The following presents a detailed analysis of our research into the visual preferences of children with Autism Spectrum Disorder (ASD) regarding the form design of companion robots. This work aims to establish a more effective evaluation framework for such designs, providing actionable insights for both designers and therapeutic support staff. Our methodology integrates objective eye-tracking data with subjective preference scoring to elucidate the relationship between specific design features and the visual engagement patterns of children with ASD.
Children with Autism Spectrum Disorder (ASD) are characterized by persistent challenges in social communication and interaction, alongside restricted and repetitive patterns of behavior. Visual perception, however, often represents a relative strength. Research consistently indicates that individuals with ASD possess a marked advantage in processing and integrating visual information compared to auditory information. Consequently, therapeutic interventions that effectively leverage visual cues tend to yield more positive outcomes. For instance, visual-gaze cues have been shown to successfully capture the attention of children with ASD and elicit more active engagement. This pronounced visual-processing ability underscores the potential for well-designed visual stimuli, such as the form of a companion robot, to serve as a powerful tool in therapeutic settings.
The use of robots in ASD intervention has a history spanning several decades. Since pioneering work in the late 1990s, studies have demonstrated that social companion robots can enhance social interaction skills, reduce stereotypical behaviors, and lower anxiety levels during therapeutic sessions. Children with ASD often exhibit a higher degree of comfort and proactive behavior when interacting with a robot compared to interactions solely with a human therapist or screen-based games. Therefore, the physical design of the companion robot is not merely an aesthetic consideration; it is a critical factor influencing the initial engagement and sustained interaction crucial for therapeutic progress.
To objectively assess how children with ASD visually perceive different companion robot designs, we employed eye-tracking technology. This method records fine-grained visual behavior—where a user looks, for how long, and in what sequence—providing an external, quantifiable window into cognitive and attentional processes. By analyzing metrics such as fixation duration and time to first fixation, we can move beyond subjective surveys and build a data-driven understanding of design effectiveness.
Research Methodology and Experimental Design
Our research process followed a structured pipeline: initial market research and sample selection, preparation of visual stimuli, execution of controlled eye-tracking experiments, collection of subjective preference data, and finally, statistical modeling to link eye-movement patterns to explicit preferences.
1. Selection and Preparation of Companion Robot Stimuli
We began by compiling a comprehensive set of 112 companion robot models from major markets, therapeutic centers, and award-winning designs. To focus on form and eliminate confounding variables, we filtered this set to 48 distinct models, ensuring a wide variation in造型特征 (modeling features). With input from design experts, we selected six final exemplars that best represented three broad stylistic categories prevalent in companion robot design:
- Basic Geometric Forms (X1, X2): Characterized by primary shapes like spheres, cubes, and cylinders with large, smooth radii and unified silhouettes.
- Biomorphic Forms (X3, X4): Incorporating recognizably human or animal-like features, such as defined facial elements, limbs, and organic proportions.
- Irregular/Complex Forms (X5, X6): Featuring complex, non-uniform shapes, often modular or assembled from disparate components, with less coherent overall silhouettes.
To isolate the impact of form from color and texture, all selected images were converted to grayscale and normalized for luminance. This ensured that the assessment focused purely on轮廓 (contours), proportions, and the dynamic flow of shapes.
2. Participant Profile
The experimental group consisted of 22 children with ASD (aged 3-6), all possessing normal visual acuity. Informed consent was obtained from their parents or guardians prior to participation. The study was conducted in a quiet, controlled environment with the assistance of familiar caregivers to ensure the children’s comfort and compliance.
3. Eye-Tracking Experiment Procedure
We used a wearable eye-tracking device (aSee Glasses) to record visual behavior. Each participant was shown the six grayscale robot images in a randomized order. Areas of Interest (AOIs) were defined corresponding to each robot image. From over 30 potential eye-tracking metrics, we focused on ten that are most frequently cited in visual cognition and design evaluation literature, detailed in the table below.
| Eye-Movement Metric | Abbreviation | Description and Interpretation |
|---|---|---|
| Time to First Fixation | TFF | Duration from stimulus onset until the first fixation on the AOI. A shorter TFF indicates higher salience and attention-grabbing potential. |
| Fixation Before | FB | Number of fixations on other areas before the first fixation on the AOI. A higher FB suggests lower initial吸引力 (attractiveness). |
| First Fixation Duration | FFD | Duration of the very first fixation within the AOI. A longer FFD suggests deeper initial cognitive processing of that area. |
| Fixation Duration | FD | Sum of all fixation durations within the AOI. Longer FD indicates greater overall interest and information processing偏向 (bias). |
| Fixations Count | FC | Total number of fixations within the AOI. More FC generally correlates with higher interest and complexity assessment. |
| Regression Time | RT | Total duration of fixations made after returning to the AOI from another area. Longer RT indicates a compelling need to re-examine. |
| Regression in Count | RC | Number of times the gaze returns to the AOI. Higher RC suggests recurring interest. |
| Proportion of Fixation Duration | PFD | PFD = (AOI FD) / (Total FD). Higher PFD means a larger share of total viewing time was spent on that AOI. |
| Regression Time Rate | RTR | RTR = (AOI RT) / (Total FD). Lower RTR may indicate less need for revisiting. |
| Regaze Ratio | RR | RR = (AOI RC) / (Total FC). Higher RR indicates a greater frequency of returns relative to overall viewing. |
4. Subjective Visual Preference Scoring
Complementing the objective eye-tracking data, we gathered explicit preference scores. Using a simplified visual scale based on the Likert method (1=Dislike, 5=Like Very Much), children indicated their preference for each companion robot after viewing it. This provided a direct measure of their stated liking, which could then be correlated with their implicit visual behavior.
