Interacting with a Robot for Individuals with ASD: Comparison
Please note this is a comparison between Version 1 by Marion Dubois-Sage and Version 3 by Marion Dubois-Sage.
Individuals with Autism Spectrum Disorder show deficits in communication and social interaction, as well as repetitive behaviors and restricted interests. Interacting with robots could bring benefits to this population, notably by fostering communication and social interaction. Studies even suggest that people with Autism Spectrum Disorder could interact more easily with a robot partner rather than a human partner. TWe will be looking at the benefits of robots and the reasons put forward to explain these results will be looked at by researchers. The interest regarding robots would mainly be due to three of their characteristics: they can act as motivational tools, and they are simplified agents whose behavior is more predictable than that of a human.
  • autism spectrum disorder
  • socially assistive robot
  • human–robot interaction
  • social skills
  • social motivation
  • social cognition

1. Introduction

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by deficits in communication and social interaction as well as restricted or repetitive patterns of behavior, interests, or activities. A neurobiological dysfunction is thought to be at the root of this disorder. Symptoms appear early in development and have a major impact on the child’s daily life in many contexts, particularly in situations of social interaction. Robotics, a currently expanding field, could be of great interest in improving support for specific populations and in particular individuals with ASD. Robots can be categorized according to the different functions they perform. Examples include social robots, which specialize in interacting with humans through gestures and speech, and assistance robots, which aim to help people with special needs. The use of robots has recently expanded into a new field of application: socially assistive robots, which aim specifically to foster engagement in social interactions with specific populations.
According to the DSM-5 classification, the two main criteria for autism are impaired communication and social interaction on the one hand and restricted interests and repetitive behaviors on the other [1]. A review of the literature shows that robots can address both symptoms: their use with autistic people is generally aimed at improving communication skills and reducing repetitive behaviors [2][16]. Thus, interventions are particularly geared towards communication and social interaction difficulties but also focus on more specific behaviors, such as learning appropriate behaviors and reducing maladaptive behaviors (stereotyped behaviors and anxiety).
To determine the benefits of robots for individuals with ASD, researchwers first conducted a search on Google Scholar using the following keywords: “autism + robot + benefits”, yielding 24,900 results. As this research view is not intended to be systematic, researcherswe were particularly interested in experimental papers dealing with the progress observed in individuals with ASD following interaction with a robot (see [3][17] for a systematic review). RWesearchers have excluded experimental papers dealing with the use of virtual robots, focusing on the use of robots in an interactive setting. In all, researcherswe selected 50 experimental studies, with publication years ranging from 2008 to 2023.
Three types of interventions can be distinguished according to the objective sought [4][18]: robots can promote communication and social interaction [5][6][7][8][13,14,15,19], supporting the learning of specific behaviors, such as emotion recognition [9][10][20,21], and reducing the frequency of behaviors deemed maladaptive, including repetitive behaviors and anxiety [11][22].

2. Fostering Communication and Social Interaction

Social assistance robots have positive effects on the social abilities of children with ASD, who generally show more social behaviors during interaction with a robot than with a human [7][12][15,23]. This benefit can be observed across a range of social behaviors impacted by ASD: eye contact, joint attention, collaborative play and activity engagement skills, touch, verbal communication, and imitation. RWeseachers will list the effects of a robot for each of these behaviors (see Table 1 for a summary).
Table 1. Benefits of robots in fostering communication and social interaction in individuals with ASD.
SectionArticleVariableEffectEffect p

