Qualitative Spatial Relations Research Articles

A machine learning based approach for generating point sketch maps from qualitative directional information

People often use qualitative relations to describe locations or directional information, especially in written communication, such as ‘the restaurant is located at the southeast corner of the square’. However, when a large number of spatial entities are involved, qualitative relations alone are not intuitive enough for people to understand a spatial configuration. In fact, many applications, e.g. pertaining to sharing travel experiences, use sketch maps, i.e. maps focusing on the main features of an area whilst abstracting exact scale measurements, to help demonstrate abstract qualitative relations with more intuitive geometric points. Current approaches for generating point sketch maps from qualitative spatial relations require a high level of expertise, face inherent difficulties with efficiently processing large-scale data in bulk, and are vulnerable to inaccurate or conflicting information contained in qualitative data. To address these limitations, by incorporating machine learning techniques, we propose to translate the problem into an optimization problem of data reconstruction, enabling a novel end-to-end approach for generating point sketch maps from qualitative directional relations in bulk. Experiments on real-world datasets show that the proposed approach has very high accuracy and is robust even with a large portion of inaccurate or incomplete information.

Location retrieval using qualitative place signatures of visible landmarks

Location retrieval based on visual information is to retrieve the location of an agent (e.g. human, robot) or the area they see by comparing their observations with a certain representation of the environment. Existing methods generally treat the problem as a content-based image retrieval problem and have demonstrated promising results in terms of localization accuracy. However, these methods are challenging to scale up due to the volume of reference data involved; and the image descriptions might not be easily understandable/communicable for humans to describe surroundings. Considering that humans often use less precise but easily produced qualitative spatial language and high-level semantic landmarks when describing an environment, a coarse-to-fine qualitative location retrieval method is proposed in this work to quickly narrow down the initial location of an agent by exploiting the available information in large-scale open data. This approach describes and indexes a location/place using the perceived qualitative spatial relations between ordered pairs of co-visible landmarks from the perspective of viewers, termed as ‘qualitative place signatures’ (QPS). The usability and effectiveness of the proposed method were evaluated using openly available datasets, together with simulated observations by considering different types perception errors.

The history and processes of Titan's equator from the geospatial-topology of spectrally distinct units

We use hyperspectral-imaging observations from the Visual and Infrared Mapping Spectrometer (VIMS) to identify and explain processes of Titan's equator through the qualitative spatial relationships between geographic features (i.e. geospatial-topology). Our geographic features are defined by their spectra and geomorphology. We use tens of millions of VIMS pixels between 30°S and 30°N with incidence and emission angles <75°, and a pixel spatial scale of 200 km or less. Our dataset is several orders of magnitude larger than previous studies. This is possible through our use of novel techniques to reduce scattering and improve inter-flyby comparisons. We validate the dataset produced by these techniques by reproducing the results of previous studies. We use vector quantization, dimension reduction, and the Monte Carlo method to identify 14–16 spectrally and spatially distinct units within our dataset, a priori maps or images. These spectral units occur in distinct sequences, indicating that there is a discrete number of spectral pathways to describe the transitions across the surface. Using the same methodology used to identify the spectral units, we determine the spectral transition between units can be explained by five spectroclines. We define a spectrocline as the geographic expression of change in spectra between two spectrally distinct units; it is the spectral equivalent of an ecocline. We compare the spectra of the five spectroclines to the USGS spectral library to identify candidates for the change in compositions across the equator. We find evidence among the spectra of the spectroclines for changes in abundance of water-ice, acetylene, benzene, and alkane species. With the help of the geomorphological units identified in Cassini RADAR images, we discuss the significance of the spectral units and the spectroclines based on their geospatial-topology including their distribution, frequency, size, patterns, and sequences. From the correlations we identify between the spectra and geomorphology, we propose mechanisms for the formation and evolution of Titan's equatorial surface features.

