We are working towards the creation of an artificial pancreas that will safely automate glucose management in people with Type I diabetes. To this end, we are developing and evaluating deep reinforcement learning (RL) algorithms for this application. We are particularly interested in developing sample efficient RL techniques that leverage a combination of both real data and data from imperfect simulators to achieve tight control in realistic settings.University of Michigan is a recipient of JDRF's research grant — Identification of Areas of Artificial Pancreas Algorithm Enhancements Through Big-Data Analysis — supported by Tidepool Big Data Donation Project.