Reinforcement learning for an intelligent pediatrician robot

Abstract

The staff shortage in the medical world needs to be addressed. In this era of mobile communication, where online consultancy or telemedicine has become one way to overcome this problem, an innovative solution is needed for addressing tasks usually dealt with by a doctor. But doctors are limited by their physical fitness, tight schedules and even by their mental stability. Therefore, the involvement of Artificial Intelligence and robotics in the medical field is quite a normal development. There is a demand for robots that can imitate the personality of a friendly pediatric doctor, even though such robots cannot yet fully substitute for human pediatricians. In this paper, we discuss several possibilities as to what type of robot behavior is suitable as a substitute for a pediatrician, and the role of reinforcement learning, training them to be as friendly as possible with children, and enabling them to mimic human doctors. This learning process will not only teach robots to impersonate the personality of a doctor, but also to provide diagnosis capabilities-one of the core skills needed by a pediatrician. This paper highlights promising directions for potential applications and research. Our simulation showed that machine learning algorithms such as neural networks could support reinforcement learning in the field of pediatric medicine, reaching a diagnostic recognition rate of 96.6% for children with a particular disability. Future work might focus on increased integration between sensor devices in a humanoid robot and the medical Internet-of-Things, with more intense training enabling such a robot to be adaptable to any change of environment.