A Primer on Embodiment - On the Interaction
of Brain, Morphology, Materials, and Environment in Adaptive Behavior
• Organizer : Rolf Pfeifer, Artificial
Intelligence Laboratory, University of Zurich, Switzerland
• Abstract : Traditionally, in artificial
intelligence, neuroscience, and robotics there has been a focus
on the study of the control or the neural system itself. Recently
there has been an increasing interest into the notion of embodiment
in all disciplines dealing with adaptive behavior, including psychology,
philosophy, and linguistics. In this tutorial, I introduce the
basic underlying principles of embodiment and explore its far-reaching
and often surprising implications. While embodiment has often
been used in its trivial meaning, i.e. "intelligence requires
a body", there are deeper and more important consequences,
concerned with connecting brain, body (morphology, materials),
and environment, or more generally with the relation between physical
and information (neural, control) processes. A number of principles
will be introduced that characterize embodied systems. For example,
morphology and materials can take over some of the functions normally
attributed to control (the principle of "ecological balance").
Also, it can be shown that through the embodied interaction with
the environment, in particular through sensory-motor coordination,
information structure is induced in the sensory data, thus facilitating
perception and learning (the principle of "information self-structuring")
which is an important way in which “the body shapes the way we
think”, so to speak. A number of case studies are presented to
illustrate the concepts introduced. Moreover, a basic theoretical
framework for the study of intelligent adaptive systems is introduced.
Multiple Object Tracking in Clutter
• Organizers : Darko Musicki, Entropy
Data Pty Ltd, Australia
• Abstract : The objective of this tutorial
is to provide an introduction to tracking of multiple objects
in clutter, starting with the problem statement and functional
requirements, leading on to advanced techniques capable of tracking
a large number of objects in clutter in real time. Some practical
issues and problems are addressed.
Remote surveillance sensors, e.g. video/IC cameras, radars and sonars, return measurements (detections) from objects we need to localize and track, as well as detections from unwanted objects, usually termed clutter. Thermal noise, multipath propagation effects, etc. may also cause clutter measurements. In addition, object measurements are present in each scan only with a probability of detection. Thus, the source of measurements provided by the sensors is unknown.
Object tracker has no a-priori information on the existence and
the number of objects. Algorithms for tracking multiple objects
in clutter need to determine the existence and number of objects,
as well as to estimate their trajectories. Automatic track initialization
uses available measurements to start new tracks. In cluttered
environment this will initialize both true tracks, which follow
an object, and false tracks, which "follow" clutter
measurements. Trackers need to distinguish between true and false
tracks in order to confirm true tracks and terminate false tracks.
Object tracking algorithms presented in this tutorial are based
on the object existence paradigm. Each track recursively updates
the probability of object existence, i.e. the probability that
it is a true track. When the probability of object existence rises
above a confirmation threshold, the track is declared to be a
true track and confirmed. When the probability of object existence
falls below a termination threshold, the track is declared to
be a false track and terminated. Thus, each track has a hybrid
state, with a binary component, object existence, and a continuous
component, which is the object trajectory state estimate.
The main problem in object tracking is the computational complexity.
Optimal object tracking has multidimensional computational complexity.
The computational complexity grows exponentially in time, and
combinatorially with the number of measurements and the number
of objects sharing the measurements. Thus, optimal object tracking
with the current state of the art of computer technology can be
achieved only for a small number of scans and in benign environment
with a small number of objects and low density of clutter measurements.
Practical object tracking algorithms are of necessity suboptimal.
This tutorial will cover a suite of object tracking algorithms
with different performance / complexity tradeoffs. Finally, some
implementation issues will be covered.
