Semantic Mapping extension for OpenStreetMap applied to
indoor robot navigation
Lakshadeep Naik, Sebastian Blumenthal, Nico Huebel, Herman Bruyninckx, Erwin Prassler
Proposed semantic mapping approach
•Indoor robots mostly rely on the spatial
representation of the environment"
•They also need semantic information to give
meaning to spatial information"
•This work presents a hierarchical &
composable graph model for creating
indoor semantic maps, which extends
Is this area
safe to drive?
Which is wall?
Which is door?
Can I wait
Can I enter
•They contain semantic information of the environment apart from spatial information"
•Most of the existing semantic mapping approaches add semantic information on top of a
•Topological graph is created based on the detected semantic features and environment
geometry (bottom-up approach)
•They lack modular and abstract design, diﬃcult to scale & update only part of the map"
•They have minimal querying capabilities for querying semantic information"
•Sensors & algorithms used for mapping introduce uncertainty"
•A robot has to deal with uncertainty every time it uses this map
•It is an open-source, collaborative mapping project"
•Its model conforms to graph model and provides lots
of semantic tags"
•It supports modelling: vector geometry, topological
graphs, semantic information, hierarchy"
•It has been successfully used for outdoor robotics
•It provides tools supporting development (mapping) &
usage (querying) of the models
model in robotics
Existing indoor OSM models"
eg. Simple Indoor Tagging, "
New models introduced in
•OSM is based on the concept of Volunteered Geographic Information (VGI) i.e. its people
who create, edit and use the maps"
•This work presents a similar approach to create a map for robots, i.e. humans add
additional information to the OSM in the robotics context so that humans & robots can use
that same map (top-down approach)"
•It provides a composable & hierarchical graph model for creating semantic maps for
indoor environment using OSM"
Modelling OSM in robotics context
Domain speciﬁc modelling - identifying additional information required for indoor robot
Indoor OpenStreetMap for Berlin Central Station
Robot speciﬁc, semantic, topological and geometric information
modelled in proposed indoor OSM model
areas introduced in
the proposed indoor
OSM model. The
ﬁgure describes how
they can be used to
model indoor traﬃc
rules for robots.
introduced in the
proposed indoor OSM
model which supports
Modelling OSM in robotics context (continued)
Logical modelling - giving abstract structure to the data models"
Technology speciﬁc modelling - representing data models using OSM data-structures"
Compatibility with existing
indoor navigation approaches
Incorporating semantics in indoor navigation
Eﬀorts required for creating OSM in robotics context
Comparison between occupancy grid maps created
using SLAM and OSM based approach. Both maps
Occupancy grid map
generated from SLAM
Occupancy grid map
generated using OSM
Path planned using
geometric information in grid
Waypoints generated using
geometric, topological and
semantic information in OSM.
Location of the way-points
ensure that robot path sticks to
the right side of corridors,
according to the modelled
semantics of ”traﬃc rules”. "
Comparison between path generated using grid maps
and proposed semantic map
Comparison between a number of OSM data-structures required to map 20m X 20m
indoor area using existing indoor OSM model (Simple Indoor Tagging) and proposed
model in a robotics context."
Results & conclusions
Map created using proposed OSM model viewed in
JOSM editor. Blue areas are wall geometries, red are
doors. The green lines indicate the connections
between corridors and rooms, while the directed
arrows indicate connections between areas at a lower
level of abstraction. "
•OSM can be successfully used to create
semantic maps for robots"
•Maps can be created using existing OSM
mapping tools such as JOSM
•Mapped data can be queried using existing
OSM querying tools such as Overpass and
•Robots require much more information then
humans, this comes at the cost of increased
•Hence there is a need to semi-automate/
automate the process to scale it for larger
This work was supported by the European
Union’s Horizon 2020 projects ROPOD
(grant agreement No 731848) and
RobMoSys (grant agreement No 732410). "
Hochschule Bonn Rhein Sieg"
•Oﬃcially supports only outdoor environments"
•Uses geographical coordinate systems"
•Made for human navigation, robots require lot more
details then humans