A developmental principle for robotic hand-eye coordination skill

Hand-eye coordination is an important skill that a person performs effortlessly. It is a typical scenario that links perception to action in an iterative manner because human vision is not quantitative. Therefore, it's an interesting topic to develop a computational principle that imitates humanlike hand-eye coordination. This paper presents a developmental framework to address the first issue of mapping the visual sensory data into the hand's motion parameters in the Cartesian space. The authors extend their original model (1996, 1997) to the case of using a quadratic model with the attempt to achieve better performance. Hand-eye coordination involves the arm's kinematics modelling. Especially, the computation of inverse kinematics is not a simple process. We propose a new scheme, named as discrete kinematic mapping, to directly map the coordinates in task space into the coordinates in joint space at different levels of granularity that can be gradually refined through the development. This new approach may serve an evidence to explain why human brain can effortlessly undertake complex motions of the two arm/hand systems regardless the complexity of inverse kinematics.