CSAIL Publications and Digital Archive header
bullet Research Abstracts Home bullet CSAIL Digital Archive bullet Research Activities bullet CSAIL Home bullet

link to publications.csail.mit.edu link to www.csail.mit.edu horizontal line


Research Abstracts - 2007
horizontal line

horizontal line

vertical line
vertical line

A Computational Approach to Emotional Robot Learning

Juan D. Velásquez

The Problem

The primary focus of this research is to investigate affect from a computational perspective by implementing models of affective processing and learning on a variety of robotic and simulation platforms.


Affect is inherently intertwined with several attributes that we associate with intelligent behavior, including multi-modal sensorimotor integration, attention, motivation, decision-making, and learning. Thus, in order to understand intelligence, we should also attempt to understand emotions and affective processing.

Building affective robots offers a unique opportunity to address scientific questions regarding the nature of affective processing in humans and animals. Given that we have a direct window into the robot’s control systems, it is possible for us to selectively manipulate our models and perform testing in a controlled and repeatable environment. This facilitates the comparison of our implementations with respect to cognitive, ethological, or neurobiological models, thus giving us some insight into the validity of those models and into affective processing in general.


Our work focuses on the design and implementation of appropriate abstractions for affective processing. Drawing inspiration from recent advances in Affective Neuroscience [1], we have created several affect-based control systems for various mobile robotic and simulation platforms. These systems range from simple affective-based reflexes, to the facilitation of attention, and affective learning.

We have also developed a methodology for building robots that follows an affect-based decomposition. This methodology stresses the use of computational models of affective processing to build and control intelligent systems that are capable of performing a variety of complex behaviors in the real world.

Our methodology accentuates the notion of building complete systems that not only integrate models of affect, but that also take into account systems that mediate perception, attention, motivation, behavior, learning, and motor control. We propose the idea of an Affect Program as a useful abstraction that offers a natural decomposition for this task. Affect Programs integrate a variety of sensory information and synchronize a number of functions in response to biologically significant events. Thus, they are well suited to act effectively as an integration mechanism by which activity in many different systems is bound together in a coherent manner.

This methodology also addresses the issues of learning and development, which we see as an extensive and gradual process by which organisms acquire increasingly elaborate behaviors and new abilities. Recent work in robotics has begun to deal with issues of cognitive development. Affective development, on the other hand, is a new challenge that lies at the core of this work.

Our approach to this end focuses on the interactions among multiple learning systems, especially those with parallel associative learning schemes that are triggered by, and require the intervention of affectively significant (or unconditioned) stimuli. This includes those types of learning commonly referred to as incentive learning and reward-based learning, which can focus the robot’s attention and reduce the learning space by providing information concerning when to learn and what to learn.


Our current work has focused on the building blocks of affective processing. Future work will concentrate on the design and implementation of effective mechanisms for incentive learning, the process through which sensory features of appetitive (or aversive) events acquire affective significance and, hence, the ability to motivate or elicit responses and actions. [2]


[1] Jaak Panksepp. Affective Neuroscience. Oxford University Press, 1998.

[2] A. Dickinson, and B.W. Balleine, B.W. The role of learning in the operation of motivational systems. In Learning, Motivation and Emotion, Volume 3 of Steven’s Handbook of Experimental Psychology (3rd edn) (C.R. Gallistel, ed.), pp. 497--533, John Wiley & Sons, 2002.


vertical line
vertical line
horizontal line

MIT logo Computer Science and Artificial Intelligence Laboratory (CSAIL)
The Stata Center, Building 32 - 32 Vassar Street - Cambridge, MA 02139 - USA
tel:+1-617-253-0073 - publications@csail.mit.edu