The development of an assistive robot for improving the joint attention of autistic children

In this paper, an assistive robot is proposed for children with autism to enhance their joint attention, the main focus of which is an interactive scenario with potential modalities to establish a learning environment for children with autism. In order to implement the above interactive scenario, the robot has to detect a child's intention at every time segment. For this purpose, a novel attention detection algorithm is proposed which is based on a mixture Gaussian based unsupervised-clustering. A unique aspect of the proposed approach is does not require the use of any training data or models to detect a child' intention. Indeed, our algorithm is capable of detecting a child's intention when that child has a complex eye gaze pattern. These features are essential in the development of a robotic system for children with autism. The system was tested with autistic children who were attending a school for the disabled. The results were quite impressive, revealing that the children were indeed attracted to the robot and had engaged in interactions with it for a lengthy period of time. Overall, the results showed that the children's joint attention skills improved when the number of interactive trials with them was processed. In addition, taking into account its joint attention performance with the children, the robot was able to change the interactive scenarios autonomously. The proposed unsupervised model was capable of detecting a child's intention at every time segment with a recognition rate of over 75%.