Implementing Selective Attention in Machines: The Case of Touch-Driven Saccades
Files
Penn collection
Degree type
Discipline
Subject
active perception
autonomous robots
machine learning
Funder
Grant number
License
Copyright date
Distributor
Related resources
Author
Contributor
Abstract
Recent paradigms in the fields of robotics and machine perception have emphasized the importance of selective attention mechanisms for perceiving and interacting with the environment. In the case of a system involved in operations requiring a physical interaction with the surrounding environment, a major role is played by the capability of attentively responding to tactile events. By performing somatosensory saccades, the nature of the cutaneous stimulation can be assessed, and new motor actions can be planned. However, the study of touch-driven attention, has almost been neglected by robotics researchers. In this paper the development of visuo-cutaneo coordination for the production of somatosensory saccades is investigated, and a general architecture for integrating different kinds of attentive mechanisms is proposed. The system autonomously discovers the sensorymotor transformation which links tactile events to visual saccades, on the basis of multisensory consistencies and basic, built-in, motor reflexes. Results obtained both with simulations and robotic experiments are analyzed.