If the robot’s face changes, do children with autism recognize emotions differently? This study put three CASTOR appearances head to head—and found that the robot-like version drew the most attention. Socially Assistive Robotics has emerged as a potential tool for rehabilitating cognitive and developmental disorders in children with autism.

Social robots found in the literature are often able to teach critical social skills, such as emotion recognition and physical interaction. Even though there are promising results in clinical studies, there is a lack of guidelines on selecting the appropriate robot and how to design and implement the child-robot interaction.

Socially Assistive Robotics has been receiving considerable attention as an intervention tool to support Autism Spectrum Disorder therapies and innovative healthcare interventions in children with Autism Spectrum Disorder. Several strategies using Socially Assistive Robotics have been developed to improve and promote the development of social skills among children with Autism Spectrum Disorder.

These skills include joint attention, facial emotion recognition, verbal and non-verbal communication, and increased self-initiated interactions. Even though the evidence for the efficacy of Socially Assistive Robotics for Autism Spectrum Disorder therapy is promising, there is still not a consensus on how interactions should be addressed and which robot appearance might be most effective.

The use of participatory design methods in technology-based design processes for healthcare allows the different experiences and attitudes from different fields to be highlighted. The stakeholders and the target populations are no longer seen only as a source of information and requirements to produce results.

Instead, they are considered partners with experience and different points of view, which can be part of the solution. The participatory design process intends to achieve products or services that represent the real needs, expectations, and desires of the stakeholders.

This work describes the fifth and last stage related to validating the design process findings to identify the best appearance of a social robot. It also assesses the acceptability, expectation, and reactions of children with Autism Spectrum Disorder toward a novel Socially Assistive Robotics tool known as the CASTOR robot.

As a first step, the enterprise Tejido de Sueños designed 50 sketches using guidelines provided by interventions with children and stakeholders. The sketches were divided into five categories: cartoon persons, traditional robots, futuristic characters, animal-like appearance, and monsters or fantasy characters.

To validate the appearance ideas with stakeholders, two stages were carried out: appearance assessment and participatory selection. The appearance assessment aimed to identify which sketches were more engaging, using an evaluation matrix with six criteria. Fifty-two volunteers, including caregivers, parents, and interested parties, participated in this process.

Children with Autism Spectrum Disorder were not included in this first step because the activity might be difficult for them. Each participant gave a score from one to five for each requirement, and nine sketches with the highest score and interest were selected for the next stage.

Figure one shows how CASTOR’s appearance was narrowed from an initial set of fifty sketches. Panel A organizes the concepts into five categories, including fantasy characters, traditional robots, futuristic characters, cartoon people, and animals; panel B shows the participatory selection of candidate designs; and panel C presents the three appearances retained after filtering.

The figure matters because it makes the robot’s final design a stakeholder-informed outcome rather than an arbitrary choice. The validation study was conducted at the Howard Gardner clinic, where CASTOR was deployed with three different appearances. Three groups were defined to evaluate each appearance separately: the Fantastic group, the Robot group, and the Human group.

The protocol included the ethics statement, the participants, the experimental design, and the experimental procedure. A total of 21 children diagnosed with Autism Spectrum Disorder were enrolled, forming three groups of seven participants. The groups were the fantastic group, the robot group, and the human group, and all children were randomly assigned to the experimental groups.

In both control and intervention phases, the children had only three attempts to perform each activity: imitation and recognition. If children succeeded on the first attempt, they received three points; if they required more than one attempt, one unit score decreased until all three attempts were completed.

The children did not know about the scoring system, no child had seen the robot before, and CASTOR’s appearance did not change within the same group. Considering the small sample size, non-parametric statistics were used to analyze the effects of appearance on performance and attention between control and intervention phases.

The Kruskal Wallis test and Wilcoxon signed-rank test were used for those analyses, while the Friedman test and Bonferroni post-hoc test were used to determine significant differences among emotions. Table two reports mean performance scores and standard deviations for children’s emotion identification and imitation tasks, organized by control and intervention groups, appearance type, and happiness, sadness, anger, and fear.

For example, in identification, the control group’s Fantastic appearance shows three point zero zero plus or minus zero point zero zero for happiness, while the intervention average is two point seven six plus or minus zero point five three. This matters because it makes performance variability visible across task types and emotions, providing the descriptive basis for the authors’ non-parametric analyses.

Table three reports Kruskal–Wallis test results comparing appearances within each group and activity: identification and imitation. The control-group identification test has a p-value of zero point eight nine eight and an H-score of zero point zero two one with two degrees of freedom, while the intervention-group identification test reports zero point eight five nine and zero point three zero four with two degrees of freedom.

For imitation, the corresponding results are zero point nine nine three and zero point zero one three for the control group, and zero point six five one and zero point eight five seven for the intervention group, each with two degrees of freedom. The robot-like outfit presented 72.56 percent of the child’s attention in the intervention phase, the maximum children’s percentage of attention during all sessions and activities.

This suggests that CASTOR’s appearance with the larger attention and acceptance was the robot-like one. For future user studies, using CASTOR’s robot-like appearance should be considered and recommendable. Figure seven compares children’s attention to the therapist, marked T, and to CASTOR, marked R, across control and intervention sessions for the human-like, fantastic-like, and robot-like appearances.

The chart shows statistically significant differences, marked by asterisks, between therapist and robot attention in the intervention condition for all three appearances. The authors highlight the robot-like appearance, which received seventy-two point fifty-six percent of children’s attention during intervention, as the appearance associated with the greatest attention and acceptance.

This study used a novel social robot with three different appearances—human-like, fantasy-like, and robot-like—in a clinical setting. The appearances were implemented and designed through an inclusive and participatory design process.

Implementing participatory design is also an opportunity to understand the community’s context and build trust and confidence between researchers and the community. The current state of the art reveals constraints involving fragile structures and high acquisition costs of existing social robots.

The study found that CASTOR’s robot-like appearance reached the highest attention percentage, while children performed better on simple emotions such as happiness and sadness. That gives future studies a concrete appearance to investigate.