A Software-Architecture for Sensor Integration in
advanced Robotic Systems
Bernhard Hulin
DB AG, Systemtechnik, TZF72, Völckerstr. 5, D-80939 München
Tel.: +49 (0)89 1308-7604, Fax: +49 (0)89 1308-2605, e-mail: bernhard.hulin@bku.db.de
The paper describes a software-architecture for sensor integration in robotic systems. Its supreme goals are flexibility,
extensibility and ease of use. These goals include hardware-aspects, such as scalability to the number of sensors, as
well as software-aspects, such as reconfigurability to parameters and algorithms.
The architecture is quite simple; it consists of two types of software structures. One is called ”enhanced-sensor”. The
other is called ”signal-merger”.
Each sensor is mapped to a software structure of the type ”enhanced-sensor”. The structure includes the digitized signal,
the calibration data of the sensor, additional meta-data and sensor-specific results as well as sensor specific functional-
ity. The ”signal-merger”, on the other hand, includes the strategy to interpret and process the signals, which are included
in the ”enhanced-sensor”-structures. In that way the processing and control logic is separated from the core data and its
functionality.
With this architecture it is possible to integrate arbitrarily sensors to a robotic system by using the same interface. For
example sensors can be cameras, radar, sonar, and inclination sensors. Due to abstraction of the architecture it is also
easily possible to use adjusting sensors or to adjust sensors during runtime, such as it is the case in active vision.
The architecture is based on current software fundamentals. It is based on the object-oriented programming paradigm
and uses extensively design patterns. It can be seen as an object-oriented framework for sensor integration.
To concretely show the benefits of this software-architecture the implementation of three cameras into a mobile obstacle
detection system is described.
1 Introduction
Developing robotic systems and sensor systems of-
ten spans a long period of time until it is reliable. During
that period normally many changes to the sensor-
hardware and improvements of the software are per-
formed. For that reason, a software architecture is
needed that simplifies changes to hard- and software.
There exist a lot of software-architectures for robotic
or control systems (e.g. [1-4]). Their architecture is for a
whole system. As a consequence most of these archi-
tectures are very complex and sometimes need an
environmental software. So potential developers do not
use them and go on developing systems in conven-
tional manner.
This is the more the case the less a complete ro-
botic or control system is needed. Especially in sensor
systems, such as in surveillance and inspection applica-
tions, less reactions are demanded. In surveillance
applications often just a warning on a display is de-
manded. For such systems an architecture for a whole
robot is too complex. Sufficient for those systems is a signal-processing-architecture which can easily be
integrated into robotic systems.
Many software-development tools exist for signal-
processing that allow quickly and simply creating pro-
grams, such as Measurement Studio [5]. An overview of
such tools for computer vision is given in [6]. These
tools are more or less function libraries, whose func-
tionality can be used in applications, but they do not
suggest a software-architecture for signal-processing
applications.
In this paper a simple software-architecture for sig-
nal-processing is described. It consists of just two types
of software structures. One is called ”enhanced-
sensor”. The other is called ”signal-merger”.
Since this software architecture is for a special prob-
lem domain it is a framework [7]. A framework generally
consists of a set of interfaces and abstract
types/classes whose interactions are predefined by the
framework.
This paper is structured as follows: First the problem
of sensor-integration is analyzed according to its needs.
Then, in section 3, the concluding framework based on
object-oriented principles is shown. Finally its benefits
and drawbacks are discussed theoretically and in the
light of a system used for obstacle detection for trains. Photomec
Second European Workshop on Photonics in Mechanical and industrial Processes
Louvain-la-Neuve, Belgium
February 20-21, 2003
