A Developmental Approach for Affordance and Imitation Learning Through Self-Exploration in Cognitive Robots
Cognitive robotics is one of the branches of robotics that is concerned with the design and implementation of robots that will accomplish cognitive tasks such as perceiving the dynamic world around them and making acceptable decisions in real-time by imitating humans. A flexible and natural way to make robots learn new skills is to implement them with the ability to learn by imitation accompanied by effective affordance learning, which would be a feasible method to automate the tedious manual programming of robotic tasks. Imitation learning offers promising directions for gaining insight into faster affordance learning with which robots would have a good likelihood of learning complex behaviors from a small set of experiences much like human beings, and ultimately develop autonomous perceptual-motor control mechanisms. Thus, the goal of the dissertation is to develop a flexible mechanism for the robot so that it can learn high level motor tasks by experiencing action outcomes via its own sensors and forming an action-perception coupling similar to what happens in human beings. In order to learn basic behaviors, the robot goes through certain experimental stages involving self-exploration, affordance learning, and imitation learning which are parallel to the developmental pattern found in babies developing new motor skills. In this dissertation, using biologically inspired cognitive mechanisms such as affordance relations and imitation learning on a humanoid robot, we propose a new developmental approach for cognitive robots to learn novel motor behaviors in virtual and real environments.