CASPER: Cognitive Architecture for Social Perception and Engagement in Robots

AbstractOur world is being increasingly pervaded by intelligent robots with varying degrees of autonomy. To seamlessly integrate themselves in our society, these machines should possess the ability to navigate the complexities of our daily routines even in the absence of a human’s direct input. In other words, we want these robots to understand the intentions of their partners with the purpose of predicting the best way to help them. In this paper, we present the initial iteration of cognitive architecture for social perception and engagement in robots: a symbolic cognitive architecture that uses qualitative spatial reasoning to anticipate the pursued goal of another agent and to calculate the best collaborative behavior. This is performed through an ensemble of parallel processes that model a low-level action recognition and a high-level goal understanding, both of which are formally verified. We have tested this architecture in a simulated kitchen environment and the results we have collected show that the robot is able to both recognize an ongoing goal and to properly collaborate towards its achievement. This demonstrates a new use of qualitative spatial relations applied to the problem of intention reading in the domain of human–robot interaction.

Reasoning over higher-order qualitative spatial relations via spatially explicit neural networks

Qualitative spatial reasoning has been a core research topic in GIScience and AI for decades. It has been adopted in a wide range of applications such as wayfinding, question answering, and robotics. Most developed spatial inference engines use symbolic representation and reasoning, which focuses on small and densely connected data sets, and struggles to deal with noise and vagueness. However, with more sensors becoming available, reasoning over spatial relations on large-scale and noisy geospatial data sets requires more robust alternatives. This paper, therefore, proposes a subsymbolic approach using neural networks to facilitate qualitative spatial reasoning. More specifically, we focus on higher-order spatial relations as those have been largely ignored due to the binary nature of the underlying representations, e.g. knowledge graphs. We specifically explore the use of neural networks to reason over ternary projective relations such as between. We consider multiple types of spatial constraint, including higher-order relatedness and the conceptual neighborhood of ternary projective relations to make the proposed model spatially explicit. We introduce evaluating results demonstrating that the proposed spatially explicit method substantially outperforms the existing baseline by about 20%.

Towards Explainable Action Recognition by Salient Qualitative Spatial Object Relation Chains

In order to be trusted by humans, Artificial Intelligence agents should be able to describe rationales behind their decisions. One such application is human action recognition in critical or sensitive scenarios, where trustworthy and explainable action recognizers are expected. For example, reliable pedestrian action recognition is essential for self-driving cars and explanations for real-time decision making are critical for investigations if an accident happens. In this regard, learning-based approaches, despite their popularity and accuracy, are disadvantageous due to their limited interpretability. This paper presents a novel neuro-symbolic approach that recognizes actions from videos with human-understandable explanations. Specifically, we first propose to represent videos symbolically by qualitative spatial relations between objects called qualitative spatial object relation chains. We further develop a neural saliency estimator to capture the correlation between such object relation chains and the occurrence of actions. Given an unseen video, this neural saliency estimator is able to tell which object relation chains are more important for the action recognized. We evaluate our approach on two real-life video datasets, with respect to recognition accuracy and the quality of generated action explanations. Experiments show that our approach achieves superior performance on both aspects to previous symbolic approaches, thus facilitating trustworthy intelligent decision making. Our approach can be used to augment state-of-the-art learning approaches with explainabilities.

Incorporating Compositions in Qualitative Approaches

Qualitative approaches to various tasks, ranging from localization and mapping to active planning, are gaining considerable momentum in recent years. These approaches represent the environment through spatial relationships between small sets of landmarks in independent local coordinate systems. An essential component in these approaches is the composition operator, enabling spatial information propagation between different sets to infer new ones. Integrating compositions within qualitative algorithms brings several difficulties. For instance, if the information required to perform a specific composition operation is unavailable, it must be inferred first, possibly via a preparatory composition operation. This recursive issue becomes more challenging as the amount of information grows. This letter addresses two main questions arising from the above, which remained open: 1. Given an initial set of qualitative spatial relationships, what new ones can be composed? 2. What is the optimal sequence of compositions operations to create a target set among all possible sequences? We provide a theoretical derivation to address the first question and a novel search algorithm to address the second.

