Analysis of Motor Signatures of Social Adaptation in Autism for Efficient Human-Centric Systems

This paper proposes a computational analysis framework to identify potential biomarkers of autism-related motor behavior, specifically focusing on how social context modulates motor imitation. The study analyzed 3D motion capture data from autistic (n=14) and neurotypical (n=20) adults performing dance imitation tasks under three conditions: baseline free movement (Body Shake), solo dance imitation (individually-framed stimulus), and duo dance imitation (socially-framed stimulus).

Methodologically, the authors used Dynamic Time Warping (DTW) to quantify within-participant trial-to-trial movement consistency, and proposed a novel per-participant metric called the Social Context Sensitivity Index (SCSI), defined as the difference in DTW consistency between the socially-framed duo condition and the solo imitation condition (SCSI = DTWduo − DTWsolo). Features were computed separately for three joint groups: upper body, core body, and lower body. These features were then used in a Support Vector Machine (SVM) with radial basis function kernel under Leave-One-Subject-Out cross-validation with nested hyperparameter tuning.

Key findings include:

• Reaction time did not significantly differ between groups across any condition, with high within-group variability, confirming it is not a reliable discriminator.

• No significant group differences in DTW consistency were found during baseline Body Shake or solo dance imitation conditions.

• During duo dance imitation, neurotypical controls showed substantially higher DTW distances (mean=399.9, SD=174.4) compared to autistic adults (mean=284.1, SD=80.2), with a large effect size (d = −0.85), indicating increased movement variability in neurotypical individuals in response to social framing.

• Joint-group analysis revealed significant differences in the upper body during duo imitation (p = 0.044, d = −0.879), with a marginal trend in the lower body (p = 0.066, d = −0.812), while core body showed no significant differences.

• The SCSI showed significant group differences in upper (Control: 90.5±70.1 vs. Clinical: 26.2±38.9; p = 0.007, d = −1.107) and lower extremities (Control: 87.0±65.2 vs. Clinical: 20.1±32.2; p = 0.015, d = −1.238), with neurotypical adults showing markedly higher SCSI values, while autistic participants exhibited SCSI values close to zero across all joint groups.

• Classification performance progressively improved from solo-only features (balanced accuracy 59.7%), to duo-only features (66.7%), to the combined feature set including SCSI (79.2% balanced accuracy, sensitivity 75.0%, specificity 83.3%).

The authors conclude that neurotypical adults tend to modulate their movement variability in response to social stimuli, whereas autistic adults maintain stable movement consistency regardless of social context. This social context sensitivity, quantified by the SCSI, constitutes a robust biomarker of autism-related motor behavior, with large effect sizes in the limbs and absent during solo or baseline conditions, suggesting it reflects a selective difference in social motor adaptation rather than general motor ability. The findings highlight the importance of social modulation in motor assessments and inform the development of inclusive human-centric technologies and medical systems.

Improvements to AI Systems:

1. Social-Context-Aware Motor Assessment Models

• Integrate the Social Context Sensitivity Index (SCSI) as a feature into AI-driven diagnostic tools for autism, enabling automated detection of atypical social motor adaptation (e.g., in telehealth or clinical motion-capture setups).

• The improved AI can distinguish between general motor ability and social-context-specific motor modulation, reducing false positives in autism screening by ignoring baseline or solo motor performance.

2. Adaptive Human-Robot Interaction (HRI) Systems

• Train AI agents (e.g., social robots or virtual avatars) to modulate their own movement variability based on the user’s SCSI, personalizing interaction styles for autistic individuals.

• The improved AI can adjust its imitation prompts or social framing in real time to maintain user engagement, avoiding overwhelming or under-stimulating motor responses.

3. Reinforcement Learning for Social Motor Learning

• Use the DTW-based consistency and SCSI as reward signals in RL agents that teach motor skills (e.g., dance, physical therapy) to autistic users.

• The improved AI can optimize teaching strategies that encourage beneficial variability in neurotypical-like responses while respecting the stable, consistent motor patterns of autistic users, leading to more inclusive motor training.

4. Explainable AI for Biomarker Discovery

• Incorporate the joint-group analysis (upper/lower/core) into explainable AI pipelines to highlight which body regions drive classification decisions, improving clinician trust.

• The improved AI can output interpretable reports (e.g., “atypical social modulation in upper limbs”) rather than black-box scores, aiding personalized intervention planning.

5. Cross-Condition Transfer Learning in Wearable Tech

• Develop AI models that use solo and duo motion data to predict social motor sensitivity, enabling lightweight wearables (e.g., smartwatches with accelerometers) to estimate SCSI without full motion capture.

• The improved AI can provide continuous, real-world monitoring of social motor adaptation in daily life, detecting early signs of social motor differences outside clinical settings.

6. Generative Models for Synthetic Training Data

• Use the reported effect sizes and DTW distributions to generate synthetic motion data that mimics autistic vs. neurotypical social modulation, augmenting small datasets.

• The improved AI can train more robust classifiers for rare populations, improving generalizability of motor-based autism screening across diverse demographics.

7. Personalized Feedback Systems for Social Motor Therapy

• Build AI-driven feedback loops that compare a user’s SCSI to their own baseline (rather than group norms), providing individualized coaching to increase or maintain social motor flexibility.

• The improved AI can deliver tailored exercises that target specific joint groups (e.g., upper body) where deficits are detected, enhancing therapeutic efficacy.