A bilingual cognitive robot that learns like a toddler

Abstract

We propose an embodied architecture featuring a developmental agent and a social robot for human-robot verbal engagement at preschool level. Initially, we modelled bilingual acquisition and demonstrated the agent’s skill to appropriately detect the spoken content and automatically match the human user’s language. We aim to contribute at designing multilingual robot agents able to understand and communicate contexts in multiple tongues without loss of meaning in translations, which can greatly affect human cognitive and interaction capabilities. Furthermore, we design a novel methodology to teach the robot how to learn gradually from experience new knowledge not specified at design time and directly apply it for task solving, without being trained anew. The demonstrated system shows promising potential to becoming a low-cost social companion that can be easily integrated in the human world.