An essential feature of agents is a learning capability to overcome the limitations of built-in knowledge and pre-programmed skills. Since very early, learning algorithms have been the focus of many researchers, aiming at providing tools to increase the agent capabilities to face new or unexpected situations. Furthermore, agents can now be endowed with powerful embodiments, i.e. robot with advanced sensing and actuation capabilities. Thus, it is only natural to explore how exhisting learning paradigms can be adapted to the presence of an embodiment, and what new paradigms the embodiment stimulates.
For example, supervisory learning algorithms have been used to extract the key features of surgical operations, by observing the actions of surgeons performing selected procedures.
The paradigm that an embodiment supports, is "active learning" which implies probing the environments for cues and further knowledge. Within this framework, "learning by experimentation" implies a scientific method in probing the environment, by carrying out experiments to verify or falsify hypothesis about the nature and the properties of the surrounding environment. This approach requires "motivation" to trigger the experiment, "experimental design" to plan the experimental steps, and "repeatable execution" to ensure data correctness.
The proposed workshop will provide:
(i) an overview of existing learning methods in specific application domains,
(ii) discuss requirements and capabilities of methods for active learning,
(iii) outline R&D challenges, and (iv) motivate the search for new
application possibilities of robotic learning technology.
Debora Botturi
Department of Computer Science
University of Verona
E-mail: debora@metropolis.sci.univr.it
Paolo Fiorini
Department of Computer Science
University of Verona
E-mail: paolo.fiorini@univr.it
Erwin Prassler
Fachhochschule Bonn-Rhein-Sieg
Department of Computer Science, Germany
E-mail: erwin.prassler@fh-bonn-rhein-sieg.de
Monica Reggiani
Department of Computer Science
University of Verona
E-mail: reggiani@metropolis.sci.univr.it