User adaptable tasks for differential teaching with applications to robotic autism therapy

摘要

Autism is characterized by limitations in social interaction, speech, imitation and motor coordination. Recent studies suggest a link between imitation and Autism. In this paper, a robot is utilized to practice imitation learning in individuals with Autism. A system is proposed to gradually improve the imitation and social interaction of patients by adaptive training, which involves repetition of tasks close to the individual capacity limit and gradually increasing the capacity by improving the difficulty level. An ideal robot motion such as a hand wave performed by a trainer is recorded using a Kinect sensor and is replayed on the small humanoid robot Zeno. The motion is encoded on the robot using the Dynamic Movement Primitives (DMP) architecture, which is a set of non-linear differential equations used to generalize a motion by just changing the time, frequency and amplitude of the trajectory. The robot can adapt its motion to match that of the subject by analyzing the subject's movements by changing the DMP trajectory parameters accordingly. In the future, it is expected that this system will help subjects of different capabilities learn in a consistent manner by adaptively adjusting to their progress.