Eldad Assa and Ehud Ahissar
Weizmann institute of science, Rehovot, Israel
We hypothesize, based on biological data, that perception is a closed-loop process implemented by motor-sensory-motor (MSM) loops. In these loops the current sensory signal is a function of the sensed object and of the sensory organ motion which is a function of the previous sensory signal. Each such loop, according to the hypothesis, underlies the perception of a single feature.
We define, based on the framework of dynamical systems, a template of differential equations that can be used to model these loops. Perception is achieved in such systems through convergence of the dynamic variables to a steady state. The information of the perceived feature is present at the array of values of the different variables at this steady state.
Using this biologically derived model of perception and corresponding mathematical tools we aim to mimic the perceptual capacities of living organisms in artificial agents. A synthetic MSM loop: A simple robot which includes a DVS camera, a pan-tilt unit and a desktop computer, is used to implement CLP models and test their performance in hardware. As an example, contrast CLP model is described, and the results of its implementation are presented and compared to the results of a computer simulation. Testing this implementation with moving shapes stimuli reveals non-designed smooth pursuit like behavior.