One of my goals is to work towards developing Cognitive Robots, especially with regard to improving the functionalities that facilitate the interaction with human beings and their surrounding objects. Any cognitive system designated for serving human beings must be capable of processing the social signals and eventually enable efficient prediction and planning of appropriate responses.

My main focus during my PhD study is to bridge the gap between the motoric space and the visual space. The discovery of the mirror neurons ([RC04]) shows that the visual perception of human motion (visual space) is directly associated to the motor control of the human body (motor space). This discovery poses a large number of challenges in different fields such as computer vision, robotics and neuroscience. One of the fundamental challenges is the understanding of the mapping between 2D visual space and 3D motoric control, and further developing building blocks (primitives) of human motion in the visual space as well as in the motor space.

First, I present my study on the visual-motoric mapping of human actions. This study aims at mapping human actions in 2D videos to 3D skeletal representation. Second, I present an automatic algorithm to decompose motion capture (MoCap) sequences into synergies along with the times at which they are executed (or “activated”) for each joint. Third, I proposed to use the Granger Causality as a tool to study the coordinated actions performed by at least two units. Recent scientific studies suggest that the above “action mirroring circuit” might be tuned to action coordination rather than single action mirroring. Fourth, I present the extraction of key poses in visual space. These key poses facilitate the further study of the “action mirroring circuit”. I conclude the dissertation by describing the future of cognitive robotics study.