2nd workshop on
"Robotic and Virtual Interactive Systems
in Therapy of Autism
and other psychopathological disorders"

A two-day workshop at Hospital La Salpêtrière, Paris
27-28 September 2002
amphitheatre Charcot

Jacqueline Nadel, UMR7593, La Salpêtrière, UPMC, CNRS Dept SDV
Arnaud Revel, UMR8051, ENSEA, Université de Cergy, CNRS Dept STIC

Friday 27 September


10h Welcome- Arnaud Revel and Jacqueline Nadel
Posters handling
10h30-11h45 Introduction – Jacqueline Nadel (CNRS UMR7593, La Salpêtrière)
Autism, development and the search for efficient therapies
11h45-12h coffee break
12h-12h45 Anja Rutten, Helen Neale, Sue Cobb, Steven Kerr, Sarah Parsons, Ann Leonard, Peter Mitchell, & Tony Glover (School of Psychology & School of Computer Science, University of Nottingham)
The AS Interactive Project: Further development and the evaluation of Ves for users with Asperger Syndrome


The AS Interactive Project aims to develop VEs to allow adolescents and adults with ASDs to practise social skills. The project was introduced to the First International Workshop on Robotics and Virtual Interactive Systems in Therapy for Autism and other Psychopathological Disorders, 2001 where developments and results of the first year of the project were presented. During the first year, emphasis was on development, utilising a user-centred design process. This resulted in production of two Single User Virtual Environments (SVEs), the Café and the Bus. These environments focused on helping users learning how to find an appropriate place to sit down. Based on the environments created during the first year, an evaluation study was carried out (Leonard et al., 2002)
During the second year of the project, SVEs were refined following a period of evaluation and implementation of changes in consultation with staff and students in a school for pupils with ASD. Several features have been added or changed following consultation with users and facilitators. The number of levels of difficulty in each of the environments was increased to provide a more challenging environment and flexibility was included, so that each time the same task was presented, the environment appeared visually distinct to prevent students from rote-learning what to do based on visual memory alone. In addition to the main task of finding a seat, in the more difficult levels the users had to queue appropriately. Evaluations of the SVE carried out in year 2 are detailed in Neale et al., (2002). Additionally, two Collaborative Virtual Environments (CVEs) have been developed - the Social Café and the Job Interview/Formal Meeting. These environments provide less structure, and potential for richer social interactions with others through the CVE, including the use of gestural behaviours. We have carried out some informal trials of CVE use within a school environment and are currently evaluating these to examine the potential of this technology for social skills practise.
Year 3 of the project will see continued evaluations of SVE and CVE scenarios. The SVEs will be developed into project deliverables and additional support material for teachers and support workers will be prepared.
This presentation will give an overview of project developments from school-based studies with adolescent AS users.

12h45- 14h Lunch
14h – 14h45 Kerstin Dautenhahn, Aude Billard, Megan Davis, Tamie Salter, Iain Werry (ASRG, School of Computer Science, University of Hertfordshire)
Children with autism interacting with robots in the Aurora project


The talk will give an update on current progress in the Aurora project which studies how to use robots in autism therapy. The chosen setup is inherently playful and unconstraint, e.g. the children are not required to solve any tasks other than playing, and the only purpose of the robot is to engage children with autism in therapeutically relevant behaviours such as turn-taking and imitation. A key issue is that the children proactively initiate interactions rather than merely responding to particular stimuli. Additionally, the chosen setup is social, i.e. it involves not only the robot and the autistic child present, but also the teacher and one or two experimenters. As we have shown previously (Werry et al 2001; Dautenhahn et al., 2002) this social setup is used by some children in a very constructive manner demonstrating their communicative competence: they use the robot as a focus of attention in order to interact and/or communicate with other people in the room.
The first part of the talk will address issues of design spaces and niches spaces of robots in autism therapy. Given the wide range of abilities of children with autism it seems unlikely that one type of robot will be the solution: rather, the design space of robotic designs (variations in behaviour as well as appearance) need to be mapped to the "niche space" of particular requirements that individual children, or groups of children with similar set of symptoms show. We therefore argue that any progress in the field needs to systematically assess how children with autism interact with robots and what the particular benefits of robots are in comparison to other non-robotic toys.
The second part of the talk will present studies where we investigated how 15 autistic children interacted with a humanoid robotic doll called Robota. The purpose of the robot was to imitate children's arm movements (Dautenhahn & Billard 2002). Different from the trials with the mobile robot, here the teacher was greatly involved in setting up and guiding the "game" that the children played with the robot. We discuss advantages and disadvantages of such a robotic design as well as first results on the analysis of the videos documenting the interactions. If time permits then the last part of the talk will summarise results from a comparative study on eye-gaze and contact-behaviour for a group of 17 autistic children where we compared how they behave towards a small mobile robot as opposed to a non-robotic toy (Werry 2002). We will briefly summarise and discuss the data.

