entropies, the number of incorrect associations can be reduced while improving the quality of the correct matches. I.

Abstract—In this paper, the performance of a topologicalmetric visual path following framework is investigated in different environments. The framework relies on a monocular camera as the only sensing modality. The path is represented as a series of reference images such that each neighboring pair contains a number of common landmarks. Local 3D geometries are reconstructed between the neighboring reference images in order to achieve fast feature prediction. This allows recovery from tracking failures. During navigation the robot is controlled using image-based visual servoing. The focus of the paper is on the results from a number of experiments conducted in different environments, lighting conditions and seasons. The experiments with a robot-car show that the framework is robust against moving objects and moderate illumination changes. It is also shown that the system is capable of on-line path learning. Index Terms—visual servoing, mapping, localization, visual memory, path following

Efficient probabilistic planar robot motion

We present an overview of FAB-MAP, an algorithm for place recognition and mapping developed for infrastructure-free mobile robot navigation in large environments. The system allows a robot to identify when it is revisiting a previously seen location, on the basis of imagery captured by the robot’s camera. We outline a complete probabilistic framework for the task, which is applicable even in visually repetitive environments where many locations may appear identical. Our work introduces a number of technical innovations- notably we demonstrate that place recognition performance can be improved by learning an approximation to the joint distribution over visual elements. We also investigate several principled approachestomakingthesystemrobustinvisuallyrepetitiveenvironments,anddefineanefficient bail-out strategy for multi-hypothesis testing to improve system speed. Our model has been shown to substantially outperform standard tf-idf ranking on our task of interest. We demonstrate the system performing reliable online appearance mapping and loop closure detection over a 1,000km trajectory, with mean filter update times of 14ms.