A new neural architecture based on ART and AVITE models for anticipatory sensory-motor coordination in robotics

Abstract

In this paper a novel sensory-motor neural controller applied to robotic systems for reaching and tracking targets is proposed. It is based on how the human system projects the sensorial stimulus over the motor joints, sending motor commands to each articulation and avoiding, in most phases of the movement, the feedback of the visual information. In this way, the proposed neural architecture autonomously generates a learning cells structure based on the adaptive resonance theory, together with a neural mapping of the sensory-motor coordinate systems in each cell of the arm workspace. It permits a fast openloop control based on propioceptive information of a robot and a precise grasping position in each cell by mapping 3D spatial positions over redundant joints. The proposed architecture has been trained, implemented and tested in a visuo-motor robotic platform. Robustness, precision and velocity characteristics have been validated.