14h45-15h coffee break
15h- 15h45 François Michaud (Mobile Robotics & Autonomous Intelligent Systems, Sherbrooke)
Mobile robotic toys in therapy of Autism


Since 1999 we, as engineers in electrical and computer engineering, have been designing a great variety of mobile robotic toys with the goal of using them as pedagogical tools for children suffering from autism or other developmental disorders. These mobile robots can move autonomously in the environment and interact in various manners (vocal messages, music, visual cues, movements, etc.) with the child. Compared to a human, a robot may be less intimidating and more predictable. It can follow a deterministic play routine, and also adapt over time and change the ways it responds to the world, generating more sophisticated interactions and unpredictable situations. This flexibility allows robotic toys to evolve from simple machines to systems that demonstrate more complex behavior patterns. In our case, the interaction framework created by our robots is to get the attention of the child, ask the child to do something, and to reward the child if the request is successfully satisfied. Since each child is a distinct individual with preferences and capabilities, it might not be possible to design one complete robotic toy that can help capture and retain the interest of every child. So our strategy is to design many different types of robots, and observe the possible factors that might influence the child's interests in interacting with a robotic toy, like shape, colors, sounds, music, voice, movements, dancing, trajectory, special devices, etc., and learning from our observations to design new robots that could in the near future be used by parents and educators.
For the workshop, we will present the particularities of our robotic toys and what they can do as pedagogical tools, what we have learned from our experiences over the last four years, and outline what we plan to do in the next two years to study more closely what impacts these devices can have on children with autism. As engineers, we need to combine our expertise with scientists in the field, in order to get interesting insights that will help guide the design of innovative new robots. And our hope is that mobile robotic toys can become efficient therapeutic tools that will help children with autism develop early on the necessary skills they need to compensate for and cope with their disability.

15h45-16h coffee break
16h- 16h45 Brian Scassellati (University of Yale)
How to use anthropomorphic robots to study social development


In the last ten years, there has been an emphasis in the robotics community on developing robots that look like people, act like people, and interact with people in the same ways that people interact with each other. This talk will examine four different ways of using these anthropomorphic robots to study the development of social skills in children. Observing how humans react when placed in a social context with machines that share some human characteristics allows us to study our own mental processes and our views of ourselves. Humanoid robots are evocative objects in that they provoke people to question and to reassess their ideas about what it is to be intelligent, to have emotions, and to be a person. Pilot research on how interactions with two anthropomorphic robots impacted concepts of identity and self in agroup of sixty children aged 6 o 14 years will be presented. (This is joint work with Cynthia Breazeal and Sherry Turkle).
A robot that is capable of perceptually detecting social cues also provides a quantifiable metric for those social cues. These metrics have potential uses in characterizing the development of social abilities and in the diagnosis of developmental disorders such as autism. While we do not claim that the metrics identified by building social robots will take the place of the clinician's judgment, these quantitative metrics may be extremely useful to the medical community in establishing a diagnosis, in tracking the success of intervention programs, and in reporting results. Joint work with Ami Klin, Warren Jones, and Fred Volkmar from the Yale Child Study Center on the application of these metrics will be presented.
Social robots that are constructed according to models of skill acquisition in children can also be used as an evaluation tool for those models. Just as simulations of neural networks have been useful in evaluating the applicability of models of neural function, these robots can serve as a test-bed for evaluating the predictive power and validity of models of human evelopment. Further, a robotic model can also be subjected to controversial testing that is potentially hazardous, costly, or unethical to conduct on humans. Research on two models of the development of theory of mind and joint attention skills that were implemented on Cog will be presented as well as a new robot that is being constructed to address issues of sensorimotor development and social development.
Finally, we speculate on the use of social robots as a therapeutic device for autism. If you could control the level of social sophistication in a robotic device, would that robot provide a crutch for learning social skills gradually? What can be learned from these intervention approaches about the structure of social skill development?