3. Fostering Specific Behaviors

The application of social robots with autistic individuals can also target the improvement of a particular behavior. The work presented in this section is of two types. Some studies promote the emergence of relevant and appropriate behaviors, while others aim to reduce maladaptive behaviors (stereotyped or repetitive behaviors) and anxiety. RWesearchers will examine the benefits of a robot on these behaviors (see Table 2 for a summary).
Table 2. Benefits of robots in fostering specific behaviors in individuals with ASD.
SectionArticleVariable-ValueRobotEffectEffect p-ValueSample Size (ASD)Mean Age (Standard Deviation)RobotSample Size (ASD)Mean Age (Standard Deviation)Functioning/Mean IQ (Standard Deviation)Duration of Robot Intervention (mn)Country
Functioning/Mean IQ (Standard Deviation)Duration of Robot Intervention (mn)Country
Eye contactBarnes et al. [13]Eye gazerobot > human
Appropriate behaviorsBharatharaj et al. [46]Touching interactionNANAO38.33 (4.04)NANANAKiliRo249.71 (3.24)1 session (NA)USA
NA1/day for 7 weeks (60 mn/session)India Bekele et al. [14]Eye gazerobot > human𝑝<0.05NAO62.78–4.9 y.o.NA1 session (30–50 mn)USA
 Costa et al. [16]Appropriate touchgentle touch > harsh𝑝<0.05KASPAR86–10 y.o.NA7 sessions (NA)UK Cao et al. [15]Eye gazerobot > human𝑝<0.05NAO154.96 (1.10)
 David et al. [19]NA1 session (NA)China
Turn-taking skillsrobot = human for 3 children𝑝>0.05NAO53–5 y.o.LF-HF8–12 sessions (10 mn/session, 1/day)Romania Costa et al. [16]Eye contactrobot > human𝑝<
 Ghiglino et al. [47]0.001Theory of Mind skillstraining with humanoid robot > non-anthropomorphic robot and traditional therapy𝑝<0.001KASPAR; 𝑝<80.01iCub, COZMO6–10 y.o.NA7 sessions (NA)UK
435.8 (1.14)71.48 (16.50) (COZMO); 71.14 (15.49) (iCub)2/week for 8 weeks (15 mn/session)Italy Costa et al. [17]Eye gazerobot > human
 Holeva et al. [48]𝑝<0.05Theory of Mind skills (NEPSY II)robot training = human; pretest < post-test𝑝QTRobot>159.73 (3.38)0.05; 𝑝<0.001IQ < 80 (n = 8); 80–120 (NAOn44 = 6); IQ > 120 (n = 1)1 session (1.5–4.3 mn)Luxembourg
9.48 (1.95)IQ > 702/week for 3 months (NA)Greece Damm et al. [18]Eye gazerobot > human
 Lakatos et al. [49]Visual Perspective Taking and Theory of Mind skills (Charlie test)𝑝<0.05FLOBI9pretest < post-test21 (NA)𝑝<0.05KASPAR13112.5 (range: 94–133)NAGermany
8.11 (1.96)79.30 (14.33); range: 60–1031 to 10 sessions (15–20 mn/session)UK David et al. [19
 ]Lee et al. [Eye contact50robot > human for 2/5 children]Proper force of touchingfeedback > no feedback𝑝=0.01NAO𝑝<0.0553–5 y.o.LF-HFTouch pad18–12 sessions (10 mn/session, 1/day)Romania
22 y.o.491 session (NA) Duquette et al. [20]Eye contactrobot > human𝛽=0.35TITO45.1 (NA)LF3/week for 7 weeks (NA)Canada
Japan
 Marino et al. [10]Recognition and understanding of emotionspretest < post-test (robot training); pretest = post-test (human training)𝑝=0.001; 𝑝>0.05NAO1473.3 months (16.1) (robot group); 82.1 (12.4) (human)NA10 sessions (90 mn/session, 2/week)Italy Scassellati et al. [8
 ]So et al. [Eye contact with other people51pretest < post-test]Recognition and production of intransitive gesturesrobot training > no training𝑝<0.01; 𝑝𝑝<0.05<0.05JIBO12NAO309.02 (1.41)IQ ≥ 7023 sessions (30 mn/session, 1/day)USA
5.10 (0.83) (experimental group); 5.8 (0.35) (control)NA4 sessions (30 mn/session, 2/week)China Shamsuddin et al. [21]Eye contactSo et al. robot > human[NANAO9110 y.o.]1071 session (15 mn/session)Malaysia
Recognition and production of emotional gesturesrobot training > no training𝑝<0.001NAO138.99 (2.14) (experimental group); 9.50 (2.42) (control)range: 49–674 sessions (30 mn/session, 2/week)China Simlesa et al. [7]Eye contactrobot > human𝑝<0.05NAO125.2 (0.63)
 So et al. [52MA: 2–3 y.o.]1 session (NA)Croatia
Recognition and production of intransitive gesturesrobot training = human𝑝>0.05NAO239.17 (1.29) (robot group); 8.92 (0.93) (human)range: 46–745 sessions (30 mn/session, 2/week)China Simut et al. [22]
 Eye contactrobot > human𝑝<0.05Takata et al. PROBO[306.67 (0.92)5391.23 (range: 70–119)1 session (15 mn/session)]Understanding of others’ feelings and behaviorspretest < post-test𝑝<Belgium
