The Landmark Spider: Representing Landmark Knowledge for Wayfinding Tasks

Mobile navigation systems are important assets for travelers visiting foreign environments, as they provide instructions on how to find the way to get to a chosen destination. Research has shown that finding ones way in a foreign environment is primarily based on cues in the environment. In this paper, we propose an algorithm that generates the clearest route in terms of spatial references and use the selected landmarks to describe the route. Wayfinding is dynamic in nature and as wayfinders move along, the cues used as reference points change, hence, raising the need for route generation methods that account for motion. Our model generalizes this dynamic task and selects spatial cues based on distance and orientation of the navigator with respect to the landmark, and the salience of spatial objects. The results of this process are represented in a spatioanalogical fashion, which diagrammatically supports wayfinding decisions. The Generation of Route Descriptions Landmarks play an important role when humans navigate through foreign environments (Lynch, 1960; May, Ross, Bayer, & Tarkiainen, 2003). For example, trying to find the way is much easier if the navigator can rely on a description of the route based on well-recognizable objects in the environment, instead of navigating solely on the basis of street names and metric directions (Tom & Denis, 2003). Landmark-based navigation applies knowledge about prominent objects in the environment to guide travelers through foreign environments (Hampe & Elias, 2004; Lee, Tappe, & Klippel, 2002). Therefore, collecting and incorporating landmarks along a route is a crucial task of navigation systems that aim at providing efficient and reliable route instructions. Several proposals have been made on how to automatically extract landmarks from data sets and how such landmarks could be used to enhance wayfinding instructions (Elias, 2003; Nothegger, 2003; Copyright © 2002, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. Raubal & Winter, 2002). So far, however, the question what routes offer the clearest cues and how to integrate these cues in the route generation process is only poorly understood. We define a cognitive model that assesses the relevance of significant spatial objects based on their features (i.e., saliency), and from a traveler’s perspective (i.e. heading, distance). The model assesses the relevance of landmarks at each node along the route with respect to the traveler, includes the selected landmarks in the route generation process, and dynamically weaves a web of landmarks along the route. These computation forms the basis on which a diagrammatic representation of the route is produced. We will term this model the Landmark Spider. Landmark-based Navigation Cognitive research has shown that the clarity of route instructions may be as important in navigational tasks as the overall length of the route. Streeter and co-authors found that human navigators were prepared to take suboptimal routes in terms of travel time, if these routes were potentially easier to describe and to follow (Streeter & Vitello, 1986; Streeter, Vitello, & Wonsiewicz, 1985). Landmarks are important elements in route instructions as they support clarity of a specific route and therefore ensure efficient and reliable navigation (Denis, Pazzaglia, Cornoldi, & Bertolo, 1999). So far, the typical approach to incorporate landmarks in route descriptions has been to enrich referential route representations with information about landmarks (Denis et al., 1999; Elias, 2003; Nothegger, 2003; Raubal & Winter, 2002; Redish & Touretzky, 1995). Our approach is different in that it uses a subset of all available landmarks, which are most prominent and easy to find, to determine the clearest route. The introduction of landmarks in the route generation process is advantageous, since it enables us to avoid areas of low landmark density and ensures consistent access to landmark information along the complete route. Diagrammatic Route Description Visual representations with overlay of diagrammatic elements are ubiquitous in decision-making (Casakin, Barkowsky, Klippel, & Freksa, 2000; Hegarty, Haarslev, & Narayanan, 2002). For instance, in location-based tasks, information services represent instructions about intended movements and plans, and monitor the progress of action on maps that contain terrain and other relevant information. In many contexts, as for instance in the military, diagrams are deemed so important that manuals have standardized the elements of such representations. Diagrams are often overlaid on top of maps, and they indicate, using a combination of iconic and spatially veridical elements, information such as movements of cars, locations and identities of business units, regions of traffic, points of interest, and so on (Hernandez & Zimmermann, 1993; Werner, 2002). Diagrams abstract away details that are not essential to a reasoning task and highlight those that are, and by means of symbolic elements (such as attached labels and iconic diagrammatic elements), they point to relevant pieces of conceptual information. Figure 1 shows an example of a route sketch as could have been supplied to a tourist asking for directions. The route sketch abstracts the reality in such a way that important cues are preserved and yet the directions are easily understandable and easy to follow. Figure 1. Example of a route sketch The whole process is so natural that we often fail to appreciate the complexity of the cognitive activities involved. Nevertheless, understanding and formalizing the perceptual and conceptual processes involved in such apparently effortless reasoning is necessary if we wish to provide effective navigation instructions. Intelligent computer support that, among other things, relieves the traveler by highlighting upcoming cues, attempts to infer the best reference points based on existent route knowledge, and presents the results in such a way that salient information visually stands out, can greatly enhance the usability, reliability, and efficiency of route descriptions. We attempt to enhance navigation as we exploit the benefits of diagrams for landmark-based navigation. Our goal is to computationally reproduce route descriptions as produced by ‘experts’ (Figure 1) by assessing the relevance of spatial objects along a route and by considering the traveler’s movement. Assessing the relevance of landmarks results in a set of objects by which the route will be described. The result of this process is represented in a spatio-analogical fashion, which diagrammatically supports wayfinding decisions navigators may have to take along the way. Highlighting what landmarks are relevant at each point along a route relieves the cognitive load put on the navigator, and hence, assists in efficiently finding the way. A Framework for Landmark Deduction In landmark-based navigation, the schematization of landmark knowledge is important for reducing the cognitive effort put on navigators when trying to find a destination. Freksa (1999) has proposed that an appropriate representation tool should include the following processes: 1) identifying and selecting relevant aspects from the physical environment, 2) choosing an appropriate structure for the inferences to be made between the represented world and the representing world, and 3) interpreting the results of the inferences. In the following sections, we present a framework based on these three processes, which sets the base for deducing relevant landmarks, integrating them in the route generation, and presenting them appropriately. It is to be noted that the main focus of this work is on the integration of landmarks in the route generation process, rather than extracting landmarks from databases. Therefore, we assume that a set of potential landmarks is specified in the area of interest and that positional information and saliencies are known.