Investigating a Robot as a Therapy Partner for Children with
Autism
Iain Werry1 & Kerstin Dautenhahn2 & William Harwin1
1. Department of Cybernetics, University Of Reading, UK
2. Department of Computer Science, University of Hertfordshire, UK
Key Words: autism, micro-behaviour, robot
Abstract:
The aurora project is investigating the possibility of using a robotic platform as a therapy aid for
children with autism. Because of the nature of this disability, the robot could be beneficial in its ability
to present the children with a safe and comfortable environment and allow them to explore and learn
about the interaction space involved in social situations. The robotic platform is able to present
information along a limited number of channels and in a manner which the children are familiar with
from television and cartoons. Also, the robot is potentially able to adapt its behaviour and to allow the
children to develop at their own rates. Initial trial results are presented and discussed, along with the
rationale behind the project and its goals and motivations. The trial procedure and methodology are
explained and future work is highlighted.
Introduction:
The term autism encompasses a range of disabilities on the Autism Spectrum,
including Aspergers Syndrome (sometimes referred to as high functioning autism),
Pervasive Developmental Disorder – Not Otherwise Specified (PDD-NOS) and
autism. The spectrum is defined by its effects on the individual, and although these
can be wide ranging and diverse between people, a few common features exist. The
most prominent of these are categorised as the Triad of Impairments by the National
Autistic Society (NAS). They are:
• Deficits in social interaction
• Deficits with social communication
• Deficits and limitations in the areas of imagination and generalisation
Autism is a disability which affects between five and fifteen people in every ten
thousand and there is currently little evidence of the reasons behind the disability or
who is most at risk from it. Autism is a present from the very first moments of a
child’s life and there is currently no cure for it, but if it is diagnosed early in the
child’s development then certain therapy techniques can help the child to compensate
and to learn to cope with the disability in the real world more effectively.
One of the more popular therapy techniques is that of TEACCH (Treatment and
Education of Autism and Related Communication Handicapped Children) (Watson et
al, 1989). This approach centres on the child being presented with a range of
situations which he should respond to. Appropriate responses are rewarded and
reinforced. The child’s lack of ability to generalise can make this process lengthy, but
the situations are controlled in order to give the child a range of similar experiences to
learn from.

---

The Aurora Project:
The prospect of using technology to enhance our understanding of autism and to help
those who suffer from it was investigated in 1976 (Weir & Emanuel, 1976). An
autistic boy was introduced to a logo turtle, operated via a set of buttons, however this
case study gives little details of the boys background and functioning ability. Other
researchers have also realised the potential of computers for enhancing the autistic
child’s interaction capabilities, for example see Murray & Lesser’s Autism and
Computing web site and Michaud et al, 2000, who developed interesting robotic
designs for children with autism.
The Aurora Project (Werry & Dautenhahn, 1999; Dautenhahn & Werry, 2000) was
started in 1998 to investigate the use of a robotic platform as a therapy aid for children
with autism. It was thought that a robot as a teaching aid for therapists holds a number
of advantages over the use of only a human therapist. While autistic children focus on
individual details of a scene, they may miss the more general picture of events due to
the number of features and articles which require their attention. A robot is able to
function and communicate in a limited number of ways, allowing the child to focus on
a few communication channels and not miss any of the details. Additionally, while the
stress of interaction and learning with a human teacher may place extra pressure on
the child, interaction with a robot – which is often a familiar and safe environment
due to exposure from television and similarities with the child’s toys – reduces this
pressure and allows the child to relax and enjoy the interaction, leading to a stronger
learning environment.
Long term goals of the project include the development of a robotic platform which
can be used in schools by teachers and therapists to allow autistic children to practise
their social development and interaction skills learned in other classes. In the short
term, the robots effectiveness must be evaluated so that we are able to determine
better the strengths and weaknesses of this approach. The level of interaction with the
robot by the children is an important factor in its success and so the children should
enjoy the robotic platform. A particular challenge in this aspect is the evaluation
methodology, since the children are able to move around the room and to interact with
the robot in any way that they feel comfortable, for example by their relative position
to the robot and their level of interaction and involvement. This leads to the difficulty
of getting quantitative results from an unrestrained process which could result in the
child performing almost any gesture or action.
Testing Methodology:
In order to gauge the effectiveness of the robot, it is tested with a number of autistic
children in a special school. These trials take place in the children’s school, providing
a familiar surrounding for them, and are video recorded for evaluation at a later time.
Trials occur in a room approximately 2 metres by 3 metres which has only enough
chairs for two experimenters and a teacher from the child’s school, on hand to provide
guidance and to ensure that the child does not become agitated or bored. Trials
typically last for ten minutes, with the first four minutes taken with either the robotic
platform or a toy truck of approximately the same size and shape. The next two
minutes involve both the toy and the robot, although the robot is now turned off and
inactive, and the last four minutes then involve the toy or the robot, which ever was

