Robot-assisted Social Skill Training for Adults with ASD

Robot-assisted Social Skill Training for Adults with ASD Frank Broz, Ayan Ghosh, Ingo Keller, Peter McKenna, Gnanathusharan Rajendran, and Ruth Aylett Heriot-Watt University Edinburgh, UK f.broz,ayan.ghosh,,ijk1,p.mckenna,t.rajendran,r.s.aylett@hw.ac.uk ABSTRACT This paper describes a user-centered approach to designing a system enabling adults with autism spectrum disorder to engage in social skills training with a robot. This system will allow people to rehearse work-relevant social skills. Using its own internal models of social behavior, the system will provide feedback on the rehearsals, supporting users’ ability to learn from episodes of human-robot interaction. CCS CONCEPTS · Human-centered computing → Empirical studies in interaction design; · Social and professional topics → People with disabilities; · Computer systems organization → Robotics; KEYWORDS social robotics, facial expressions, autism ACM Reference Format: Frank Broz, Ayan Ghosh, Ingo Keller, Peter McKenna, Gnanathusharan Rajendran, and Ruth Aylett. 2018. Robot-assisted Social Skill Training for Adults with ASD. In Proceedings of 13th Annual ACM/IEEE International Conference on Human Robot Interaction (HRI’2018). ACM, New York, NY, USA, 3 pages. https://doi.org/10.1145/nnnnnnn.nnnnnnn 1 INTRODUCTION According to 2011 UK census figures, Autism Spectrum Disorder (ASD) affects 547,000 people over the age of 18 (1.3% of working age adults) [9]. These adults encounter serious difficulties in securing and maintaining employment. The unemployment rate among adults with ASD is higher than 85%, which is nearly double the unemployment rate of 48% for the wider disabled population and compares to the UK unemployment rate of 5.5%. One reason for this is that people with ASD struggle to correctly interpret social signals, i.e., the expressive behavioural cues through which people manifest what they feel or think (facial expressions, vocalisations, gestures, etc.). This leads to difficulties in correctly interpreting interactions with co-workers and supervisors. Behavioural Skills Training (BST) [7] is recognized as one of the most effective training approaches for the effects of an ASD. BST is a behaviourist training approach involving phases of instruction, modelling, rehearsal, and feedback in order to teach a new skill [6]. It has been used to teach social skills to people both with Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s). HRI’2018, March 2018, Chicago, Illinois USA © 2018 Copyright held by the owner/author(s). ACM ISBN 978-x-xxxx-xxxx-x/YY/MM. https://doi.org/10.1145/nnnnnnn.nnnnnnn and without disabilities [11]. However, BST is too labour-intensive to be widely applied. If robots could be used to help deliver BST (particularly the rehearsal and feedback phases), this could reduce the effort required from human therapists and lower the cost of providing BST. We are developing a system to allow people to rehearse social skills with an expressive, autonomous social robot in realistic workplace scenarios. Because the robot will be equipped with a model of interaction for those scenarios that links expressive behavior to underlying intention, the system can provide feedback to the user by exposing its own decision-making process after interaction. While focusing on a particular case of BST for high-functioning adults with ASD, this research contributes to the long-term vision of social robots able to seamlessly integrate into our everyday life, opening the way to a multitude of domestic, educational and assistive applications. We argue that the development of successful long-lived human-robot relationships requires transparency of the robot’s motives, goals and plans so that its intentional stance is clear to human interaction partners. 2 PRODUCING EXPRESSIVE BEHAVIOR We will be looking at office-based scenarios, aiming to train highfunctioning ASD individuals to decode communication signals from their employer or co-worker. We are focusing on broad groups of emotions such as approval (positive) and disapproval (negative) expressions [10]. Our work on this to date has concentrated on evaluating the interpretability of facial expressions for our robot with an undiagnosed population. We have identified which mappings of facial action units onto our low dof robot face are well-categorized [8]. Following on this work, we are investigating the impact of ASD on expression recognition. 