Long Term Symbol Anchoring for Dynamic and Realistic Environments

One of the most important challenges that intelligent systems are now facing is to acquire the ability to operate in realistic environments and interact with humans for long periods of time. This project considers a key aspect of this challenge namely to give intelligent systems acting in physical environments the flexibility to adapt to changes in the environment, and to acquire dynamically new knowledge needed in the interaction with humans. In particular, we concentrate on the natural interaction and common changes that pertain to objects which are used in everyday activities. The intelligent system needs to be able to refer to these objects in the physical world and to associate to them both the general common sense knowledge about the object and the specific knowledge that is relevant for the task at hand. This aspect is part of the study of anchoring, that is, the process to establish and maintain in time the correspondence between symbolic knowledge and sensory information about physical objects.

This continuation project builds on the results achieved until now on anchoring and extends it by leveraging upon methods like knowledge retrieval from the Web, activity recognition, and reasoning. Besides exploring theoretical aspects, we will test and validate our research in an implemented and integrated system in a smart home for elderly. An apartment in an elderly residential complex will be used for realistic interactions with users.