A head-neck-eye system that learns fault-tolerant saccades to 3-D targets using a self-organizing neural model.

Abstract

This paper describes a head-neck-eye camera system that is capable of learning to saccade to 3-D targets in a self-organized fashion. The self-organized learning process is based on action perception cycles where the camera system performs micro saccades about a given head-neck-eye camera position and learns to map these micro saccades to changes in position of a 3-D target currently in view of the stereo camera. This motor babbling phase provides self-generated movement commands that activate correlated visual, spatial and motor information that are used to learn an internal coordinate transformation between vision and motor systems. The learned transform is used by resulting head-neck-eye camera system to accurately saccade to 3-D targets using many different combinations of head, neck, and eye positions. The interesting aspect of the learned transform is that it is robust to a wide variety of disturbances including reduced degrees of freedom of movement for the head, neck, one eye, or any combination of two of the three, movement of head and neck as a function of eye movements, changes in the stereo camera separation distance and changes in focal lengths of the cameras. These disturbances were not encountered during motor babbling phase. This feature points to general nature of the learned transform in its ability to control autonomous systems with redundant degrees of freedom in a very robust and fault-tolerant fashion.