For children with Autism, speech and behavioral therapy is recommended early on to bring improvements in their development caused by social, emotional and communication challenges. Many of these children are recommended to get 20 hours of therapy a week. However, these therapies can get really expensive. Moreover, standardized therapy is difficult to design and implement since every child has varying symptoms and behavioral patterns.
Since a lot of these therapies involve repetitive training activities, “socially assistive robots”, powered by artificial intelligence, can not only reduce costs related to therapy but also bring therapy to a child’s own environment at home, make them more comfortable, personalize training by using data collected during previous trainings, and understand their behavioral patterns.
Recently, a team at University of Southern California, led by Maja J. Matarić, created a machine learning model that uses audio and video data collected from dialogue and eye contact children make with the robot, to predict if the child is engaged in a certain training activity. If they are not, the robot changes directions and re-engages with the child in other ways. While testing this model, the team achieved an accuracy of 90% in predicting the child’s engagement.
Even though the robot asks participating children to play math games, what it really teaches them is social interactions (taking turns to play with robot; making eye contact with robot when it’s talking). In these children’s case, the robot becomes a friend and helps them improve their skills over time.
Watch the video below to learn more about these socially assistive robots. Make sure to hit the source link too to read more about the testing, especially some of the challenges the team faced during that process. [Spoiler: participating children’s siblings started playing games with the robots too and skewed the data!]
The team also published a paper with their findings recently.
Source: MIT Technology Review