0.01Sota, CommU, A-Lab android ST1417.57 (3.39)89.50 (10.95)5 sessions (1 h/session, 1/day)Japan Tapus et al. [23]Eye gazerobot > human𝑝<0.05NAO44.2 (1.67)NA7–13 sessions (8 mn/session, 2/day)Romania
 Wood et al. [54]Theory of Mind skills (Charlie test)pretest < post-test for 7/12 children𝑝<0.05KASPAR1211–14 y.o.MA: 6–14 y.o.2–10 sessions (NA)UK Wainer et al. [24]Eye contactrobot > human𝑝<0.05KASPAR6
Reducing maladaptive behaviorsBharatharaj et al. 6–8 y.o.[NA2 sessions (15 mn/session)UK
46]Stress levelpretest > post-test𝑝<0.05KiliRo249.71 (3.24)NA1/day for 7 weeks (60 mn/session)IndiaJoint attentionAnzalone et al. [25]Joint attention
 Costa et al. [17robot < human𝑝<0.01NAO169.25 (1.87)]73 (14)1 session (NA)France
Stereotyped behaviorsrobot < human𝑝<0.05QTRobot159.73 (3.38)IQ < 80 (n = 8); 80–120 (n = 6); IQ > 120 (n = 1)1 session (1.5–4.3 mn)Luxembourg Cao et al. [15]Joint attentionrobot = human𝑝>0.05NAO154.96 (1.10)NA1 session (NA)China
 Kumazaki et al. [39]Stress levelrobot < human𝑝<0.01ACTROID-F2929.1 (2.6)IQ ≥ 701 session (25 mn)Japan Cao et al. [26]Joint attentionrobot < human𝑝<0.001NAO27
 Pop et al. 46.37 months (4.36)[MA: 42 months max2 session (NA)The Netherlands
34]Stereotyped behaviorsrobot < human𝑝<0.05PROBO114–7 y.o.IQ > 708 sessions (1 mn/session)Romania Ghiglino et al. [6]Joint attentionrobot = human𝑝>
 0.05Cozmo245.79 (1.02)58.08 (19.39)5 weeks (10 mn/session)Shamsuddin et al. [Italy
21]Stereotyped behaviorsrobot < humanNANAO110 y.o.IQ = 1071 session (15 mn)Malaysia Kajopoulos et al. [27]Joint attentionpretest < post-test𝑝<0.05CuDDler74–5 y.o.NA6 sessions over 3 weeks (20 mn/session)Singapore
 Shamsuddin et al. [11]Stereotyped behaviorsrobot < humanNANAO68.9 (NA)range: 46–785 sessions (15 mn/session)Malaysia Kumazaki et al. [28]Joint attentiontime +; robot > human𝑝<0.01CommU2870.56 months (6.09) (robot group); 69.00 (4.39) (control)NA1 session (15 mn)Japan
 So et al. [29]Joint attentionpretest < post-test (robot)𝑝<0.05HUMANE387.51 (0.87) (robot group); 7.91 (0.89) (human group)LF6 sessions (30 mn/session)China
 Taheri et al. [30]Joint attentiontime +𝑝<0.05NAO/ALICE-R5066–15 y.o.LF-HF12 sessions over 3 months (30 mn/session)Iran
 Warren et al. [31]Joint attentiontime +𝑝<0.01NAO63.46 (0.73)NA4 sessions over 2 weeks (NA)USA
 Wiese et al. [32]Gaze cueing effect
 Stanton et al. [37]Stereotyped behaviorsrobot < toy𝑝=0.06AIBO115–8 y.o.NA1 session (30 mn)USArobot > human𝑝<0.01EDDIE1819.67 (1.5)NA1 session (15 mn)Germany
InteractionGhiglino et al. [6]Social interaction initiationrobot > human𝑝<0.05Cozmo245.79 (1.02)58.08 (19.39)5 weeks (10 mn/session)Italy
 Kim et al. [33]Social behaviors towards peerrobot > human𝑝<0.05PLEO244–12 y.o.NANAUSA
 Pop et al. [34]Collaborative gamerobot > human𝑝<0.05PROBO114–7 y.o.IQ >708 sessions (1 mn/session)Romania
 Pliasa et al. [35]Social interaction initiationrobot > human𝑝<0.05DAISY66–9 y.o.NA2 sessions (20 mn/session)Bulgaria
 Rakhymbayeva et al. [36]Engagement timetendency time 2 > time 1; familiar > unfamiliar activities𝑝=0.05; 𝑝<0.05NAO76.1 (2.7)LF7–10 sessions (15 mn/session)Khazakstan
 Stanton et al. [37]Social interaction initiationrobot > human𝑝<0.05AIBO115–8 y.o.NA1 session (30 mn)USA
 Wainer et al. [24]Cooperation in gamerobot > human𝑝<0.01KASPAR66–8 y.o.NA2 sessions (15 mn/session)UK
TouchCosta et al. [16]Spontaneous touchrobot > humanNAKASPAR86–10 y.o.NA7 sessions (NA)UK
 Simlesa et al. [7]Touchrobot > human𝑝<0.05NAO125.2 (0.63)MA: 2–3 y.o.1 session (NA)Croatia
CommunicationFarhan et al. [38]Verbal and non-verbal communicationtime 4 > time 1NANAO45, 12, 13, 24 y.o.range: 41–474 sessions (NA)Bangladesh
 Huskens et al. [5]Self initiated questionspretest < post-test; robot = human𝑝<0.05; 𝑝>0.05NAO63-14 y.o.IQ >804 sessions (10 mn/session)Netherlands
 Kim et al. [33]Speechrobot > human𝑝<0.05PLEO244–12 y.o.NANAUSA
 Kumazaki et al. [39]Posture, gaze, facial expressionsrobot > human𝑝<0.001ACTROID-F2929.1 (2.6)IQ ≥ 701 session (25 mn)Japan
 Simlesa et al. [7]Vocalizationrobot < human𝑝<0.01NAO125.2 (0.63)MA: 2–3 y.o.1 session (NA)Croatia
 Stanton et al. [37]Speechrobot > human𝑝<0.05AIBO115–8 y.o.NA1 session (30 mn)USA
 Syrdal et al. [40]Communicationnumber of interactions +𝑝<0.05KASPAR192–6 y.o.NANAUK
 
