MIT scientists have developed a new "deep learning" network that can help robots measure the quality of their interactions with children with autism spectrum conditions by using data for each child. Autism Spectrum Disorder is a condition associated with brain development that affects how one person perceives and socializes with others, causing problems in social interaction and communication. The term "spectrum" in Autism Spectrum Disorder refers to the wide range of symptoms and severity.
Armed with personalized "deep learning", the kid-friendly robot NAO can estimate the commitment and interest of each autistic child based on data uniquely for that person, based on a study conducted on 35 autistic children. The new development can make your life easier.
"The long-term goal is not to create robots that replace human therapists, but to supplement them with key information that will allow therapists to personalize the therapy content and also do more captivating and naturalistic interactions between robots and children with autism "said Oggi Rudovic, first author of the study.
The perception of the robot's answers agreed with the estimates of human experts with a correlation of 60 percent to the scientists. "The challenge of creating machine learning and AI [artificial intelligence]which works on autism, is particularly annoying because the usual AI methods require a lot of data that is similar for each category learned." In autism, where heterogeneity prevails, fail the normal AI approaches, "said Rosalind Picard, co-author of the study.
In robot-assisted therapy, a human therapist displays flash cards of various faces that express different emotions such as happiness, sadness, anxiety, and a programmed robot or NAO that shows the child the same feelings. The therapist watches as the child grapples with the robot and receives feedback on how to continue the lesson.
NAO is a two-foot-tall robot that resembles an armored superhero or droid and conveys various emotions through the change in its eye color, limb movement and the tone of its voice. "Therapists say it can be a big challenge for them to attack the child for just a few seconds, and robots attract the child's attention," said Rudovic.
"Humans change their expressions in many different ways. Robots always do it in the same way, and this is less frustrating for the child because the child learns in a very structured way how the expressions are shown," he said. The system created by these researchers includes not only personalization, but also KI-driven deep learning.
Deep learning has been used in automatic speech and object recognition programs and is therefore well suited to a problem such as searching for several features of the face, body and voice that serve to create a more abstract concept such as the binding of one Child to understand.
"Deep learning allows the robot to extract the most important information directly from these data without requiring people to manually create these properties," said Rudovic. For the therapy robots, a personalized framework was created that could learn from data collected on each individual child.