Quickly locating POIs in large datasets from descriptions based on improved address matching and compact qualitative representations

AbstractLocating points of interest (POIs) from descriptions can support intelligent location‐based services. Available research achieves it through address matching and spatial reasoning. However, semantic characteristics and spatial proximities of address fields are usually neglected in address matching; current applications of spatial reasoning represent qualitative spatial relations in semantic networks for efficient queries, but they do not yet scale to large datasets for qualitative direction reasoning due to massive qualitative direction relations between objects; moreover, spatial reasoning on various quantitative distances should be optimized. This study proposes a method that improves the accuracy of address matching by combining multiple similarities and enables quick spatial reasoning through the faster relation retrieval of compact qualitative direction representations implemented on global equal latitude and longitude grids (ELLGs) and the ELLG‐based quantitative calculations. The proposed method has been verified by two real‐world datasets and proven to be efficient and accurate when locating POIs in large POI datasets from descriptions.

Learning Micro-Management Skills in RTS Games by Imitating Experts

We investigate the problem of learning the control of small groups of units in combat situations in Real Time Strategy (RTS) games. AI systems may acquire such skills by observing and learning from expert players, or other AI systems performing those tasks. However, access to training data may be limited, and representations based on metric information -- position, velocity, orientation etc. -- may be brittle, difficult for learning mechanisms to work with, and generalise poorly to new situations. In this work we apply \textit{qualitative spatial relations} to compress such continuous, metric state-spaces into symbolic states, and show that this makes the learning problem easier, and allows for more general models of behaviour. Models learnt from this representation are used to control situated agents, and imitate the observed behaviour of both synthetic (pre-programmed) agents, as well as the behaviour of human-controlled agents on a number of canonical micro-management tasks. We show how a Monte-Carlo method can be used to decompress qualitative data back in to quantitative data for practical use in our control system. We present our work applied to the popular RTS game Starcraft.

Guided navigation from multiple viewpoints using qualitative spatial reasoning

ABSTRACT Navigation is an essential ability for mobile agents to be completely autonomous and able to perform complex actions. However, the problem of navigation for agents with limited (or no) perception of the world, or devoid of a fully defined motion model, has received little attention from research in AI and Robotics. One way to tackle this problem is to use guided navigation, in which other autonomous agents, endowed with perception, can combine their distinct viewpoints to infer the localization and the appropriate commands to guide a sensory deprived agent through a particular path. Due to the limited knowledge about the physical and perceptual characteristics of the guided agent, this task should be conducted on a level of abstraction allowing the use of a generic motion model, and high-level commands, that can be applied by any type of autonomous agents, including humans. The main task considered in this work is, given a group of autonomous agents perceiving their common environment with their independent, egocentric and local vision sensors, the development and evaluation of algorithms capable of producing a set of high-level commands (involving qualitative directions: e.g. move left, go straight ahead) capable of guiding a sensory deprived robot to a goal location. In order to accomplish this, the present paper assumes relations from the qualitative spatial reasoning formalism called StarVars, whose inference method is also used to build a model of the domain. This paper presents two qualitative-probabilistic algorithms for guided navigation using a particle filter and qualitative spatial relations. In the first algorithm, the particle filter is run upon a qualitative representation of the domain, whereas the second algorithm transforms the numerical output of a standard particle filter into qualitative relations to guide a sensory deprived robot. The proposed methods were evaluated with experiments carried out on a 2D humanoid robot simulator. A proof of concept executing the algorithms on a group of real humanoid robots is also presented. The results obtained demonstrate the success of the guided navigation models proposed in this work.