 Taheri et al. [30]Verbal communicationtime +𝑝<0.05NAO/ALICE-R5066–15 y.o.LF-HF12 sessions over 3 months (30 mn/session)Iran
ImitationConti et al. [41]Imitationtime +NANAO65 and 10 y.o.LF15 sessions (6–8 mn/session)Italy
 Costa et al. [17]Imitationrobot = human𝑝>0.05QTRobot159.73 (3.38)IQ < 80 (n = 8); 80–120 (n = 6); IQ > 120 (n = 1)1 session (1.5–4.3 mn)Luxembourg
 Duquette et al. [20]Imitation of facial expressions; of words and gesturesrobot > human; robot < humanNATITO45.1 (NA)LF3/week for 7 weeks (NA)Canada
 Pierno et al. [42]Imitation velocityrobot > human𝑝<0.001ROBOTIC ARM1211.1 (NA)HF1 session (NA)Italy
 Simlesa et al. [7]Imitationrobot = human𝑝>0.05NAO125.2 (0.63)MA: 2–3 y.o.1 session (NA)Croatia
 Soares et al. [43]Imitation of emotionsrobot > human; post-test > pretest (robot), post-test = pretest (human)𝑝<0.05; 𝑝<0.05; 𝑝>0.05Zeno456.8 (1.5) (robot group); 7.5 (1.4) (human group); 7.8 (1.2) (control)HF2/week for 3 weeks (5–15 mn/session)Portugal
 Taheri et al. [44]Imitationrobot < human𝑝<0.001NAO204.95 (2.01)NA1 session (NA)Iran
 Zheng et al. [45]Imitation qualityrobot > human𝑝<0.05NAO63.83 (0.54)NA2 sessions (NA)USA
HF: High-Functioning Autism; IQ: Intellectual Quotient; LF: Low-Functioning Autism; MA: Mental Age; NA: Not Available; y.o.: Years Old.
HF: High-Functioning Autism; IQ: Intellectual Quotient; LF: Low-Functioning Autism; MA: Mental Age; NA: Not Available; y.o.: Years Old.

