Type 1 Diabetes is a chronic condition characterized by the lack of insulin-producing beta cells in the pancreas. The artificial pancreas is a system of devices which has the potential to act as a healthy pancreas, thereby alleviating the burdens of self-management of Type 1 Diabetes. While the physical components of the artificial pancreas -- the continuous glucose monitor and insulin pump -- have experienced rapid advances, a technological bottleneck remains in the control algorithm, which is responsible for translating information from the glucose monitor into instructions for the pump. A machine learning technique known as reinforcement learning is a natural candidate for designing an adaptive, data-driven algorithm for blood glucose control. Learning is generalized using advanced functional regression tools, exploiting the full information contained in the glucose curve (an infinite-dimensional object) rather than the conventional reduce-then-design paradigm. The framework has broader implications for reinforcement learning methodology when the state space is functional.