This is the first chapter of the third section. It describes the kinds of mathematical models usable by a mobile robot to represent its spatial reality, and the reasons by which some of them are more useful than others, depending on the task to be carried out. The most common metric, topological, and hybrid map representations are described from an introductory viewpoint, emphasizing their limitations and advantages for the localization and mapping problems. It then addresses the problem of how to update or build a map from the robot raw sensory data, assuming known robot positions, a situation that becomes an intrinsic feature of some SLAM filters. Since the process greatly depends on the kind of map and sensors, the most common combinations of both are shown.
TopChapter Guideline
Top1. Introduction
Some approaches to autonomous robots intend not to use any explicit representation of the environment, even not to use any representation at all, aiming at employing the environment itself as its own best model and considering the robot as part of it (Brooks, 1991). However, having an internal model of space—a map—is currently the only known practical way of efficient and optimally planning actions (taking decisions) to operate in the long-term. Furthermore, maintaining such a map has not been ruled out as a real possibility by the modern embodied cognition paradigm (Anderson, 2003), which has extended the seminal work based on not employing any model at all. It has been shown that humans use a so-called cognitive map to plan routes and locate in our environments (Tolman, 1948); a different story is whether the map is explicitly stored in our brains—up to date this possibility seems controversial at least—or emerges from a set of complex interactions with the environment, if we choose a developmental perspective. In state-of-the-art mobile robot mapping and localization, the former is the dominant approach, and consequently that is the one followed in this book.
While discussing robot localization in previous chapters we already faced the most common metrical environment representations, namely grid maps and landmark or feature-based maps. This revealed the tight coupling existing between localization and mapping: the robot cannot localize well if it does not have a good map, but on the other hand, a map cannot be accurately reconstructed if the robot is poorly localized. In this chapter, we will widen and organize this perspective on maps by describing the best-known types of explicit spatial representations for robots. A few of them are quite common in the robotic localization and mapping literature, while others have very restricted niches of usability, i.e., they respond to very specific sensors, environments or tasks. The existing variety makes evident the diversity and complexity of the problems arising when an automatic mobile device is intended to operate autonomously in a given spatial region: no single map representation seems to be universally valid for all the tasks of a given robot; in fact, the choice of the map has important implications in these tasks (e.g. navigation, manipulation, etc.) that extend well beyond the issues of localization and mapping addressed in this book.