3.1. Supporting the Learning of Appropriate Behaviors

Interactions with robots offer the possibility of designing specific remediation to support particular learning, such as learning turn-taking [19][55][30,64] or non-verbal language gestures [9][51][20,65], emotion recognition [10][21], or regulating touch and physical contact [16][27].
In addition to using robots to increase the occurrence of certain behaviors, other protocols use robots to reduce the occurrence of maladaptive behaviors.

32.2. Reducing Maladaptive Behaviors: Repetitive Behaviors and Anxiety

The use of the NAO robot could reduce the percentage of stereotyped behaviors in children aged 5 to 13 [11][21][22,32]. In other studies, children also show fewer stereotyped behaviors when performing an activity with a robot partner than with a human partner [17][34][28,45]. The frequency of autistic behavior tended to decrease when children interacted with an AIBO dog robot rather than a mechanical dog toy [37][48]. In addition, using a KILIRO parrot robot for 60 min a week for 7 weeks helps both children and adolescents (aged 6–16) to reduce their stress levels [46][66]. In adults with ASD, the stress reduction observed during job interview training was greater for training with an android robot than with a human [39][50], suggesting that interacting with a robot is less stressful than interacting with a human for individuals with ASD. It is worth noting that the most anxious children display more stereotyped behaviors [56][80]. Stress levels could be linked to the occurrence of global and motor stereotyped behaviors (but not to verbal stereotyped behaviors) [57][81]. Reducing the stress involved in interacting with a human by using a robot as a partner could therefore help to reduce stereotyped behaviors in individuals with ASD.
In this way, socially assistive robots can promote communication and social interaction in autistic children, support specific social learning, and reduce repetitive behaviors. However, the results are mixed when rwesearchers compare the effectiveness of training with a robot with that of training with a human. While some studies show that robots are more effective than humans (e.g., [6][7][13][19][45][47][11,14,15,24,30,56]), others show that they are comparable (e.g., [5][17][48][13,28,67]) or even less effective [25][26][36,37] than humans. This heterogeneity in results can be partly explained by the wide variety of skills targeted by robot intervention. A meta-analysis confirms the benefits of robots on the social development of children with ASD, but not on motor or emotional aspects [3][17].

43. A Preference for Interacting with Robots Rather than Humans in Individuals

with Autism Spectrum DisorderD?

