Beyond Geometric Mapping
Henrik I Christensen
Centre for Autonomous Systems
Royal Institute of Technology
SE-100 44 Stockholm, Sweden
hic@kth.se
Abstract
Autonomy for mobile systems inherently requires localisation of a
system to ensure that it is not getting lost. In addition localisation
is also an important service for goal achievement. For operation in
natural environments -- be it domestic houses or outdoors -- there is
often a need to perform concurrent localisation and mapping. The
problem is typically referred to as Simultaneous Localisation and
Mapping (SLAM). So far most of the research in SLAM has emphasized a
purely geometric approach to mapping. Such an approach is well suited
for fully autonomous systems, but when a system cooperates with a human
there is a need to provide a common representation for the mapping.
Euclidean geometry (SE(3)) is not necessarily the best possible basis
for such a representation. People typically refer to places and
locations in terms of their location relatively to landmarks or
relatively to local topological regions. In addition such places and
locations are referred to through a semantic label -- "the kitchen",
"Henrik's office", etc. Such representations might not be suitable for
robot localisation. There is consequently a need to establish a common
basis in which both objectives can be accommodated. We will refer the
overall process as Human Augmented Mapping. Neither the robot, not the
human operator is perfect, and there is thus a need for a dialogue
between the user and the robot to resolve ambiguities.
The presentation will discuss a SLAM based representation using
graphical models, associated topical information and annotated semantic
information. Inference and uncertainty modelling within the
representation is briefly outlined. In addition integration of a
language model for description of spatial references is introduced to
allow "discussion" of ambiguous situations. The system has been tested
across a number of different environments and users. In addition to a
model of the overall system and its architecture, empirical results are
also presented for three different robots, in four environments across
10 users.
Biography
Henrik I Christensen is the director of the Centre for Autonomous
Systems at the Royal Institute of Technology, and a chaired professor
of computer science specialising in autonomous systems, with the
Department of Computer Science and Numerical Analysis.
His research is on mobile robotics, autonomous systems, computer
vision, and biologically inspired robot systems. The overall emphasis
is on a holistic approach to design of systems, including
mathematically well defined methods for design, analysis and
implementation of systems. A fundamental idea is that methods should be
evaluated in realistic settings which involves an interesting scenario
and a full systems context.
He is a co-founder of the company Intelligent Machines and serve as a
scientific advisor to Evolution Robotics. In addition he is actively
involved in a number of community efforts in particular as the
coordinator of the EU network of excellence in Robotics - EURON, that
coordinates research, education, industrial links and publishing across
Europe. He is a fellow of the International Foundation of Robotics
Research and serves as an IEEE RAS
distinguished lecturer (2004-2006).