An Inhibition of Return Mechanism for the Exploration of Sensorimotor Contingencies

Abstract

The modelling of cognition is fundamental to designing robots that are increasingly more autonomous. Indeed, researchers take inspiration from human and animal cognition in order to endow robots with the ability to learn and adapt to their environment. In specific cases, the robot has to find the right compromise between exploring the environment, or exploiting its own experience to advance its knowledge of a skill. Our approach considers a neurally-inspired model to learning sensorimotor contingencies based on exploration and exploitation. For the exploration, an inhibition of return mechanism is implemented that generates new actions. In this work, we investigate how the tuning of the inhibition of return affects the exploratory behavior. To do so, we set up an experiment where a 3D printed humanoid robot arm GummiArm has to learn how to move a baby mobile toy with only a visual feedback. The results demonstrate that the tuning of the inhibition of return influences the exploratory behavior, leading to a faster learning of sensorimotor contingencies as well as the exploration of a reduced motor space.