The benefits for individuals with ASD of interacting with a robot (at least in terms of social development) raise several questions. First, it is worth asking whether this type of interaction is more attractive to individuals with ASD than interaction with a human, as has been proposed in several studies [7][47][58][59][11,15,82,83]. In the next section, rwesearchers will look at how individuals with ASD perceive robots, and what characteristics they attribute to them.

54. Reasons Provided to Explain the Benefits of Robots

 Several explanations can be proposed based on different theories aimed at explaining the social difficulties observed in autism. ResearchersWe have already observed that people with ASD have social difficulties and show less interest in social stimuli than Typically Developing (TD) D individuals. The link between social skills and social motivation seems well established in children with ASD: those with the least social motivation have more severe social difficulties [60][61][62][148–150]. Nevertheless, the meaning of this association remains to be determined.

Two main theories have been proposed to explain the association between social difficulties and reduced social interest observed in autism [63][151]: Social Motivation Theory [64][152] and Social Cognition Theory [65][66][74,153]. RWesearchers will analyze the two theories that seek to explain the causality between these two characteristics of ASD and build on these theories to explain the benefits brought about by robots. 

65. Conclusions

In conclusion, the use of robots with people with ASD appears to be beneficial in encouraging communication and social interaction, supporting the learning of specific social behaviors, and reducing maladaptive behaviors. Interaction with robots could improve the social skills of individuals with ASD [3][17]. Children with autism touch and look more at a robot than at a human [7][17][19][15,28,30]. They also show a greater tendency to follow a robot’s gaze than a human’s gaze [32][43]. Children’s participation in an activity is increased when a robot is present [34][36][45,47], thus increasing interaction initiation [6][35][14,46] and speech production [33][44]. During an action imitation task performed by a human or robotic arm, children with autism perform better with the robotic arm, while TD children perform better with the human arm [42][53]. Furthermore, when ToM training is implemented, children with ASD make more progress with a humanoid robot than with a human [47][11], and the same results are observable in emotion recognition training with an iconic robot [10][21]. Finally, studies suggest that the stress felt by individuals with ASD during social interactions could be reduced with a robotic partner [39][50], leading to a decrease in the occurrence of stereotyped behaviors [17][34][28,45]. Further studies are needed to confirm the influence of robots on stereotyped behavior, but the benefits of robots on social development have been confirmed by a recent meta-analysis [3][17].

Although autistic children may categorize robots in the same way as TD children, they seem more attracted to robots and attribute more human characteristics to them. Some studies even suggest that theyare more interested in robots than in humans. RWesearchers sought to explain the benefits of robots by drawing on two theories: Social Motivation Theory and Social Cognition Theory. On the one hand, considering Social Motivation Theory, individuals with ASD may show more interest in a non-human agent than in a human one. This increased interest in interaction would then allow them to accumulate social experience, which could consequently reduce their social difficulties. On the other hand, as Social Cognition Theory argues that people with ASD engage less in interactions due to difficulties in understanding the social world, it is conceivable that a simpler, more predictable agent such as a robot could reduce these difficulties. As a result, this type of agent would encourage autistic individuals to engage more fully in social interactions. The robot, as a more attractive, simplified, and predictable social agent than a human, would thus encourage social interactions in autistic individuals. For children with significant social difficulties, robot-assisted training could thus constitute a step of intermediate difficulty compared to training with a human [67][182].

It therefore seems appropriate to rely on this type of agent when assessing the social skills of individuals with ASD: by reducing the difficulty of the interaction, a robot could enable children to better mobilize their abilities, leading to better estimation of their social skills. This suggests that psychology tests might be more successful if the experimenter is a robot rather than a human, as has already been shown in TD children [68][69][70][71][72][247–251]. The robot would then provide a more accurate means of assessing social–cognitive functioning [73][74][252,253], which would be particularly relevant for children with ASD [75][254].