Semantic Localization for IoT
381
relationship) and topology [28]. RCC laid the foundation for the GeoSPARQL standard [13], which is today widely (but incompletely) implemented by modern semantic
repositories
5 to leverage RCC relationships for queries on geospatial data sets.
4 Related Formal Structures from AI and Robotics
Pereira’s BigActor model [29] gives a formalism for mapping with many similarities
to our approach. Specifically he defines two kinds of spatial structures: a logical-space
model with a rough correspondence to what we would call a relational ontology, and
a physical-space model corresponding to a coordinate-based physical map. Pereira
requires the same relations hold true between the same objects in physical and logical
space. The model-theoretic proposal for semantic localization in this chapter can be
seen as a generalization of Pereira’s approach to include more diverse kinds of spatial
structures.
Similar ideas to relational ontologies have been around in the world of AI and
robotics research for some time [22, 30, 31]. However, where semantic localization
is designed to integrate modeling and programming for heterogenous IoT systems,
the focus of research in this domain is commonly inference and autonomous decision
making. As an additional point of contrast, spatial modeling in robotics is usually
from the perspective of a robot as it moves from place to place, but spatial modeling
for localization systems is usually from the perspective of a place as people (or
robots) move within.
6
The distinction between absolute and relative space is raised by Vieu [31]. The
elements of a spatial ontology are Basic Entities (is the space composed of points
or basic regions?), Primitive Notions (topology: relating to contact and part-whole
relationships; orientation: absolute, intrinsic, and contextual; distance: metric functions and discrete distance notions), and bounded/unboundedness. Vieu goes on
to overview actual approaches researchers have used to represent space. Vieu also
examines the difference between 3D space composed with time and 4D views.
Kuipers [22], in a classic robotics paper, introduces an ontology for spatial information flow from sensor values to, ultimately, 2-D geometry. His ontology allows
information to be incomplete at different levels. For example the graph-topological
connections between different maps may be known even if each of the maps hasn’t
been entirely fleshed out.
An example of the advantages of combining physical maps with relational information for robotic localization was demonstrated by Atanasov et al. [30]. The authors
use set-based identification of semantically interesting indoor objects such as chairs
5 Essentially a semantic repository is a database for relational data.
6 We refer to models of things moving through space as “Lagrangian Models” and models of space
with things moving within as “Eulerian Models”. The terminology comes from the analysis of fluid
flows.
381
relationship) and topology [28]. RCC laid the foundation for the GeoSPARQL standard [13], which is today widely (but incompletely) implemented by modern semantic
repositories
5 to leverage RCC relationships for queries on geospatial data sets.
4 Related Formal Structures from AI and Robotics
Pereira’s BigActor model [29] gives a formalism for mapping with many similarities
to our approach. Specifically he defines two kinds of spatial structures: a logical-space
model with a rough correspondence to what we would call a relational ontology, and
a physical-space model corresponding to a coordinate-based physical map. Pereira
requires the same relations hold true between the same objects in physical and logical
space. The model-theoretic proposal for semantic localization in this chapter can be
seen as a generalization of Pereira’s approach to include more diverse kinds of spatial
structures.
Similar ideas to relational ontologies have been around in the world of AI and
robotics research for some time [22, 30, 31]. However, where semantic localization
is designed to integrate modeling and programming for heterogenous IoT systems,
the focus of research in this domain is commonly inference and autonomous decision
making. As an additional point of contrast, spatial modeling in robotics is usually
from the perspective of a robot as it moves from place to place, but spatial modeling
for localization systems is usually from the perspective of a place as people (or
robots) move within.
6
The distinction between absolute and relative space is raised by Vieu [31]. The
elements of a spatial ontology are Basic Entities (is the space composed of points
or basic regions?), Primitive Notions (topology: relating to contact and part-whole
relationships; orientation: absolute, intrinsic, and contextual; distance: metric functions and discrete distance notions), and bounded/unboundedness. Vieu goes on
to overview actual approaches researchers have used to represent space. Vieu also
examines the difference between 3D space composed with time and 4D views.
Kuipers [22], in a classic robotics paper, introduces an ontology for spatial information flow from sensor values to, ultimately, 2-D geometry. His ontology allows
information to be incomplete at different levels. For example the graph-topological
connections between different maps may be known even if each of the maps hasn’t
been entirely fleshed out.
An example of the advantages of combining physical maps with relational information for robotic localization was demonstrated by Atanasov et al. [30]. The authors
use set-based identification of semantically interesting indoor objects such as chairs
5 Essentially a semantic repository is a database for relational data.
6 We refer to models of things moving through space as “Lagrangian Models” and models of space
with things moving within as “Eulerian Models”. The terminology comes from the analysis of fluid
flows.