16h45-17h coffee break
17h-18h Isabelle Viaud-Dalmon (CNRS UMR7593, La Salpêtrière)
Virtual reality as a tool for rehabilitation in Psychiatric disorders (+ demonstration)


Virtual reality (VR) represents a set of computer technologies, which allow users to interact with a three-dimensional, computer-generated environment in real time. VR is starting to be used in psychophysics experiments as well as in psychological therapy around the world. VR provides a way to immerse a user in an environment in which all the parameters can be measured, and in which the interaction between different sensory modalities can be controlled.
Therefore VR represents an interesting tool to study the integration of space-related multisensory information in human and its disorders. For example, we have used VR to study the adaptation to incoherent visual-vestibular stimulation, in a task in which subjects had to control their whole-body rotations with a joystick. In another experiment, we studied the effect of sensory conflict both on sensorimotor control and on the stored representation of a path.
However, before this tool can be established as a standard and can be used on an everyday basis for therapy, several of its aspects need to be thoroughly studied. VR accounts on the presence of users in the virtual environment: the user has to believe that he is actually in the virtual world and not anymore in the physical world. This first aspect is far from being trivial since it invokes derealisation experiences. Another aspect is linked to the fact that the interaction with any kind of VR system necessitates an adaptation of the subject. The effectiveness of VR for experimental and therapeutic purposes will be discussed in this framework.


Saturday 28 September


9h15-10h Lola Canamero (Department of Computer Science, University of Hertfordshire) & Philippe Gaussier (Group Signal and Image Processing, University of Cergy/CNRS)
Emotion understanding : robots as tools and models


Affective Computing is a new research area that aims at endowing robots and computers with emotional capabilities (e.g., to express, recognize, or "have" emotions) in order to make them more life-like and better adapted to interact with humans. Whereas the perspective of having artifacts that can display emotional expressions, respond to and adapt to our emotional states on a superficial level seems increasingly appealing, there is much scepticism regarding whether artifacts can "have" emotions in a deeper sense, since this is often considered as a unique feature of the human (and some other animal) species.
In our opinion, this and other fundamental questions that stem from it (as for example in what cases it makes / does not make sense to give our artifacts emotions, and what aspects of them) must be thoroughly investigated if affective computing is to become a serious discipline that can contribute to our understanding of emotional phenomena. Indeed, robots offer an excellent platform for this investigation, since they can be used not only as tools (with easily modifiable parameters) to support research in other disciplines, but also as (synthetic, implemented, and working) models of emotional systems in non-human, non-biological species, hopefully shedding light on some integral elements and aspects of emotions. The current state of the art in affective computing research is still far from this objective, not only due to the early age of the discipline, but in particular to the complexity of emotional phenomena and our still limited understanding of them. Work in this area has tended to focus on only one of the aspects that seem more apparent regarding the "dual nature" of emotions:
a) "Internal" robotic/agent architectures integrating emotional elements for behavior modulation and control – what we would term emotions as "second order" control or monitoring mechanisms.
b) "External manifestations" of emotions (e.g., facial displays) that can be used as signals for social interaction and communication – for example, work on expressive robots.
However, although the use of emotions can certainly provide novel solutions, the problems underlying these lines of research – behavior control and social interactions – are classic problems in robotics that have largely been tackled without resorting to emotional mechanisms. We will therefore review some representative problems and (the achievements and the limitations of) classical solutions in these areas, in order to better appreciate what the roles and contributions of emotions have been / can be with respect to particular problems in robotics. We will finally sketch some ideas on how this "dual aspect" of emotions can be meaningfully integrated in robots if we want to use them as tools and models to investigate and understand emotions.
To conclude our talk, we will illustrate how we are undertaking the implementation of these ideas in, and some of the problems raised by, an expressive robotic head designed to investigate emotion understanding – currently recognition and imitation of facial displays – in typical and autistic children.

