Cognitive architecture for an Attention-based and Bidirectional Loop-closing Domain (CABILDO)

• Autores: Antonio Jesús Palomino López
• Directores de la Tesis: Rebeca Marfil Robles (dir. tes.), Juan Pedro Bandera Rubio (dir. tes.)
• Lectura: En la Universidad de Málaga ( España ) en 2014
• Idioma: inglés
• Número de páginas: 222
• Tribunal Calificador de la Tesis: Jorge Manuel Miranda Dias (presid.), Juan Antonio Rodríguez Fernández (secret.), Rubén Martínez Cantín (voc.), José Ramón Fernández Vidal (voc.), Antonio José Rodríguez Sánchez (voc.)
• Enlaces
  • Tesis en acceso abierto en: RIUMA
• Resumen

  • This Ph. D. Thesis presents a novel attention-based cognitive architecture for social robots. The architecture aims to join perception and reasoning considering a double and simultaneous imbrication: the ongoing task biases the perceptual process to obtain only useful elements whereas perceived items determine the behaviours to be accomplished. Therefore, the proposed architecture represents a bidirectional solution to the perception-reasoning-action loop closing problem. The basis of the architecture is an Object-Based Visual Attention model. This perception system draws attention over perceptual units of visual information, called proto-objects. In order to highlight relevant elements, not only several intrinsic basic features (such as colour, location or shape) but also the constraints provided by the ongoing behaviour and context are considered. The proposed architecture is divided into two levels of performance. The lower level is concerned with quantitative models of execution, namely tasks that are suitable for the current work conditions, whereas a qualitative framework that describes and defines tasks relationships and coverages is placed at the top level. Perceived items determine the tasks that can be executed in each moment, following a need-based approach. Thereby, the tasks that better fit the perceived environment are more likely to be executed. Finally, the cognitive architecture has been tested using a real and unrestricted scenario that involves a real robot, time-varying tasks and daily life situations, in order to demonstrate that the proposal is able to efficiently address time- and behaviour-varying environments, overcoming the main drawbacks of already existing models.