Visual Attention Mechanisms Revisited

摘要

Currently robots are evolving into an increasing complexity and have to support a large workload that directly affects their functions. To compensate this situation they must make a better use of their available resources while behaving in a reliable way. The goal of this project is to endow Shelly, the social robot created by RoboLab with a predictive system of visual attention that allows it to maintain an updated internal representation of its environment, providing it with a basic sense of awareness. This improvement allows the robot to foresee simple facts, react to unpredicted situations and integrate changes of the environment in its internal memory. To achieve this level of functionality we have combined overt and covert head movements with an updatable internal model of the environment through a predictive and dynamic attention loop. The system has been developed using the Robo-Comp framework [21] and the new components have been integrated in the CORTEX cognitive architecture. The implementation is available for public use.