10h-10h15 Coffee break
10h15-11h Gerardo Guttierez (Robotic Institute, University of Valencia, Spain)
& Rita Jordan (School of Education, Birmingham)
Virtual reality for understanding imagination in people with autism


Difficulties and delay in understanding symbolism, especially in relation to symbolic play, have long been documented as characteristic of people with autistic spectrum disorders (ASDs). It is not clear whether such difficulties and delays represent a core deficit in imagination, as some have proposed, or whether they result from other aspects of autism (Jordan, in preparation). Nor is it clear whether the problems lie with all aspects of play or with the aspect of pretend play referred to as 'symbolic play' only. Leslie (1994) suggests three categories of symbolic play: object substitution, attribution of false properties and reappearance/ disappearance. There have been many attempts to teach symbolic play to children with ASDs, and a recent attempt (Sherratt, 2002) attributes the comparative success of the programme (with children with both autism and severe learning difficulties) to the use of structure, repetition and affective engagement. Virtual Reality (VR) has been claimed to provide a particularly facilitatory environment for people with ASDs in that it also offers structure, opportunities for repetition, affective engagement and, additionally, control of the learning environment. Virtual reality shares the advantages of computer-based learning, and has the additional advantage of making it more likely that the results will generalise to real-word
This study, is an attempt to use a virtual reality environment to develop understanding of symbolic representation and imagination, within a 'familiar, yet playful environment. It also attempts to evaluate the contribution of virtual reality to any observed gains through comparison with traditional teaching approaches. In this paper we present the design and software developed under project INMER, which is currently being used and evaluated with a sample of people with autism. The project is used to ensure understanding over a range of 'teaching steps' leading towards symbolic understanding, with the VR tool being used to elucidate the symbolic and imaginary aspects, when appropriate. These steps cover: functional use of objects, functional play, imaginary play (involving object substitution at two different levels of difficulty), actual (or in this case VR) transformation of objects, 'magic' transformations and imaginary transformations. This careful stepped approach to teaching, ensuring understanding at each stage, is aimed at avoiding confusion, which could result from the premature use of a VR tool. The evaluation will also include an evaluation of generalisation. At this preliminary stage there are no results to report. However, the paper concludes with some of the limitations that have already become apparent, such as the lack of opportunities for the people with ASDs to themselves be creative and the lack of social stimulation to spark creativity; the authors suggest some future developments in VR that might address these.

11h-11h15 Coffee break
11h15- 12h Arnaud Revel ( Group Signal and Image Processing, ENSEA/CNRS, Cergy),
Jacqueline Nadel, Marie Maurer & Pierre Canet (CNRS, UMR7593, & ITIN group, Cergy)
VE: a tool for testing imitative capacities of low-functioning children with autism