FUZZY POSITIONING MODELING OF NATURAL LANGUAGE LOCATION DESCRIPTION

Abstract. The development of cognitive technology and natural language understanding has made the interaction between humans and machines more mature. Natural language interactive location services through language or text are more in line with human cognitive habits, and it is the future development direction of location service techniques. Different from accurate GPS location, the uncertainty of human cognition causes the natural language position description is of uncertainty. How to take the uncertainty of position description into account to model it, and further establish a positioning computing framework that supports the expression of uncertainty, is a difficult problem in this field. At present, quantitative spatial relationship models can be well applied to navigation and positioning. However, these models cannot be directly transferred to deal with qualitative spatial relationships, such as positioning based on natural language descriptions. An effective solution is to establish a method to transform the qualitative natural language positioning into a quantitative one. To build this transformation, we proposed a fuzzy positioning model based on fussy mathematics theory and methods. The proposed model was validated using indoor and outdoor positioning experiments. The experiments showed that it could achieve a high positioning accuracy both indoor and outdoor.

Relations of land cover, topography, and climate to fire occurrence in natural regions of Iran: Applying new data mining techniques for modeling and mapping fire danger

In recent years, land uses have been changing and aridity has been increasing in the forests and rangelands of central Koohdasht which is a region in the forests of the Zagros Mountains in western Iran. Consequently, the number of fires has also increased. This study employs data-mining techniques to model fire danger using information regarding land cover, climate, topography, and other fire-danger influencing factors. A land cover map was prepared using Sentinel-2A satellite images and a maximum likelihood (ML) algorithm. Digital data describing other factors that influence fire danger (slope angle, aspect, elevation, climate, topographic wetness index, and distances from rivers and roads) were compiled from several sources and imported into a GIS. The locations of past fires in the study area were also determined from MODIS satellite images and data acquired from the region’s fire service. The quantitative and qualitative spatial relationships between effective factors and patterns of fires were investigated to model fire danger. A new machine-learning algorithm (the Boruta algorithm) was used to assess the relative importance of the fire-danger factors. Fire danger maps were created using several new data-mining algorithms including support vector machine (SVM), generalized linear model (GLM), functional data analysis (FDA), and random forest (RF). All were run in R 3.3.3 software. Finally, the fire danger maps were validated with several indices to determine the model that best predicts the fire danger in Koohdasht County. The results reveal that fire locations were determined mostly by elevation (low), aspect (south and southwest facing slopes), and aridity (semi-arid regions). Most fires occurred in non-natural landscapes: residential areas (46.74% of fires), agricultural lands (25.77%), and gardens (5.42%). In total, 77.93% of fires occurred in non-natural landscapes and within 500 m of roads. Only 22.07% of fires occurred on rangelands and forests. Three factors (distance from roads, climate, and aspect) were the strongest predictors of fire locations in the study area. Furthermore, area-under-the-curve (AUC) values indicate that the FDA (0.777) and GLM (0.772) algorithms generated the most accurate fire danger maps. These results have practical implications for fire danger management in the Zagros forests and provide baseline information for forest managers about the most important factors affecting fire danger in the similar regions. This methodology can be used by forest managers to predict the areas with greatest fire danger to prevent future fires through land use management, planning, and strategic decision-making. The results enable forest managers to find the best methods to monitor, manage, and control fire occurrence based on fire danger maps in the forests of western Iran, or in forests of other regions with similar conditions.

Learning Dynamic Spatio-Temporal Relations for Human Activity Recognition

Human activity, which usually consists of several actions (sub-activities), generally covers interactions among persons and/or objects. In particular, human actions involve certain spatial and temporal relationships, are the components of more complicated activity, and evolve dynamically over time. Therefore, the description of a single human action and the modeling of the evolution of successive human actions are two major issues in human activity recognition. In this paper, we develop a method for human activity recognition that tackles these two issues. In the proposed method, an activity is divided into several successive actions represented by spatio-temporal patterns, and the evolution of these actions are captured by a sequential model. A refined comprehensive spatio-temporal graph is utilized to represent a single action, which is a qualitative representation of a human action incorporating both the spatial and temporal relations of the participant objects. Next, a discrete hidden Markov model is applied to model the evolution of action sequences. Moreover, a fully automatic partition method is proposed to divide a long-term human activity video into several human actions based on variational objects and qualitative spatial relations. Finally, a hierarchical decomposition of the human body is introduced to obtain a discriminative representation for a single action. Experimental results on the Cornell Activity Dataset demonstrate the efficiency and effectiveness of the proposed approach, which will enable long videos of human activity to be better recognized.

