Towards a Mirror System for the Development of
Socially-Mediated Skills
Yuval Marom5 George Maistros and Gillian Hayes
Institute of Perception, Action, and Behaviour,
Division of Informatics, University of Edinburgh,
5 Forrest Hill, Edinburgh, EHl 2QL, UK
{yuvalm, georgem, gmh}Odai.ed.ac.uk
Abstract
Wc present a system that attempts to model
the functional role of mirror neurons, namely
the activation of structures in response to
both the observation of a demonstrated task,
and its generation. Through social situated-
ness and a sot of innate skills, perceptual and
motor structures develop for recognition and
reproduction of demonstrated actions. Wo be-
lieve this is an implementation towards a mir-
ror system, and wo tost it on two platforms,
one in simulation involving imitation of object
interactions, the second on a physical robot
learning from a human to follow walls.
1. Introduction
Epigonotic robotic systems should contain compo-
nents that are capable of being shaped by the interac-
tions of an agent with its built environment as well
as by the social interactions between agents. Our
work deals with a particular social learning model
that incorporates learning by imitation, and tem-
poral attention. The experience of a learner robot
is governed by the actions of a demonstrator, so
that the learner gets to sample only those particular
parts of the perceptual space pertaining to the skills
the teacher is demonstrating. Thus, the social sit-
uatedness of the learner and teacher crucially influ-
ences the structures that develop within the learner’s
‘brain’.
Wo are inspired from the tight coupling between
perception and motor control found in mirror neu-
rons. Mirror neurons wore found in the macaque
monkey brain and they wore shown to have both vi-
sual and motor properties. In fact, single neuron
studios by Gallcsc ct al. (1996) and Rizzolatti ct al.
(1996) explored further the properties of mirror neu-
rons and exposed a strong relationship between per-
ception and motor control. Mirror neurons fire both
when the monkey performs an action and when it
observes another monkey or the experimenter per-
form that same action. Based on those properties
it is believed that such a mirror system may form
the fundamental basis for imitation (Rizzolatti ot al.,
2000).
The architecture presented in this paper, shown in
Figure 1, is an attempt to model the mirror system,
i.o. the functional role of mirror neurons. However,
there is no evidence from Nouroscionco on how the
mirror system is learnt or built. Thus, rather than
arbitrarily building a mirror system a priori, wo use
a machine learning approach for this purpose, itself
inspired from Psychology and Biology.
Initially the mirror system contains no structures,
and wo believe that a socially situated agent can de-
velop such structures from observation of a demon-
strated action, and then utilise them to reproduce
that action. Our architecture relies on the existence
of simple innate reactive skills for motor control.
Those are merely responsible for the inverse kine-
matics of the physical system, handled by an inverse
model.
2. Mirror System
As shown in Figure 1, the mirror system is formed
by the coupling of perceptual structures (nodes) and
motor structures (schemas). The input to the mir-
ror system is a continuous perceptual stimulus that
involves both the observed demonstrator and the im-
itator’s perceived physical environment (e.g. objects
and walls). The output of the mirror system ex-
presses the desired target state of the agent (e.g. pos-
tural targets) and goes directly to the motor system
for execution.
Each perceptual structure is used to recognise a
temporal chunk of the perception of the agent, and
is hard-wired to a single motor structure that holds
motor targets. The targets can potentially achieve
the recognised part of the action, using the innate
skills stored in the inverse model. The coupling be-
tween a perceptual and motor structure expresses the
ability of the agent to perform a particular part of the
recognised action. Currently, those associations fol-
low a one-to-one relationship, and in on-going work