Equipped with our knowledge of developmental indices (Nadel & Butterworth, 1999), we have started an exploration of imitative capacities in low-functioning children with autism. Such an exploration is needed, since results in this area are controversial, with authors claiming that children with autism have specific impairments in the domain of imitation, others (saying that the imitative deficits are not specific to autism but more generally include children with different developmental impairments with dysphasia and more specifically with language impairments and still others like us denying noticeable deficits in low-level imitation in children with autism. Our stance is based on the idea that a hierarchy of mechanisms are at play when we imitate according of the kind of imitation we use, from low-level use of mirror neurons in perception-action coupling to high-level mechanism involved in the representation of actions or program of actions (Rizzlolatti et al., 2002). The relatively late diagnosis of autism suggests that early motor development is not specifically impaired. We thus postulate the integrity of perception-action coupling in autism.
The major impairment of children in autism lies in social component, therefore, it is of major interest to try to distinguish what in imitative performance is due to motor and cognitive capacities and what is due to capacities to relate with partners. We propose an experiment with 3 interactive conditions: an on-line condition, where a real partner asks the child to do like him/her, an off-line condition, where a televised partner asks to do like him/her, a virtual environment condition where the virtual partner asks the child to do like him/her. The imitative performance of the children is the discriminant variable.
Such a project would need us to answer to several pre-requisites: is it possible for low-functioning children with autism as well as for young infants to discriminate between virtual reality and real life? To our own knowledge the attempts to make persons with autism interacting in a virtual environment were always performed with adults or adolescents and with high functioning persons. With young and low-functioning children with autism however, the mastery of such complex situations is far from being obvious and we need at first to process an analysis of the different components of the context which can lead to differentiate virtual environment from real life. This is the reason why the virtual environment we will test is very simple. It is composed of only one avatar displayingtwo facial expressions with no eye-to-eye contact and proposing several simple actions, with only one sentence "do like me". A demonstration of the virtual environment design will be provided by ITIN group.

12h-14h Lunch
14h – 18h

Session of Posters and Demonstrations:

Begonia Pino: Use of computers to enhance social engagement and social understanding in children with autism –poster–



Children with Autistic Spectrum Disorders (ASD) have difficulties understanding social situations and displaying appropriate behaviours. Many educational programmes have attempted to decrease these difficulties by teaching social skills to these children. Although these programmes have successfully taught social rules in many cases, children fail to transfer them to daily life. A social interaction project focused on social understanding succeeded because children had the opportunity to interact in an 'almost' natural setting (Dunlop et al., 2002). Besides, children with ASD tend to enjoy working or playing with computers. Murray (1997) shows that computers are not threatening, non judgemental, predictable, reliable, etc., thus, providing a safe environment in which social interaction can take place. This research investigated the use computers as an environment to teach and practice social understanding. The focus on social understanding was inspired by a social interaction project where children were involved in different real life activities, such as games, snacks, and outings. The rationale was that computer provided a 'real life' environment, and a shared interest as well as a motivational and safe tool around which construct a relationship.
In the first stage the goal was to observe whether the computer fostered greater social engagement: child involved in more and longer interactions, initiating more, increased eye contact, etc. Child and experimenter carried out an activity (playing a game) in a computer version and later in a non-computer version (half of subjects started with non-computer version). The second stage intended to improve children's social understanding by putting social rules into practice with the mediation of the computer, for a longer period (a series of weekly sessions) where the activity was tailored to the child's interests and assisted by the experimenter. The expected results of the first stage were, first, that there would be more interactions in the computer version, but maybe less eye contact, which may suggest that children with ASD become more socially engaged when using computers with another person. Differences between the children with ASD and the control group could pinpoint the adequacy of the use of computers to enhance social interaction in the population with ASD in particular.

  Yufang Cheng, David Moore & Paul McGrath: Virtual learning environments for children with autism –poster-


Autism is a neurodevelopmental disorder characterised by a triad of impairments: in communication, social understanding, and rigidity of thought (Wing 1996). It is often held children with autism are poor at mind-reading, have a limited understanding of emotional expressions of both others and themselves.
An interesting possibility is that the use of Collaborative Virtual Environment (CVE) technology may be able to help children with autism counter these difficulties. CVE can be defined as a computer-based, distributed, virtual space or set of spaces, in which people can meet and interact with others. Given this definition, the prima facie argument for CVE for people with autism is clear: a CVE can potentially provide a means by which people with autism can communicate with others (autistic or non-autistic) and thus circumvent their social and communication impairment and sense of isolation. The technology can also be used for purposes of practice and rehearsal. A key aspect of CVE is that users are represented in the environment by their personal "avatar" (Cassell et al, 2000). If autistic children are to benefit from CVE, therefore, it is important that they are able to interact successfully with their own and other people's avatars. Further, it may be that working with avatars that represent the emotions of their users, helps combat any theory of mind deficit. Our research interest, then, concerns how people with autism interact with avatar representations. Given this, we have built a system, utilising avatar representations of emotions based on work by Fabri (2001), that requires users to work through three stages. In stage 1 an avatar is presented in isolation, for the emotions happy, sad, angry and frightened. Stage 2 represents the same emotions in the context of a social story, since this may help users infer the likely emotion caused by certain events. In stage 3 the user is given an avatar representation of one of the emotions and asked to select what event caused this emotion; the argument here is that inferring the possible cause of a displayed emotion is likely to be essential when using a CVE for communication. Users' responses to the system are recorded by the software for subsequent analysis.