Qualitative Spatial Representation for the Humanities

‘Qualitative spatial reasoning and representation’ is a range of techniques developed in Artificial Intelligence to meet the need for a computational treatment of qualitative spatial relations. Examples of such relations include ‘next to’, ‘overlapping’, ‘to the left of’, ‘separate from’, ‘including’, and so on. These relations occur within the data found in the spatial humanities, but the computational techniques described here do not appear to have been used in connection with this context. While Geographical Information Systems (GIS) are widely used as a means of visualizing and exploring material in the spatial humanities, GIS technology is acknowledged to be ill-suited to information that is vague, uncertain, ambiguous, imprecise or having other qualities that in a scientific setting could be regarded as imperfections. In the humanities such ‘imperfections’ are of course important, and qualitative spatial relations are one source of data that challenges scientifically based GIS. This article reviews the origin of qualitative spatial reasoning and representation in A. N. Whitehead's mereotopology and argues for exploring how these methods could complement GIS as a computational technique in the humanities. Qualitative representation is applicable to modelling spatial arrangements in many domains, not just geographical space. This is demonstrated through an example of spatial relations in lines of printed text.

Geographies of the Holocaust: Experiments in GIS, QSR, and Graph Representations

This article responds to the widespread and oft-noted challenges digital humanists face in working with data that is uncertain and characterised by complex narratives. Using an example drawn from the vast archives of post-war interviews with Holocaust survivors, it draws on approaches developed in Qualitative Spatial Representation (QSR) to explore how two key spatial aspects of survivor's narratives – their uncertain wartime trajectories and the slippage in scales as these are retold – can be represented. Spatial information in narratives tends to not provide the exact coordinates necessary to store the information in geospatial databases. Instead, narratives rely much more on often less precise qualitative spatial relations such as ‘near’, ‘next to’, ‘at the corner of’ without precise geometric interpretations. Given that relational databases are ill-equipped to store this kind of relational spatial knowledge from natural language sources, the article argues for a need for digital humanists to return to first principles and reconsider database design. Descriptive triple-based graph representations, which have been devised to accommodate this kind of highly irregular, semi-structured relational knowledge, have the potential to work with, rather than against, the grain of the narrative sources that underlie the work of much of the digital humanities.

Reasoning about Qualitative Direction and Distance between Extended Objects using Answer Set Programming

In this thesis, we introduce a novel formal framework to represent and reason about qualitative direction and distance relations between extended objects using Answer Set Programming (ASP). We take Cardinal Directional Calculus (CDC) as a starting point and extend CDC with new sorts of constraints which involve defaults, preferences and negation. We call this extended version as nCDC. Then we further extend nCDC by augmenting qualitative distance relation and name this extension as nCDC+. For CDC, nCDC, nCDC+, we introduce an ASP-based general framework to solve consistency checking problems, address composition and inversion of qualitative spatial relations, infer unknown or missing relations between objects, and find a suitable configuration of objects which fulfills a given inquiry.

