Human-Robot Interactions as a Cognitive Catalyst for the Learning of Behavioral Attractors

We address in this paper the problem of the autonomous online learning of a sensory-motor task, demonstrated by an operator guiding the robot. For the last decade, we have developed a vision-based architecture for mobile robot navigation. Our bio-inspired model of the navigation has already proved to achieve sensory-motor tasks in real time both in unknown indoor and outdoor environments. We propose to bootstrap the underlying PerAc architecture in order to control the sensory-motor learning. The interaction leads the robot to autonomously build a precise sensory-motor dynamic approximating the behavior of the teacher. A real dialog based on actions imposed by the teacher and those proposed by the robot emerges, which catalyzes the learning of the robot. The architecture is finally tested in real indoor and outdoor environments.