2.1 Using the AQ Rather than working directly with an ASD population, we have used the Autism-spectrum Quotient (AQ) [2] to explore potential correlations between the prevalence of autistic traits and the ability to categorize the robot’s expressions. The AQ has been shown to be an effective screening tool for ASD [13] which gives a score of 0-50 indicating the prevalence of autistic-type traits in an individual. Research has been conducted showing a degradation in performance in non-ASD individuals corresponding to AQ score for tasks in which ASD is associated with degraded performance (concept formation) [5]. The use of high-AQ individuals as a proxy for those diagnosed with ASD is supported by the literature and simplifies recruiting a representative sample to evaluate our system, as it can be difficult to reach sufficient numbers of adults with ASD to participate. HRI’2018, March 2018, Chicago, IllinoisFrank USA Broz, Ayan Ghosh, Ingo Keller, Peter McKenna, Gnanathusharan Rajendran, and Ruth Aylett Preliminary results did not show an impact of AQ score on expression recognition for expressions that were well-categorized by an undiagnosed population [1]. This suggests that the robot’s simplified expressions may be more easily interpretable than human facial expressions across the autism spectrum. Of course, human expressive behavior is multi-modal and involves other social signals such as prosody, gesture, and proxemics. Capturing multi-modal expression within a specific social context will be addressed through human-human data collection as described in the next section. 3 MODELLING WORKPLACE SCENARIOS It is critical to the success of this project that the workplace scenarios chosen be both ecologically valid and useful to adults with ASD. In order to ensure this, we are consulting with stakeholders on the selection and design of the scenarios prior to modeling. 3.1 Involving Stakeholders Through connections with local charities that specialize in supporting ASD adults in the workplace, we are planning focus groups with both the workers who provide this support and with adults with ASD themselves. The purpose of these focus groups are to identify: which social skills are most problematic for ASD adults in the workplace, what scenarios would best allow people to practice these skills, and how the system we are developing could support rehearsal of those scenarios in an ecologically valid manner. 3.2 Once the scenarios are identified, the next step is to model the interactions that take place in these scenarios. In order to develop rich models of expressive behavior relating to these scenarios, we will collect data from human-human roleplay. For each episode of interaction, roles will be assigned to participants in a manner similar to that described in Broz et al. [4]. While each participant’s intention is given for an episode of interaction, their behavior is not prescribed and they are unaware of the given intention of their interaction partner. This roleplaying allows intentions to be linked to realistic interaction data and also enables the collection of subjective data about the quality of interaction. Additionally, we will administer the AQ to roleplay participants so we are able to categorize behavior as originating from high-AQ or low-AQ individuals. The information available from episodes of roleplay may also be enhanced using annotation by human experts. 4 Figure 1: The proposed system architecture. Human-Human Roleplay SYSTEM IMPLEMENTATION The robot’s control architecture will need to manage the production of low-level expressive behavior as part of high-level behavioral policies that will determine the robot’s responses to the human interaction partner during BST. Our approach to producing policies for the robot is based on the prior work of Broz et al., which used partially observable Markov decision processes (POMDPs) to model socially acceptable behavior for human-robot interaction [3]. The modelling approach taken in this work links a human partner’s observable behavior to the unobservable intentions motivating that behavior, allowing the robot to act based on beliefs about the partner’s current intention. The agent’s own intentions are represented in the reward structure of the model. A major component of this interactive system will be algorithms that interpret human behavior and provide input on relevant cues using the methodology of social signal processing [12]. The accurate detection of social cues is important both so that the system can respond in real time and so that feedback can be given after interactions where the participant failed to give expected cues or did not respond as expected. Just as behavior production must be multimodal in these scenarios, the recognition of expressive behavior must use a range of modalities as input. A high-level representation of the proposed architecture can be see in Figure 