Nicole Oudin, Jacqueline Nadel & Joëlle Proust: Computarized facilitated communication for nonverbal children with autism –demonstration--


  Hideki Kozima:The Infanoïd - demonstration -


We are designing a robot that can help children, either normal or autistic, lean to communicate with others. Communication is one form of the social interaction in which one predicts and controls someone else's behavior by using social clues like bodily/facial gestures and speech. This project note describes our on-going exploration on the design principle of a robot with which normal or autistic children can play contingency-detection game. In the game, the robot reacts to such social clues that the children will make and displays such social clues that will induce some response in the children, possibly forming social interaction. As a possible embodiment, we introduce our infant-robot, Infanoid, which is currently capable of primordial attentional interaction with humans.

  Caroline Potier, Daniel Viezzi, Jacqueline Nadel & Philippe Gaussier: Neonatal imitation modelled by a robot –demonstration-

Abstract :

In the context of an interdisciplinary cooperation, we have conceived a robotics "mouth" to study imitation of tongue protrusion and mouth opening as performed by human newborns in a human context. Results obtained in this framework will suggest further robotics developments which could help us in the understanding of the mechanisms involved in imitation. The robotics mouth can be programmed simply by the final user which can specify both the amplitude and the speed of the movements. The minimalist program is written in C under linux and works in text mode. This implies no specific expansive powerful equipment. The study of tongue protrusion is simplified by the ability of the system to create patterns of actions, or sequences of patterns in correspondence with psychological protocols.
One of the main interest in the robotic solution in comparison with the human one is the precision (up to 1 millisecond) and the reproducibility of the sequences. This allows to compare results obtained during several experiments very precisely. The use of a robotic solution allows to keep an identical referential both in time and space since we can test the same protocol whenever and wherever. Besides, compared with a real mouth, the robotic one can be schematised at will in order to test which features are really important in a human mouth. Conversely, compared with a simulated mouth displayed on a 2D monitor, the robotic mouth is 3D, and more than that, it is really "embedded" and "situated".
The basic question for developmental psychologists is to explore whether neonatal imitation in humans is a selective process that requires biological modelling or whether it is an elective process likely to occur in front of animated though non biological stimuli.

  Ronald Kemeling: Mimic approach –demonstration-

Abstract :

Children with a development problem such as autism rarely get the opportunity to explore their environment and, even if they are able to do so, positive feedback is often lacking. Their body scheme awareness is also often poorly or insufficiently developed. With the MIMIC program an environment can be developed which gives children full control of what happens. Moreover, they receive multichannel feedback. Studies in Sweden and the US have shown that this form of feedback can be very effective. MIMIC is a unique multimedia computer program with which a fully interactive multisensory development and learning environment can be created in a simple way. The principle is very simple. The computer observes the space by means of the camera. When any movement is observed on a pre-defined spot, the computer responds with an action. The action depends on what the counsellor has programmed in the computer. All kinds of movements are possible by means of which concepts such as high, low, left-right, large-small, in-out can be visualised. A number of persons can simultaneously use this environment and in this way they can make music or participate in an interaction. Colours and subsequently emotions can be linked to a movement towards a specific spot. Language and communication exercises can be composed. Spatial orientation exercises, body scheme development, matching exercises, behavioural therapeutic approaches, eliciting of movements are only some of the possibilities of MIMIC. I have worked with children of two different schools for special education (SLD). Now I am working on a content with videos of a 15 year old girl with autism. I wish to know whether this form is appropriate for her. I wish to show you several aspects of the program's operation and I have some video material. I would like to set up a collaborative study of the effects and development of content.

18h Final comments and what next?



Haut de la page haut