. 2019 May 10;21(5):e13203. doi: 10.2196/13203

Table 2.

Child interventions delivered via robot for autism spectrum disorder.

Author	Sample	Design, conditions (n)^a	Duration	Measures	Outcomes^b
Kim et al [45]	24 US children with autism-spectrum disorder (ASD); 21 male, 88%) aged 4-12 years (mean 9.4, SD 2.4); Autism Diagnostic Observation Schedule (ADOS): 20 met criteria for autism, and 4 for autism spectrum disorder	Random order within subject: Pleo robot interaction, adult interaction and computer game	1 session: 3 × 6-min interactions, each separated by 6 min of interview and play	Verbalization (number of utterances produced)	Robot segment: > total speech versus adult, Computer game*; >speech to confederate versus Adult, computer game*; > speech to Pleo than computer game*, Pleo versus Adult not significant (ns)
Huskens et al [46]^c	6 Dutch males with ASD aged 8-12 years (mean 10.50, SD 1.37); all had Social Communication Questionnaire (SCQ) >15 (range 18-32)	Random order within subject: robot: NAO; human trainer; robot and human made statements inviting a question and performed requested actions (eg, dance)	Introduction to robot (2 sessions); baseline—4 robot, 4 human 10-mi training sessions; and follow-up 2 weeks after last training	Question-asking (number of self-initiated questions) in 3-5 × 10-min sessions with human assessor at baseline and follow-up	Both conditions, baseline session 1 versus intervention and follow-up^d: >correct questions during training, maintained at follow-up
Pop et al [47]	20 Romanian children (sex not stated) with ASD aged 4-9 years; no significant between-group differences on Children’s Autism Rating Scale (CARS)	Randomization in clusters of 3: story telling; Probo robot-assisted therapy social stories (SS-RAT, 7); computer-presented social stories (SS-PC, 6); and control (7)	SS-RAT and SS-PC: 6 sessions^e; control: 4 × 10-min observations on different days	Social expression (degree of prompt required for social response)	SS-RAT versus control at post^e: > social expression*; (SS-PC versus control ns)
Peca, Simut [48]	27 Romanian children^f (22 male, 82%), 18 with ASD, 9 with pervasive developmental disorder (PDD), aged 4.5-8 years (mean 6.2, SD 1.0). No significant between-group difference in mean ADOS (Robonova: 15.00; adult: 15.09)	Contingent (imitating child) and noncontingent play, with: Robonova robot (12) and adult (9)	1 session: 2 × 80-second segments^g separated by a 5-min pause	Social intention (eye gaze, positive affect, initiations, intention testing, tests per initiation frequency); contingent (mirrored behavior)	Robot versus adult: > eye gaze*** (contingency ns)
Srinivasan et al [38,39,41,49]	36 US children with ASD (32 male, 89%) aged 5-12 years (mean 7.6, SD 2.2) ADOS-2 range 6-10 (means—Robio: 8.5, rhythm: 7.9, and control: 8.4)	Robot (12): NAO and Rovio, whole-body imitation and interpersonal synchrony games; rhythm (12): human, singing and whole-body imitation games; and control (SC, 12): tabletop activities (academic, communication, and fine motor)	32 sessions over 8 weeks (post at 10 weeks)	Joint Attention Test (JTAT); social verbalization; imitation, praxis, interpersonal synchrony; Bruinicks-Oseretsky Test of Motor Proficiency (BOT); Repetitive and maladaptive behaviors; and Affective states	Robot versus control^h: >attention to human partner, elsewhere*ⁱ; <attention to objects; > spontaneous human attention; > self-directed vocalization; < human social vocalization **; < spontaneous human social vocalization; < sensory behaviors in late session*; > negative affect; < interested affect; and < fine motor control at Post. Robot versus Rhythm, during session^h: <attention to human partners and elsewhere **ⁱ; < spontaneous human attention* *; > self-directed vocalization; < spontaneous human social vocalization; and < positive affect in mid & late sessions*. Group x Early, Mid, Late Session: Words in response to questions*** (only Rhythm rose). Robot, pre and post: > body co-ordination* and > imitation*. Robot, early versus late session: <positive affect and > time in-synchrony*
Costescu et al [44]	27 Romanian children with ASD (20 male, 74%) aged 6-12 years (mean 8.7, SD 1.8); ADOS-Generic (mean 10.32)	Robot-enhanced therapy (14, RET)^j: My Keepon, distinguishing emotions from 15 social situations; discussion: cognitions, emotions and behavior connections; adaptive strategies for anger, self-control and control (n=15, standard care [SC])	RET: 6 × 2-hour weekly group sessions	Frequencies of correct strategies in a social situation; rational or irrational beliefs; adaptive behaviors; and emotional intensity	RET versus control, post (controlling pre): >rational beliefs and <(negative) emotion intensity*