---

not present initially. However, this schedule is occasionally altered by the teacher if it
becomes apparent that the child is bored, irritated or is not enjoying the interaction.
The video record of the trial is evaluated according to a set of criteria. The trial is
broken into one second intervals, and each second is evaluated for micro-behaviours,
following from Tardif et al, 1995. The behaviour parameters consist of two
categories, those which depend on the focus of the behaviour, for example the
distinction between the child looking at the robot and looking at an environmental
feature is important, and those parameters which are relatively independent of focus,
there may be instances where the child produces speech and that it is impossible to
determine the focus of this speech, although the occurrence is an important event.
The behaviour parameters are:
Focus Specific: eye gaze, eye contact, operate, handling, touch, approach,
move away, attention.
Focus Independent: vocalisation, speech, verbal stereotype, repetition, blank
Eye Gaze measures the amount of time that the child spends looking at the robot or
toy, while Eye Contact is an indication of how much the child looks at the face or eye
area of the robot or toy. Autistic children show an avoidance of eye contact, and the
children in trials identify the heat sensor of the robot as its head. In this way, we are
able to gauge how willing the children are to make ‘eye contact’ with the robot – or
toy in terms of the front windscreen. The parameters ‘operate’, ‘handling’ and ‘touch’
can be grouped together to give the total contact time, where ‘operate’ is interacting
with the robot through its sensors and so is impossible for the toy. However, we
expected that the inability to ‘operate’ the toy would be balanced by an increased
‘handling’ factor, such as pushing it around the room. ‘Verbal stereotype’ includes
echolalia, while repetition focuses on, but is not limited to, autistic behaviours such as
spinning wheels. ‘Blank’ is the amount of time that the child sits idle, seeming not to
interact or notice the external world in any way. Additionally, notes are kept to
document any behaviours which do not fall into existing categories.
Results:
Four children participated in the trials, with the trial for Child C cut short with the toy
as the child was obviously bored and showed no interest in it. Figure 1 shows the
percentage of trial time which was spent by the child touching, handling and operating
both the robot and the toy. It also shows the contact time and the time spent looking at
the robot and toy as a percentage of the total trial time.
Touch
Handle
Operate
Seconds
Contact
Gaze
Child A
Robot
26.33%
42.70%
0.00%
452
69.03%
81.64%
Toy
11.79%
45.12%
-
246
56.91%
40.24%
Child B
Robot
18.61%
5.28%
18.06%
360
41.95%
60.56%
Toy
3.33%
57.22%
-
360
60.55%
71.67%
Child C
Robot
11.26%
72.64%
0.00%
435
83.90%
93.33%
Toy
0.00%
2.99%
-
134
02.99%
14.18%
Child D
Robot
1.93%
0.23%
17.08%
363
19.24%
53.99%
Toy
19.33%
37.67%
-
300
57.00%
60.33%
Figure 1: The percentage of time for behaviour parameters. Contact time is the total of touch, handle
and operate.
Trial times here are generally greater for the robot, showing that the children seem
able to interact and play with it for a substantial time comparative to ‘normal’ toys

