A robot can teach children to read emotions—but does its face change what they learn? This study found a striking difference in attention, but not the improvement many people might expect. Social robots are being explored as tools for helping children with autism build social skills, including recognizing emotions and physical interaction.

But promising clinical results have not answered a practical question: how should someone choose a robot, or design the child-robot interaction? The robot’s appearance was designed through inclusive, participatory techniques, involving the expectations, sensations, perceptions, and reactions of children with autism spectrum disorder, their caregivers, and the CASTOR team.

The process included the expectations, sensations, perceptions, and reactions of children with autism, their caregivers, and the team building the robot. The study then tested the design findings to identify the best appearance for a social robot and to assess children’s acceptance, expectations, and reactions toward CASTOR.

So the study focused on validating the robot’s design: identifying an appropriate appearance and assessing children with autism spectrum disorder’s acceptability, expectations, and reactions to CASTOR. The children completed recognition and imitation activities with each robot appearance, and each activity allowed only three attempts.

The children did not know the scoring system, no child had seen the robot before, and its appearance did not change within the same group. The robot-like outfit received the highest share of the children’s attention during the sessions, with seventy-two point five six percent of attention in the intervention phase.

This suggests that the robot-like appearance had the greatest attention and acceptance, and the study recommends considering it in future user studies. Here is the surprise: performance did not significantly differ between the control and intervention groups.

The results say the robot did not change or improve performance. CASTOR did not improve the children’s task performance: the study found no statistically significant difference between the robot and control groups, and reported that performance was not changed by the robot.

Even so, CASTOR may still serve as a therapy aid and help as an assistive tool when used in traditional methods. The study supports continuing research on using social robots to promote social interaction with children with autism, while longer research is needed to identify differences between children.

Future work will test physical interaction, more complex robot behavior, and whether robot-assisted intervention improves outcomes compared with traditional intervention without robots. A robot-like appearance drew more attention from the children, yet the robot did not improve their task performance.

That makes appearance useful for engagement, not proof of better learning.