Yun et al [40]	15 Korean males with ASD aged 4-7 years (mean 5.8, SD 0.9). No significant between-group differences on ADOS subscales or current SCQ (lifetime SCQ higher** and IQ lower* in robot group)	Social skills training (eye contact and reading emotions). Robot (8): iRobiQ (4 weeks), CARO (4 weeks) and human trainer (7)	8 × weekly 30-40 min sessions (post at week 9)	Autism Diagnostic Observation Schedule (ADOS, by blind rater); Vineland Adaptive Behavior Scale (Korean version); Social Communication Questionnaire; Social Responsiveness Scale; and Child Behavior Checklist (Korean version, CBCL)	No differences robot, human; both (versus pre): < (better) ADOS Play; <CBCL Internalizing at post (Depression and Anxiety, Withdrawal subscales); >frequency of eye contact, Session 8; >recognition accuracy of most difficult facial emotions by Session 4
So et al [42]	13 Hong Kong children (10 males, 77%) with ASD aged 6-12 (mean 9.0, SD 2.4) ADOS scores not reported (nr)	NAO Robot (7); control (6): educational videos; for both, phase I: Recognize 8 gestures; phase 2: Produce 8 gestures	In each 6-week phase: 4 × 30-min sessions over 2 weeks; tests pre, post, and 2 -week follow-up	Phase 1: Recognize gestures; phase 2: Produce gestures; tested on 2 trained gestures, 2 untrained; 20% of ratings rescored by a blind rater	Phase 1 recognition, robot versus control: pre and post: >scores on trained*, generalized, human-to-human gestures; post follow-up: ns; phase 1 production, robot versus control: pre and post: >scores on trained, generalized^k, human-to-human ns; post follow-up: ns
So et al [43]	45 Hong Kong (Cantonese-speaking) children (36 males, 80%), aged 4-6 years; 30 with ASD (3 female): intervention (mean 5.8, SD 0.8) waitlist (mean 5.7, SD 0.4); 15 age-matched controls (6 female) (mean 5.3, SD 0.7); and ASD severity nr	NAO robot demonstrates and elicits gestures while narrating stories; intervention (15); waitlist (15); age-matched, no ASD control (15)	Over 9 weeks: 4 × 30-min training sessions for 14 gestures (2 sessions per week); tests at pre, post, 2-week follow-up (2 test sessions each)	Gestural production in training, novel stories (10 seconds to respond, prompt and another 10 seconds if no response); gestural recognition; psychoeducational—third edition; Bruininks-Oseretsky Test of Motor Proficiency 2nd Edition (BOT); and Attention Network Task (ANT)	Gestural production (pre, post, follow-up), controlling for language and developmental age, BOT, ANT, gestural recognition^k: Group × Time*, Group × Training and Novel , Group × Time ×Training and Novel** control > Intervention, Wait List at Pretrained: intervention > waitlist (post*, follow-up); > control (Post*, Follow-up); -Novel: Intervention > Wait List (Post*, Follow-up); = Age-matched controls; Intervention versus Wait List (Pre to Follow-up) with covariates as above: Verbal imitation: Group × Time* (only Intervention group increasing*)

^aRandomized controlled trial with the individual participant as the unit of randomisation unless labeled otherwise.

^bEffects on measures not reported under results were not statistically significant. Some results that did not involve the robot condition are omitted. Results where the robot did significantly worse than the comparison condition are italicized.

^cDifferences between effects of the 2 conditions were only reported descriptively.

^dAnalyses of changes within conditions are reported separately, as are effects for each individual.

^eTotal period of training and timing of post not reported.

^fAn additional 6 children were excluded because they refused to undertake the tasks.

^gThe paper refers to the session segments as sessions.

^h Results from these studies were incompletely reported, and some reporting is ambiguous. Effects are across sessions unless otherwise stated.

ⁱThe attention target analysis appears inappropriate (only the robot group could have attention to the robot, affecting analysis of condition effects). “Elsewhere” is attention other than to the human partner, robot, or objects.

^jAnalyses were on 12 RET (2 withdrew); 15 control participants.

^kRecoding for gestural appropriateness rather than strict accuracy was interpreted as supporting these results, but only gave Group effects (using pre and follow-up only).

*P<.05.

**P<.01.

***P≤.001.