Data Analysis and Model Construction
The collected data from the eye-tracking experiment and preference scoring were analyzed using statistical software (SPSS 22.0). The average values for the key eye-tracking metrics across the six companion robot samples are summarized below.
| Sample | TFF (s) | FB | FFD (s) | FD (s) | FC | RT (s) | RC | PFD | RTR | RR |
|---|---|---|---|---|---|---|---|---|---|---|
| X1 (Geometric) | 0.02 | 2.34 | 0.32 | 1.58 | 8.06 | 1.26 | 7.06 | 0.34 | 0.27 | 0.26 |
| X2 (Geometric) | 0.03 | 3.23 | 0.37 | 1.68 | 9.32 | 1.31 | 8.32 | 0.35 | 0.27 | 0.29 |
| X3 (Biomorphic) | 0.03 | 4.17 | 0.24 | 1.26 | 7.41 | 1.02 | 6.41 | 0.26 | 0.21 | 0.23 |
| X4 (Biomorphic) | 0.07 | 8.48 | 0.17 | 0.97 | 5.68 | 0.79 | 4.68 | 0.20 | 0.17 | 0.16 |
| X5 (Irregular) | 0.11 | 13.08 | 0.15 | 0.72 | 4.71 | 0.57 | 3.71 | 0.15 | 0.12 | 0.13 |
| X6 (Irregular) | 0.07 | 7.98 | 0.19 | 0.90 | 5.76 | 0.70 | 4.76 | 0.19 | 0.15 | 0.17 |
The average subjective preference scores were: X2 (4.00) > X1 (3.60) > X3 (3.00) > X4 (2.75) > X6 (2.63) > X5 (1.78). This clearly indicates a preference hierarchy: Basic Geometric > Biomorphic > Irregular/Complex.
To establish a predictive model, we employed a stepwise multiple linear regression analysis. The goal was to identify which subset of eye-movement metrics most strongly predicted the subjective preference score (the dependent variable, $y$). The general form of the model is:
$$ y = \sigma_0 + \sigma_1 x_1 + \sigma_2 x_2 + … + \sigma_n x_n $$
where $x_1…x_n$ are the independent variables (eye-movement metrics).
The stepwise procedure identified three metrics as the most significant predictors: Proportion of Fixation Duration (PFD, $x_1$), Time to First Fixation (TFF, $x_2$), and Fixations Count (FC, $x_3$). The resulting regression equation was:
$$ y = 1.990 + 2.439x_1 – 5.478x_2 + 0.088x_3 $$
The model’s fit was satisfactory (Adjusted R² = 0.313, p < 0.001), and the Durbin-Watson value (1.555) indicated independent residuals.
The absolute values of the standardized coefficients reveal the relative influence of each predictor: $|\sigma_2| > |\sigma_1| > |\sigma_3|$. This means that Time to First Fixation (TFF) had the strongest negative impact on preference (shorter TFF leads to higher preference), followed by Proportion of Fixation Duration (PFD), and then Fixations Count (FC).
Model Validation and Integrated Findings
We validated the regression model using a new group of 12 children with ASD. Their eye-tracking data (PFD, TFF, FC) for the six companion robots were input into the equation. The calculated preference scores closely matched the actual scores provided by this new group, confirming the model’s reliability and generalizability.
Integrating the eye-tracking data, model results, and design analysis yields clear, actionable conclusions for the design of companion robots for ASD children:
- Key Predictive Metrics: For evaluating visual preference towards a companion robot design, three eye-tracking metrics are paramount: Time to First Fixation (TFF), Proportion of Fixation Duration (PFD), and Fixation Count (FC). Designs should aim to minimize TFF and maximize PFD and FC.
- Optimal Form Characteristics: The companion robot forms that performed best—eliciting quick attention (low TFF), sustained viewing (high PFD), and repeated looks (high FC)—were those with basic geometric forms. These designs feature clean, unified silhouettes composed of primary shapes (circles, rounded squares, smooth cylinders) with minimal intricate detailing and harmonious transitions between parts (e.g., head to torso).
- Hierarchy of Preference: A clear preference hierarchy emerged: Basic Geometric > Biomorphic > Irregular/Complex. While biomorphic forms with moderate仿生 (biomorphic) features (like simplified faces) held some interest, they were less immediately engaging than geometric ones. Highly complex, irregular, or mechanically busy forms were the least preferred and failed to sustain visual attention.
Conclusion
This study demonstrates the efficacy of combining eye-tracking technology with traditional evaluation methods to create a robust, data-informed framework for assessing companion robot design. By moving beyond subjective opinion, we identified specific, measurable visual behaviors (TFF, PFD, FC) that are strongly linked to the stated preferences of children with ASD. The findings strongly advocate for the use of simple, bold, and coherent geometric forms in the造型设计 (modeling design) of therapeutic companion robots. Such designs align with the visual processing preferences of children with ASD, facilitating quicker engagement and longer sustained attention—critical prerequisites for successful human-robot interaction and therapeutic progress. Future work will involve testing physical companion robot prototypes to account for factors like three-dimensional presence, texture, and color, further refining these design guidelines for the optimal companion robot.