A developmental algorithm for ocular–motor coordination

Abstract

This paper presents a model of ocular-motor development, inspired by ideas and data from developmental psychology. The learning problem concerns the growth of the transform between image space and motor space necessary for the control of visual saccades. An implementation is used to produce experimental results and these are presented and discussed. The algorithm is simple, extremely fast, self calibrating, adaptive to change, and exhibits emergent stages of behaviour as learning progresses.

Learning the visual–oculomotor transformation: Effects on saccade control and space representation

“…In robotics, both encodings have been used to represent such a transformation. For example Marjanovic et al (1996); Chao et al (2010); Shibata et al (2001) employed the monocular encoding, while Schenck and Moller (2004);Forssén (2007) used the binocular one. Only Bruske et al (1997) compared the two approaches but only in terms of saccade accuracy.…”

“…In its original formulation, the direct inverse modeling consists in performing random movements and then learning the inverse association between the motor command and its perceptual outcome. This technique was employed in the learning of saccade control in several works together with some ad hoc strategies to reduce the exploration process (Schenck and Möller, 2006;Chao et al, 2010;Antonelli et al, 2013a). The main drawbacks of this approach are that it does not cope with redundant systems, and it is not goal-directed.…”

“…The accuracy of this approach depends on the granularity of the cells and suffers the curse of dimensionality of the input space. This approach has been used in saccade control to map the monocular visual position of the stimulus (2 dimensions) into an eye movement (Marjanovic et al, 1996;Chao et al, 2010).…”

“…For example, Chao et al (2010), in their algorithm for visualoculomotor coordination, considered a target to be in the fovea when it was inside a radius of 20 pixels from the center of the image, that in their case had a resolution of 300 pixels. In our experiments, considering a fovea with a radius of 20 pixels in an image with a resolution of 1024 × 768 pixels, a saccade brings the target inside the fovea in the 97.9% of the trials for the monocular case and in the 99.99% for the binocular one.…”

An Infant Development-inspired Approach to Robot Hand-eye Coordination

“…These maps and the link mechanism are described in the following sub-sections. The visuo-motor system and the hand sensory-motor system are implemented by respective sensory-motor coordination models, whose prototypes have been used in our previous work [31,35]. The sensor-motor coordination models are based on a basic sensory motor map structure.…”