---

and that they are not afraid of it. Also, the robot is able to engage them for this
amount of time and is sufficiently interesting for them.
Average Behaviour Times
40
35
30
Contact Toy
25
Contact Robot
20
Gaze Toy
15
10
Gaze Robot
Average Time
5
0
Child A
Child B
Child C
Child D
Children
Figure 2: The average time of behaviours for toy and robot.
Figure 2 shows the average behaviour times for each of the four behaviours (contact
time with the toy, eye gaze at the toy, contact time with the robot and eye gaze with
the robot) for each of the four children. It can be seen that when comparing the
average behaviour times for robot and toy, the times are similar, except for cases
where the average time for the robot are markedly greater, in particular for eye gaze,
for example Child B.
The evaluation data for the trials also need to examine the combination of behaviours,
since isolated incidences may not give accurate results. For example, Child 4
interacted with the robot by allowing it to approach him. When the robot was very
close, the child took several steps backwards and again waited for the robot to follow
him. Then, when the child could not go back any further due the room’s walls, he
took several small steps into the robots space, and waited for the robot to reverse
away from him before repeating this. In one instance, when the robot turned too far
and so the child was no longer in its sensor arc, the child stepped sideways to continue
the interaction. Also, when this child was given the toy truck, he spent most of the
interaction time simply lifting it, and the rest of the trial time ignoring it.
Discussion:
The results thus far are encouraging in that they indicate that the children not only
enjoy interacting and playing with the robot at various levels, but that they focus
attention on the robot for longer than the toy truck. The children seem able to form
very simple bonds with the robot and even to understand the basic interactions
involved. The children spend time actively touching and making contact with the
robot and they are not afraid or wary of it at all.
These factors are important for future work, since if the child does not feel
comfortable interacting with the robot and if he is not happy to focus attention on it,
then any amount of development would be wasted. Future work will concentrate on
improving the complexity of the robot, while bearing in mind that some children will

---

require more complex interaction than others. The evaluation methodology and
behaviour parameters are in a constant evolutionary development and so we expect
these to adapt and change as a result of further work. The ability of the robot to adapt
to the interaction and to the individual needs of the child is obviously desirable as it
would allow the robot to be used with a variety of children of different abilities, and
additional features for the robot enhance the children’s enjoyment are also under
development. It is envisioned that the robot will be able to develop a storyline in order
to guide the children through situations in a linear way, with the child able to control
the speed of this progression.
References:
Aurora Web Site: http://www.aurora-project.com – Last Accessed: 13/12/2000
Dautenhahn, K. & Werry, I. (2000). Issues of Robot-Human Interaction Dynamics in
the Rehabilitation of Children with Autism. Proc. From Animals To Animats, The
Sixth International Conference on the Simulation of Adaptive Behavior (SAB2000),
Paris, France (Sep, 2000).
Michaud, F. & Clavet, A. & Lachiver, G. Lucas, M. (2000). Designing Toy Robots to
help Autistic Children – An Open Design Project for ECE Education. In: Proceedings:
ASEE, 2000.
Murray, D. & Lesser, M. Autism and Computing Web Site:
http://www.shifth.mistral.co.uk/autism/NAS/ - Last Accessed: 14/12/2000
Tardiff, C. & Plumet, M-H. & Beaudichon, J. & Waller, D. & Bouvard, M. &
Leboyer, M. (1995). Micro-Analysis of Social Interactions Between Autistic Children
and Normal Adults in Semi-Structured Play Situations. International Journal of
Behavioural Development, 18 (4), 727 – 747.
Watson, L. R. & Lord, C. & Schaffer, B. & Schopler, E. (1989). Teaching
Spontaneous Communication to Autistic and Developmentally Handicapped Children.
Irvington Publishers, Inc. New York.
Weir, S. & Emanuel, R. (1976). Using Logo to Catalyse Communication in an
Autistic Child. D.A.I. Research Report, University of Edinburgh.
Werry, I. & Dautenhahn, K. (1999). Applying Mobile Robot Technology to the
Rehabilitation of Autistic Children. SIRS'99 Proceedings, 7th International
Symposium on Intelligent Robotic Systems, Coimbra, Portugal, July 1999.
Acknowledgements:
We gratefully acknowledge the support of the NAS and, in particular, Patricia
Beevers and the staff and pupils of Radlett Lodge School. The robotic platform used
in this project is kindly donated by Applied AI, Inc and the project is supported by an
EPSRC research grant (GR/M62648).

---