WHAT:
Guy Hoffman:
"Ensemble: Fluency and Embodiment for Robots Acting with Humans"

WHEN:
Thursday, July 26, 2007 3:30pm - 5:30pm

WHERE:
Bartos Theatre, MIT Media Lab (E15)

WEBCAST:
http://www.media.mit.edu/events/movies/video.php?id=guy-2007-07-26
The link will become active on the date and time scheduled for this event.

DISSERTATION COMMITTEE:
Cynthia Breazeal
LG Career Development Professor of Media Arts and Sciences
MIT Media Laboratory

Alex (Sandy) Pentland
Toshiba Professor of Media Arts and Sciences
MIT Media Laboratory

Lawrence Barsalou
Samuel Candler Dobbs Professor of Psychology
Emory University

ABSTRACT:
This thesis is concerned with the formulation and exploration of an embodied cognitive architecture for robotic agents, enabling them to perform fluently with their human counterparts. Thus, our effort is to steer away from the stop-and-go rigidity present in virtually all human-robot interaction to date.

We define fluency as the ethereal yet manifest quality existent when two humans perform together at an impressive level of coordination, flexibility, and adaptation, in particular when they are well-accustomed to the task and to each other. In these cases their timing is precise and efficient, they alter their plans and action appropriately and dynamically, and this behavior emerges often without exchanging much verbal information.

We argue that one of the keys to this goal is the adaptation of an embodied approach to robot cognition. Based on mounting psychological and neurological evidence, embodied views of human and animal intelligence are gaining support. We show how central ideas from this theory are applicable to robot cognition and present a cognitive architecture making use of perceptual memory, simulation, emulation, and perception-action loops.

In addition, we demonstrate that anticipation of perceptual input, and in particular of the actions of others, are an important ingredient of fluent joint action. We show results from a study of the effects of anticipatory action on fluency and teamwork, and use these results to suggest benchmark metrics for fluency. We also show the relationship between anticipatory action and the above-mentioned simulator/emulator approach to perception, through a comparative study of an implemented architecture on a the robot AUR, a robotic desk lamp.

A result of this work is modeling the effect of practice on human-robot joint action, and similarly that of rehearsal on a human-robot joint performance. We argue that mechanisms that govern the passage of cognitive capabilities from a deliberate and flexible, yet comparatively slower system to a faster, sub-intentional, and more rigid one, are crucial to fluent joint action in well-rehearsed teams and ensembles. Also, we claim that the reinforcement of anticipatory action plays a role in practice.

Theatrical acting theory is a major inspiration for this work. We show how lessons from acting methods can be applied to fluent human-robot interaction, as well as demonstrate some of the methodology developed in a joint human-robot theatrical performance.