Combining Deep Learning and Qualitative Spatial Reasoning to Learn Complex Structures from Sparse Examples with Noise

Many modern machine learning approaches require vast amounts of training data to learn new concepts; conversely, human learning often requires few examples—sometimes only one—from which the learner can abstract structural concepts. We present a novel approach to introducing new spatial structures to an AI agent, combining deep learning over qualitative spatial relations with various heuristic search algorithms. The agent extracts spatial relations from a sparse set of noisy examples of block-based structures, and trains convolutional and sequential models of those relation sets. To create novel examples of similar structures, the agent begins placing blocks on a virtual table, uses a CNN to predict the most similar complete example structure after each placement, an LSTM to predict the most likely set of remaining moves needed to complete it, and recommends one using heuristic search. We verify that the agent learned the concept by observing its virtual block-building activities, wherein it ranks each potential subsequent action toward building its learned concept. We empirically assess this approach with human participants’ ratings of the block structures. Initial results and qualitative evaluations of structures generated by the trained agent show where it has generalized concepts from the training data, which heuristics perform best within the search space, and how we might improve learning and execution.

Identifying Surrounds and Engulfs Relations in Mobile and Coordinate-Free Geosensor Networks

This article concerns the definition and identification of qualitative spatial relationships for the full and partial enclosure of spatial regions. The article precisely defines three relationships between regions—“surrounds,” “engulfs,” and “envelops”—highlighting the correspondence to similar definitions in the literature. An efficient algorithm capable of identifying these qualitative spatial relations in a network of dynamic (mobile) geosensor nodes is developed and tested. The algorithms are wholly decentralized, and operate in-network with no centralized control. The algorithms are also “coordinate-free,” able to operate in distributed spatial computing environments where coordinate locations are expensive to capture or otherwise unavailable. Experimental evaluation of the algorithms designed demonstrates the efficiency of the approach. Although the algorithm communication complexity is dominated by an overall worst-case O ( n 2 ) leader election algorithm, the experiments show in practice an average-case complexity approaching linear, O ( n 1.1 ).

Grounding semantic maps in spatial databases

Semantic maps add to classic robot maps spatially grounded object instances anchored in a suitable way for knowledge representation and reasoning. They enable a robot to solve reasoning problems of geometrical, topological, ontological and logical nature in addition to localization and path planning. Recent literature on semantic mapping lacks effective and efficient approaches for grounding qualitative spatial relations through analysis of the quantitative geometric data of the mapped entities. Yet, such qualitative relations are essential to perform spatial and ontological reasoning about objects in the robot’s surroundings.This article contributes a framework for semantic map representation, called SEMAP, to overcome this missing aspect. It is able to manage full 3D maps with geometric object models and the corresponding semantic annotations as well as their relative spatial relations. For that, spatial database technology is used to solve the representational and querying problems efficiently. This article describes the extensions necessary to make a spatial database suitable for robotic applications. Especially, we add 3D spatial operators and a tree of transformations to represent relative position information. We evaluate the implemented capabilities and present real life use cases of SEMAP in different application domains.

A Bayesian based Methodology for Indirect Object Search

The main goal of this paper is to propose a Bayesian based methodology for implementing robot informed search for objects. The methodology uses convolutions between observation likelihoods of secondary objects and spatial relation masks for estimating the probability map of the object being searched for, and also a search procedure that uses this probability map. A method for computing complex spatial relation masks by using a basis composed of basic relation masks and a database of co-occurrences of objects is used. Each basic relation mask corresponds to a qualitative spatial relation (QSR), such as: ‘very near’, ‘near’, or ‘far’. The search procedure takes into account the probability that the main object can be in different regions on the map and the distance to those regions. Also, the object search procedure is able to detect objects and generate new plans while moving. The proposed methodology is compared with uninformed and alternative informed search approaches using simulations and real-world experiments with a service robot. In simulations, the use of the proposed methodology increases the detection rate from 28% (direct uninformed search) to 79%, when the main object can be detected within a maximum distance of 1 meter. In the real world experiments, the use of the proposed methodology increases the detection rate from 40% (direct uninformed search) to 87% when using convolutions with soft masks, global search, and information on the positive detection of secondary objects. The detection rates obtained when using the proposed methodology are also much higher than those obtained by alternative informed search methods, both in the simulated and in the real-world experiments.