A Framework for Anchoring In Hybrid Deliberative Systems

A designer of robotic systems that act in the real world is faced with the problem of symbol anchoring. In a general theory symbol anchoring would consist of three distinct problems: 1) detection of the symbol, 2) extracting its properties, and 3) updating these properties over time. Solving all the three problems for every symbol is very hard and computationally expensive. We claim that this general endeavour can be simpli ed by dividing symbols into di erent classes based on the paradigm of hybrid deliberative architectures. Such an architecture functions on two di erent levels of abstraction: a reactive level and a deliberative one. A framework is proposed, where the anchoring is considered separately for each of these levels, since they need di erent types of symbols. On both levels the anchoring problem can be simpli ed not solving all of the three tasks above, which signi cantly decreases its complexity. We verify the functionality of our framework by designing a navigation system for domestic, large-scale environments. This system successfully achieves its missions with simple anchoring processes, little a priori knowledge about the environment, and limited computational complexity.