Konferenzbeitrag | Links: | BibTeX-Datensatz | |
---|---|---|---|

Autoren: | |||

Quelle: | Proceedings of the IEEE Conference on Cognitive Methods in Situation Awareness and Decision Support, 2012. | ||

Konferenz: | IEEE Second International Multi-Disciplinary Conference on Cognitive Methods in Situation Awareness and Decision Support (CogSIMA), New Orleans, USA, 6.-8. März 2012 | ||

Modern autonomous systems are challenged by complex, overwhelming
computer processing power, though, time critical tasks. Handling
of reactive and proactive activities in real time requires an exceptionally
well designed autonomous system for constant situation awareness
and decision support. The basis for such situation awareness and
decision support is a robust and comprehensive representation of
the environment of the autonomous system, called world modeling.
The world modeling sub-system is responsible for a representation
of the current state of the environment, as well as a history of
past states and forecasts for possible future states. It receives
information from sensors, processes it and fuses into existing environment
description. Since the incoming information contains uncertainties
and can be treated, for example, by means of Degree-of-Belief (DoB)
distributions, powerful statistical methods can be employed for the
information fusion process (e.g. Bayesian fusion). The history of
past states allows for advanced information analysis, such as qualitative
situation estimation. On the other hand, a direct analysis of the
DoB distributions, for example, information entropy calculation,
gives a quantitative estimation of situations. The future states
can be predicted on the basis of known evolution parameters of the
environment, for example, by attributes and objects aging modeling.
The qualitative and quantitative situation estimations, as well as
the comprehensive environment description itself allows for permanent
situation awareness and intelligent support for decision making sub-systems.
Both information flow and modeling situation can be evaluated numerically
with the information entropy calculation. The difference between
entropies of an attribute before and after the observation fusion
gives a numerical estimation for the information gain. On the other
hand, the evaluation of entropies of all attributes can give an overall
estimation of the object representation. Extending entropy analysis
on groups of objects and their relations allows for numerical estimation
of situations.
In order to numerically estimate attribute sets of all modeling objects,
the entropy calculation must be unified for both discrete and continuous
DoB cases. In order to overcome the infinite discrepancy between
the entropy of quantized random variables and the entropy of discrete
random variables, the unification introduces a notion of the least
discernible quantum (LDQ). The LDQ defines the utmost precision for
any operation over the attribute.
The proposed analysis was developed within the German Research Foundation
(DFG) Collaborative Research Center (SFB) 588 ``Humanoid Robots --
Learning and Cooperating Multimodal Robots''. The main goal of the
project is to build a humanoid assisting robot. For development and
tests, a kitchen environment has been created as a test field. Within
this environment, several humanoid robots are cooperating with humans
and performing complex tasks, e.g. interactive objects and concepts
learning, context recognition, analysis of situations and intentions.