Robotic Assistance For Children With Cerebral Palsy Based On Learning From Tele-cooperative Demonstration

Physical interaction with environment and object manipulation play an important role in development of children’s cognitive and perceptual skills. For children who have severe physical impairments, one of the biggest concerns is the loss of opportunities for meaningful play. Assistive robots can enable children to engage in play activities. In this paper, we focus on robotic assistance for position-following play activities such as pick and place. This task is done via a master-slave teleoperation system with the master robot in the child’s hand and the slave robot performing the task in the environment. In the demonstration phase, a therapist (or, in general, a helper such as a parent) holds the slave robot in the task environment to modify and assist the child’s movements as the child controls the master robot. A Learning from Demonstration (LfD) technique, which utilizes Gaussian mixture models (GMM) and Gaussian Mixture regression (GMR), is used to learn the helper-administered assistance to the child for completing the task. These probabilistic models provide insight into how the helper assisted the child by analyzing the multiple trials of demonstration in the presence of the helper. In the robotic assistance phase, the robot will utilize the learned data to assist the child in the helper’s absence and on a child-specific and as-needed basis. The efficacy of this framework is validated through experimental conducted involving a 2D play environment.