Context preference-based deep adaptive resonance theory network for the cognitive memory architecture of intelligent robots

Abstract

Intelligent robots need to acquire knowledge gradually as time goes by in order to perform complex tasks. Thus, a memory model should be incorporated to the intelligent robots for enhancing its cognitive abilities. To address this challenge, we propose a cognitive memory architecture, called Context Preference-based Deep Adaptive Resonance Theory (CPD-ART) network. CPD-ART network integrates the models of human episodic and semantic memories along with a preference-based decision making module. Taking advantage of the episodic memory encoding and retrieval frameworks, our cognitive memory architecture could learn complex tasks and recall the complete sequence of the events perfectly. The semantic memory provides the conceptual meaning of every action, object, or environment information required by the episodic memory and the preference-based decision making module while retrieving the events sequence of the learned tasks. On the other hand, the preference-based decision making module affects the cognitive memory retrieval decision by introducing new choices to alter the task accordingly considering the current external situations such as weather or time. The simulation and experiment to verify the effectiveness of the proposed architecture are also presented.