Stuffed with Sensors

Chances are that you had a favorite stuffed toy as a child. It may have been a bear, mouse, lion, or something else entirely, but in any case, that favorite animal was actually much more than a toy, whether you realized it or not. By playing with, and caring for, a stuffed toy, children learn social skills, empathy, kindness, and language skills.

So when a child does not seem to be developing as expected, stuffed toys can turn into valuable tools. By observing how children interact with their toys, parents and child development experts can diagnose a wide variety of conditions, and sometimes even provide customized interventions that could prevent the emergence of more serious problems in the future.

Watching a child play with their toys for hours on end is not everyone’s idea of a good time — and anyway, the act of observing can itself change behavior — so instrumented smart stuffed toys are often used to collect data. These smart toys have sensors, like pressure sensors, embedded within them to reveal how children interact with them. However, given that these sensors are rigid and need to be near the surface, which changes the feel of the toy, they typically have very few (maybe one or two) sensors.

Such sparse data cannot tell a person very much about a child’s behavior. Fortunately, researchers at the University of Massachusetts Amherst have found a way to pack in more sensors — way more sensors — without taking the all-important squish out of a stuffed toy. Using a special type of sensor, they have made it possible to pack 24 pressure sensors into a small teddy bear.

The researchers’ invention, called FabToy, uses fabric-based pressure sensors instead of traditional rigid ones. These sensors, made from a textile material coated with an ion-conductive polymer, are highly sensitive and can detect even the most subtle interactions. This allows FabToy to capture everything from a gentle tickle to a firm squeeze without altering the toy’s soft and huggable texture.

Unlike conventional pressure sensors that use metal or plastic components, FabToy’s soft fabric sensors change their electrical resistance when pressed. The ion-conductive coating enhances this effect, allowing for greater sensitivity across a range of pressures. This means that whether a child is patting, holding, rubbing, or tickling the toy, the system can detect and classify each action with high accuracy.

FabToy’s data processing system works in two ways — locally, using a small, low-power nRF52840 microcontroller, and remotely, by transmitting compressed data via Bluetooth Low Energy to a computer or smartphone. Local processing leverages machine learning models that are optimized to recognize common interactions while keeping computational demands low. Meanwhile, remote processing uses an autoencoder — a type of artificial intelligence algorithm that reduces data size — to send only the most important information, cutting down on battery drain and data transmission costs. Once this information is passed to the remote device, larger models can be executed, offering more detailed and accurate results.

Because FabToy can sense interactions across its entire surface, it provides a much richer dataset than previous smart toys. Traditional instrumented toys, with just a few sensors, often miss complex interactions where a child is using both hands or different levels of force at different locations. FabToy’s ability to track detailed interactions in real time makes it a powerful tool for child development studies.