Towards Practical Deployment of a Robotic Mobile System for Early Detection of Cerebral Palsy in Infants.

Our research investigates methods and systems to allow for early detection of cerebral palsy in infants and innovative interventions with the goal of improving long-term outcomes. Cerebral Palsy is a development disorder that may be predicted by observing the spontaneous kicking patterns of an infant. Our previous work includes a robotic baby mobile and 3D camera system for detecting and motivating infant kicking motions using the stimuli modalities of the baby mobile. Infant kicking patterns can provide clues that indicate causes for concern for future risk of cerebral palsy. We have also investigated learning baby mobile stimuli preferences of individual infant using a Markov Decision Process to develop an optimal policy. The hypothesis is that the optimal policy will maximize the kicking instances of each infant, which will be beneficial for creating more opportunities to detect kicking abnormalities and also provide encouragement during physical therapy sessions. This work continues our progression by investigating techniques for replacing the 3D RGB camera with a low cost 2D camera, which will improve the practicality of the system for in-home deployment. Additionally, because of the restrictions of the global pandemic, we have developed a baby kicking simulator to test the effectiveness of using a Markov Decision Process to calculate an optimal policy for encouraging increased kicking actions by infants. This technique can be applied to multiple infants, with policies tailored to the preference of each child. In this paper, we describe the techniques for translating the 3D computer vision system to a 2D system and evaluate the accuracy. We also describe the design of the baby kicking simulator and the optimal policies devised by implementing a Markov Decision Process.