Engagement Is Not Transfer: A Withdrawal Study of a Consumer Social Robot with Autistic Children at Home

A research team from Tsinghua University led by Yibo Meng conducted a groundbreaking 8-week randomized controlled trial to test whether engagement with AI-powered social robots translates to improved human social skills in autistic children. The study involved 40 children aged 5-9 using the commercial Qrobot social robot at home, with families assigned to either continued access or complete withdrawal of the device after an initial period. Quantitative measures and caregiver interviews assessed anxiety, social motivation, emotion inference, and empathic behaviors throughout the trial period.

Results revealed a surprising counterintuitive finding: while continued robot access significantly reduced children's anxiety (confirming strong affective benefits), the withdrawal group demonstrated substantially greater improvements in social motivation, emotion understanding, and empathic behaviors toward both caregivers and peers. Qualitative analysis uncovered a "handoff versus siloing" pattern—withdrawal appeared to promote reorientation toward human social interaction, while continued access concentrated engagement within the child-robot dyad, limiting transfer to real-world contexts. The researchers interpret these results as evidence that high engagement with social robots doesn't guarantee social skill transfer, suggesting that strategic withdrawal periods might be more effective for developing human-directed social abilities.

The study, accepted for publication at IDC 2026, challenges conventional assumptions about therapeutic technology deployment and raises important questions about how to balance the clear anxiety-reducing benefits of social robots with the need to foster human social skill development. This research provides crucial evidence for clinicians, educators, and parents making decisions about incorporating AI companions into therapeutic and educational interventions for neurodiverse children.