1. Using the same POMP-based representations of belief about the state of social interaction that will be used in planning and execution, the system will expose the agent’s reasoning process to the human partner after an episode of interaction. This will allow an autistic person to compare their interpretation of the intentions motivating the interaction to the system’s and to correct and learn from misunderstandings during episodes of rehearsal. Exactly how these belief-state traces can best be presented to a human in order to summarize the critical aspects of an interaction is an open question for investigation. This review could be aided by video playback of the interaction itself in order to allow the user to review the expressive behaviors displayed and focus on important details. 5 CONCLUSIONS This paper presents a proposed system for BST-based rehearsal and feedback for adults with ASD through interaction with an autonomous robot. We will collect human interaction data through roleplay and use the AQ to categorize participants based on their level of autistic traits. Our approach seeks to create ecologically Robot-assisted Social Skill Training for Adults with ASD valid training scenarios that will allow autistic individuals to repeatedly roleplay common workplace interactions and practice recognizing and interpreting social behavior in these contexts. The robot will facilitate this learning by exposing its decision-making process (in the form of beliefs about the state progression) after an episode of interaction, revealing to the user why certain expressive behaviors were selected. ACKNOWLEDGMENTS This work was supported by the Research Councils UK, project EP/N034546/1, SoCoRo (http://socoro.net). REFERENCES [1] 2018 (to appear). Robot Behaviour and Autistic Traits. In International Conference on Autonomous Agents and Multiagent Systems. [2] S Baron-Cohen, S Wheelwright, R Skinner, J Martin, and E Clubley. 2001. The autism-spectrum quotient (AQ): evidence from Asperger syndrome/highfunctioning autism, males and females, scientists and mathematicians. J. Autism Dev. Disord. 31, 1 (feb 2001), 5ś17. http://www.ncbi.nlm.nih.gov/pubmed/ 11439754 [3] Frank Broz, Illah Nourbakhsh, and Reid Simmons. 2013. Planning for HumanRobot Interaction in Socially Situated Tasks. Int. J. Soc. Robot. 5, 2 (2013), 193ś214. https://doi.org/10.1007/s12369-013-0185-z [4] Frank Broz, Illah R Nourbakhsh, and Reid G Simmons. 2011. Designing {POMDP} Models of Socially Situated Tasks. In IEEE Int. Symp. Robot Hum. Interact. Commun. 39ś46. https://doi.org/10.1109/ROMAN.2011.6005264 [5] Hollie G. Burnett and Tjeerd Jellema. 2013. (Re-)conceptualisation in Asperger’s Syndrome and Typical Individuals with Varying Degrees of Autistic-like Traits. J. Autism Dev. Disord. 43, 1 (jan 2013), 211ś223. https://doi.org/10.1007/ s10803-012-1567-z [6] Nancy Dib and Peter Sturmey. 2012. Behavioral Skills Training and Skill Learning. In Encycl. Sci. Learn., Norbert M. Seel (Ed.). Springer US, Boston, MA, 437ś438. https://doi.org/10.1007/978-1-4419-1428-6_644 [7] A. Hillier, H. Campbell, K. Mastriani, M. V. Izzo, A. K. Kool-Tucker, L. Cherry, and D. Q. Beversdorf. 2007. Two-Year Evaluation of a Vocational Support Program for Adults on the Autism Spectrum. Career Dev. Transit. Except. Individ. 30, 1 (jan 2007), 35ś47. https://doi.org/10.1177/08857288070300010501 [8] Peter Edward McKenna, Mei Yii Lim, Ayan Ghosh, Ruth Aylett, Frank Broz, and Gnanathusharan Rajendran. 2017. Do you think I approve of that? Designing facial expressions for a robot. Springer, 188ś197. https://doi.org/10.1007/ 978-3-319-70022-9_19 [9] ONS. 2016. 2011 Census aggregate data. (2016). https://doi.org/10.5257/census/ aggregate-2011-1 [10] James A. Russell and James A. 1980. A circumplex model of affect. J. Pers. Soc. Psychol. 39, 6 (1980), 1161ś1178. https://doi.org/10.1037/h0077714 [11] Kelise K Stewart, James E Carr, and Linda A LeBlanc. 2007. Evaluation of familyimplemented behavioral skills training for teaching social skills to a child with Asperger’s disorder. Clin. Case Stud. 6, 3 (2007), 252ś262. https://doi.org/10.1177/ 1534650106286940 [12] Alessandro Vinciarelli, Maja Pantic, and Herv?? Bourlard. 2009. Social signal processing: Survey of an emerging domain. Image Vis. Comput. 27, 12 (2009), 1743ś1759. https://doi.org/10.1016/j.imavis.2008.11.007 [13] M R Woodbury-Smith, J Robinson, S Wheelwright, and S Baron-Cohen. 2005. Screening adults for Asperger Syndrome using the AQ: a preliminary study of its diagnostic validity in clinical practice. J. Autism Dev. Disord. 35, 3 (jun 2005), 331ś5. http://www.ncbi.nlm.nih.gov/pubmed/16119474 HRI’2018, March 2018, Chicago, Illinois USA