Hopfield Net spreading activation for grounding of abstract action words in cognitive robot

Abstract

Human intelligence is the main inspiration for the cognitive robotics field, and language is one of the unique qualities of humans. The increasing interest in acquisition of language in cognitive agents/robots with grounded phenomenon facilitates robotic services in real scenarios in real means through stepping into the shoes of human intelligence. In this paper, a cognitive robotics model is proposed for the grounding of a higher-order concept in the sensorimotor experience of humanoid robot. In the proposed model, robot's conceptual knowledge is hierarchically organized in a semantic network by using the inputted linguistic description of these words. The meaning of the higher-order word is indirectly grounded in sensorimotor primitives through a top-down activation process. Specifically, in the described model, the meaning of the abstract action word is acquired through the combination of already grounded action primitives. Hence, the proposed model with the strength of symbolic computation extends the grounding mechanism to the abstract words category, and also provides a tool (i.e. a practical way) to investigate the link that exists between sensorimotor representation and abstract conceptual knowledge of the cognitive robot. For verification of the proposed model, simulation- experiments have been conducted by using iCub robot data.