Maintaining Attentional Capacity in a Social Robot

Yuval Marom and Gillian Hayes

Abstract

Attention as the perception of change, or event detection, is important for an agent interacting with its physical and social environments. Internal modifications of the controller, in terms of adaptation of a decision threshold used to detect changes, is used to control the level of detail attended to, or the attentional effort. By maintaining effort within some capacity bounds, the agent maintains an attentional drive, which results in not only a `comfortable' level of processing, but also in desired performance. We present results from simulations of robotic learning by imitation, where an agent learns a task by following a capable teacher agent around the environment.

The preprint is available as gzipped dvi (13 kB) and postscript (77 kB) files.

Alternatively, you can request a copy by e-mailing me.