Brief Report: A Novel System to Evaluate Autism Spectrum Disorders Using Two Humanoid Robots.

Kumazaki et al. (2019) set up two small humanoid robots that could talk and move together. They asked children with autism to play while the robots watched and scored the kids' social communication.

The team then compared the robot scores to standard ADOS ratings to see if the machines caught the same behaviors.

The two-robot scores lined up well with the ADOS. A higher robot score meant a higher ADOS score, showing the setup can spot social communication differences.

The result supports using robot pairs as a fresh, extra way to assess autism traits.

So et al. (2024) extends this idea. They used one robot, not two, and reached high accuracy (kappa 0.90) by tracking eye contact and inattentive time. Their strong numbers push the field from 'it can correlate' to 'it can reliably screen.'

Zhou et al. (2025) sweep in 13 trials and say XR tech shows clear medium benefits for autistic kids, while robot evidence is still thinner. This meta-view includes the 2019 study and flags the same gap: robots look promising but need firmer proof.

Rojahn et al. (2012) warned years earlier that robot autism tools were exploratory and weak on methods. Kumazaki et al. (2019) answer that call by adding a controlled, ADOS-linked design, showing the field is slowly moving from caution to cautious optimism.

You now have peer-reviewed evidence that a quick, playful two-robot interaction can yield scores that mirror the ADOS. This opens the door for faster, less stressful screenings in clinics or even schools. While you should not swap out ADOS yet, you can start thinking of robot pairs as a supplemental data point—especially for kids who tense up around adult evaluators.