Classification, Sensor Fusion, and Tracking
The MAP (MAP Anchors Perceptions) module allows to build the
environment model on the basis of the data acquired
through sensors. MAP contains a hierarchical conceptual model, in which
are specified the classes of objects that can be perceived.
The anchoring process is divided into three phases:
In the classification phase (carried out by the TIGER sub-module),
MAP receives from sensors descriptions
of perceived objects in terms of percepts (symbolic features),
and identifies the concept that has the best matching degree.
At the end of the classification phase, the TIGER produces the
conceptual instances related to the perceived objects and their
degree of reliability.
Since the same physical object may be perceived at the same time
by distinct sensors, the output of the TIGER may contain several
conceptual instances that are related to the same physical
object. In the merging phase, the FUSION sub-module merges those
conceptual instances that are supposed to be originated from the
perception of the same object by different sensors.
The tracking phase of anchoring consists of maintaining in
time a coherent state of the model and a correct classification
of instances. This phase tries to match the conceptual instances
perceived in the past with those generated by the latest
perception (data association). Then, the dynamic properties of
the conceptual instances are updated through Kalman filtering.
This approach is well suited for multi-agent domains. It is expected
that in a multi--agent context each agent could take advantage of data
perceived by its teammates. In fact, each agent can be seen as an
intelligent sensor that, instead of producing features, generates
conceptual instances that can be processed directly by the FUSION
sub-module of another agent.
Bonarini, A., Matteucci M., Restelli M. (2001) Anchoring: do we need new solutions
to an old problem or do we have old solutions for a new problem? Proceedings
of the AAAI Fall Symposium on Anchoring Symbols to Sensor Data in Single and
Multiple Robot Systems, AAAI Press, Menlo Park, CA, 79-86.
Bonarini, A., Matteucci, M. and Restelli, M. (2001) A framework for robust
sensing in multi--agent systems. In A. Birk, S. Coradeschi (Eds.) RoboCup
2001--Robot Soccer World Cup V, Lecture Notes in Computer Science, Springer
Verlag, Berlin, D.