Towards Direct Policy Search Reinforcement Learning for Robot Control

This paper proposes a high-level reinforcement learning (RL) control system for solving the action selection problem of an autonomous robot. Although the dominant approach, when using RL, has been to apply value function based algorithms, the system here detailed is characterized by the use of direct policy search methods. Rather than approximating a value function, these methodologies approximate a policy using an independent function approximator with its own parameters, trying to maximize the future expected reward. The policy based algorithm presented in this paper is used for learning the internal state/action mapping of a behavior. In this preliminary work, we demonstrate its feasibility with simulated experiments using the underwater robot GARBI in a target reaching task

A Gain Scheduling Model Predictive Controller for Blood Glucose Control in Type 1 Diabetes

This paper presents a control strategy for blood glucose(BG) level regulation in type 1 diabetic patients. To design the controller, model-based predictive control scheme has been applied to a newly developed diabetic patient model. The controller is provided with a feedforward loop to improve meal compensation, a gain-scheduling scheme to account for different BG levels, and an asymmetric cost function to reduce hypoglycemic risk. A simulation environment that has been approved for testing of artificial pancreas control algorithms has been used to test the controller. The simulation results show a good controller performance in fasting conditions and meal disturbance rejection, and robustness against model–patient mismatch and errors in meal estimation