Knowledge acquisition through introspection in Human-Robot Cooperation

Abstract

When cooperating with a team including humans, robots have to understand and update semantic information concerning the state of the environment. The run-time evaluation and acquisition of new concepts fall in the critical mass learning. It is a cognitive skill that enables the robot to show environmental awareness to complete its tasks successfully. A kind of self-consciousness emerges: the robot activates the introspective mental processes inferring if it owns a domain concept or not, and correctly blends the conceptual meaning of new entities. Many works attempt to simulate human brain functions leading to neural network implementation of consciousness; regrettably, some of these produce accurate model that however do not provide means for creating virtual agents able to interact with a human in a teamwork in a human-like fashion, hence including aspects such as self-conscious abilities, trust, emotions and motivations. We propose a method that, based on a cognitive architecture for human-robot teaming interaction, endows a robot with the ability to model its knowledge about the environment it is interacting with and to acquire new knowledge when it occurs.