Towards an AI-driven soft toy for automatically detecting and classifying infant-toy interactions using optical force...

Area of Science:

• Biomedical Engineering
• Developmental Pediatrics
• Machine Learning

Background:

• Early identification of neurodevelopmental disorders in infants is critical for effective intervention and improved long-term outcomes.
• Quantitative assessment of infant-toy interactions during natural play offers a promising approach for early detection of developmental delays.
• Prior research indicates distinct toy interaction patterns between full-term and at-risk preterm infants.

Purpose of the Study:

• To design and develop an automated toy system for quantitative assessment of infant-toy interactions.
• To detect infants at risk for motor delays through analysis of play interaction data.
• To evaluate the efficacy of soft force sensors and machine learning in classifying infant actions.

Main Methods:

• Development of a novel toy incorporating soft, lossy force sensors made from optical fibers.
• Collection of an interaction database (2480 interactions) from 15 adults simulating infant play (touches, weak grasps, strong grasps, kicks).
• Utilizing a machine learning model to analyze sensor data and classify interaction types.

Main Results:

• The configuration of 6 soft force sensors on the toy generated unique activation patterns.
• The machine learning algorithm successfully identified distinct interaction types from the collected data.
• These findings suggest the toy's potential for accurate classification of infant actions.

Conclusions:

• The developed sensorized toy demonstrates potential for automated, quantitative assessment of infant-toy interactions.
• The system shows promise for early detection of neurodevelopmental and motor delays in infants.
• Future work includes full toy sensorization and testing with infants.